Acknowledgements Many people, in many ways, have helped me to complete this book. I would like to thank them all. My fri...
66 downloads
647 Views
1MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
Acknowledgements Many people, in many ways, have helped me to complete this book. I would like to thank them all. My friends, who provided me with the so often sorely needed times of relaxation. My fellow-PhD-students, for sharing and smoothening frustrations and problems on the road. My family, who have always supported me and who have shown great interest in my work. My colleagues, for stimulating discussions and coffee breaks. Special thanks are due to members of the TOSCA team with whom I had the privilege to work: Henk Barkema, Jan Cloeren, Pieter de Haan, and Vera Kamphuis, and most importantly my supervisors Jan Aarts and Nelleke Oostdijk, who introduced me to the fascinating world of syntax and corpus linguistics and who have greatly inspired this work through many discussions and through their comments on this book. I would also like to thank Loe Boves, Martin van ‘t Hof, and Flor Aarts, who read an earlier draft of this book and made valuable comments and suggestions, and Frans van der Slik, who helped me with parts of the statistics. Finally, I would like to thank André Olthof. First of all, for the time and effort he spent in implementing the elicitation program, but secondly, and most importantly, for his unfailing support through the years.
Chapter 1 The description of the English noun phrase 1.1
Introduction
In this chapter, I discuss the description of the English noun phrase (NP) as it can be found in the modern descriptive tradition, and arrive at an integral description of the NP: the prototypical noun phrase structure. This NP description is implicitly based on the idea that constituents are built up of a continuous sequence of words. However, even in a relatively rigid word order language such as English, constituents have a certain kind of ‘mobility’; they can occur in different positions from the one they typically occupy. Until now little is known about the mobility of phrasal constituents. In this book, I investigate the mobility of immediate constituents of the noun phrase in contemporary British English. At the same time, the study of mobility of constituents in the NP is used as a case study for a multi-method approach to the data. I argue that, from a methodological point of view, descriptive studies improve considerably if they use a multi-method approach to the data. More specifically, if they use a combination of corpus data and experimental data. 1.1.1 English descriptive linguistics English has a long-established tradition of descriptive studies. With the publication of the reference grammars of The Great Tradition at the beginning of this century, there came an end to a long period in which predominantly prescriptive grammars prevailed.1 Unlike their predecessors, grammarians such as Hendrik Poutsma (A Grammar of Late Modern English, 1904-1926), Etsko Kruisinga (A Handbook of Present-day English, 1909-1932) and Otto Jespersen (A Modern English Grammar on Historical Principles, 1909-1949) focused on the description of the (common core of) the language as evidenced in authentic texts. While the texts were used almost exclusively for the purpose of exemplification, they had an important influence on the descriptive analysis. In the first place, the nature of the data (literary citations) explains the orientation toward written, literary language, which makes up only part of the actual use of language. The degree of variation in structure which they came across was bound to be small. To obtain a fairly complete description, large bodies of text should be studied which contain a great many different varieties of the language, both written and spoken. Secondly, the often dated nature of the texts resulted in descriptions of rather obsolete use. In the third place, traditional grammarians did not (have an adequate method to) study the texts in a systematic manner. They 1
The term ‘The Great Tradition’ originates from F. Aarts (Aarts, 1975).
Chapter 1 - The English NP
2
only describe the frequent, more basic patterns and structures together with some unusual patterns which happened to attract their attention. As a consequence, some features were necessarily missed out and their grammars are not quite complete. In addition, although these grammars can be said to be predominantly descriptive, they include traces of prescriptivism by commenting on what is considered to be ‘proper usage’ rather than actual language use. The traditional descriptive approach as it is described here can also be found in more recent reference grammars. Here we refer to the two grammars by Randolph Quirk, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik, viz. A Grammar of Contemporary English (1972) and A Comprehensive Grammar of the English Language (1985). These grammars are based on contemporary English as sampled in the Survey of English Usage (SEU). 2 The SEU contains a large range of varieties of ‘educated English’, both written and spoken.3 All samples contained in the corpus are post-1950. In this respect the grammars of Quirk et al. differ considerably from the ones discussed above. Also, these modern reference grammars are not based solely on the conventional tradition of the older grammars, but use new insights developed in the field of language description, e.g. notions from discourse analysis and information processing. However, despite these differences from the earlier works, the grammars of Quirk et al. are traditional in their general approach – the description is informal and the descriptive model remains implicit. Observations about the (relative) frequency of occurrence and distribution of constructions are generally absent; where such observations are made, they tend to remain limited to predominantly impressionistic statements about what is considered to be common practice in the use of constructions and what is thought to be appropriate in a given context. For example, consider Quirk et al.’s description of discontinuous modification: Some adjectives that take complementation simultaneously function as premodifiers. Compare:
can
This result is different from yours. similar to hers. This is a
different result from yours. similar result to hers.
In cases like these last, discontinuity is felt to be quite normal and the position is similar to that of the correlative item in comparative sentences. (Quirk et al., 1985: 1400; underlining IdM) 2
Cf. Quirk (1968) By ‘educated English’ is meant “the repertoire of educated native speakers of British English: their linguistic activity from writing love letters or scientific lectures to speaking upon a public rostrum or in the relaxed atmosphere of a private dinner party.” (Quirk and Svartvik, 1979: 204) 3
1.1 - Introduction
3
Although the grammars of Quirk et al. include an extensive coverage of the structures and phenomena found in English, their description is still, at points, incomplete and much is left implicit. What is more, owing to their use of impressionistic statements and their lack of a formal approach, the description is often inconsistent. On these grounds, the grammars of Quirk et al. can be considered a continuation of the reference grammars of The Great Tradition rather than a break with that tradition. We can conclude that, while the grammars in The Great Tradition are comprehensive descriptions of the English language, they are incomplete, not just in terms of coverage (i.e. the constructions and phenomena described) but also in terms of detail (under what conditions, in which contexts, genres, etc. are constructions used). In particular, the description of what are perceived to be less common (variants of) constructions shows numerous lacunae. In order to further develop the description of the English language, traditional grammars should be supplemented and made more explicit. One way of reaching this goal is by systematically testing the traditional descriptions on careful selections of vast amounts of varieties of texts, i.e. on carefully sampled corpora. Modern advances in computational science in general, and computational linguistics in particular, have made such a systematic testing of traditional descriptions possible. The strand of computational linguistics which uses modern computer facilities to arrive at a complete and systematic description of a language, and which can thus be considered a continuation of traditional descriptive linguistics, is known as ‘corpus linguistics’. A corpus-based description of the English language not only describes the basic patterns, but also those patterns that are less frequent because they only occur either under specific conditions or in restricted language varieties. 1.1.2 Corpus linguistics Corpus linguistics is the branch of (computational) linguistics that is concerned with the study of actual language use on the basis of (text) corpora. In the modern conception of corpus linguistics, the use of the computer has made it possible to process large amounts of data and to analyse the data efficiently and consistently so that the data can be used by linguists for the study of (the structure of) language. This has its impact on the methodology that is adopted. In the first place, the corpus, the data-collection, has to be stored in a way that enables efficient retrieval for language research. This not only requires corpora to be computer readable, but also for corpora to be stored in an advanced and userfriendly database management system. Secondly, the raw texts should be enriched with linguistic information in a fashion that makes it possible to meaningfully explore the data. The linguistic information should thus ideally possess a high level of detail and correspond to those descriptive notions and methods that are commonly accepted. The best way to meet the last requirement is to follow the descriptions found in grammatical handbooks of the language. There are essentially two ways of annotating a corpus: manually and automatically. Manual analysis has the disadvantage of being time-consuming and inconsistent even if the analyses are assigned according to one and the same descriptive model. Automatic analysis is therefore preferred. The corpus can be
4
Chapter 1 - The English NP
automatically analysed by means of a ‘grammar-based parser’.4 To start with, the linguist’s hypotheses, based on literature and his/her own intuitions and knowledge of the language, are contained in a formal grammar. This grammar is then automatically converted into a parser so that it can be used for the analysis of a corpus. This way, the formal grammar serves a double purpose: it is the place where the linguist formulates his/her intuitions and after conversion into a parser, it can serve as a tool for the analysis of corpora. Thus, corpus linguistics is, in fact, a formalized approach to descriptive linguistics. The automatic analysis of the (raw) corpus can be used to test the coverage and validity of the formal grammar; i.e. the linguist’s hypotheses laid down in the formal grammar. In practice, it turns out that the corpus contains numerous structures that have not yet been included in the description. On the basis of the parsing results the grammar can be revised until, in a cyclic process of (re)analysis and revision, the description reaches the desired level of completeness. When the corpus is fully analysed, it constitutes a ‘linguistic database’. With the developments that have taken place in corpus linguistics in recent years, corpora are becoming available that have been enriched with detailed linguistic information. These linguistic databases can be explored so as to yield insights into the actual use of constructions, their frequency of occurrence and their distribution. The use of corpora takes a central position within the tradition of descriptive linguistics. Even the older reference grammars of English discussed in the previous section are partly based on collections of texts. Of course, the role corpora play nowadays differs considerably from the role they had at the beginning of this century. The earlier use of corpora is characterized by the manual collection of passages from mainly literary texts, which are then used for exemplification of a structure or phenomenon. The more ‘modern’ approach to corpus linguistics involves large collections of texts, both written and (transcribed) spoken, available in machine-readable form. For English, the era of modern corpus linguistics started in the early nineteen sixties when the Brown Corpus (Standard Corpus of Present-Day Edited American English, Ku era and Francis, 1967) was compiled. It contains approximately one million words of written American English and is available in its raw version and in a tagged version in which each word has been assigned a ‘part of speech’ code. The British counterpart, LOB, (Lancaster-Oslo/Bergen Corpus of British English, Johansson et al., 1978) followed almost ten years later. Both corpora attempt to represent a cross-section of written American (BROWN) or British (LOB) English printed in 1961, by including a great many different text categories, such as press reportage, scientific writing, science fiction, etc. The modern approach to corpus linguistics not only requires the collection of large, representative bodies of machine-readable texts, but also computational 4
I here reflect the Nijmegen view of corpus linguistics. This approach differs considerably from, for example, the probabilistic approach adhered to by the Unit for Computer Research on the English Language (UCREL) of Lancaster University. For a discussion see Aarts and van den Heuvel (1985), Leech (1987) and Oostdijk (1991).
1.1 - Introduction
5
tools for storing, retrieving and enriching corpora and for facilitating language research for large groups of users. Related to this is the need for standardisation in the mark-up of corpora and of the linguistic information that is included. Such standardisation will ensure the compatibility of corpora and at the same time the comparability of linguistic research based on these corpora. At the moment, (further) development of taggers and parsers for the morpho-syntactic analysis of corpora and discussions on standardisation are in full swing and will continue for years to come. So far, developments in computer science and computational linguistics have resulted in broadly two types of new corpora. In the first place, corpora have become available (and are becoming available) which are relatively small, but which have been enriched by detailed linguistic information profiting from (ongoing) developments in parsing techniques. Good examples of such corpora are the Nijmegen Corpus and the TOSCA Corpus5, but also the ICE Corpus which is still under development.6 Secondly, large-sized corpora are now becoming available which include all kinds of language varieties, including spoken data. A good example is the British National Corpus, which comprises 100-million words, including 10 million words of spoken English.7 The entire corpus has been tagged and a small selection has been syntactically analysed.8 To conclude, advances in hardware and software over the last three decades have effected the re-establishment in descriptive linguistics of the study of language use on the basis of corpus data.9 What is more, corpus data are now also widely used by theoretical linguists and the use has spread to other research fields such as speech technology, sociolinguistics and lexical studies. With ongoing developments in corpus linguistics, it is likely that the use of corpus data will spread even further, since more data are becoming available that can be readily accessed by many researchers in various fields. Thus, through the production of (linguistic) databases, corpus linguistics supports research in other (linguistic) fields. 1.1.3 Word order variation In section 1.1.1 we concluded that, while the traditional grammars are comprehensive, they are incomplete in terms of coverage and detail of the 5
TOSCA stands for TOols for Syntactic Corpus Analysis. Both the Nijmegen Corpus and the TOSCA Corpus were compiled and analysed at the University of Nijmegen. The tools used for the analysis were also developed there. Cf. Oostdijk (1991). 6 ICE stands for International Corpus of English. It is a corpus of national and regional varieties of English which facilitates comparative studies between varieties of English. The ICE project was initiated by Sidney Greenbaum. Cf. Greenbaum (1988, 1992). The first part, the British variety (ICE-GB), has been tagged and parsed and is available on CDROM from the Survey of English Usage. 7 Cf. Aston and Burnard (1998). 8 Only part of the corpus has been checked for the correctness of the tagging. 9 In the interval, the focus of description was mainly on ‘competence’. Where corpus data were used, the studies are characterized by the manual analysis of (relatively) small amounts of corpus data.
6
Chapter 1 - The English NP
description. One of the areas where traditional descriptions are incomplete is word order variation. Where such variation is mentioned the description often lacks (or provides only tentative) information on the frequency of occurrence and the conditions under which variation occurs. For example, consider Kruisinga’s remark on a modifying (attributive) adjective to a noun: When an attributive adjective is accompanied by a plain noun (always expressing measure) the attributive group generally follows the noun (a), but pre-position occurs sometimes in spoken English (b). a. A ditch five feet wide. A child three years old. b. A three-year-old child. (Kruisinga, 1909: 217; underlining IdM) Since occurrences of grammatical constructions actually found in corpora display much more variation than traditional descriptions lead us to believe, a syntactically analysed corpus forms the pre-eminent means to test such descriptions, so that they can be supplemented and/or made more explicit. So far, studies of word order variation have been mainly concerned with (immediate) constituents at the level of the sentence. However, the immediate constituents of the phrase have, in principle, a greater potential for mobility: not only can they change places among themselves at phrase level, they can also occur outside the boundaries of the phrase at sentence or clause level. Quirk et al. (1985: 48) use the term ‘mobility’ to refer to permutations in sentence/clause structure - that is, to sentence or clause constituents that occur at unusual positions in the sentence or clause. Until now little is known about the mobility of phrasal constituents. In this book, I investigate the mobility of the immediate constituents of the noun phrase in English. A study of (syntactically) analysed corpus data, will provide insight into the nature and frequency of possible variants. Furthermore, it gives an impression of the lacunae in traditional grammars. As such, the research is meaningful for descriptive linguistics in general, since the results that emerge from this study can contribute to making existing descriptions more explicit and more complete. Within the field of corpus linguistics, and more specifically grammarbased corpus analysis, the lack of insight into the mobility of constituents is experienced as an important complication, where the control of ambiguity in a formal grammar is concerned. Since corpus linguistics aims at the full coverage of a language combined with detail and correctness of analysis, a formal grammar, which plays a central role in the analysis, should describe all possible regular structures as well as their variants. Without further restrictions, such a grammar gives rise to an excess of ambiguity in the analyses. Therefore, insight into the nature of the mobility phenomenon, the frequency with which it occurs, and the conditions that play a part in it forms an important step towards optimizing the formal grammar.
1.2 - English linguistics and the NP 1.2
7
English linguistics and the noun phrase
In the descriptive tradition of English, many grammarians from different backgrounds and approaches have contributed to the description of English syntax. A sentence is not just a string of words in a random order; it has (a complex) internal structure. From the time linguists began to recognize larger patterns in the structure of language, the description of some kind of noun group or noun phrase has played an important role. Here I discuss the treatment of the noun phrase as it can be found in the literature, resulting from the various linguistic theories which have contributed to the description of English syntax. My aim is to achieve an integral description of the NP which can then be used as a basis for the corpus study. The description of the NP in traditional grammars will function as a starting-point. 1.2.1 The NP in The Great Tradition The grammars of The Great Tradition have already been discussed in section 1.1.1. The first three are those by Poutsma, Kruisinga, and Jespersen. Although these grammars can be said to be predominantly descriptive, they lack an adequate descriptive model for their analyses. They start by describing the various parts of speech and then work toward the sentence as a unit, but they often fail to recognize the intermediary larger patterns. This is especially true of Poutsma. Of the three, Jespersen is the only one who uses the term ‘phrase’: “A phrase is a combination of words which together form a sense unit, though they need not always come in immediate juxtaposition” (Jespersen, part II, 1914: 15). Jespersen does not, however, mention the existence of a ‘noun phrase’ as such. For his description of syntax he divides the words of the sentence into three classes: primary words (or: principals) for the main sentence elements; secondary words (or: adjuncts) for the modifiers of principals; and tertiary words (or: subjuncts) for the modifiers of adjuncts. As a basis for this classification he uses semantic rather than formal relationships between constituents. Kruisinga distinguishes ‘syntactic groups’: “A syntactic group is a combination of words that forms a distinct part of the sentence” (Kruisinga, 1909: 177). One of the syntactic groups is the ‘noun group’ which can have the same functions in the sentence as the noun (or: substantive). It consists of a head plus modifier(s). The modifier can be a group in itself (Kruisinga’s sub-group). What is striking about his description of the noun group is that he only recognizes ‘premodifiers’, i.e. words (or groups) that precede the head, but no ‘postmodifiers’, which follow the head. Poutsma, Jespersen and Kruisinga have all compiled a large collection of carefully documented literary citations, which they use in their grammars for the purpose of exemplification. Because of the limited nature of the texts, the variation in structure which they came across was bound to be small. Therefore, in their descriptions they are preoccupied with the more frequent structures and those oddities in language which are most striking. As a consequence, many features were necessarily missed out so that their grammars are incomplete. Also,
8
Chapter 1 - The English NP
because they depend rather heavily on semantic criteria and lack an adequate descriptive model, the descriptions are implicit and informal. In section 1.1.1, I argued that the grammars of Quirk et al. (1972, 1985) could also be ranked with the grammars of The Great Tradition, although they include a more systematic account of the English language. Quirk et al. classify the units of the sentence in four ways. In the first place, a unit is classified according to the word class or word group to which it belongs (noun, noun phrase, etc.). Secondly, a unit can serve a range of ‘constructional functions’ within the grammar (subject, verb, object, etc.). In the third place, units may have a ‘participant role’ (or ‘semantic function’) in the sentence (e.g. agentive, recipient, etc.). And last, units may be viewed “as items which can be manipulated within the structure of sentences for different kinds of prominence, serving the total sequential organisation of the message” (Quirk et al. 1972: 937) (e.g. focus, theme, etc.). However, the four classifications are not used consistently in the grammar. Units are (sometimes) almost randomly referred to by their word class or by their function in the sentence. The descriptive system put forward by Aarts and Aarts (1982) is very similar to that of Quirk et al. in that they use categories and functions to describe the units of the sentence. However, their description is much more consistent. Their approach is characterized by an immediate constituent analysis representing a rank-hierarchy of units of description and the linear order of constituents. Every constituent, at every level of analysis, can be considered as an element that plays a role in a larger structure (its function) and as something that has its own individual characteristics which it shares with other units of the same kind (its category) (cf. Aarts and Aarts, 1988: 10-14). When we adopt the descriptive model as it was introduced by e.g. Quirk et al. (1972) and explicated by Aarts and Aarts (1982), the noun phrase can be described as a headed phrase in which the head is the only obligatory constituent, while there are optional function slots for the determiner (DET), the premodifier (PREM), and the postmodifier (POM).10 The occurrence of the determiner is dependent on the realization of the head, while there may be zero or more preand/or postmodifiers. The structure can be represented as in Figure 1-1 (p. 9).11 It is a surface structure account in which the constituents are described in terms of the position in which they actually occur in language use. In terms of realization, the head is typically realized by a noun (common or proper) or a pronoun, but it may also be realized by a numeral, formal it, the proform one, or a (nominal) adjective. The function of modifier in the NP can be realized by an adjective phrase, adverb phrase, noun phrase, prepositional phrase, and clause. Adjective phrases, adverb phrases, and noun phrases commonly precede the head, while they can also occur in postmodifying position. Prepositional phrases and clauses follow the head. The determiner function is typically realized by an article
10
I use the term ‘optional’ to refer to constituents (functions) whose presence is not required in the higher order constituent (category) to form a grammatical construct. It is not the realization, but the occurrence of the function slot which is optional. 11 Brackets indicate optionality, the asterisk indicates possible multiple realizations.
1.2 - English linguistics and the NP
9
(definite or indefinite), but it can also be realized by a numeral, pronoun, or (genitive) NP. NP (DET)
(PREM)*
HEAD
article pronoun numeral NP
AJP AVP NP
noun pronoun proform
(POM)*
PP CL AJP AVP NP Figure 1-1: Noun phrase structure following Quirk et al. (1972, 1985)12 In their discussion of the complex noun phrase, Quirk et al. range the determiners among the premodifiers. However, elsewhere in their grammar they distinguish three determiner ‘sub-functions’ (as do Aarts and Aarts): the predeterminer, central determiner and postdeterminer. This more detailed analysis is based on mutual exclusiveness and word order. If the determiner function is realized, it will minimally consist of a predeterminer, a central determiner or a postdeterminer. In principle, however, all three sub-functions are optional. Predeterminer Central determiner - pronoun: exclamatory, - article universal, quantifying - pronoun: demonstrative, - numeral: possessive, negative, multiplicative, fraction assertive, non-assertive, relative, interrogative - genitive NP (specifying)
Postdeterminer - numeral: ordinal, cardinal - pronoun: quantifying
Items that can occur as predeterminers are generally mutually exclusive. They precede other determiners whenever these are present. (1)
all the children twice this area one-third the amount
The central determiners are also mutually exclusive. They may follow the predeterminers and/or precede the postdeterminers, but can occur on their own as well: (2)
12
both her questions his many qualities
those three hours all her five fingers
Personal pronouns typically realize the function of head, while universal, demonstrative, possessive and quantifying pronouns typically function as determiners in the NP.
Chapter 1 - The English NP
10
Among the items that can realize the function of central determiner are also (specifying) genitive noun phrases. A difference should be made between the specifying genitive functioning as central determiner and the classifying genitive functioning as premodifier in the NP. Consider: (3)
a. my mother’s car b. a mother’s heart
In (3b) the determiner a refers forward to the head heart and mother’s is said to be a classifying genitive functioning as premodifier. In (3a) my refers to the genitive mother’s and the NP my mother’s functions as central determiner in the NP my mother’s heart. Postdeterminers follow (predeterminers and) central determiners, when present. They are not mutually exclusive, but can be found to co-occur. For example, (4)
all these three girls every last day
the first two weeks the last few things
We can now account for complex determiner sequences like: (5)
both my two brothers twice your last offer
But there is still something missing. Consider: (6)
how many people around two minutes not all the children
twenty odd years my own problem his every thought
The elements in (6) precede or follow the determiner sequence. They can only cooccur with other determiners, which they seem to modify. Oostdijk (1993) in her description of the NP solved this problem by distinguishing a category ‘determiner phrase’ (DTP) that realizes the function of determiner in the noun phrase structure.13 Within the determiner phrase she distinguishes five function slots: determiner phrase premodifier (DTPR), predeterminer (DTPE), central determiner (DTCE), postdeterminer (DTPS), and determiner phrase postmodifier (DTPO). In theory all functions in the DTP can co-occur, while none of the functions is realized obligatorily. However, the DTPR and DTPO require the cooccurrence of one of the other DTP functions, i.e. if the DTP is realized it minimally consists of a DTPE, a DTCE, or a DTPS. Only the DTPS can have multiple realizations. The structure of the determiner phrase can be represented as in Figure 1-2 (p. 11). 13
From here on the term determiner will only be used to indicate the function, and not the categories that realize it. In this I differ from Quirk et al.
1.2 - English linguistics and the NP
11
(DET) DTP DTPR
DTPE
DTCE
DTPS
DTPO
how around not about almost nearly
PN(univ) ART NUM(ord) more NUM(mult) PN(dem) NUM(card) own NUM(frac) PN(poss) PN(quant) every PN(ass) PN(nonass) PN(inter) PN(neg) PN(rel) GEN Figure 1-2: Determiner phrase structure following Oostdijk (1993) 14 1.2.2 The NP in structural grammar In the development of descriptive linguistics, structural linguistics has made an important contribution. In particular, the introduction of immediate constituent analysis by Bloomfield has immensely influenced grammarians. While Bloomfield himself paid fairly little attention to the development of his approach to syntax, his suggestion that formal definitions of constructions and form classes should be preferred to the use of more philosophical definitions based on meaning was followed up and systematized by the Post-Bloomfieldians such as Harris, Wells, Fries, and Nida. Bloomfield was an active supporter of a purely scientific approach to linguistics. In such an approach, the use of meaning to define the fundamental word-classes is not suitable: ... class-meanings are not clearly-definable units which could serve as a basis for our work, but only vague situational features, undefinable in terms of our science. (Bloomfield, 1933: 267-268) However, Bloomfield does recognize the usefulness of the parts of speech in his constituent analysis: The syntactic form-classes of phrases, therefore, can be derived from the syntactic form-classes of words: the formclasses of syntax are most easily described in terms of wordclasses. (Bloomfield, 1933: 196) 14
The lists of items that can occur as DTPR and DTPO are merely meant as an indication of the type of element that can occur at this position. They are not intended to be exhaustive. Brackets to indicate optionality are missing in this structure. In principle, all functions are optional, but the DTP will minimally consist of a predeterminer, a central determiner, or a postdeterminer.
Chapter 1 - The English NP
12
The first to apply the structuralist methods to the description of English syntax were Fries (1952) and Nida (1966). In his book The Structure of English Fries gives an account of modern English speech based on recordings of conversations on a large variety of subjects. He rejects the traditional parts of speech. Instead, he defines four major form classes and fifteen function classes, based on substitution in ‘minimum free utterances’ and formal criteria such as word endings or co-occurrence with function words. He also rejects traditional meaning definitions for functional relations such as modification. In his approach modifiers are treated relative to their head (or nucleus) which can be realized by one of the form classes or function classes. The contrast between the older traditional procedure of grammatical analysis and the approach used here lies in the fact that the conventional analysis starts from the undifferentiated total meaning of an utterance and raises the question, “What names apply to various parts of this meaning?” whereas our analysis starts from a description of the formal devices that are present and the patterns that make them significant and arrives at the structural meaning as a result of the analysis. (Fries, 1952: 56-57) As modifiers of the noun (Class 1 word) Fries considers those words that are placed between the determiner and the Class 1 word and those words or wordgroups immediately following the Class 1 word. Preceding the noun are: -ed/-ing participles, adjectives, nouns and adverbs. Following the noun are: ‘prepositional phrases’ (function word of Group F with Class 1 word) and ‘clauses’ (function word of Group J with included sentence). In Fries’ analysis of English sentences in ten steps, the analysis of the ‘noun phrase’ is described in step 8. “Postmodifiers are cut off first, beginning with the last one. Word groups as modifiers are treated on the level as whole units in relation to the head to which they are attached” (1952: 267/8). Next, the words preceding the head are cut off, starting with the one which is most remote from the head. The analysis of the phrase an oral examination of the students which is thorough can be represented as follows (Fries, 1952: 266): an
oral
examination
of the students
which is thorough
In A Synopsis of English Syntax Nida rigorously applies immediate constituent analysis to the description of English structures. Although his description is not meant to be comprehensive, he covers a fair amount of sentence types. His goal is to show the suitability of immediate constituent analysis as an analysis tool. He starts from the sentence as a unit, working his way down to the ultimate constituents, dividing every constituent into two immediate constituents.
1.2 - English linguistics and the NP
13
In the structure of the noun phrase he recognizes five ‘function slots’: the predeterminer, determiner, postdeterminer, head, and postposed attributive.15 The predeterminer precedes the determiner and can be an attributive to the head (nearly a pint), to the postdeterminer (quite a large house), or to everything from the determiner to the head (so little a place). They differ in realization from the predeterminers which were discussed in section 1.2.1, and in function from determiner phrase premodifiers. A determiner phrase premodifier only modifies one of the other determiners and not the premodifier and/or head of the noun phrase. Nida’s predeterminer thus does not form part of the determiner phrase represented in Figure 1-2 (p. 11). Instead, it seems to constitute an immediate constituent of the NP. In many respects, Nida’s predeterminer is similar to Stageberg’s (1965) ‘restricter’ and Fries’ (1972) and Oostdijk’s (1993) ‘limiter’. In my (integral) description of the NP I will include a ‘limiter’ as a separate function slot in the structure of the NP (see also section 1.3). The limiter can be realized by an adverb phrase only, thus excluding Nida’s last occurrence where an adjective phrase precedes the determiner. Additionally, the limiter has as its scope the whole following NP, thereby it also excludes Nida’s second example (quite a large house). The limiter function will be further discussed in section 1.3.1, while the two excluded structures will be dealt with further in section 1.4. Following Nida’s analysis, the phrase quite a large house to live in can be represented as in Figure 1-3. A line with an arrow represents an endocentric, subordinate relation with the arrow pointing towards the head. A line with a cross represents an exocentric relation. quite a large house to paint
Figure 1-3: NP analysis following Nida (1966) Stageberg (1965) also uses a binary constituent analysis. His contribution to the description of syntax has been the application of a threefold classification for words or word groups: they are classified by form, by position and by function. For example, a word can be an adjective by form, an adjectival by position and a modifier by function. Stageberg distinguishes a ‘noun cluster’ which “consists of a noun and all the words and word groups that belong with the noun and cluster around it. The noun itself is called the headword or head, and the other words and word groups are modifiers of the noun” (Stageberg, 1965: 163). Thus, in their attempt to make linguistics a scientific discipline, Bloomfield and the Post-Bloomfieldians have left their mark on descriptive linguistics. From then on, descriptions had to be explicit, consistent and based on observable facts.
15
Nida’s ‘postdeterminer’ function corresponds to the ‘premodifier’ function in the traditional NP description.
Chapter 1 - The English NP
14 1.2.3 The NP in generative grammar16
Within generative grammar the study of the internal structure of the NP has mostly served the wider discussion on the theory of syntax and has not been a topic of study in its own right.17 Within early X-bar theory, every maximal projection (XP) must have a head (X). Determiners and modifiers to the NP are included by adding a specifier and complement position to the basic structure or, in the case of external arguments, by adjunction.18 Specifiers are optional, as is adjunction. Complements are regulated by the subcategorization principle. In English, internal arguments (e.g. subcategorized PPs) are projected to the right of the head in D(eep)-structure, while external arguments (e.g. predicative adjectives) are licensed on the left. Constituent order can change under the influence of movement requirements. Figure 1-4 gives an example of an NP analysis in the generative framework. NP Spec
N’ AP
a
N’
good
N
PP
example
of an X-bar tree
Figure 1-4: NP structure according to X-bar theory Under the influence of Case Theory the structure of the NP is now sometimes considered as a projection of functional determiners selecting a fully lexical NP complement, resulting in the following enriched structure:19 DP Spec
D’ D
NP
Figure 1-5: DP structure 16
I use the term ‘generative grammar’ as a collective name for all theories that form part of the generative framework, starting with Chomsky’s Syntactic Structures. 17 Notable exceptions are Radford (1988) and Giorgi and Longobardi (1991). 18 Adjunction is a transformational rule that ties a node A to a node B by copying B so that A can be attached to the ‘new’ B node. The binary branching principle is then still satisfied. See also Chapter 5, section 5.3. 19 See Abney (1987).
1.2 - English linguistics and the NP
15
This structure not only creates the possibility of assigning case to genitive NPs, it also offers the possibility to regulate the distribution of ‘bare’ NPs (without a determiner) and to extend the internal structure of the determiner phrase. 1.2.4 The NP in more recent approaches to descriptive linguistics Over the last thirty years, generative grammar has greatly developed and it is still, in its variant forms, one of the most important approaches to linguistics. It has influenced and inspired many other approaches of which only Functional Grammar (FG), which rejects most of the tenets of the transformationalgenerative tradition, will be discussed here. Of the more recent approaches, FG stands out in that it focuses on the (communicative) use of language. In this respect it offers a different approach to a study of the mobility of constituents. Unlike approaches in the formal paradigm, where language is looked upon as an abstract, formal object, FG considers language as an instrument of social interaction between speakers, it is part of the speaker’s communicative competence. In FG it is not enough to give the rules and principles that underlie the structure of language; these rules have to be explained according to their function in language use. From a functional point of view, there is no such thing as autonomous syntax. Both syntax and semantics should be studied in the wider frame of pragmatics. Within Functional Grammar, modifiers are subjected to different syntactic treatments depending on the communicative function they perform. Within the theory of Dik’s Functional Grammar (Dik, 1989), for instance, the noun phrase (or ‘term’) structure is produced and analysed according to the following general scheme: (Wxi:F 1(xi):F 2(xi):...:F n(xi)) Here W stands for one or more term operators. Term operators function as discourse-pragmatic devices and semantic features. Among others they include distinctions such as ‘definite’ (d) versus ‘indefinite’ (i), ‘specific’ (s) versus ‘generic’ (g), and ‘proximate’ (prox) versus ‘remote’ (rem), and number markers. The variable xi symbolizes the intended referent. Every F(xi) is a “predication open in xi” and represents a restrictive modifier. They function as meaningrestricting elements in the NP, narrowing down the set of possible referents of the term. Non-restrictive modifiers are not included in this scheme. (7)
Those two intelligent young girls with blue eyes standing in the middle of the room are my sisters.
(8)
(d rem 2 xi: girlN(xi): youngA(xi): intelligentA(xi): {imxj: eyeN(xj): blueA(x j)}(x i): Pres Part [standV(xi)Ag, {((d1xk: middleN(xk): {d1xl: roomN(xl))poss}(xk))Loc}])
The subject NP in (7) can now be represented as in (8). The combination of the operators ‘d rem 2’ yields the determiner ‘those’. In addition ‘2’ is represented by
16
Chapter 1 - The English NP
the numeral ‘two’ and brings about the plural form of ‘girl’. There are two adjectival restrictors ‘young’ and ‘intelligent’. The other two predications state that the girls have eyes with the property ‘blue’ and are standing on the location ‘middle’ which is a property of ‘room’. The order of the constituents is determined by constituent ordering principles. These principles are divided into general principles, reflecting the ordering typology of all natural languages and specific principles, specifying the actual constituent ordering patterns of individual languages. 1.2.5 Discussion of the different approaches The description of the noun phrase in traditional grammar (section 1.2.1) is an example of what Gleason (1965), among others, calls the ‘slot-and-filler technique’. In the slot-and-filler approach sentences are first of all divided into their major (functional) elements, e.g. subject, verb and object. These elements are not necessarily single words, but can also be realized by phrases or sentences (clauses), which then require further analysis. This process is repeated until a detailed analysis into single words (ultimate constituents) is reached. A noun phrase is considered as having a number of slots or positions, for each of which there can be specific appropriate fillers. The slot-and-filler technique has no restrictions as to the number of (immediate) constituents in a construction. Another commonly used technique to divide sentences into their meaningful parts is the ‘binary-branching’ technique.20 This technique is used by the structuralists and also in the generative framework. In this approach constructions at each stage of the analysis are divided into two parts until ultimate indivisible units are reached. An important difference between the traditional and structuralist approach on the one hand and the generative approach on the other is that the first have only one level of representation for structural description (in other words, a single P(hrase)-marker), while transformational grammar has two: a set of underlying Pmarkers, defined by the phrase structure rules, and a derived P-marker, which represents the surface constituent structure. The derivation from the underlying strings is dealt with in a transformational component which can express the relations between sentences in a unified, natural and economic way. Both the slot-and-filler technique and the binary-branching technique are dependent on reasonable and consistent procedures for dividing constructions into their immediate constituents. Structuralists have tried to give a formal definition of (immediate) constituents based on their distribution. The most important methods were substitution (see e.g. Harris, 1951 and Fries, 1952) and expansion (Wells, 1947). The method of substitution says that, if a complex sequence can be represented (substituted) by a simpler sequence, the complex sequence constitutes one immediate constituent (IC). Expansion is largely based on substitution. A sequence is analysed into ‘expansions’ which form the ICs of that sequence. If 20
Gleason (1965) refers to this technique as the ‘immediate constituent’ (IC) technique. I prefer the term ‘binary-branching’ technique, since I also consider the slots in the slot-andfiller technique to constitute the immediate constituents of the higher level.
1.3 - The structure of the NP
17
two sequences A and B can occur in the same context and A contains at least as many morphemes as B and is structurally diverse from B, A is called an expansion of B. For example, in the sentence The king of England opened parliament, ‘the king of England’ is an expansion of ‘John’ and thus an IC of the sentence. Although both the slot-and-filler technique and the binary-branching technique may result in the same divisions into parts, the binary-branching technique will (generally) go through more steps, since it is more rigidly dichotomous. The divisions may be simpler than with the slot-and-filler technique, but they are not necessarily more relevant. The slot-and-filler technique applies layering only when its necessity is observed in the data. Despite their differences, both approaches take as their starting-point the belief that the syntax of natural languages is amenable to a description in terms of immediate constituents (ICs), where the rules of the grammar are all of the form: X®Y where X is a single element and Y is a string of elements (possibly a null string). This is also known as a phrase structure model of syntax. 1.3
The structure of the noun phrase
We can now come to an integral description of the noun phrase, based on the various descriptions discussed in the previous section. As we saw, all approaches are based on a description in terms of immediate constituents. The adequacy of IC analysis for the description of syntax has been questioned on several occasions. The first to criticize strict IC analysis was Chomsky (1957), who tried to formalize immediate constituent analysis in terms of context-free rewrite rules (phrase structure grammar). He proposed a more complex formalism for syntax, namely transformational grammar (TG), in which phrase structure rules produce a ‘deep structure’ representation of the sentence, which is then subject to certain transformational rules (obligatory and optional) resulting in a ‘surface structure’ representation. One of the advantages of this approach over IC analysis, as Postal (1967) points out, is that it can deal with discontinuous constituents. Constituent analysis depends heavily on word order. Whenever non-adjacent elements (seem to) form one immediate constituent, a description of discontinuity is required. Although (structural) grammarians often did recognize discontinuous structures, these could not be assigned by phrase structure grammar. Next to clarity in formulation Postal considers the treatment of discontinuity to be one of TG’s greatest merits. One of the virtues of TG is that it provides a straightforward formalization of the notion ‘discontinuous constituent’. In TG, discontinuities are for the most part produced by the operation of permutation transformations. That is, if in some sentence there is a sequence DAE and D and E are
Chapter 1 - The English NP
18
discontinuous constituents of some higher order constituent B, then there is some P-marker for that sentence in which D and E are continuous constituents of B, ... (Postal, 1967: 67) However, the problem created by the description of discontinuous constituents has never really been tackled in transformational grammar. Moreover, it has been shown by for example Harman (1963, 1966) that discontinuity could also be formalized in non-transformational terms. Indeed, many IC approaches did discuss a solution to discontinuity, but failed to formalize it. Take for example the following discussion by Fries (1972) on discontinuity in the noun phrase: If the filler of a Loose Knit Modifier is complex, various portions of the filler may be permuted to different positions within the noun phrase. Thus, the phrase a dangerous crime to indulge in would be analyzed as containing the adjective phrase dangerous to indulge in within the Loose Knit Modifier, with to indulge in permuted to a position following the filler of the Head tagmeme of the Noun Phrase. (Fries, 1972: 222) There is thus no reason to reject a surface structure approach to syntax in favour of a transformational or deep structure (D-structure) approach. What is more, in the most recent development in the generative framework, the Minimalist theory, D-structure has been eliminated as a separate level of representation (Chomsky, 1995). One of the reasons for this is that it has been demonstrated that (restrictions on) movement transformations can be accounted for without the use of a D-structure level of representation.21 Of the descriptions in section 1.2 only Functional Grammar will allow for a description of mobility in terms of ordering principles which are not directly (or indirectly) based on syntactic principles. Since the mobility of constituents may well be explained by discourse principles (how do language users present information) or processing principles (how do we interpret a sentence), FG may form an appropriate starting point for the integral description. Unfortunately, however, the descriptive model of FG is still very much under development. Elaboration of the model falls outside the scope of the current research. The integral description will therefore be a surface structure approach following a formal paradigm. Of the descriptions discussed in section 1.2, the one of Quirk et al. (1972, 1985) was the most elaborate. Also, the descriptions contained in the grammars of Quirk et al. are subscribed to by a great many linguists. For these reasons, their NP description will be used as a starting point for the integral description.
21
See also Chapter 5, section 5.2.1.
1.3 - The structure of the NP
19
Within the structure of the noun phrase, all categories can be assigned a ‘typical’ function. In this representation adjective, adverb and noun phrases function as premodifiers, while prepositional phrases and clauses occupy the postmodifier position. Applied to the NP description of Quirk et al. represented in Figure 1-1 (p. 9) and with the addition of the function ‘limiter’ (LIM) as discussed in section 1.2.2, this results in the following prototypical NP structure: NP (LIM)
(DET)
(PREM)*
AVP
DTP
AJP AVP NP
HEAD noun pronoun proform
(POM)* PP CL
Figure 1-6: Prototypical NP structure Variations of this ‘prototypical’ NP structure are possible, but are considered to constitute ‘minor types of modification’ (Quirk et al., 1972). These variations can be explained by either the nature of the structure (the occurrence of a specific category or even a specific lexical item), or contextual conditions (e.g. the internal structure of a modifier), or a combination of both. An adjective phrase, for instance, typically occurs as postmodifier when the head of the NP is realized by a (non-)assertive or negative pronoun (ex. 9), when the adjective belongs to the subclass of marked attributive adjectives or is one of a limited number of items that in their bare form can occur both in premodifying and in postmodifying position (ex. 10), and/or when the adjective phrase includes a postmodifier to the adjective phrase head (exs. 11-12). (9) (10) (11) (12)
somebody important the items concerned/available circumstances peripheral to the core money sufficient to buy the house
1.3.1 The limiter In the structure of the noun phrase the limiter is realized by an adverb phrase. The adverb phrase consists of a single adverb or an adverb modified by not: (13)
only these people particularly my father
not just my opinion not even the poor citizens
The limiter is similar to the determiner phrase premodifier (see section 1.2.1) in that it precedes other determiners and is realized by an adverb phrase. Moreover, a limiter cannot co-occur with a determiner phrase premodifier. That they are recognized as two separate functions can be explained from their different syntactic and prosodic properties. A determiner phrase premodifier can only
Chapter 1 - The English NP
20
occur if either the predeterminer, the central determiner, or the postdeterminer slot of the DTP is also realized (ex. 14). A limiter is not dependent on the realization of the DTP and can in fact occur when the determiner slot in the NP is not filled (ex. 15). With the determiner phrase premodifier the stress lies on the following item, whereas the limiter receives stress itself. The determiner phrase premodifier only modifies the pre-, central, or postdeterminer with which it cooccurs. The limiter restricts the reference of the (rest of the) NP (ex. 16). (14) (15) (16)
not áll people around thrée weeks mérely small-boy junk hárdly anybody júst my luck at léast a small domestic problem
1.3.2 The determiner The realization of the determiner function has already been discussed in section 1.2.1. The function of determiner is unique to the noun phrase. The possible realizations of the determiner function constitute a closed set of items. The articles are central to this set in that they have no other function independent of the noun they precede, whereas the other items in the set can also occur independently (e.g. pronouns) and/or fulfil other functions in the sentence/clause or in other phrases (e.g. adverbs). Not every realization of the determiner can cooccur with every realization of the head of the noun phrase. They are subject to certain co-occurrence restrictions. For example, an indefinite article cannot occur with a plural noun, a proper noun, or a pronoun. Also, certain realizations of the head cannot occur without a determiner, e.g. singular count nouns such as book, shirt, tree, etc. While other realizations of the head can never co-occur with a determiner, e.g. proper nouns (unless they have generic reference). In other words, the optionality (and realization) of the determiner is largely dependent on the realization of the head. 1.3.3 The head The head is the dominant member of the noun phrase. It can replace the entire phrase (unless the noun requires a determiner). The head is typically realized by a noun (common or proper) but it may also be realized by a pronoun, a numeral, formal it, the proform one, or an adjective (often referred to as ‘nominal adjective’). The nominal adjective is usually preceded by a definite article: (17)
the poor the anguished
1.3.4 The modifier Modifiers can create a subclass of the class denoted by the head of the noun phrase. Such modifiers are known as ‘classifying’ or ‘restrictive’ modifiers.
1.3 - The structure of the NP
21
Another function of modifiers is to describe the referent of the noun phrase in terms of a particular quality it possesses. These modifiers are known as ‘descriptive’ modifiers: (18)
a pretty boy a Belgian boy
descriptive classifying
Modifiers can either precede or follow the head. The former are known as premodifiers and the latter as postmodifiers. The premodifier Clearly the most common realizations of the premodifier are the adjective phrase and the noun phrase (including the classifying genitive noun phrase): (19)
a very cold day some quite amazingly beautiful pictures a white house an extremely unpleasant look
a coming events notice a long distance call the corner grocery a doctor’s degree
-ed and -ing participles can also be considered as adjective phrases22: (20)
a rather recently published article a properly conducted affair
a floating stick a rapidly turning structure
Also recognized as a premodifier by almost all linguists is the adverb phrase. Only a limited class of adverbs occurs in premodifying position (unless the head of the NP is realized by an adjective): (21)
the above statement the then duke
the very old the badly injured
Premodifiers can co-occur, i.e. there can be an (unrestricted) sequence of adjectives and/or nouns. However, the order in which premodifiers occur is not (entirely) free. The relative ordering is dependent on the syntactic and semantic class membership of the items concerned. This topic has received a considerable amount of attention in the literature. All studies point in the same direction:
22
But a distinction can be made between -ed and -ing participles functioning as premodifier and fully adjectival -ed and -ing forms functioning as premodifier, in that only fully adjectival forms can be modified by adverbs as very, so, more, most, too, etc. Also, the fully adjectival forms can get the prefix un- as in uninterested where there is no verb to uninterest. Contrast e.g. an accomplished person with an accomplished task.
Chapter 1 - The English NP
22
... the general rule is to place the more objective and undisputable qualifications closer to the noun, and the more subjective, opinionlike ones farther away. (Hetzron, 1978: 178) In practice, this means that nouns are closer to the head than adjectives and that descriptive adjectives will precede classifying adjectives. The relative order of premodifiers in the NP can be specified even further by referring to semantic classes only (e.g. Dixon, 1982) or semantic and syntactic classes (e.g. Quirk et al., 1972): Dixon: 1. Value (e.g. good, bad) 2. Dimension (e.g. big, large, little) 3. Physical Property (e.g. broken, hot) 4. Speed (e.g. fast, quick, slow) 5. Human Propensity (e.g. mad, happy) 6. Age (e.g. new, young, old) 7. Colour (e.g. black, white, red) 8. Origin/Composition (e.g. Dutch, wooden) 9. Purpose/Benificiary (e.g. dog food)
Quirk et al.: 1. General (e.g. hectic, small, heavy) 2. Age (e.g. new, young, old) 3. Colour (e.g. black, white, red) 4. Participle (e.g. running, carved) 5. Provenance (e.g. Gothic, Chinese) 6. Noun (e.g. London, jade) 7. Denominal (e.g. social, moral)
head The postmodifier In the prototypical structure of the noun phrase, one postmodifier slot is distinguished. This slot can be filled by a prepositional phrase or a clause. The prepositional phrase includes possessive constructions: (22)
the man with the gun the children aboard the ship a day in December
a time of innocence a pot of coffee the house of my father
The clauses can be divided into finite and non-finite clauses. The class of finite clauses consists of restrictive relative clauses (ex. 23), non-restrictive relative clauses (ex. 24), and appositive (or: subordinate) clauses (ex. 25). The relative pronouns (or adverbs) introducing restrictive relative clauses may be ellipted (when they do not function as subject in the clause). (23) (24) (25)
the boy that/who was standing near the table the woman (who(m)/that) I met yesterday This is John, who(m) I mentioned to you yesterday He met his father, who(m) he hates more than anyone the belief that he would soon come the fact that she knew no-one at the party
1.3 - The structure of the NP
23
The class of non-finite clauses consists of: -ing participle clauses (ex. 26), -ed participle clauses (ex. 27), and infinitive clauses (ex. 28). All of these clauses can be either restrictive or non-restrictive. (26) (27) (28)
a child living with his parents the boy, hopping from one leg to the other the phenomena described the town centre, crowded with tourists the place to be his proposal, to come and live with him
Categories can co-occur in postmodifying position. One general (syntactic) ordering principle that has often been recognized is that the structurally most complex postmodifier follows the less complex one; i.e. that postmodifying phrases precede postmodifying clauses.23 (29)
the girl [with the blue skirt] [standing in the corner] [talking to John]
1.3.5 Coordination Coordination is a special syntactic means to combine elements in the language. For example, noun phrases can be coordinated at sentence level (ex. 30). Within the structure of the noun phrase, coordination most frequently involves two immediate constituents with the same function and realized by the same category (ex. 31). In addition, two (or more) functions can be taken together to form one part of the coordination (ex. 32). Occasionally, the two coordinated parts are ‘unbalanced’, i.e. consist of different functions and/or categories (ex. 33). (30) (31)
(32) (33)
John and Mary came to the party. They bought two books and some papers. one or two thoroughly comfortable chairs this spoilt and sulky child a certain ease and authority threats about Robin’s safety and about his life a sick or a very tired man his mingled cleverness and emotional immaturity your host and your hostess of the evening an accident or sudden illness the guts or the good judgment to go to the police a kidnapper or gang of kidnappers under threat of being cornered the thousand Turkish workers and their families who lived there
Coordination is often used as a test for constituency. A string of words is considered to form a constituent of the language if it can be coordinated with a 23
For an extensive discussion on the order of postmodifiers in the noun phrase see Oostdijk and Aarts (1997).
Chapter 1 - The English NP
24
similar string. The examples in (32) and (33) show that this test does not always support the prototypical structure adopted in this study. 1.3.6 Apposition So far, we have discussed two ways to ‘combine’ noun phrases. In the first place, a noun phrase can occur inside another noun phrase as a modifier to the head. Secondly, two noun phrases can be coordinated. The first is an instance of subordination (an NP is embedded in a higher order NP), while the second is an instance of concatenation (the two NPs occur at the same level in the structure of the sentence/clause or phrase). Another instance of the concatenation of noun phrases which has not been discussed so far is apposition (ex. 34). Two NPs that occur in (full) apposition have identical reference, i.e. they refer to one and the same person or thing. Within the structure of the sentence/clause or phrase, an appositive noun phrase is considered as one constituent which realizes one function. NPs in apposition can have any internal structure, including postmodification (ex. 35). However, two NPs in apposition can also share a postmodifier (ex. 36). (34) (35) (36)
1.4
And to this his wife Irene very adequately matched up. His wife’s wish - Grace’s wish - might be sacred to him. Among the new boys there was a child from a Research Station, Sam. They walked silently through the garden - a great warm walled space already heavy with the scents of evening. At Feather’s there is just one daughter, Simona, who is frightfully stylish and deb. She married her cousin Bobby Angrave, who did seem a likely heir. Variant noun phrases
The prototypical description of the NP which we have discussed in section 1.3 can account for a large proportion of the NPs found in actual language use. However, when we study the occurrences of NPs in a corpus, we find variants which are not adequately described by the surface structure description of traditional grammars and which pose problems for theories that use general movement or placement rules such as Transformational Grammar and Functional Grammar. In order to be able to include these NPs in the existing descriptions, detailed information is needed about the nature of such ‘variant’ NPs. In this research, all variations on the prototypical NP structure represented in Figure 1-6 (p. 19) are considered to be ‘variant NPs’. Variation with respect to the prototypical structure occurs for instance when additional elements are introduced into the NP, such as an emphasizer: (37)
I myself The people themselves
1.4 - Variant NPs
25
But the majority of variant NPs can be explained by the ‘mobility’ of the (immediate) constituents of the NP. In theory, an (immediate) constituent of the NP can ‘move’ in two directions. In the first place, a constituent can occur to the left of its prototypical position. This type of mobility will be indicated by the term ‘fronted’. In the second place, a constituent can occur to the right of its prototypical position. This type of mobility will be indicated by the term ‘deferred’. If a constituent occurs in a position outside the boundaries of its mother constituent (either fronted or deferred) it is referred to as ‘floating’. If a phrase has one or more ‘floating’ immediate constituents is called ‘discontinuous’, i.e. not all constituents of the phrase are adjacent to each other. The (prototypical) position of a constituent of the NP is generally expressed with regard to the position it takes in relation to the head of the NP. A premodifier precedes the head of the NP, while a postmodifier follows it. In the same way the mobility of constituents can be expressed with regard to the position a constituent has in relation to the head. The definitions ‘fronted’ and ‘deferred’ can then be further specified by stating that a fronted constituent always precedes the head of the NP, while a deferred constituent follows it. In doing so, the sequence modifier-determiner-head is unambiguously a case of fronted modification, whereas otherwise it could also have been interpreted as a deferred determiner. For the structure of the noun phrase, the potential mobility of phrase constituents has two consequences. In the first place, the immediate constituents of the NP other than the head can occur at various positions within the NP, or they can occur outside the boundaries of the NP. In the second place, the immediate constituents of a phrase modifying the NP can be subject to mobility, resulting in a discontinuous modifier of the NP, either within or across the NP boundaries. If we take the prototypical NP structure as a starting point, we can, in theory, come across (variant) NPs with a (floating) deferred of fronted modifier, a (floating) deferred or fronted determiner, a (floating) deferred limiter, a discontinuous modifier, and/or a discontinuous determiner. In the corpus, we find instances of most of these variant NPs. They can be classified using the following nine types: Type A: NPs with a deferred modifier: a modifying adjective phrase (AJP, exs. 912), adverb phrase (AVP, exs. 38-39) or noun phrase (exs. 40-41) follows the head in postmodifying position; (38) (39) (40) (41)
the space below the reason why a man his age the paper this morning
Type B: NPs with a floating deferred modifier: a modifying clause or phrase occurs outside the boundaries of the NP, following the head, at sentence or clause level; for example,
Chapter 1 - The English NP
26 (42) (43) (44)
A rumour was spread that the president was dead. He gave the order to his commanders to withdraw. You have no means, I suppose, of retrieving it?
Type C: NPs with a fronted modifier: a modifying clause precedes the head of the noun phrase in premodifying position; (45) (46) (47)
a do-it-yourself job a not-to-be-forgotten scene the Insert date/time icon
or a modifying adjective phrase precedes the determiner; for example, (48) (49) (50)
… although I wish I did as useful a one. It was too fine a day for the car. I had never done anything on so small a scale before …
Type D: NPs with a discontinuous modifier: the immediate constituents of a modifying adjective phrase do not occur adjacent to each other. Four different occurrences of a discontinuous AJP can be distinguished24: (a) the AJP (premodifier plus) head precedes the head of the NP (but follows the determiner, if present) while the AJP postmodifier follows the head of the NP; examples are: (51) (52)
Spinoza is not an easy philosopher to grasp. This is a much higher frequency than would be expected.
(b) the (premodifier plus) head of the AJP precede the determiner plus head of the noun phrase, while the AJP postmodifier follows the head of the NP; for example, (53) (54)
24
But standing on the touch line as he did (if the figure wasn’t too weird a one to entertain) he had to await what the general disorganisation cast up to him. It appears to be the 60S components of the ribosomes which associate with the membrane in such a way that the 40S particles protrude into the cell sap.
I found two occurrences of a discontinuous adjective phrase in my corpus data that do not conform to this description of discontinuous modification. In both NPs the first part of the discontinuous AJP occurs in postmodifying position. In the first example discontinuity is caused by the floating of the adjective phrase postmodifier toward the end of the sentence, whereas in the second example it occurs at the beginning of the sentence. It is striking that in both examples the head of the NP is realized by a (non-) assertive pronoun. - But there was something more positive to it than that. - Than these last five words he positively felt that he had never heard anything odder in his life.
1.4 - Variant NPs
27
(c) the premodifier of the AJP precedes the determiner, while the head of the AJP follows the determiner (but precedes the head of the NP); for example, (55) (56)
Bobby has rather a freakish sense of humour. Evidently, structures like mitochondria and chloroplasts have quite a high degree of functional autonomy.
(d) the AJP premodifier precedes the determiner, the AJP head follows the determiner but precedes the head of the NP, the AJP postmodifier follows the head of the NP; for example,25 (57)
He seemed quite a different man from the one she had met the other day across Mr. Crowther’s desk, quite different from the man who talked to Jenny this afternoon.
Instead of following the head (or postmodifier) of the NP immediately, the postmodifier of a discontinuous AJP can occur outside the boundaries of the NP; for example, (58) (59)
And Harry Mack was in better shape, medically speaking, than I expected. Oh, I feel that if a man has sufficient violence in him to slit his own throat, he’s certainly capable of slitting another’s.
Type E: NPs with a deferred determiner: the determiner occurs in a position following the head, adjacent to the other constituents in the NP; for example, (60) (61)
Make it quite clear to us all. How are you both?
Type F: NPs with a floating deferred determiner: the determiner occurs outside the boundaries of the NP, following the head; for example, (62) (63)
They were both frightfully rich. This is all very interesting.
Type G: NPs with a discontinuous determiner: (a) the immediate constituents of the determiner phrase are interrupted by a modifying AJP (exs. 64-65), (b) the predeterminer follows the head of the NP (possibly outside NP boundaries), while the rest of the determiner phrase precedes the head (exs. 66-67), or (c) the determiner phrase postmodifier follows the head of the NP (possibly outside NP boundaries) (exs. 68-70);
25
Compare Nida’s analysis of quite a large house to paint as discussed in section 1.2.2.
Chapter 1 - The English NP
28 (64) (65) (66) (67) (68) (69) (70)
Mayo leant over the desk and pointed an angry two fingers at the Canon. … we will be looking back at an equally fruitful and enjoyable 12 months. The children both love playing games. The substituent amino acids have all been identified. The one anguished parent deserves as much sympathy as the other. The people themselves don’t actually get many more days than six a year. Enough documentation has already been assembled to warrant drawing a new composite profile of leadership.
Type H: NPs with a deferred limiter: the limiter occurs in a position following the head, adjacent to the other constituents in the NP; for example: (71) (72) (73)
Mayo looked at her as though she had not said anything at all. I have no standing in the matter whatever. … although Appleby was inclined to think that this notion was the issue only of his own sustained professional acquaintance with human misconduct.
As was said above, part of the variant NPs can be explained by the introduction of a non-prototypical constituent into the structure of the NP. These introduced constituents are themselves also subject to the mobility phenomenon. An example is the emphasizer. If an emphasizer occurs adjacent to the other constituents of the NP, it usually takes the last position in the noun phrase (ex. 37). In theory, it can also be fronted or occur outside NP boundaries, either preceding or following the head of the NP. In the corpus we find only one type of mobility for emphasizers: Type I: NPs with a floating deferred emphasizer: the emphasizer occurs outside NP boundaries; for example, (74) (75) (76)
1.5
I would not myself have dreamt of saying a word to Charles. It is postulated that the structural genes for these two enzymes are themselves controlled by a master gene. Amanda was elegant herself in a formal housecoat tailored in billiard-table cloth.
Conclusion
In the present chapter I have discussed the description of the English noun phrase as it can be found in the modern descriptive tradition, and have arrived at an integral description of the NP: the prototypical noun phrase structure (Figure 1-6, p. 19). This integral description is characterized by an immediate constituent analysis representing a rank-hierarchy of units of description and the linear order
1.5 - Conclusion
29
of constituents. Every constituent, at every level of analysis, is described for its function (the role it plays in a larger structure) and its category (based on characteristics it shares with other constituents of the same type/class). This NP description is implicitly based on the idea that constituents are built up of continuous sequences of words. However, even in a relatively rigid word order language such as English, constituents have a certain kind of ‘mobility’; they can occur in different positions from their usual one. If an NP constituent or part of an immediate constituent of the NP occurs in a different position from the one it occupies in the prototypical NP, this results in a ‘variant’ NP structure. In this book, I investigate the mobility of immediate constituents of the noun phrase in English. The integral description forms the starting point for the corpus study. A corpus study enables us to test the adequacy of this description on the occurrences of NPs in actual language use. The purpose of the corpus study is to make an inventory of the frequency and distribution of variant NPs in contemporary British English. This means that there are two important restrictions to the study of mobility in the English noun phrase. First, it aims to be a description of language use. Corpora are intrinsically manifestations of the way people use a language, either in written or spoken form. In other words, a corpus contains evidence of ‘performance’ or ‘externalized language’ and not of ‘competence’ or ‘internalized language’. The description is thus a ‘surface structure’ description of the noun phrase structure. The second restriction is that the current study aims for a description of mobility in ‘contemporary British English’ only. This means that the corpus that is used has to be characteristic of this particular variety of English. However, a corpus study alone will not suffice. Every corpus study is finite and cannot guarantee a complete picture. Not only is the non-occurrence of a certain structure no guarantee for its ungrammaticality or unacceptability, neither does the occurrence of a single structure in isolation give information about its possible variants. By systematically testing a theoretical description, such as the prototypical NP structure, on the occurrences in a corpus, the corpus study will give rise to (strong) hypotheses about the characteristics of the mobility of NP constituents. These hypotheses can subsequently be tested by means of another approach to the study of language use. Following others (e.g. Quirk and Svartvik, 1966, 1979; Greenbaum and Quirk, 1970; Greenbaum, 1973, 1984), I use elicitation data to supplement the corpus data. Chapter 2 discusses the empirical approach which was adopted for the present research. This approach, which I refer to as ‘the multi-method approach to descriptive linguistics’ combines judgments of native speakers with the quantitative and qualitative data obtained from a corpus. In chapter 3, the results of the corpus study are discussed. The purpose of the corpus study was to make an inventory of the frequency and distribution of NPs in contemporary British English in order to arrive at a detailed description of NP structures. Special attention is paid to the occurrence of variant NPs. The results of the corpus study give rise to hypotheses on the possible word order variation in the NP. The hypotheses for the fronted modifier, the discontinuous adjective phrase, and the floating deferred modifier were subsequently tested in
30
Chapter 1 - The English NP
an elicitation experiment. The results of the elicitation experiment are discussed in chapter 4, as well as their combination with the data obtained from the corpus study. In the past, some occurrences of discontinuity in the noun phrase have been studied from various theoretical backgrounds. In chapter 5, I give an overview of some of the existing literature on discontinuity in the noun phrase and of research into the mobility of constituents in language in general. Research in the fields of language typology, discourse analysis, psycholinguistics and descriptive linguistics has already revealed various conditions that can influence the mobility of sentence/clause (and phrase) constituents. In this chapter, I discuss some of the approaches for their applicability to the mobility of noun phrase constituents. In chapter 6, finally, I summarize the main findings and conclusions of the present study and formulate some suggestions for further research.
Chapter 2 Methodological considerations 2.1 Introduction The present study on the mobility of constituents in the English noun phrase is a study of language use, rather than a study of the language faculty or of ‘competence’.1 For the study of language use, corpora constitute the primary source of data, since a corpus is a reflection of language production in a context of language use. A second reason why for the present study a corpus is the obvious source of data is that a corpus study is useful for testing the adequacy of a given (traditional) description. In the present study, the prototypical description of the noun phrase is tested for its adequacy on occurrences of NPs in language use. Confrontation with the corpus will provide the researcher with information regarding the structures that are described by the grammar and also regarding the structures that the grammar fails to describe. However, a purely corpus-based approach to descriptive linguistics has some clear disadvantages. For one, a corpus has to be very large in order to contain a fair amount of structures in the language and give a reliable impression of the relative likelihood of structures. It is hard to predict beforehand exactly how large a corpus has to be, since the size which is minimally required is determined by the frequency of occurrence of the phenomenon under investigation. A second disadvantage of the exclusive use of corpus data is related to the inherent restrictedness of corpora. Although a corpus study will provide the researcher with (strong) hypotheses on the characteristics of the phenomenon under investigation, it is never clear whether these characteristics hold for the language as a whole, or whether they are a consequence of the restrictedness of the corpus. At the same time, the non-occurrence of a construction does not yield a conclusive answer as to whether the construction is really unacceptable or ungrammatical. These disadvantages of the exclusive use of corpus data can be overcome by supplementing corpus data with another approach to the study of language use. In this chapter, I discuss an empirical approach which combines corpus data and intuitive data. This approach, which I refer to as ‘the multi-method approach to descriptive linguistics’ combines judgments of native speakers with quantitative and qualitative data obtained from a corpus. In principle, the multimethod approach is independent of the type of descriptive study that is conducted. In this book, however, it is illustrated by means of the study of the mobility of constituents in the English NP. The present research starts with a 1
Parts of this chapter have previously been published in de Mönnink (1997a/b) and de Mönnink (1999).
32
Chapter 2 - Methodological considerations
corpus study of the frequency and distribution of NPs in British English. An elicitation experiment is then used in order to supplement the corpus data and to test the hypotheses that resulted from the corpus study. This specific implementation of the multi-method approach has some consequences for the elaboration of the method discussed here, more specifically for the design of the corpus which is discussed in section 2.5, and the design of the elicitation experiment which is discussed in section 2.6. 2.2 Corpus data versus intuitive data Within linguistics there has often been a clash between the use of intuitive data on the one hand and corpus data on the other. This clash is partly caused by the different fields of interest that exist in the wide linguistic discipline. If we place these fields on a data scale, one end is occupied by linguists who want to account for the mental processes involved in language acquisition, language production and language processing. In other words, they investigate a person’s implicit knowledge of language, his/her competence. Most of these linguists find no use for corpus data.2 They are not interested in frequencies of occurrence or the distribution of structures. Instead, they use intuitive data. Intuitive data is largely based on the intuition of the linguist and, perhaps more importantly, judgments of native speakers. One of the main arguments against intuitive data is that no methodology has been formulated of sentence judgments (cf. Schütze, 1996). The gathering of intuitive data has been largely informal, while very little use is made of the procedures of psychological experimentation. A second, related argument against intuitive data is that the types of test used to tap into the linguistic knowledge of a speaker in fact measure performance, since giving judgments is a production task which reflects language behaviour in a concrete situation (cf. Nagata, 1989; Shohamy, 1996). For descriptive studies which aim at the description of a language or a language phenomenon, intuitive data have the additional drawback that judgments are subjective and that descriptions will remain incomplete since no person can think of all possible occurrences in a language. Also, the tests often bring to light a speaker’s generalizations about his or her language use, instead of actual language use. At the other end of the data scale we find linguists who use corpus data to the exclusion of other types of data, either because they aim for a ‘purely scientific’ approach to descriptive linguistics in which intuitive data have no place, or because they are interested in quantitative data only.3 The use of corpus data has some clear advantages for descriptive studies. First of all, corpora are reflections of actual language use. They are collections of language data as produced by the native speakers of a language in everyday life. Corpora are thus valuable for collecting objective data on language use. This also implies that 2
This is mainly true for theoretical linguists. It does not apply to most linguists in the fields of applied linguistics and psycholinguistics who are interested in first and second language acquisition, language production and/or language processing. 3 This last group is also known as statistical or quantitative linguists.
2.2 - Corpus data versus intuitive data
33
corpus data can be used by non-native speakers of a language, for whom introspective data are not directly accessible. Secondly, corpus data give access to language variation (Oostdijk, 1991). A corpus gives insight into the distribution of linguistic features across, for example, text types or communicative situations. A third aspect of corpora is that they provide contextual information for phenomena. Finally, corpus data are verifiable, which is an important requirement for a scientific approach to linguistics. However, the use of corpora as the exclusive source of data for descriptive linguistics has been rightfully criticized. No corpus, however large, can contain every possible structure of a language in all possible contexts. Thus, a description of all and only the occurrences in a corpus may be objective, but it will not result in a complete description of the language. Toward the middle, but still at the corpus end of the data scale, we find the field of corpus linguistics.4 Corpus linguistics is the branch of (computational) linguistics that is concerned with the study of actual language use on the basis of corpora. The position of corpus linguistics (CL) and corpus data in present-day descriptive linguistics is expressed by Aarts (1995) as follows: At the moment we can say that corpora have been reinstated as respectable sources of linguistic data. But, in contrast to the early structuralists, we are now able to formulate linguistic hypotheses initially on the basis of intuitive data and to use corpus data as the empirical test bed for these hypotheses. The second factor that influences the nature of present-day CL, is the fact that CL has developed into a computational discipline. As a consequence, the linguistic hypotheses that are formulated in a CL context have to be expressed in an explicit and formal way. Together, these two characteristics make CL into a formal, empirical, languagespecific discipline. (Aarts, 1995: 566) Corpus linguistics is not placed at the end of the data scale because intuitions play a significant role in arriving at a description of language use. Two different types of intuition are involved. In the first place, the linguist will have ‘linguistic’ intuitions on the possible structures in a language. These intuitions form the starting point of every descriptive study in that they are the basis for the linguist’s hypotheses. Secondly, we need intuitions on language use to judge which of the occurrences that we come across in a corpus should be included in our description, and perhaps more importantly, which constructions that were not encountered in the data should be included all the same. To provide this latter type of intuition, the linguist should (ideally) be a native speaker of the language under investigation, or else use the intuitions of other native speakers. This second need of intuitive data has been recognized by several corpus linguists (e.g. 4
See also chapter 1, section 1.1.2 on corpus linguistics.
34
Chapter 2 - Methodological considerations
Aarts, 1991; Fillmore, 1992), but the practical and theoretical implications for linguistic methodology have so far not been worked out. 2.3 A multi-method approach to descriptive linguistics For descriptive studies neither the exclusive use of intuitive data, nor the exclusive use of corpus data is adequate. Descriptive studies aim at providing a comprehensive description of the use of a specific language. In this section, I argue that descriptive linguistics would gain considerably from a multi-method approach to the data, i.e. a combination of both intuitive data and corpus data.5 The corpus linguistic approach forms a good starting point, since it already combines both types of data. However, in a profound and empirical approach to descriptive linguistics, the use of both types of data should not be a linear and/or uni-directional process, starting with intuitive data and ending with corpus data. Instead, a descriptive study should be conducted following a cyclic process. The data cycle for descriptive linguistics adhered to in this study is represented in Figure 2-1. literature Hypothesis/-es
intuitive data
corpus data
Hypothesis/-es literature
Figure 2-1: Data cycle for descriptive linguistics A methodologically sound descriptive study can start and end anywhere on the data cycle as long as the whole round is completed at least once. For most studies the easiest point to start will be with intuitive data which forms the basis for the research hypotheses. It is, however, also possible to use corpus data for exploratory purposes. The corpus data can point to interesting research questions 5
The idea of combining corpus data with intuitive data is not new. Back in the nineteensixties, Quirk and others already envisaged that for a comprehensive description of English Grammar, corpus data should be supplemented by elicitation data (cf. Quirk and Svartvik, 1966, 1979; Greenbaum and Quirk, 1970; Greenbaum, 1973, 1984).
2.3 - A multi-method approach to descriptive linguistics
35
and can help decide what categories are relevant to classify one’s research problem and to formulate hypotheses. It may even be the case that hypotheses are already implicitly or explicitly available. This is the case, for example, with descriptions of structures or phenomena which have been the subject of previous descriptive studies. The linguist may put the appropriateness of the description to the test by using both types of data. In many cases it will not be clear precisely where in the data cycle the research starts. Most descriptive studies start from a vague idea a linguist has about a specific phenomenon or structure in a language. But where did this idea come from? Was it the result of a spontaneous thought (introspection)? Was it something that the linguist read in some publication about a structure or phenomenon that triggered his attention (hypothesis)? Or did he come across a striking occurrence while reading the morning paper, which somehow got stuck in his mind (corpus/textual data)? In each of these cases, the actual research process will start from a hypothesis formulated by the linguist, regardless of how the hypothesis came into being. Every time the research reaches a ‘hypothesis/es’ point in the cycle, it should ideally be the case that one or more hypotheses can be formulated, accepted, rejected or reformulated on empirical grounds. The process should only stop when all (remaining) hypotheses have been either accepted or rejected. Of course, following the data cycle is in itself no guarantee for a sound empirical approach. It largely depends on the way in which the intuitive and corpus data are obtained and analysed. For corpus data, the main concern is the design of the corpus which is used. For intuitive data we first have to find an appropriate way to collect the data. Simple introspection by one person is all right as a starting point of the description, because it will then be tested on objective data as it is found in the corpus. However, simple introspection should never be the final point in the process, since it would interfere with the empirical grounding of the research. Also, it is often hard for a linguist to distinguish between his or her intuitions as a native speaker of a language or his or her linguistic intuitions. The closer the research gets towards a complete description of a phenomenon in a language, the less the introspection of one person will have to offer. At a later point in the process we therefore need a carefully designed experiment by means of which we can elicit intuitive data in a way that guarantees the collection of objective empirical evidence. 2.4 Combining corpus data with elicitation data For the study of mobility in the English noun phrase, the starting point is the prototypical description of the NP as discussed in chapter 1. The study starts from the following (null) hypothesis: the prototypical description of the noun phrase accounts for all occurrences of the NP in British English. This hypothesis is tested on corpus data. The corpus yields data which indicate that the null-hypothesis has to be rejected, since variants of the prototypical description are found. Subsequently, new hypotheses regarding possible variants are formulated. These hypotheses are all based on one general hypothesis: in theory, any noun phrase
36
Chapter 2 - Methodological considerations
constituent (or part of an immediate constituent) can be mobile in any given context. The corpus study shows which of the theoretical possibilities actually occur in the corpus, and which do not. In order to test whether the ‘missing’ variants are in fact grammatical or ungrammatical in language use, a separate hypothesis is formulated for each missing construction, which predicts that the construction is unacceptable in British English. All hypotheses are then tested in an elicitation experiment. As a result a hypothesis is either accepted or rejected. The cycle is then completed. In this way, the elicitation experiment is used to supplement the corpus data, yielding additional noun phrase structures which are acceptable in English language use, although they were not encountered in the corpus. Besides supplementation, the combination of corpus data and elicitation data can be used to determine which constructions have to be incorporated in the description of the phenomenon under study and which can be left out. As Aarts (1991) points out, acceptability judgments are important for the corpus linguist, since it helps him make decisions about whether or not a construction found in the corpus should be incorporated in the grammar.6 Aarts makes a distinction between intuition-based and observation-based grammars. An intuition-based grammar “is written on the basis of the linguist’s intuitive and explicit knowledge of the language and whatever is helpful in the literature” (1991: 46). It will describe the sentences that are both grammatical and acceptable, the ‘clear cases’. The observation-based grammar is the result of the confrontation of the intuitionbased grammar with the sentences contained in the corpus. A corpus will contain both acceptable and unacceptable sentences. The corpus linguist has to make decisions about which of these structures should and which should not be included in the grammar. Clearly, unacceptable, ungrammatical sentences should not be accounted for by the grammar. Aarts calls these utterances ‘metagrammatical’. They are often meant to be ungrammatical by the writer for stylistic or other reasons, and do not belong to the set of linguistically acceptable sentences yielded by a grammar. With acceptable, ungrammatical sentences, however, it proves more complex to come to a decision. Aarts gives two possible criteria by which to decide whether a construction should be incorporated or not: is the construction intuitively judged to be a current one, and can it be incorporated in the grammar? The issue of incorporation is a secondary criterion based on practical considerations as well as questions about the suitability of the descriptive framework. Not every new structure the linguist comes across should necessarily lead to an adaptation of the description. Currency is the primary criterion. But how can currency be defined? It has to do with general acceptability, and that can be defined in terms of frequency of occurrence and normalcy. Data about the 6
In this article Aarts restricts himself to a discussion of formal grammars which can be used for the linguistic annotation of the corpus, but for the most part his assertions hold for descriptive grammars in general. The formalization of a grammatical description (for instance for the purpose of applying it in the process of the automatic annotation of corpora) has the advantage of bringing to light inconsistencies and lacunae contained in the description.
37
2.4 - Combining corpus data with elicitation data
frequency of constructions is within easy reach but, as Aarts points out, it is not necessarily a good indication for acceptability. The notion of normalcy, on the other hand, is much harder to capture. There is as yet no empirical method to determine normalcy. Thus, “we have no other alternative than to rely on our intuitions with respect to the currency of a particular construction, until we can give the notion an empirical grounding” (1991: 59). Intuitions about and literature on grammaticality and acceptability
Is the construction grammatical?
no/doubt
Does it occur in corpus data?
no
yes
Is it acceptable?
no
yes
yes
Is it current?
no
yes
Can it be incorporated?
yes
purely intuitionbased grammar
no
discarded
purely observationbased grammar GRAMMAR
Figure 2-2: Process of grammar construction The process of constructing a grammar as described in Aarts (1991) is represented in Figure 2-2.7 This process is repeated until the grammar reaches the desired level of completeness. It is clear that, in deciding which structures should be included and which should not, acceptability judgments play an important role. Once a sentence is described by the grammar it is grammatical in the formal sense of the word. In this sense, ‘grammaticality’ is a theoretical term and, at best, 7
What is called ‘grammar’ in Figure 2-2 is referred to as ‘observation-based grammar’ by Aarts (1991).
38
Chapter 2 - Methodological considerations
reflects a native speaker’s judgments about the well-formedness of sentences. It is not directly accessible through the intuitions of the speaker. Acceptability, on the other hand, refers to the judgments speakers have about the (contextual) appropriateness of sentences in their own language. In other words, acceptability judgments reflect a form of (linguistic) behaviour and this behaviour is influenced by many attitudinal, cognitive and pragmatic factors (Newmeyer, 1983). Carefully designed elicitation tests, in which all of these factors are considered, might just give this process the ‘empirical grounding’ it needs. The process of grammar construction represented in Figure 2-2 (p. 37) is not in accordance with the multi-method approach. Constructions which do not occur in the corpus but which might be acceptable all the same are simply discarded. In the multi-method approach these constructions would also be subjected to (acceptability) judgments of native speakers. The process in Figure 2-2 is easily adapted to the multi-method approach by subjecting constructions which did not occur in the corpus to the same judgment process as to which the ungrammatical sentences which did occur were subjected. This adaptation to the process is represented in Figure 2-3. …
Does it occur in corpus data? yes/no
Is it acceptable?
no
…
Figure 2-3: Process of grammar construction adapted to the multi-method approach So far, we have considered two reasons for extending a corpus-based study with elicitation tests. First, experimental data can serve to supplement corpus data and second, the corpus linguist can use the acceptability judgments of informants in order to decide which constructions to incorporate in the grammar. Greenbaum (1984) points to a third advantage: Elicitation tests have been devised to resolve questions raised during the analysis of corpus material. Although their function is primarily supplementary, the results may also pose questions for further investigation through corpus searches or for additional elicitation experiments. (Greenbaum, 1984: 200)
2.5 - The corpus study
39
This is in line with the adoption of a cyclic process for descriptive studies. The corpus data yield hypotheses which are tested in an elicitation experiment. The results of the elicitation experiment may yield new hypotheses which can then be tested by means of a new, more detailed or more extensive, corpus study. For the present study on mobility of NP constituents, supplementation of the corpus data is the most important reason for combining corpus data with experimental data. However, the elicitation data are also used to make decisions on incorporation into the description of the NP and for enabling a more extensive corpus search. 2.5 The corpus study The purpose of the corpus study in the present research is to make an inventory of the frequency and distribution of NPs in contemporary British English and to arrive at a detailed description of the syntactic characteristics of these structures with special emphasis on the mobility of constituents. Corpus-based studies generally comprise a qualitative and a quantitative analysis. The qualitative analysis aims at a detailed description of the phenomenon under study. The quantitative analysis gives a precise picture of the absolute and relative frequency of occurrence of the particular phenomenon. For a quantitative and qualitative study on the mobility of NP constituents the corpus material used has to meet certain requirements as to the design of the corpus, its size and the level of delicacy in the annotation. This section discusses the design issues for the present study. 2.5.1
Design of the corpus
Not simply any collection of texts is considered to constitute a corpus. A corpus is a collection of texts that has been built according to explicit design criteria and (often) for a specific purpose. An important aspect of corpus design is therefore the problem of representativeness. Ideally, a qualitative study of a phenomenon of language use should give insight into the linguistic variation of that phenomenon in a specific language or language variety. Therefore, the corpus has to be representative of that language (variety). For the present study this means that the corpus should ideally include samples of a number of well-defined written samples (text categories) and (transcribed) spoken samples representative of contemporary British English. Questions as to exactly how many and which text categories should be distinguished or how many samples are required to represent a particular text category have been and still are subject to discussion.8 In practice, the amount of time and money available are often decisive for the design of the corpus. Also subject to discussion is the question of required corpus and sample size. In theory, corpus size is determined by the frequency of the phenomenon under study and the design of the corpus. For quantitative analyses, use is made 8
For a discussion on issues of representativeness of corpus design see Biber (1990, 1993), Biber and Finegan (1986), and Oostdijk (1988a/b, 1991).
40
Chapter 2 - Methodological considerations
of generally accepted statistical methods. However, for some statistical tests to be reliable a specific minimum frequency is needed (see e.g. de Haan, 1992). The chi-square test is often used to determine the distribution of a phenomenon in a corpus. For a chi-square test to be reliable the number of expected observations of a single variable cannot be lower than five. If one wants one corpus to contain various text categories to identify linguistic variation among texts, sample sizes have to be sufficiently large to fully represent the phenomenon under study (Biber, 1993; de Haan, 1992). Exactly how large the corpus as a whole has to be depends on the actual frequency of the phenomenon under study and the number of variables involved. It is, however, difficult to predict beforehand how large each sample (and thus the corpus as a whole) has to be. Again, in practice corpus size is often a question of the availability of material. The present corpus study can be divided into two parts. The first part involves an exhaustive inventory of the frequency and distribution of (variant) NPs. The occurrences have been collected from the corpus described in Appendix A; in the rest of the book this corpus will be referred to as ‘Corpus A’. Corpus A comprises approximately 170,000 words and includes five distinct genres: fiction, non-fiction, drama, scripted speech, and spontaneous speech. The genres can be put on a scale of formality with the non-fiction samples being the most formal and the unedited, spontaneously spoken samples the least formal (Figure 2-4). The fiction samples include some written dialogues, while the drama samples are meant to reflect everyday speech. The number of samples per category (genre) is fairly small (varying from 2 to 16), while the average size of the five categories is approximately 30,000 words.9 most formal
non-fiction scripted speech fiction drama spontaneous speech
least formal Figure 2-4: The scale of formality The second part of the corpus study is an inventory of occurrences from the corpus described in Appendix B, from now on referred to as ‘Corpus B’. The samples in corpus B were chosen for the fact that their detail of annotation enabled an automatic search for NP structures. The automatic searches were not checked for completeness. Corpus B comprises approximately 220,000 words and includes fiction and non-fiction samples. Since the inventory is not (necessarily) exhaustive, this second part of the corpus study cannot be used to make statistical claims about the occurrence of variations on the traditional NP structure. The 9
Exceptions to the average size are fiction with 61,835 words and drama with 19,664 words.
2.5 - The corpus study
41
occurrences were studied for their distribution and for the possible realizations of their functional constituents. 2.5.2
Annotation and exploration
The required nature and delicacy of annotation of the corpus data is determined by the nature of the corpus study. For studies on the occurrence of lexical items in their context, raw corpora or corpora annotated for their parts of speech are often sufficient. However, a qualitative study of a syntactic phenomenon, like the mobility of NP constituents, requires a fully parsed corpus, that is, a corpus which has been annotated with detailed syntactic information. From the previous discussion it can be concluded that, for a profound corpus-based study on the nature and frequency of variant NPs in contemporary British English, one would ideally want to have access to a large corpus designed to include a variety of samples representative of this subset of English. The individual samples should be sufficiently large to obtain ample quantitative data for various statistical techniques. Unfortunately, everyday practice often does not reflect the ideal situation. For the present study, corpus design and size are determined by the availability of corpus data annotated with detailed information about the syntactic structure of sentences. To date, the availability of such corpora still leaves much to be desired. All samples in Corpus A have been annotated according to the descriptive framework that is used in the analysis of the ICE corpora.10 An example of a sentence analysis according to this descriptive framework is given in Figure 2-5 (p. 42).11 The example is represented in indented tree format. Each blank space from the left-hand side represents a deeper level in the tree structure. Each line begins with a function label (in capitals), followed by a comma and a category label (in capitals). Where appropriate, the category label is followed by attributes (between brackets, separated by commas). Word leaves follow the attributes and are enclosed in parentheses. All utterances in the corpus have received a full syntactic analysis and are stored in a syntactic database. For the present study, all analyses were checked manually, especially with regard to the function and structure of noun phrases. Where necessary, analyses were changed and from time to time extra annotation was included to facilitate automatic exploration of the data.12 The analyses are made accessible for further research by means of the Linguistic DataBase system (LDB, van Halteren and van den Heuvel, 1990; van Halteren, 1997). The LDB is a database management system which was specially developed to handle syntactic databases. It enables the user to exploit corpora that have received a full syntactic analysis as represented in Figure 2-5. Queries are expressed by means of exploration schemes. Each scheme consists of a pattern and one or more activities. The pattern describes the constituent structure or 10
International Corpus of English, cf. Greenbaum (1988). A key to the abbreviations is given in Appendix C 12 For a complete description of the annotation added to the TOSCA/ICE analyses see section 3.1. 11
42
Chapter 2 - Methodological considerations
phenomenon that is being studied, while the activities specify the effect(s) the search should have. In this way any node (or combination of nodes) in a syntactic tree can be queried for the function it has, its realization, its frequency of occurrence or co-occurrence with other nodes in a specified context. ^ 1 a n a l ys i s 1 a n a l ys i s i n 0 . 6 s e con d s w i t h T O S C A - IC E / v0 . 2 . 9 3 0 4 2 3 # NOFU,TXTU() U T T , S ( a ct , d e cl , i n d i c, m ot r , p as t , u n m ) S U , N P () N P H D , N ( pr op , s i n g) { D r F e l l } V , V P ( a ct , i n d i c, m ot r , p as t ) M V B , LV ( i n d i c, m ot r , p a st ) { h a d } OD,NP() D T, D T P ( ) D T C E , AR T ( i n d e f) { a } N P P R , AJ P ( a t t r u) AJ H D , A D J ( a t t r u) { b r u s q u e } N P H D , N ( com , s i n g) { m a n n e r } P U N C , P M ( p er ) { . }
Figure 2-5: Representation of a TOSCA/ICE sentence analysis (in indented tree format) 2.6 The elicitation experiment In section 2.2 I argued that a profound descriptive study should be based on both corpus data and intuitive data. I also argued that a carefully designed experiment would be needed to elicit the judgments of native speakers in order to establish an empirical approach to descriptive linguistics. Corpus data represent a more naturalistic approach to data gathering since it deals with observations of actual language use. Elicitation data, on the other hand, are (generally) more artificial in nature, which allows a better control over the variables that can influence the occurrence of a structure, such as sentence type. And while corpus data provide information on overt variables such as frequency and distribution, elicitation experiments can also provide insight into covert variables such as language attitude. In this section I discuss the design of the elicitation experiment. 2.6.1
Design of the experiment
The design of the elicitation experiment which was used in the present study is an adapted version of the design used in the Survey of English Usage (SEU).13 The SEU design comprises two main types of test: performance tests and judgment tests. Both performance and judgment tests have the purpose of eliciting 13
See Quirk and Svartvik (1966, 1979) and Greenbaum and Quirk (1970).
43
2.6 - The elicitation experiment
acceptability judgments of native speakers about possible structures in their language. In addition, some types of performance test can be used to elicit structures which were not encountered in the corpus, but which are expected on the basis of literature and/or the linguist’s intuitions. Also, because of the relatively spontaneous character of the answers, performance tests may yield constructs which were not envisaged by the linguist, but which are relevant to the research. Both performance and judgment tests have the purpose of eliciting acceptability judgments. However, with performance tests the judgments remain more implicit since they are obtained through the production of sentences by informants. With judgment tests information on the general acceptability of structures is obtained in a more explicit manner. Compliance Operation Performance Completion Elicitation
Selection Forced-choice selection Word placement Composition
Evaluation Rating Judgment
Preference Ranking Similarity
Figure 2-6: SEU design of elicitation tests A fair amount of attention has been paid to the design of elicitation tests. Most studies which have used elicitation data have given an explicit description of the design of the experiment, including the raw data, the number and nature of the informants, and the types of test used. Within the field of corpus linguistics, the SEU design (SEUD) as represented in Figure 2-6 is the most important example. Greenbaum in particular has paid attention to general design issues (1973, 1976). But for a more elaborate discussion on and accounts of experimentation with design issues, we have to look to studies in the fields of psycholinguistics, sociolinguistics and (second) language acquisition. Here the discussion has often focused on the reliability of intuitions in linguistic research. Within the generative tradition the view is taken that linguistic intuition is a uniform component of native speakers’ competence and can be brought to light through grammaticality judgments.14 Nagata, however, has indicated through several experiments with design issues that ‘speakers do not possess the invariant criterion for judging the grammaticality of sentences, but rather the criterion changes depending on the strategy subjects adopt in the judgments’ (Nagata, 1989: 468). This suggests that a judgment test is a practical production task and 14
I here use the term ‘grammaticality judgment’ because it is used as such in literature. It should be clear, however, that I prefer and will further use the term ‘acceptability judgment’ since in my opinion only acceptability is directly accessible to the intuition of the speakers of a language (see also Newmeyer, 1983).
44
Chapter 2 - Methodological considerations
as such involves psychological processes and is affected by the mental state of a subject and the strategy he adopts to perform the task. Thus, in the design of the experiment these factors should be controlled as much as possible. Table 2-1: The design of the elicitation experiment15 Performance tests: 1 Composition, Rating Construct as many sentences as possible from the (scrambled) words and/or word groups which are presented.16 2 Composition, Select one Construct one sentence from the (scrambled) words and/or word groups which are presented. 3 Operation, Compliance Change the indefinite article into the definite article and rewrite the sentence. 4 Operation, Selection Rewrite a sentence starting with This is (a). 5 Completion, Forced-choice Fill in the gap, choosing from two options. 6 Completion, Word placement Rewrite the sentence using a given word/phrase. 7 Completion, Supplementation Complete the sentence (open ended). Judgment tests: 1 Evaluation Judge a single sentence on its acceptability. 2 Preference, Rating Judge sentences on their acceptability. The sentences are offered in pairs. 3 Preference, Ranking Rank several sentences according to their acceptability. 4 Similarity Judge the semantic relationship between two sentences. 5 Frequency Judge the frequency of sentences. The sentences are offered in pairs. 6 Normalcy Judge the normalcy of sentences. The sentences are offered in pairs. Except for some minor additions and changes to the design of performance tests, I have used the SEUD as a basis for the elicitation of variant NPs. The SEUD performance tests are restricted to making changes in (operation), or additions to (completion) given sentences. When the primary aim of the elicitation experiment is to obtain additional structures to the ones encountered in the corpus, the usefulness of this design is fairly restricted. Of course, totally 15
With ‘Operation, Compliance’ and ‘Operation, Selection’ an example of a possible assignment is given in the second column to explain the test. For all other tests, a general description of assignments is provided. 16 A ‘word group’ does not necessarily constitute one (immediate) constituent. It is not a word group in the linguistic sense of the term.
2.6 - The elicitation experiment
45
unrestricted production is not suitable, but production through the composition of given parts of sentences seems a good addition to the SEUD. 17 This composition test can take two forms: one can ask subjects to give all possible sequences (rating) or you can ask them to give only the one that comes to mind first (select one). When the aim is to get an impression of the currency of a structure, the SEUD does provide the linguist with information on the general acceptability, but does not directly ask the informants for a judgment on the normalcy or the frequency of occurrence of a structure. Greenbaum (1984) adds the elicitation of the relative frequency of sentences to the SEUD. In this context a direct judgment of normalcy seems an interesting third addition to the original SEU design. The design of the present elicitation experiment is represented in Table 2-1 (p. 44). In all, there are thirteen types of test: seven types of performance test and six types of judgment test. An illustration of the use of the types of test is given in Appendix F. 2.6.2
Presentation of the experiment
From previous experiments reported in the literature various design variables can be extracted regarding the linguistic material to be used in the experiment and the procedures to be followed in presenting and administering the task.18 In this section I sum up my approach to some of these variables. In the actual experiment the thirteen types are divided into three test batteries to keep the variation in the types of tests restricted. Each battery includes both performance and judgment tests. The batteries are subsequently divided into three versions (A, B and C) to control order-of-presentation effects. The Bversion differs from the A-version in that the order of the types of test has been changed as well as the order of the sentences or constituents within a single question (when appropriate). The C-version differs from the A-version in that the order of the sentences within one type has been changed. For example, the production of sentence (1) was aimed at by means of a Composition Rating test (CR). This type of test formed part of the first battery. In the A version, CR was the first type of test with which the informants were presented. Sentence (1) was the fourth question/task in its type and the order of the scrambled parts was as in (1a). In the B version, CR was the second type of test, (1) was still the fourth question in its type, and the order of the scrambled parts was as in (1b). In the C version, CR was the first type of test again, (1) is now the first question in its type, and the order of the scrambled parts is as in (1a) again. (1) We gave a completely different department from ours the assignment a) from ours, department, a, we gave, completely different, the assignment b) we gave, from ours, department, completely different, a, the assignment 17
In the SEUD the term ‘composition’ is used to indicate a type of completion test where subjects are given an unfinished sentence and are instructed to complete the sentence in any way they like. For the SEU term ‘composition’ I want to introduce the term ‘supplementation’. 18 See Chaudron (1983) and Ellis (1991).
46
Chapter 2 - Methodological considerations
The batteries are presented in their various versions to comparable groups of informants. For the present study all informants were native speakers of British English, both male and female. They were all young adult speakers with a university background. None of the informants had had any linguistic training. Reports in the literature on elicitation experiments point to the problem of variability in judgments between subjects but also within subjects. Snow and Meijer point out that “testing even a relatively large group of subjects, all of them relatively intelligent and language-conscious, does not assure internally consistent judgements concerning the relative acceptability of sentences” (1977: 172). This variability can be ascribed to various cognitive and linguistic factors but also to the informants’ behaviour in a test environment. It is therefore important to control environmental factors as much as possible and keep test conditions similar. In order to keep test conditions as constant as possible, the elicitation is performed by means of a computer program.19 The program is Windows-based and provides a subject with one task at a time. Figure 2-7 (p. 47) is a screen print of a performance test (Composition, Select one), while Figure 2-8 (p. 48) is an example of a judgment task (Preference, Rating). The performance tests require informants to type in their answer(s). With judgment tests they can indicate their judgments on a seven-point scale by simply clicking the appropriate number with the mouse. The SEU mostly uses a three-point scale, but previous experiments have indicated that “the findings of those who used larger rating scales appear to be more statistically manipulable and revealing” (Chaudron, 1983: 368). The sevenpoint scale is interpreted as shown below. The interpretation for normalcy is the same as for acceptability, but substitute ‘acceptable’ with ‘normal’ and ‘unacceptable’ with ‘abnormal’. score 1-2 2-3 3-4 4-5 5-6 6-7
acceptability perfectly acceptable20 acceptable controversial unacceptable highly unacceptable extremely unacceptable
similarity exactly the same similar meaning controversial fairly different different (meaning) totally different
frequency extremely frequent frequent fairly infrequent infrequent very infrequent does not occur
The Preference Ranking test receives a different interpretation. Here informants are asked to rank five sentences according to their relative acceptability, a ranking of 1 indicates that the sentence is judged to be the most acceptable of the five, a ranking of 5 indicates that it is the least acceptable.
19
I am grateful to André Olthof for his implementation of the program. Judgments of control sentences such as John went with Mary to the park yesterday have shown that the optimum average acceptability score is approximately 1.3. 20
2.6 - The elicitation experiment
47
Figure 2-7: Screen print from the elicitation program of a production test (Composition, Select one) Variation of structures within a test is attained by alternating the elicitation of the different types of variant NP. Still, the target NPs are very specific and thus perhaps easy to recognize. In view of this fact, distractor items are included to prevent the informants from identifying the target structures. These distractor items resemble the target structures in that they show variation in word order. Unlike the target structures, however, they do not deal with NP structure. Apart from diverting the attention from the aim of the experiment, distractor items can also be used to test the suitability of a candidate informant. For this purpose the distractor sentences have to contain unproblematic constructions. If a subject fails
48
Chapter 2 - Methodological considerations
Figure 2-8: Screen print from the elicitation program of a judgment test (Preference, Rating) to perform a task for such a sentence correctly, his or her test should be excluded from the test results. An example of distractor sentences is given in (2). In these sentences only the position of the adverbials is variable. (2) Tom hit the dog yesterday. John went with Mary to the park yesterday.
2.7 - Combining two different types of data
49
Also, clear and explicit instructions can contribute to better results. The instructions should be such that it is fairly easy for an informant to perform the task. Informants can change their answers while still in the same window, but they can never go back. If informants feel that they are unable to answer a question they can click the ‘skip’ button (bottom right in the example screens). This will evoke a new window in which they are asked to give their reason for skipping the question. The batteries are designed in such a way that it will take an informant approximately twenty minutes to complete the experiment. One hypothesis about complex structures is that they are harder to process for a speaker than their regular counterparts. The amount of time needed to complete every single task is recorded, since it may give insight into the correctness of this hypothesis. While record is kept of the time an informant needs to answer, no time limit is imposed, so that every informant can finish the experiment. The actual experiment was conducted with (mostly) students at the University of Liverpool over two rounds.21 During the first round 45 students took part in the experiment, 15 informants for each battery. The second round followed 5 months after the first. In this round of elicitation 75 people took part, 25 for each battery.22 Since the elicitation is done by means of a computer program the main manner of presentation is written. However, the second round of elicitation also included an interview which followed the computerized part. In the interview informants were asked, among other things, to judge the acceptability of sentences including variant NPs and additionally give reasons for the (un)acceptability of the sentences. 2.7 Combining two different types of data If we postulate that the design of the corpus and the experiment are such that they yield reliable data, how can we combine the experimental data with the corpus data? Although these two types of data may appear heterogeneous at first sight, they are in fact more compatible than expected. Both approaches to data gathering are concerned with language use. Of corpora it is generally accepted that they are collections of actual language use. However, the range of linguistic facts which a corpus contains is influenced by its size and by the sampling method used to compile the corpus. The main problem is how to deal with the non-occurrence of constructions in a corpus but also how to interpret the occurrence of constructions which are not expected on the basis of the literature or (informants’) intuitions. Furthermore, as argued by Oostdijk (in press), the data (automatically) available to the researcher is largely determined by the nature and detail of the linguistic annotation used to enrich the corpus and by the researcher’s interpretation of that data. So, while a corpus is unquestionably a 21
I am grateful to Antoinette Renouf and her team at the Research and Development Unit for English Studies who made it possible for me to do the elicitation experiments at the University of Liverpool. 22 The informants received a compensation of £3 in the first round and £5 in the second.
50
Chapter 2 - Methodological considerations
good source for an objective study of language use, the role of the linguist in interpreting the data remains an important factor. Elicitation data are not always considered to be a reflection of language use. Many linguists still consider judgment data the primary source of linguistic evidence for the study of grammar. In their view, judgments of native speakers on the acceptability or grammaticality of a sentence are a direct reflection of their internal grammar (their competence). Others, however, have indicated that native speakers are incapable of judging grammaticality because it is not directly accessible through their intuition and that giving judgments is a production task (e.g. Newmeyer, 1983; Nagata, 1989; Shohamy, 1996; Schütze, 1996). I subscribe to the view that both production and judgment tasks are a form of language behaviour and thus a reflection of performance and not of competence directly. Elicitation data comprise instances of language use produced in an experimental setting. As with corpus data, the input of the researcher plays a vital role in the interpretation of the data. So, although there is no doubt a strong relationship between occurrence in a corpus and grammaticality on the one hand, and acceptability judgments and grammaticality on the other, neither corpus data nor elicitation data give direct insight into the ‘grammar’ of a language. Instead they reflect language use and are both methods which can help the linguist make decisions on which structures to incorporate in the description of the language under study.
Chapter 3 Results of the corpus study 3.1 Introduction Following the multi-method approach described in chapter 2, the study of mobility in the noun phrase can be said to start from the following (null-) hypothesis: every noun phrase which is acceptable in English language use can be described by the prototypical NP structure. The prototypical NP structure is based on a consensus structure of the noun phrase as found in grammatical handbooks of English. On the basis of intuitions, observations in the literature and previous work on corpus data, I expect this null-hypothesis to be rejected. As a result of the confrontation of the prototypical NP structure with the NPs found in a corpus, I expect to arrive at a detailed description of the frequency and distribution of those NPs that can be described by the prototypical structure and those that cannot. In chapter 2, some general issues on the design of corpora were discussed and an account was given of the design of the corpora that are used in the present study. Let me just recapitulate and elaborate on what was said in section 2.5 about the compilation and annotation of the corpora. I use the plural ‘corpora’ to point to the fact that the corpus study can be divided into two parts based on two separate corpora: Corpus A and Corpus B (cf. Appendix A and B). However, of these two, only Corpus A constitutes a corpus in the corpus linguistic sense of the word in that it was built according to explicit design criteria. Allowing for practical restrictions on the availability of material, Corpus A was designed to be illustrative of contemporary British English. It comprises 170,000 words and includes five distinct text categories: fiction, non-fiction, drama, scripted (prepared) speech, and spontaneous speech.1 All samples have been fully annotated and the analyses were checked manually. Where necessary the annotation was changed and/or extra annotation was included to facilitate the study on mobility of constituents in the NP. Corpus B, on the other hand, includes only fiction and non-fiction samples.2 The samples were chosen for the fact that
1
The samples in Corpus A were taken from five different corpora. They include five samples from the Nijmegen Corpus: Cell Biology (non-fiction), The Bloody Wood (fiction), The Mind Readers (fiction), Stop It (drama), and Nil Carborundum (drama); one fiction sample from the TOSCA Corpus: Carson’s conspiracy; three samples of computer manuals (non-fiction) from the IPSM Test Corpus: Lotus, Dynix and Trados; sixteen samples of spontaneous speech from the ICE-GB corpus; and five stretches of prepared speech from the Spoken English Corpus (SEC). See also Appendix A. 2 All samples in Corpus B were taken from the TOSCA corpus. See also Appendix B.
52
Chapter 3 - Results of the corpus study
their detail of annotation enabled an automatic search for NP structures but the analyses were neither checked, nor added to. The samples in Corpus A have all been annotated according to the descriptive model used in the analysis of the ICE corpora (International Corpus of English).3 The model can be characterized as an immediate constituency model in which every immediate constituent is assigned a function label, which indicates its relation to the other elements within the same superordinate (mother) constituent, and a category label, which indicates its individual characteristics. The major units of description are the word, the phrase and the clause/sentence. In the analyses, an attempt has been made to conform to familiar notions as much as possible. For that purpose, the notions and descriptions are derived from traditional descriptive grammars of English. It is therefore not surprising that the description of the noun phrase in the TOSCA-ICE analysis is very similar to the prototypical noun phrase structure described in Chapter 1, since the latter was also based on traditional descriptions of the NP. The prototypical NP structure is repeated here as Figure 3-1 for convenience. NP (LIM)
(DET)
(PREM)*
AVP
DTP
AJP AVP NP
HEAD noun pronoun proform
(POM)* PP CL
Figure 3-1: Prototypical NP structure Since the description is based on an immediate constituency model, there is no straightforward way to deal with discontinuous constituents.4 As we saw in section 1.4, many of the variations on the prototypical NP structure can be explained by the mobility of a constituent of the NP, often resulting in a discontinuous NP (NPs with a floating postmodifier, floating determiner or floating emphasizer) and/or a discontinuous constituent of the NP (NPs with a discontinuous adjective phrase or a discontinuous determiner phrase). For some types of variant NP, a description was already provided in the TOSCA-ICE analysis. This was the case for: -
3
Type A: NPs with a deferred modifier: apart from its more common realization by means of a prepositional phrase and/or clause, the postmodifier
For the analysis of the ICE-GB corpora, use is made of the TOSCA-ICE parser. Underlying this parser is the TOSCA-ICE grammar which is based on the TOSCA descriptive model developed by N. Oostdijk (cf. Aarts et al., 1998). The samples in Corpus B have been analysed by means of the TOSCA-parser. See also section 2.5.2 for an example analysis. 4 See also section 1.3.
53
3.1 - Introduction
function can be realized by an adjective phrase, adverb phrase and/or noun phrase. -
Type B: NPs with a floating deferred modifier: a category that functions as modifier and that follows the head of the NP but does not occur adjacent to the head or another constituent of the NP receives the function label FNPPO. The FNPPO node can be attached anywhere in the tree, but is mostly attached at the same level as the NP it modifies. For example, in the analysis in Figure 3-2 both the NP and its floating postmodifier are immediate constituents of the S(entence):5
The question arises whether it is possible to combine the two theories .
Figure 3-2: Analysis tree of a sentence in which an NP with a floating postmodifier occurs. -
Type C: NPs with a fronted modifier where a modifying clause precedes the head of the NP in premodifying position: apart from its more common realization by means of an adjective phrase, adverb phrase and/or noun phrase, the premodifier function can be realized by a clause.
-
Type E: NPs with a deferred determiner: in the analyses the function label DTDE is included which is used to indicate a deferred determiner.
-
Type H: NPs with a deferred limiter: in the TOSCA/ICE analyses, a limiter is also allowed to occur in any position following the head of the NP. No extra function label is introduced.
5
The analysis in Figure 3-2 is represented in graphical tree format. In principle, it contains the same information as the indented tree format which was discussed in section 2.5.2. Here, however, the attribute information is left out for sake of clarity. A key to the abbreviations is given in Appendix C.
54
Chapter 3 - Results of the corpus study
For the other types of variant NP extra annotation was introduced in order to be able to retrieve and query the variant NPs automatically. They are treated as follows: -
Type C: NPs with a fronted modifier where a modifier precedes the determiner: for modifiers that precede the determiner an extra function label was introduced (PRFR).
-
Type D: NPs with a discontinuous modifier: for modifiers of which the immediate constituents are not adjacent to each other a new function label (MODI) was introduced. However, since discontinuity breaks up a modifying (adjective) phrase of the NP into two or three parts, we also need to introduce three new category labels (AJPD1, AJPD2, and AJPD3). AJPD1, AJPD2, and AJPD3 are ‘ad hoc’ categories for the first, second, and third parts of the discontinuous modifier. This way an extra level is introduced in the analysis of a modifier in the NP. The extra level is necessary to indicate that the discontinuous modifier is realized by an AJP which is divided in two or three parts, and to explicate the realizations of those parts (which is not necessarily one immediate constituent of the AJP). An example of how an NP with a discontinuous modifier is analysed is given in Figure 3-3. Actually , you have a much more macabre mind than I have .
Figure 3-3: Analysis tree of a sentence in which an NP with a discontinuous AJP occurs. -
Type F: NPs with a floating deferred determiner: deferred determiners that occur outside NP boundaries receive the function label FDTDE.
-
Type G: NPs with a discontinuous determiner: the treatment of the discontinuous determiner is threefold. In the first place, the determiner phrase premodifier can occur after the pre- or central determiner in the determiner phrase. Secondly, a noun phrase premodifier can interrupt the determiner structure, by occurring between the first part, the pre- or central determiner, and the second part, the postdeterminer. In the third place, the
3.2 - Frequency and distribution of NPs
55
determiner phrase postmodifier can float away from the rest of the DTP. For the floating determiner phrase postmodifier an extra function label is introduced (FDTPO). The FDTPO can be attached inside the NP or outside at clause/sentence level. -
Type I: NPs with a floating deferred emphasizer: for emphasizers the function label EMPH was included in the TOSCA/ICE description. For floating deferred emphasizers an extra function label (FEMPH) was introduced.
The first part of the corpus study, which is based exclusively on Corpus A, involves an exhaustive inventory of the frequency and distribution of NPs. The NPs are studied for their functional structure and the realization of those functions. A study of NPs in Corpus B is used exclusively to collect extra occurrences of variant NPs in their contexts. In the rest of this chapter, I discuss the results of the corpus study and the implications these results have for the continuation of the data cycle. That is, I introduce the hypotheses that resulted from the corpus study and which were tested in an elicitation experiment. The results of the elicitation experiment are discussed in chapter 4. 3.2 Frequency and distribution of noun phrases Corpus A comprises a total of 54,517 NPs. These also include ‘nested NPs’ where an NP occurs at some level inside a higher order NP, either as an immediate constituent (exs. 1-2) or more deeply embedded in the NP structure (exs. 3-4).6 Example (4), for instance, would count as four NPs in all.7 (1) (2) (3) (4)
[a [day] conference] [sp3-a0019] [the meeting [this afternoon]] [sp3-j0233] [the door of [his car]] [bw-0018] [the roughly formed arch [which] constituted [the entrance to [the grotto]]] [bw-0563]
The embedded noun phrase in (1) is an example of a noun phrase premodifying another noun. In other descriptions this may be treated as a compound noun.8 In 6
Of the NPs 10,113 (18.5%) are embedded in another NP. The percentage of embedded NPs is much higher in the formal categories (33.0% in non-fiction and 26.1% in prepared speech) than in the more informal categories (14.9% in fiction, 11.0% in spontaneous speech and 8.1% in drama). 7 From this point onward, all the examples which are taken from a corpus are followed by a code indicating the origin of the example. The letters before the dash are an abbreviation of the name of the corpus. The abbreviations are explained in Appendix D. The numbers following the code indicate the number of the sentence in the corpus. 8 The main reason being that such sequences have uneven stress. Compare e.g. stress pattern of the compound blackbird with the modifier + head sequence black bird. The stress pattern of day conference is similar to that of the compound.
56
Chapter 3 - Results of the corpus study
my analyses, noun sequences are analysed as modifier + head sequences as much as possible. There are several reasons for doing so. First of all, modifying noun phrases can occur in coordinations (exs. 5-6). They can also be coordinated with an adjective (ex. 7). Secondly, a sequence of nouns can have a clear internal structure, which could otherwise not be expressed (exs. 8-9). And in the third place, the first noun in the sequence is clearly a noun phrase and not part of a compound noun when it can itself be modified by an adjective (ex. 10), have a separate determiner (ex. 11), or even be postmodified (ex. 12). (5) (6) (7) (8) (9) (10) (11) (12)
the [gas] and [water] installation [mr-1057] either [protein] or [pyrimidine] synthesis [cb-0006] [property], [financial], and [income] funds [sp3-f0060] a tiny [[war] wound] pension [mr-0720] [[country] style] chicken a la Ken Hon [sp2-220037] [[good] quality] [virgin] oil [sp1-610284] a [[hundred and fifty] pound] job [sp2-070109] a [middle [of the road]] Social Democrat from the industrial heartland [sp3-a0282]
According to the prototypical NP structure, a complex prototypical (or non-variant) NP can in theory have a variety of structures, ranging from only a determiner plus head, to an NP with a limiter, a determiner, an indefinite number of premodifiers, a head, and an indefinite number of postmodifiers. In practice, NP structures are restricted in their realizations. Table 3-1 (p. 57) gives the distribution of the prototypical NP structures that were found in Corpus A.9 The maximum number of modifiers that occurs is four: either two premodifiers and two postmodifiers (exs. 13-14); three premodifiers and one postmodifier (exs. 1516); or one premodifier and three postmodifiers (ex. 17). In addition to these four modifiers, the NP can have a limiter (ex. 18). (13) (14) (15) (16) (17) (18)
9
the [proposed] [mysterious] confrontation [with Dr Fell] [in the presence of his uncle] [bw-1334] the [historic] [opening] session [on Thursday night], [which both sides desperately wanted reported, as much in Europe and the United states as anywhere] [sp3-a0079] [great] [big] [life-size] photographs [of relatives] [sp2-070156] the [new] [post-revolutionary] [Russian] art, [heralded as the ‘new machine art’ by placards at the exhibition] [sp3-d0082] [special] mutants [of bacteriophage] [(called ‘ambre’ and ‘ochre’ mutants)] [in which premature termination of the synthesis of some proteins occurs] [cb-0408] almost the [only] [ordinary] [normal] person [in the place] [mr-0851]
For an explanation of the abbreviations used in Table 3.1 (p. 57) see Appendix C. For a full frequency distribution of the prototypical NP structures in Corpus A see Appendix E.
57
3.2 - Frequency and distribution of NPs Table 3-1: Prototypical NP structures in Corpus A # of prototypical NPs 30,243
% of the total # of NPs 55.47
DT NPHD
9,377
17.20
DT NPPR NPHD
3,484 414 27
6.39 0.76 0.05
3,155 222 14
5.79 0.41 0.03
1,264 136 7 91 1 10
2.32 0.25 0.01 0.17 0.00 0.02
105
0.19
28 7
0.05 0.01
43 4
0.08 0.01
LIM DT NPPR NPHD NPPO NPPO
17 1 1
0.03 0.00 0.00
LIM NPHD
99
0.18
LIM NPPR NPHD LIM NPPR NPPR NPHD
11 1
0.02 0.00
LIM NPHD NPPO
27
0.05
3
0.01
2,174 207 15
3.99 0.38 0.03
1,563 69 2
2.87 0.13 0.00
346 35 3 23 1
0.63 0.06 0.01 0.04 0.00
53,230
97.64
NPHD
DT NPPR NPPR NPHD DT NPPR NPPR NPPR NPHD DT NPHD NPPO DT NPHD NPPO NPPO DT NPHD NPPO NPPO NPPO DT NPPR NPHD NPPO DT NPPR NPPR NPHD NPPO DT NPPR NPPR NPPR NPHD NPPO DT NPPR NPHD NPPO NPPO DT NPPR NPHD NPPO NPPO NPPO DT NPPR NPPR NPHD NPPO NPPO LIM DT NPHD LIM DT NPPR NPHD LIM DT NPPR NPPR NPHD LIM DT NPHD NPPO LIM DT NPHD NPPO NPPO LIM DT NPPR NPHD NPPO LIM DT NPPR NPPR NPPR NPHD NPPO
LIM NPPR NPHD NPPO NPPR NPHD NPPR NPPR NPHD NPPR NPPR NPPR NPHD NPHD NPPO NPHD NPPO NPPO NPHD NPPO NPPO NPPO NPPR NPHD NPPO NPPR NPPR NPHD NPPO NPPR NPPR NPPR NPHD NPPO NPPR NPHD NPPO NPPO NPPR NPHD NPPO NPPO NPPO
Total prototypical NPs
58
Chapter 3 - Results of the corpus study
The NPs in Corpus A range from 1 to 57 words, with an average of 3.4 words. The sentence containing the longest NP occurs in the fiction sample ‘The Mind Readers’. It is given in (19). (19)
Once upon a time he had done a great deal of work for that curious Alicein-Wonderland body which is purely civilian and contrives to have no specific master, no powers, and no means of defending itself, save by adroit evasion, but which exists to protect the Realm within the Realm for just those precise periods when it can be shown to be in danger and for not one instant longer. [mr-1411]
Table 3-2 gives the distribution of NPs in corpus A.10 Of the NPs in Corpus A, 55.47% are realized by a head only (the class of simple NPs). This group is uninteresting for the present study, since it is not subject to the mobility phenomenon. The mobility of a constituent is expressed with regard to the position it takes in relation to the head of the noun phrase and/or other constituents (not necessarily immediate constituents of the NP). An NP should thus comprise at least two constituents in order to show mobility of a constituent. In the rest of this study only the group of complex NPs is taken into account.11 Table 3-2: Distribution of NPs in Corpus A # NPs Fiction Non-fiction Drama Spont. speech Scripted speech Total
18,824 9,064 6,494 9,721 10,414 54,517
% simple NPs 57.59 40.27 69.37 67.86 44.65 55.47
% complex NPs 42.40 59.74 30.63 32.14 55.35 44.53
% variant NPs 2.52 2.02 1.80 2.52 2.57 2.63
% NPs # NPs with with mobility mobility 2.14 402 1.65 150 1.72 112 1.93 188 2.05 214 1.96 1,066
The group of ‘variant NPs’ includes all NPs that do not conform to the prototypical description of the NP such as represented in Figure 3-1 (p. 52). This group comprises 1,287 NPs, 2.36% of the total number of NPs (or 5.3% of all complex NPs). NPs do not conform to the prototypical description when: a.
additional elements are introduced into the NP, such as e.g. an emphasizer (exs. 20-22), but also hesitation markers (exs. 23-25), interjections (ex. 26), and inserted remarks (exs. 27-28):
(20)
he himself
10
[bw-1834]
The distribution over genres is further discussed in section 3.4.1. The group of complex NPs consists of all NPs minus those realized by a head only (inclusive of all variant NPs). 11
3.2 - Frequency and distribution of NPs (21) (22) (23) (24) (25) (26) (27) (28)
59
the scientists themselves [sp3-a0023] the wind itself [sp3-a0028] your husband’s … ah … place of work [mr-0156] the uhm marches of well Brittany [sp1-250232] all the stages er of it being renovated [sp1-090080] the export of uh Edam cheese to Tanzania and Andorra yeah which wouldn’t actually be accounted for by normal sort of standards [sp1-610331] William – or was it John? – Halfpenny’s Rural Architecture in the Chinese Taste [bw-1230] those ‘dangerous reasoners’ (the phrase is Rousseau’s) who told them that man is no more than a machine [sp3-d0234]
b.
the NP includes a coordination of which the coordinated parts are not immediate constituents of the NP. In the prototypical description of the NP all functions can be realized by a coordination of categories. However, only phrases and/or clauses can be coordinated. That is, the coordinated parts have to be immediate constituents of the NP. Coordinations like the following are thus not accounted for in the prototypical description12:
(29) (30) (31) (32) (33) (34)
the pulses, the circulation of the blood [mr-1728] a wary as well as a very intelligent young man [bw-1455] this grease and stinking petrol [cc-0450] a vital hour or half hour [cc-1784] an alteration or a great diminution of it [cb-0346] a lot of grumbling and mumbling and threats of never again from the legislative council [sp3-a0457]
c.
a constituent occurs in a position different from the position it occupies in the prototypical description of the NP; it is ‘mobile’. These NPs are the subject of study in this research.
As was expected, the null-hypothesis can be rejected on the basis of the corpus data: NPs which do not conform to the prototypical NP description do occur. What is more, the majority of variant NPs can be explained by the mobility of constituents. What I investigate in this research is under what conditions the mobility of constituents in the English noun phrase occurs and how these can be described. For this purpose, I compare the realization and distribution of variant NPs with that of complex non-variant NPs. In chapter 1, I gave an overview of the types of variant NP that were found in the corpus and that were the result of the mobility of a constituent. In the rest of this chapter, each type of variant NP is discussed separately for its distribution and its syntactic characteristics.
12
See also section 1.3.5. On the whole, these unbalanced coordinations are no problem for the TOSCA-ICE analysis (see also Oostdijk, 1991).
60
Chapter 3 - Results of the corpus study
3.3 Types of variant NP In this section nine different types of variant NP will be discussed in detail: deferred modification, floating deferred modification, fronted modification, discontinuous modification, deferred determiners, floating deferred determiners, discontinuous determiners, deferred limiters, and floating deferred emphasizers. 3.3.1
The deferred modifier
In the prototypical noun phrase, modifying adjective phrases, adverb phrases and noun phrases precede the head of the noun phrase. However, occasionally these phrases are also found to follow the head. That is, they occur in postmodifying position. This occurred 468 times in Corpus A. In other words, in 5.29% of the NPs with one (or more) premodifier(s) a premodifier has been deferred. An adjective phrase, for instance, will occur in postmodifying position when the head of the NP is realized by a (non-)assertive or negative pronoun (exs. 35-37), when the adjective is itself postmodified (exs. 38-39), and/or when the adjective belongs to a limited number of items that in their bare form can occur both in premodifying and in postmodifying position (exs. 40-41).13 (35) (36) (37) (38) (39) (40) (41)
something extremely disturbing which came to my notice only today [bw-1384] anything valuable [mr-0801] nothing very amazing [mr0461] a soul distinct from our physical body [sp3-d0225a] a casualty as familiar and distinctive to the group in the Rectory as any gang of mods and rockers out for a bash [mr-1293] the DNA present [cb-0140] the money required (to ransom the boy) [cc-1731]
But there are also occurrences of deferred adjective phrases which are not so easily explained. In these cases the adjective phrase could also be placed in premodifying position, without changing the actual meaning of the noun phrase. For example: (42) (43) (44) 13
a circumstance wildly improbable an old friend just deceased metal more attractive
[bw-1386] [bw-1766] [cc-0150]
With this last class, there is often a difference in meaning between premodifying and postmodifying position. Compare for example: (i) the people concerned (ii) the concerned people
3.3 - Types of variant NP
61
For these last examples deferral of the AJP seems to give it extra prominence within the NP. Quirk et al. discuss the topic of information focus and state that the “neutral position of focus is what we may call end-focus, that is (generally speaking) chief prominence on the last open-class item or proper noun in the clause” (Quirk et al., 1972: 938). If we extend this notion of end-focus to the structure of the phrase, typical premodifying categories, such as the adjective phrase, may be deferred in order to receive end-focus.14 The general rule for adverb phrases seems to be that intensifying adverbs precede the head (exs. 45-47), while descriptive adverbs of time and place follow the head (exs. 48-50). However, when the head is realized by a (non-) assertive or negative pronoun, all adverbs, including intensifying adverbs, follow the head (exs. 51-52). The corpus contains one exception to the rule that a descriptive adverb of place follows the head. In example (53), above can occur in premodifying as well as in postmodifying position. (45) (46) (47) (48) (49) (50) (51) (52) (53)
the very beginning a little more the best loved a brisk step forward evenings here the day before someone else nothing at all the above example
[cb-0920] [bw-1051] [mr-0162] [mr-0225] [bw-0174] [cc-0532] [sp2-130085] [cc-0367] [cb-0244]
Similarly, noun phrases that denote time or place occur in postmodifying position (exs. 54-55). Other occurrences of postmodifying noun phrases are much like prepositional phrases, with the article a(n) functioning as preposition (in the meaning of per) (exs. 56-57). A third occurrence of postmodifying noun phrases could also be analysed as an instance of apposition (exs. 58-59). (54) (55) (56) (57) (58) (59) 3.3.2
the meeting this afternoon the place next door a quid an hour a day a week figure 8.12 the symbol ‘S’
[sp3-j0233] [sp2-070034] [cc-0485] [cc-0473] [cb-0401] [sp3-d0341]
The floating deferred modifier
A constituent is considered to be a floating deferred modifier if it clearly functions as a modifier to the noun phrase (head), follows the head, but is not 14
For a more elaborate discussion on possible motivations of mobility in the noun phrase see chapter 5.
62
Chapter 3 - Results of the corpus study
immediately adjacent to the head or other postmodifiers. In other words, a floating deferred modifier occurs outside the boundaries of the NP. Under this definition, the second or third postmodifier in a continuous sequence of postmodifiers is not analysed as a floating deferred modifier. Neither is a nonadjacent constituent which is not a modifier to the NP directly, but to an immediate constituent of the NP (such as the floating AJP postmodifier which will be discussed in section 3.3.4) analysed as a floating deferred modifier of the NP. The NP and its floating deferred modifier may be interrupted by immediate constituents of the sentence or clause in which they are embedded (exs. 60-62), or they may be interrupted by the insertion of metacomments (exs. 63-65).15 In one case the NP occurs as a discontinuous modifier in an adverb phrase (ex. 66). (60) (61) (62) (63) (64) (65) (66)
The question arises whether it is possible to combine the two theories. [cb-0760] Mitle, what happened today that I don’t know about. [si-2644] Right let’s see how many words we can think of beginning with D. [sp1-850254] Just one of the ups and downs, I suppose, of covering a revolution. [sp3-a0083] Once the possibility of extortion vanishes, it is a matter - one may say - of cut and run, … [cc-1793] … there was something about him - Appleby reflected, not for the first time - that didn’t quite cohere with the character. [bw-0101] With just a shade more of evidence to connect Punter with the crime, it might be possible to take him in and intimidate or bribe him into grassing on his pals. [cc-1798]
In corpus A, 168 NPs were found that have one floating deferrred modifier and four that have two floating deferred modifiers; the second directly following the first (exs. 67-68).16 In other words, in 2.39% of all NPs with one (or more) postmodifier(s) a postmodifier occurs floating. A floating deferred postmodifier can also co-occur with an adjacent postmodifier (exs. 69-70). The number of words that occur between the NP and its floating deferred modifier ranges from one to fifteen (with an average of 3.2). The number of constituents ranges from one to four. In 40.1% of the cases, the material between the NP and the floating deferred modifier consists of a single adverbial (exs. 71-72). In 15.1% it consists of a verb (exs. 62 and 73) and in 9.3% it consists of a combination of a verb and 15
See also Bunt (1996). Metacomments differ from inserted remarks (see examples (27) and (28), p. 59) in that inserted remarks clearly refer (back) to part of the NP, while metacomments are more general and have no particular reference to a part of the NP. 16 The corpus also includes one instance where the floating deferred modifier is realized by a coordination of two prepositional phrases (ex. i). This example is counted as an instance of an NP with one floating deferred modifier. (i) There was no fraternising obviously with female teachers, even with European teachers. [sp3-j0273]
3.3 - Types of variant NP
63
an adverbial (exs. 63 and 74). Subjects (ex. 64), subject complements (ex. 75), and direct objects (ex. 76) can also occur in between the NP and its floating deferred modifier (11.6%). In 25.6% of the cases the constituent that occurs between the NP and the floating deferred modifier is realized by an NP itself. Occasionally, this may ‘trigger’ the reader to assign the modifier to the last NP before connecting it correctly to the preceding NP. In the great majority of cases (78.5%) the floating deferred modifier occupies the last position in the sentence or clause in which it occurs. (67) (68)
(69) (70) (71) (72) (73) (74) (75) (76)
I mean there was a large picture of your mother’s mother wasn’t there [in a sort of wig] [looking as fierce as anything] [sp2-070163] Yet this is not an adequate explanation, for it would suggest the whole alliance (as I described it earlier) [between radical politics and radical aesthetics], [which constituted the brief but vigorous life of Berlin Dada], was merely the matter of expedience for the former. [sp3-d0129] At the last sitting Carson gave me, he said he’d had a cable from the lad a day or two before saying that he was more or less on his way. [cc-0871] His list of titles grew almost weekly, including a run of six tournaments in a row. [sp3-f0141] I had a notion last night that, if we hadn’t just been in a corner of the music room before bed time, and with people drifting around, Diana might have come out with something. [bw-0661] We should maybe just leave a message here saying head over [sp1-390161] If an exact match is not found for the term you enter, the system displays the portion of index list where the term would fit alphabetically.[cm-d026] … but evidence has also been adduced recently to suggest that they may have an effect on protein synthesis, through the genetic mechanism. [cb-0856] Nothing wrong with’em that hard work wouldn’t mend. [si-0477] Nothing woke me but the desire to see you again. [mr-1625]
Table 3-3 (p. 64) gives the realizations of the normal, adjacent NP postmodifier and of the floating deferred modifier as found in Corpus A.17 Here we see that exactly half of the 172 floating deferred modifiers are realized by a clause, while the others are realized by a phrase (or a combination of a phrase and a clause). If we compare the realizations of the floating deferred modifier with those of the adjacent noun phrase postmodifier, we see that the percentage of modifiers realized by a clause is much lower for the adjacent NP postmodifier (24.7%) than for the floating deferred modifier (50.0%). This indicates that clauses are more likely to occur as floating deferred modifier than phrases.18
17
In Table 3-3 I use the abbreviation PHR to refer to any kind of phrase (AJP, AVP, NP, or PP). For an explanation of the other abbreviations see Appendix C. 18 Pearson’s chi-square yields a significant score of 51.6 (DF=1). This is in accordance with explanations of the occurrence of floating deferred postmodification in terms of
64
Chapter 3 - Results of the corpus study
Also, the corpus yields no instances in which the floating deferred modifier is realized by an adverb phrase (AVP) or a noun phrase. Table 3-3: Realization of floating deferred modifiers as compared with adjacent postmodifiers in Corpus A Adjacent postm. # % AJP AVP NP PP CL PP + PP PP + CL CL + CL PHR + PHR PHR + CL PHR + PHR + PHR PHR + PHR + CL PHR + CL + CL COORD
164 156 96 4,770 1,885 182 206 22 42 22 14 10 3 54 7,626
2.15 2.04 1.26 62.55 24.72 2.39 2.70 0.29 0.55 0.29 0.18 0.13 0.04 0.71 100.00
Floating postm. # % 3 0 0 78 86 1 3 0 0 0 0 0 0 1 172
1.74 0.00 0.00 45.36 50.00 0.58 1.74 0.00 0.00 0.00 0.00 0.00 0.00 0.58 100.00
It is not always easy to distinguish between a noun phrase postmodifier and an adverbial at sentence or clause level. For a study of the floating deferred modifier the problem is twofold. Firstly, it may not be clear whether the floating clause or phrase itself is a modifier or an adverbial (exs. 77-78). And secondly, the NP head can be followed by two constituents of which only the second is a clear modifier. In this case, the analysis of the second constituent as either a floating deferred modifier or a regular postmodifier is dependent on the unclear status of the constituent closest to the head. If there are two or more phrases or clauses following the head of a noun phrase, the status of the one closest to the head may not be clear (exs. 79-80). (77) (78) (79)
The fighters looked a sorry sight with their suitcases and bedding rolls, their carrier-bags and kalashnikovs. [sp3-a0048] His list of titles grew almost weekly, including a run of six tournaments in a row. [sp3-f0141] For some time I’ve had it in mind to have a word with Judith about your Solo. [cc-0932]
syntactic ‘weight’. This will be further discussed in Chapter 5, section 5.5.3. All statistical calculations wer done by means of the statistical package SPSS version 8.0 for PC.
3.3 - Types of variant NP (80)
65
He revealed a charming deference towards her which transformed his peaky face. [mr-0191]
3.3.3 The fronted modifier In the prototypical description of the noun phrase, prepositional phrases and clauses follow the head of the NP, while adjective phrases, adverb phrases and noun phrases precede the head, but follow the determiner (if present). In the corpus, one occurrence was found in which a clause occurs in premodifying position (ex. 81). However, the clause could also be looked upon as a name or title for the icon (which the capital of Insert seems to suggest). In that case, Insert date/time as a whole can be analysed as a proper noun premodifying the noun icon, a prototypical structure. This analysis is further justified by the fact that, if the clause follows the head, the interpretation of the noun phrase is more ‘appositional’ (ex. 81’) and could be analysed in the same way as examples (60) and (61) above. (81) the Insert date/time icon (81’) the icon Insert date/time
[cm-l143]
Quirk et al. (1972, 1985) and Nida (1966) also mention sentences (clauses) as possible, although rare, premodifiers (exs. 82-83). In these examples, the hyphenation suggests that the sequences can be interpreted as fixed sequences which have lost the normal characteristics of a sentence/clause. Also, they cannot occur in postmodifying position. Such sequences can be analysed as single tokens. (82) (83)
his pop-down-for-the-weekend cottage a do-it-yourself job
In theory, we may also expect a prepositional phrase to be fronted. In practice, however, no occurrences were found of a PP in premodifying position. A second instance of fronted modification arises when a modifier is found to precede the determiner. We will refer to this type of fronted modification as ‘fronted premodification’ to distinguish it from the type where typical postmodifying categories occur in premodifying position. Aarts and Aarts (1988) say the following about fronted (or in their terminology ‘shifted’) premodification: Adjective phrases do not always follow items realizing the determiner function. This deviation from normal wordorder, which may be called ‘shifted premodification’, occurs in noun phrases containing the indefinite article as central determiner under either of the following conditions:
66
Chapter 3 - Results of the corpus study 1. 2.
the adjective phrase contains one of the following intensifying adverbs: as, so, how, however, ever so, that, this, too, enough, more and less; the head of the adjective phrase is in the comparative degree and preceded by no, much and far.
Examples: how strange a story however brave a soldier ever so slight a foreign accent too hot a day no worse a plan far cheaper a method (Aarts and Aarts, 1988: 109-110) In corpora A, seven occurrences of fronted premodification were found (of which three are also discontinuous). That is, in 0.08% of all NPs with a premodifier, the premodifier has been fronted. In corpus B, an additional ten occurrences were found. All fall under ‘condition one’ as formulated by Aarts and Aarts. They consist of an absolute adjective which is premodified by one of the intensifying adverbs so, as or too (exs. 84-86). No occurrences were found in which the adjective is premodified by how(ever), ever so, that, this, enough, more or less. Nor were occurrences found of a fronted adjective in the comparative degree (‘condition two’ in Aarts and Aarts). (84) (85) (86)
It wasn’t on so informal an occasion, and unmannerly thing to do. [cc-0880] the homeward girls seemed to be giving him as wide a birth as the crowds permitted [fhum03-0550] If the parasite cells have taken on too strong a hold in the brain, then new blood will be of no use at all [fhor02-1248]
Premodifying AJPs which occur in the prototypical position in the NP can also be modified by an additive (ex. 87), general (ex. 88) or particularizing adverb phrase (ex. 89). No such realizations were found in fronted position in the corpus. (87) (88) (89)
He just asked an equally stupid question. He wrote a surprisingly good play. That is a particularly nasty remark.
3.3.4 The discontinuous modifier Discontinuous modifiers are phrases of which the immediate constituents are not adjacent to each other. In the corpus, we only found discontinuous adjective phrases. Corpus A contains 68 discontinuous AJPs. This means that in 0.77% of all NPs with a premodifier, the modifier is discontinuous. Four different types of
3.3 - Types of variant NP
67
discontinuous AJP can be distinguished. In the most frequent type (67.65% of all discontinuous AJPs) the head (and, when present, the premodifier) of the AJP precedes the head of the NP, while the adjective phrase postmodifier follows the head of the NP. In section 3.3.1, we saw that an adjective phrase tends to occur in postmodifying position in the NP if the adjective is itself postmodified. In this case, part of the adjective phrase occurs in premodifying position in the NP, while the other part occurs in postmodifying position (possibly floating). The part which occurs in postmodifying position cannot occur independently of the part which stands in premodifying position. As Oostdijk and Aarts (1997) point out, the first part ‘triggers’ the second part. The second part is either triggered by the head of the AJP (exs. 90-91) or alternatively by the adjective phrase premodifier (exs. 92-93). Of this type of discontinuous adjective phrase 46 occurrences were found in Corpus A. (90) (91) (92) (93)
Martine was a very different sort of person from her aunt. [bw-0205] They are there because it’s an easy thing to do, … [sp3-j0507] With maybe the most beautiful balls in the world. [fpsy02-0412] But I had a sense that a more complex piece of playacting than that was going on. [cc-0977]
Alternatively, the premodifier plus head of the AJP can precede the determiner of the NP, while the adjective phrase postmodifier follows the head of the NP. Corpus A contains three such structures, in all cases the AJP postmodifier is triggered by the AJP premodifier (exs. 94-95). Although not found in Corpus A, constructions in which the postmodifier is triggered by the head of the AJP are also possible (ex. 96).19 (94) (95) (96)
… - she was yet so nice a specimen of her particular world that one would have felt wholly churlish in thinking to scratch a lacquer so elaborately contrived for one’s delectation. [bw-0740] The flattery was hardly subtle and Lampart was too clever a man to miss it. [fcri03-0460] …, but it was so different a procedure from the rest of the operation … [BNC-FAH-83]
Table 3-4 (p. 68) gives the frequency of occurrence of deferred continuous AJPs which contain an AJP postmodifier and the first two types of discontinuous AJP in Corpus A. The realizations of the pre- and postmodifiers in those AJPs are specified. In general, deferred continuous AJPs containing a postmodifier (107 occurrences) occur slightly more frequently as modifier in the NP than the first two types of discontinuous AJPs which also contain an AJP postmodifier (62 occurrences). We also find that adjective phrases of which the postmodifier is realized by a clause (cf. ex. 97) are more apt to occur in their discontinuous version than those whose postmodifier is realized by a phrase (ex. 98) and that 19
When an example is taken from the BNC the code between dashes is a sample reference code from the BNC. For an explanation see Burnard (1995).
68
Chapter 3 - Results of the corpus study
discontinuous AJPs tend to have a premodifier which is realized by an adverb phrase. AJPs with multiple postmodifiers are not found in discontinuous form, but are not very frequent in their regular form either. Oostdijk and Aarts do give an example of multiple ‘triggered’ postmodifiers (ex. 99). (97) (98) (99)
This is a particularly complicated benefit to calculate, … [sp3-f0031] We are a larger party than yesterday, and this evening we must be quite gay. [bw-1156] the most unsuitable woman in the world for him
Table 3-4: Realizations of the postmodifier and premodifier in discontinuous and continuous AJPs which are found as modifiers in the NP in Corpus A. AJP postm. CL PP AVP NP PP + PP PP + CL total
Cont. AJP # % 16 15.0 79 73.8 7 6.5 1 0.9 2 1.9 2 1.9 107 100.0
Disc. AJP # % 26 53.1 20 40.8 3 6.1 0 0.0 0 0.0 0 0.0 49 100.0
AJP prem. no prem. AVP NP
Cont. AJP # % 71 66.4 35 32.7 1 0.9
Disc. AJP # % 19 38.8 30 61.2 0 0.0
total
107 100.0
49 100.0
It is possible for the adjective phrase postmodifier to occupy a position in the sentence or clause which is not directly adjacent to the other immediate constituents of the NP (exs. 100-103). In most cases, the postmodifier has floated towards the end of the sentence/clause, whereas in one case it has floated towards the beginning (ex. 102). In two cases, the floating of the adjective phrase postmodifier is the cause of discontinuity of the AJP (exs. 102-103). In their continuous form, these would be cases of deferred modification.20 The last two examples also stand out in that the head is realized by a non-assertive and an assertive pronoun respectively. In all other occurrences of a discontinuous AJP the head of the noun phrase is realized by a common noun, either singular or plural, or by the proform one. (100) No better summary can be provided than that of A.D. Neale, who observed: … [neco02-0219] (101) If however women want slightly different characteristics of men than men want of women, the pressures on the two sexes will be different. [nwom02-0836] (102) Than these last five words he positively felt that he had never heard anything odder in his life. [cc-2214] (103) But there was something more positive to it than that. [cc-0570] 20
See also footnote 23 in chapter 1.
3.3 - Types of variant NP
69
A third type of discontinuous modification does not involve an adjective phrase postmodifier. In this type, the AJP premodifier precedes the determiner of the NP, while the head of the AJP follows the determiner in normal premodifying position in the NP. In Corpus A, 19 such discontinuous modifiers were found (exs. 104105). (104) Well, she sometimes teases Bobby Angrave quite effectively, if in rather a childish way. [bw-0653] (105) Well, I suppose, dressing was quite an important factor. [SP2-610348] This type can also be combined with the first type in which case we get a discontinuous modifier in three parts. The AJP premodifier precedes the determiner, the head of the AJP occurs in regular NP premodifying position, and the AJP postmodifier follows the head of the NP (exs. 106-107). However, in the corpus no such occurrences were found. (106) “Where is he now, Mum?” he said one night when Mrs Hooper seemed in rather a better mood than usual. [BNC-A1C 224] (107) …; and she eventually became quite a pleasant horse to ride. [BNC-ADF 1596] In all occurrences of fronted premodification the (central) determiner was realized by an indefinite article and in relation, the head was always realized by a singular common noun (except for two occurrences of the proform one). Overall, the structure of NPs with a fronted premodifier is relatively simple. The fronted premodifier is never found to co-occur with a limiter or with another premodifier (in prototypical position). Apart from the modifying AJP no other possible premodifiers (adverb phrase, noun phrase) are found in fronted position. 3.3.5
The (floating) deferred determiner
In some cases, words which typically function as determiner in the noun phrase follow the head of the NP instead of preceding it. In the literature, this phenomenon is also known under the name quantifier floating or quantifier stranding.21 In English, items which can occur as deferred determiner include the universal pronouns both, all, and each and the quantifying pronoun enough. A deferred determiner either occurs in the final position in the NP, or it floats away from the NP toward the end of the sentence, in which case it is non-adjacent to the constituents of the NP to which it belongs (floating deferred determiner). In Corpus A, 20 occurrences were found of deferred both of which seven are nonadjacent to the NP (ex. 108). Of deferred all 119 occurrences were found, of which 58 are non-adjacent (exs. 109-110). Only one occurrence was found of deferred enough (ex. 111). The 140 NPs with a deferred determiner constitute 0.75% of all NPs with a determiner. Surprisingly, no occurrences were found of 21
See for example Sportiche (1988). For a more elaborate discussion of the literature on quantifier floating/stranding see chapter 5.
70
Chapter 3 - Results of the corpus study
deferred each, although a quick search in the British National Corpus yields both adjacent and non-adjacent occurrences of deferred each (exs. 112-113). (108) They were both frightfully rich, although hardly more than twenty-one. [bw-0169] (109) Well, bless ‘em all, I say [nc-0717] (110) ‘You must all read this,’ she said. [mr-1816] (111) Queer cattle, times enough, Sir John. [cc-0687] (112) they each said everything was fine [BNC-AB5-1609] (113) we must pay £2 each to the National Trust [BNC-ARE-157] Deferred determiners typically occur when the NP consists of a head only, which is realized by a personal pronoun (132 occurrences, 94.29% of all cases). In these cases the deferred determiner cannot be placed in normal determiner position, since this would cause an ungrammatical construct (ex. 114). However, the deferred determiner can also belong to a more complex noun phrase or demonstrative pronoun (8 occurrences, 5.71% of all cases). In those cases the deferred determiner can be placed in normal determiner position (exs. 115-116). (114) *That should cover both us. (115) The groups of workers employing genetic and biochemical methods both predicted at an early stage that the code would prove to be degenerate, … [cb-0385] (116) This is all very interesting. [nc-1035] A floating deferred determiner is never separated far from the rest of the NP. In all but two instances the determiner was separated from the NP by only one verb, either lexical or auxiliary. The two deviations are given in examples (117) and (118). They are both examples of prepared speech. In example (118) it is not quite clear whether all is an instance of a deferred determiner, or whether it functions as the subject of a verbless clause (all were rather nervous). (117) These, of course, all give exact sexagesimal fractions. (118) There were three of us in the car, all rather nervous. 3.3.6
[sp3-d0315] [sp3-a0129]
The discontinuous determiner
In Corpus A, 44 occurrences of a discontinuous determiner were found. This means that in 0.24% of NPs with a determiner, the determiner is discontinuous. In 18 cases the discontinuity is caused by the floating of the determiner phrase postmodifier to a position after the head of the NP. The floating determiner phrase postmodifier can either follow (ex. 119) or precede (ex. 120) the NP postmodifier(s), if present. It can also be non-adjacent to the other constituents of the NP further towards the end of the sentence (ex. 121). In all cases, the floating of the postmodifier is obligatory. The constituent is recognized as a determiner phrase postmodifier by the fact that its existence is licensed by either the postdeterminer (exs. 120-121) or the determiner phrase premodifier (ex. 122).
3.3 - Types of variant NP
71
(119) Carl Carson was no doubt a somewhat egocentric chap, inclined to believe that his affairs made more impact on others than in fact they did. [cc-0227] (120) While all this had been going on, Ian Botham, England’s star allrounder, had been out of the side, until the last test of the season, against New Zealand at the Oval; [sp3-f0124] (121) I’ve more money in this town than most, and money talks. [si-0063] (122) He refused politely and set about taking his leave with as much grace as possible. [mr-0159] In an additional 19 instances, an adjective occurs between the central determiner and the postdeterminer (realized by a numeral). In the structure of the determiner phrase (as described in chapter 1, section 1.2.1) no position is available to account for these occurrences. Some of the occurrences can be accounted for by taking the postdeterminer and the head of the NP together as a compound noun realizing the function of head. The adjective can then simply be regarded as a normal NP premodifier (exs. 123-124). For four of the 17 cases it seems more expedient to analyse the adjective as a (deferred) determiner phrase premodifier, since it modifies only the postdeterminer and not the rest of the NP (exs. 125-126).22 But for the majority of cases, the determiner sequence is interrupted by a noun phrase premodifier occurring between the pre- or central determiner and the postdeterminer (exs. 127-128). In these cases the order of the postdeterminer and the adjective can also be reversed. (123) … this includes an abiding Israeli affection for the dollar - the unofficial first currency here. [sp3-a0310] (124) Many thought Norman’s course-equalling second round 63 was the turning point. [sp3-f0150] (125) … - the Martineaus, like many prosperous people, having been for a good many generations inclined to acquire costly and handsome objects … [bw-0179] (126) …, he may find it convenient to keep a mere few thousand pounds in the form of cash in hand. [cc-0986] (127) The other two genes are the z gene or structural gene for b-galactosidase, and the i gene or regulator gene. [cb-0070] (128) …, and that is about two dozen times in the past two years, … [sp3-a0319] The last 11 instances of the discontinuous determiner are very similar to the (floating) deferred determiners discussed in the previous section. In these cases a predeterminer (all or both) follows the head of the noun phrase (possibly floating), while the rest of the determiner phrase precedes the head of the NP (ex. 129).
22
These four cases are not included in the class of discontinuous determiners.
72
Chapter 3 - Results of the corpus study
(129) It was not until the front door had closed behind him and the others were all in the room that he was mentioned at all. [mr-0171] 3.3.7
The deferred limiter
As we saw in chapter 1, section 1.3.1, a limiter is an adverb phrase which occurs NP initially and which restricts the reference of the NP. A deferred limiter follows the head of the noun phrase, either preceding (ex. 130) or following (ex. 131) any noun phrase postmodifier(s), if present. In corpus A, 65 deferred limiters occurred, 15.78% of all NPs with a limiter. Postponement of the limiter is either optional or compulsory. Some lexical items that function as limiters can only follow the head. Examples are at all, whatever and whatsoever (exs. 132134).23 These three limit negative noun phrases and are the most frequent among the deferred limiters (41.5% of all deferred limiters). Other limiters that follow the head are alone and as well (exs. 135-136). Some examples of optional postponement are given in (130) and (137). In these cases the limiter can also be placed in the first position of the NP, without changing the meaning of the NP. The limiter does seem to receive more stress or prominence when it follows the head. (130) … - although Appleby was inclined to think that this notion was the issue only of his own sustained professional acquintance with human misconduct. [bw-1336] (131) As Neugebauer remarks, this system of tables alone would put the Babylonians ahead of all numerical computers in antiquity. [sp3-d0318] (132) Now there’s no flash at all. [mr-0343] (133) Nothing whatever, Bobby. [bw-1629] (134) He’s got no authority whatsoever over you. [nc-2639] (135) For instance, kidney and liver tissue can be made to disaggregate by this means alone. [cb-0751] (136) Of course we’ll have to ask him as well. [cc-0054] (137) …, which might have captured the imagination of some, at least, of the dissillusioned South Westers, … [sp3-a0356] 3.3.8
The floating emphasizer
Emphasizers are optional pronouns that have no other function than to place extra emphasis on a nominal element with which it co-occurs. An emphasizer can either be adjacent to the other constituents of the NP to which it belongs, in which case it occupies NP final position (exs. 138-139), or it can be non-adjacent (exs. 140-141). These last are examples of floating emphasizers. In corpus A, 125 NPs contain an emphasizer, of which 60 are floating (48%). In all cases, the emphasizer occurs further towards the end of the sentence, not necessarily at the same level of the rest of the NP. The floating emphasizer frequently occurs inside 23
I analyse combinations like at all, in particular, as well, at least, etc. as compound adverbs.
3.3 - Types of variant NP
73
the following verb phrase (ex. 140), but it can also occur inside other immediate constituents of the sentence or clause (ex. 141). (138) He was displeased - and, like Mrs Gillingham herself, he was puzzled. [bw-0939] (139) It was a filthy business, as chill and stinking as dry rot itself. [mr-1582] (140) The ‘improvements’ about which the young husband and wife had talked in the little belvedere long ago must themselves have been of a conservative order. [bw-0220] (141) You must know that, Judith, being yourself connected with it as you are. [bw-1137] For restrictions on the mobility of emphasizers the relation of co-reference between the NP and the emphatic pronoun seems to play an important role. The material that occurs between the two can contain other noun phrases as long as the relation of co-reference is not disturbed. In example (142), for instance, the occurrence of me is unproblematic, since himself can only be co-referential with he. In general, the reader tries to attach the emphasizer to the nearest possible (usually) preceding noun phrase. Given a certain context, this may lead to an ambiguous interpretation. In example (143), the nearest possible NP to which himself can be attached is Mr Mayo. The context, however, suggests that himself is co-referential with the boy (Edward), who, if woken, could himself give his treasures to Mr Mayo. (142) However, he told me himself or I shouldn’t have believed it. [mr-0149] (143) ‘Of course, I could have insisted that we woke Edward and that the boy gave his treasures to Mr Mayo himself,’ he said at last. [mr-1395] For all occurrences in the corpus, the material between the NP and the floating emphasizer contains a full verb phrase or part of a verb phrase (the first auxiliary). However, this may also be due to the fact that NPs with a floating emphasizer always function as the subject of the sentence or clause and are thus most frequently directly followed by a verb phrase. Indeed, in a question, the emphasizer can be separated from the noun phrase without a verb phrase occurring in between (ex. 144). Quirk et al. (1985: 359) state that the combination of a personal pronoun and an emphatic pronoun is avoided outside the subject position. This is indeed supported by the findings in corpus A. However, when the head is realized by a common or proper noun, an NP with emphasizer can occur in other sentence and phrase functions as well (exs. 145-147). But even for NPs with a common or proper noun as head, it seems impossible (or at least strange) to have a floating emphasizer outside the subject function (exs. 148-149). (144) Are you in the mood for ice-cream yourself? (145) You can ask the young man himself, when he finally turns up and is introduced to us. [cc-0083]
74
Chapter 3 - Results of the corpus study
(146) I myself am just waiting for Tommy to turn up a little more evidence before I tackle Carson himself and try to persuade him to cooperate. [cc-1748] (147) Visibility was down to zero at Oxford Circus itself, … [sp3-b0018] (148) a. ? You can ask the young man, when the car finally turns up, himself. b. You can ask the young man, when the car finally turns up, yourself. (149) a. ? I hit Carson in the face himself. b. I hit Carson in the face myself. Quirk et al. (1985: 361) also suggest that it is possible for an emphasizer to precede the NP to which it belongs. Such occurrences were not found in the corpus. However, a quick search in the British National Corpus does provide some examples (exs. 150-152). (150) Felix said, ‘If there is too much air, close the window; myself I like to feel it blowing in.’ [BNC-CB7-766] (151) They’re never away from each other, erm whereas myself I was always out. [BNC-HEN-53] (152) There were two of them, and themselves they looked young and harmless … [BNC-CE5-2332] 3.3.9
Such
In the previous sections I have discussed nine types of variant NP. The nine types were distinguished by the type of mobility (fronted, deferred, floating, and/or discontinuous) and the function to which it applies (limiter, determiner, modifier, emphasizer). So far, the discussions on fronted premodification and discontinuous modification have known an important omission. Nothing was said about constructions which involve the lexical item such, although it can clearly play a role in discontinuous structures (exs. 153-154), or resemble the fronted premodifier (exs. 155-156). (153) Nevertheless, the early atmosphere was probably still strongly reducing, and contained such gases as methane, carbon dioxide, hydrogen, ... [cb-0863] (154) …, but most Catholic scholars accepted a system of Christian apologetics which incorporated such Cartesian notions as innate ideas and the dualism of mind and body. [sp3-d0153] (155) It would have been such an affront to Grace that he would never forgive the boy. [bw-1012] (156) Thank you for being such a comfort, Mitle. [si-2642] Let us first consider examples of such occurring before the determiner as possible fronted premodifiers. Aarts and Aarts do not include these occurrences in their description of fronted premodification. Instead, they treat such as a predeterminer in these cases. However, they also observe that “the status of the item such is problematic” (Aarts and Aarts, 1988: 108). Or as Altenberg observes:
3.3 - Types of variant NP
75
The analysis of such is obviously complicated by conflicting semantic and syntactic criteria and its word-class status is far from clear-cut. (Altenberg, 1994: 226) In recent grammars such is mostly treated as a predeterminer (like all, both and half), although its functional similarity to the degree adverbs quite and rather is also mentioned.24 On the basis of a corpus-based study of such, Altenberg suggests that a categorial distinction should be made between identifying and intensifying such: while identifying such has the properties of an indefinite demonstrative, intensifying such behaves like a modifying adverb.25 Both can operate in the predeterminer position, but only identifying such can act as a postdeterminer and occur with quantifying determiners. The interpretation of such is dependent on the presence or absence of two features: a defining referent in the context and a gradable element in the noun phrase.26 When there is a defining referent but no gradable element, such is interpreted as identifying (exs. 157-158). When there is no defining referent, but a gradable element does occur in the NP, such is interpreted as intensifying (exs. 159-160). When there is both a defining referent in the context and the noun phrase contains a gradable element, the interpretation is ambiguous (ex. 161).27 (157) And Fell has been and gone. He has discharged himself from the case.’ ‘Charles, pull yourself together. What you say is impossible. No doctor could do such a thing.’ [bw-1253/57] (158) These fields can contain all kinds of elements going beyond pure text: […] How does the Workbench treat such subsegments? [cm-t127/8] (159) I’ll not ask you in, seeing it’s such a lovely day. [si-0334] (160) You’re such a snob honestly [sp2-070178] (161) Luke’s love affair […] was the only subject upon which the two friends had never achieved complete understanding. […] A wave of irritation at himself for choosing such an unfortunate subject had passed over Campion. [mr-0548/50] Identifying such can have both anaphoric and cataphoric reference. When such has anaphoric reference, it points back in the discourse. It can either refer back to a previous noun phrase (explicit reference, ex. 161) or to a ‘situation’ as sketched in the preceding context (implicit reference, ex. 157). When such has 24
Quirk et al. (1972, 1985), Aarts and Aarts (1988), Greenbaum and Quirk (1990). The distinction between identifying such and intensifying such is also made by Bolinger (1972). In his definition identifying such refers to something ‘of X identity’, while intensifying such (additionally) refers to something ‘of X magnitude’. 26 Like demonstratives, such ‘identifies a noun phrase with a specific referent in the linguistic or situational context’ (Altenberg, 1994: 229). This specific referent is referred to as the ‘defining referent’. 27 Note that while in theory both features might be absent, this is in fact impossible since the head noun can always function as the defining referent in the NP. 25
76
Chapter 3 - Results of the corpus study
cataphoric reference, it points forward in the discourse to information expressed in a following as (to)-clause or that-clause (exs. 162-163). Intensifying such can also have cataphoric reference (ex. 164). (162) It appears to be the 60 S components of the ribosomes which associate with the membrane of the reticulum in such a way that the 40 S particles protrude into the cell sap. [cb-0480] (163) Had Appleby felt himself to be in charge, he could no doubt have assembled in ten minutes such preliminary facts as there were. [bw-1841] (164) Quite suddenly, there’s such pressure on one from this and that that one hardly knows how to find the time for it all. [cc-0243] Although I agree with Altenberg that a distinction should be made between identifying and intensifying such, I do not agree with his conclusion that identifying such is a predeterminer. Altenberg himself already notes that identifying such is an interesting determiner in that it is the only determiner that can be both predeterminer (e.g. such a thing) and postdeterminer (e.g. no such thing). This complicates the analysis of such in cases like such things all such things Do we analyse such here as a predeterminer (occurring before an indefinite ‘zero’ article) or as a postdeterminer (occurring after ‘zero’)? (Altenberg, 1994: 231) Altenberg decides to “follow Quirk et al. (1985: 258) and classify such in both these cases as a predeterminer, thus recognizing all such things as an exception to the rule that predeterminers are mutually exclusive”. Although such always follows a quantifying central determiner (no/some/any/each such thing) or postdeterminer (two/many/several such things), Altenberg rejects the fact that it can also be analysed as an adjective because it can precede and co-occur with other which he considers to be a postdeterminer (ex. 165). As Altenberg points out, identifying such establishes ‘comparative’ reference since it can be replaced by the paraphrases like that or of that kind, and as Quirk et al. (1985: 1136) point out, other can also be used as a comparative item (ex. 166). Comparative other differs from other postdeterminers in that it can be postponed; it shares this characteristic with identifying such and other adjectives when followed by a postmodifying clause. I want to argue that on these grounds such and other should not be analysed as postdeterminers, but as adjectives functioning as premodifier in the NP.
3.3 - Types of variant NP
77
(165) The relationship between education and earnings, however, is complicated by the influence of such other factors as intelligence and social class, and such personal qualities as motivation and perseverance. [BNC-FR4-149] (166) a. I don’t have any other cups than those I have in the sink. b. I don’t have any cups other than those I have in the sink. Now that we have established that identifying such, when it occurs after a central determiner, should be analysed as a premodifying adjective rather than as a postdeterminer, we should reconsider its status of determiner in all other positions as well. Regarding identifying such as an adjective in all cases has some clear advantages. The analysis of such things and all such things is now no longer ambiguous. Furthermore, it explains the fact that, like other adjectives, such can occur with a non-adjacent postmodifier (i.e. such can be the head of a discontinuous AJP functioning as modifier in the NP). Also such can (only) be postponed if an as (to)- or that-clause follows it. This clause then is not a noun phrase postmodifier, but part of a discontinuous adjective phrase. And the result of the postponement of such is not a multiword subordinator, but simply an adjective phrase consisting of a head plus postmodifier (exs. 167-168). (167) a. … there is no such thing as a reasonably reticent wire to that island, … [mr-1756] b. … there is no thing such as a reasonably reticent wire to that island, … (168) a. These are coded in such a way as to highlight aspects which suggest that further investigation is called for. [neco02-1026] b. These are coded in a way such as to highlight aspects which suggest… Like other adjectives, identifying such can also be fronted (exs. 157, 162). The examples differ from other occurrences of fronted premodification by the fact that the adjective is not preceded by an intensifying adverb. Not just any intensifying adverb qualifies to fill the function slot of ‘adjective phrase premodifier’ in a fronted AJP; it can only be one of the adverbs as, so, how, however, ever so, that, this, too, enough, more, and less.28 These adverbs add a degree of (relative) comparativeness to the adjective. As pointed out above, identifying such also establishes ‘comparative reference’. However, other adjectives which occur in the comparative degree are preceded by no, much, or far when they are fronted. Why is this not the case for such? What all the adverbs mentioned here have in common is that they are intensifying (except maybe no). Perhaps identifying such also has the characteristics of an intensifier. Indeed, this possibility is supported by the fact that in some cases ambiguity may arise between an identifying and an intensifying interpretation of such (ex. 161). All this suggests that, in order for an adjective to become fronted, it must be both comparative and intensifying. These features can be inherent in the adjective (as with such) or they can be added by means of a modifying adverb.
28
See also section 3.3.3.
78
Chapter 3 - Results of the corpus study
Let us now get back to intensifying such. Altenberg remarks that “unlike identifying such, intensifying such only occurs in the predeterminer slot in the noun phrase”. If it co-occurs with a determiner, this determiner is always realized by an indefinite article and such precedes the article. Intensifying such can also occur in an NP without a determiner. In that case such directly precedes the gradable element, which it seems to modify. What is more, when intensifying such has cataphoric reference, it cannot be postponed in order to form a continuous modifier of the NP together with the as (to)- or that-clause.29 Intensifying such is closely linked with the intensifying adverb so. Often, when intensifying so occurs in a premodifying AJP, a paraphrase with intensifying such is possible, and vice versa. If cataphoric such occurs with a gradable adjective, it is even possible to postpone the adjective by replacing such by so (ex. 169). (169) a. Dad was making such a rude joke that he shocked everyone present. b. Dad was making a joke so rude that he shocked everyone present. When intensifying such precedes the central determiner (such a hot day), it can, in theory, either be analysed as a determiner phrase premodifier (DTPR), as a limiter (LIM), or as the first part of a discontinuous AJP. Since such does not modify the indefinite article, the DTPR option is invalid. If the gradable element with which such co-occurs is the head of the NP, it cannot be part of a discontinuous AJP either. In these cases such functions as a limiter in the NP (ex. 170). However, if the gradable element is a modifying adjective, an analysis as part of a discontinuous modifier seems the most expedient, since such then clearly functions as a modifier to the head of the AJP (exs. 171-172). Indeed, intensifying such resembles other adverbs that can either occur as limiter in the NP or as part of a discontinuous AJP. Compare: (170) such a snob (171) such a small house (172) such a rude joke that it shocked me
rather a bore quite a large apartment almost an impossible task to perform
To sum up then, identifying such is analysed as an adjective and can function as the head of a fronted premodifier, but also as the head of a discontinuous AJP. Intensifying such, on the other hand, is analysed as an (intensifying) adverb and can function as a limiter in the NP or as a modifier in an (NP premodifying) AJP. In this last function it can be part of a discontinuous AJP in the NP. Including the structures with such in the findings for Corpus A gives 23 extra occurrences of a discontinuous AJP, 18 of a fronted modifier, and 4 of both a fronted and discontinuous AJP. 29
Although it may be possible to postpone intensifying such from a purely syntactic point of view, once postponed it will lose its intensifying characteristic.
3.4 - The distribution of variant NPs
79
3.4 The distribution of variant NPs In section 3.3 all types of NP have been discussed in which an immediate constituent or part of an immediate constituent occurs in a position that is different from the one it occupies in the prototypical NP structure. The realizations of the mobile constituents were given and, where appropriate, other characteristics of the NP structure or, in the case of floating constituents, of the interrupting material were also discussed. In this section, the distribution of the variant NPs will be discussed. 3.4.1
The distribution over genres
Table 3-5 (p. 80) gives the distribution of these types of NP over the genres distinguished in Corpus A. In chapter 2, section 2.5.1, the five genres of corpus A were introduced. It was mentioned that the genres can be put on a scale of formality with the non-fiction samples being the most formal and the unedited, spontaneously spoken samples the least formal. Interestingly, the distribution of NPs with a mobile constituent follows this formality scale, with more occurrences in the least formal genres and fewer occurrences in the formal genres. Can we establish on the basis of these findings that a dependency relation exists between the genre and the occurrence of variant NPs? We can use a chi-square test to determine whether the two variables are (in)dependent of each other, using the number of complex NPs as an estimate for the expected number of variant NPs. The chi-square test yields a significant score of 75.0 (DF=4).30 Significantly fewer variant NPs are found in non-fiction and scripted speech (standardized residuals are -5.7 and -2.5 respectively) and significantly more in fiction, drama and spontaneous speech (standardized residuals are 2.7, 2.6 and 4.3 respectively).31 However, we have to ask the question whether the significant relation between genre and the occurrence of variant NPs is real, or whether it is caused by the high number of complex NPs in the cells. If the number of complex NPs is so high that the sample approaches population measures, the chi-square test is no longer reliable as an estimate of population properties. We can determine the validity of the chi-square test in this case by calculating the contingency coefficient for the two variables, which is an indication of the strength of the relation between the two variables. The value for the contingency coefficient is 0.056 indicating that the relation between the two variables is indeed weak and that the significant scores in the chi-square test are most probably caused by the high number of complex NPs. We can conclude that the relation between genre 30
As a value for the chi-square I took Pearson’s chi-square. If the value for the standardized residual is lower than -2.0 or higher than 2.0, the observed frequency deviates significantly from the expected frequency. 31
80
Chapter 3 - Results of the corpus study
and the occurrence of variant NPs is so weak that we do not have to include the genre distinction in every other (quantitative) research aspect. Table 3-5: Distribution of NPs with a mobile (part of a) constituent in Corpus A Spontan. Drama Fiction Scripted Non- Total Speech speech fiction 468 Deferred modifier 92 38 168 90 80 169 Floating deferred modifier 30 12 54 42 31 22 Fronted modifier 4 1 14 0 3 88 Discontinuous modifier 10 5 34 25 14 75 Deferred determiner 21 15 29 8 2 65 Floating deferred determiner 18 18 21 8 0 42 Discontinuous determiner 2 4 9 20 7 64 Deferred limiter 8 12 26 12 6 60 Floating emphasizer 1 6 41 6 6 7 Fronted + discont. modifier 1 0 5 0 1 1 Float. mod. + disc. mod. 0 0 0 1 0 1 Disc. mod + def. limiter 0 1 0 0 0 2 Def. mod.+ float. mod 1 0 1 0 0 2 Def. mod.+ disc. determiner 0 0 0 2 0 Total 188 112 402 214 150 1066 % of the total # of complex NPs 6.01 5.63 5.04 3.71 2.77 4.39
3.4.2
The distribution over functions
While prototypical complex NPs have a wide distribution over sentence/clause and phrase functions, they typically realize the functions of subject (20.1%), direct object (18.6%), and prepositional complement (39.1%). If we compare the distribution of the variant NP types with those of prototypical NPs, we find that in general variant NPs share this distribution, although they occur more frequently in the function of subject than prototypical complex NPs (36.5%) and less frequently as a prepositional complement (24.6%). Table 3-6 (p. 81) gives the distribution of NPs over functions.32 The variant NPs are distinguished according to the type of mobility (fronted, deferred, floating, discontinuous). When compared to the distribution of complex NPs, we find the least variance in distribution for NPs with a deferred constituent. NPs containing a fronted premodifier have a preference for the direct object position and occur relatively frequently as subject complement, while they never occur as subject.33 NPs with a discontinuous constituent occur relatively 32
A frequency presented in bold deviates significantly from the frequency for complex NPs given in the last column. 33 In Corpus B, however, occurrences of NPs with fronted premodifier functioning as subject are found.
3.5 - Conclusion
81
frequently as subject complement and relatively infrequently as prepositional complement. NPs with a floating modifier on the other hand, show a very different distribution. They have a clear preference for the subject position (and sentence/clause functions in general) and occur relatively infrequently as a prepositional complement. This corresponds with the findings of Aarts (1971) that simple noun phrases are favoured in subject function, since the structure of the part remaining in subject position is relatively more simple when the postmodifier occurs outside NP boundaries. Table 3-6: Distribution of NPs over sentence/clause and phrase functions in corpus A fronted deferred floating discont. total complex n=29 n=612 n=297 n=134 n=1072 NPs A 0.0 2.4 0.0 0.7 1.5 1.7 APPOS 0.0 2.6 1.0 3.5 2.2 2.8 CJ 0.0 3.7 1.3 2.1 2.9 4.8 CO 0.0 0.3 0.0 0.0 0.2 0.2 CS 6.8 3.6 7.6 6.6 18.2 24.3 OD 18.6 14.5 22.9 18.4 18.6 40.9 OI 0.0 0.0 0.0 0.0 0.0 0.1 PC 40.9 39.1 31.4 6.9 28.5 24.6 SA 0.0 0.0 0.0 0.0 0.0 0.1 SU 0.0 24.8 16.0 20.1 70.7 36.5 other phrase func. 0.0 1.5 1.0 0.0 1.2 2.5 other S/CL func. 0.0 7.8 1.0 2.1 4.9 3.4 Total 100.0 100.0 100.0 100.0 100.0 100.0 NPs with a fronted, floating and/or discontinuous constituent all have in common that no occurrences are found in the corpus in which they function in the sentence or clause as indirect object, object complement, or subject attribute. However, considering the low frequencies of the prototypical complex NPs in these functions, it is hardly surprising that we found no occurrences of variant NPs in the corpus occupying these functions. In such cases, the elicitation experiment can be used to help decide whether the non-occurrences are most likely due to low frequency, or whether they are actual (syntactic) impossibilities. 3.5 Conclusion In this chapter, the results of a quantitative and a qualitative corpus study of the English NP were discussed, with special emphasis on the occurrence of variant NPs. The purpose of the corpus study was to make an inventory of the frequency and distribution of NPs in contemporary British English in order to arrive at a detailed description of NP structures, especially for those NPs that deviate from the prototypical structure through the mobility of an immediate constituent or part of a constituent. Nine different types of variant NP were discussed in detail:
82
Chapter 3 - Results of the corpus study
deferred modification, floating deferred modification, fronted modification, discontinuous modification, deferred determiners, floating deferred determiners, discontinuous determiners, deferred limiters, and floating deferred emphasizers. The ultimate goal of the present study is to investigate what conditions exist on the mobility of constituents in the English noun phrase. The results of the corpus study give insight into the possible word order variation in the NP. In some cases more than one variant is possible, and the choice for a specific word order is up to the language user. In other words, the description of the language must include more than one rule to account for the variants, and the language user can choose a rule to apply to the structure in the context at hand. In other cases, only one variant is possible to the exclusion of others. In these cases the description must include a rule which restricts the mobility of an NP constituent in a given context. A fronted premodifier, for example, can only occur before an indefinite article followed by a common singular noun or the proform one, it is mutually exclusive with a limiter and a (prototypical) premodifier, and the NP in which it occurs cannot function as adverbial, indirect object, or object complement. Although the results of the corpus study point to certain conditions on the mobility of constituents, we cannot obtain the statistical support to accept the findings as reliable data. The frequencies of occurrence of the different types of variant NP are so low that the expected cell frequencies would drop below five for most variables, in which case statistical techniques such as chi-square are no longer reliable. How then can we determine whether the conditions hold for the type as a whole or whether they are distorted by the restricted number of occurrences found in the corpus? In theory, there are several ways in which we can deal with this problem. First of all, of course, we can enlarge the corpus until we find evidence against the conditions. The problems with this approach are apparent, the main problem being that it is a theoretically unlimited process, since it will only end when and if the conditions are falsified. An additional problem for the research at hand is that for an investigation of most types of variant NP only corpora which have received a detailed syntactic annotation can be used. At present such corpora are not widely available. For the study of floating/deferred determiners, limiters, and emphasizers, however, a less sophisticatedly annotated corpus could in principle also be used, since these are closed classes, which enable a lexical search in a raw corpus or a corpus tagged for word class information. In chapter 2, the advantages of using a multi-method approach to descriptive studies were discussed in detail. For the present study, then, the combination of corpus data with a different type of data is not only advisable, but necessary, given the low frequency of occurrence of variant NPs. To follow the data cycle from this point onwards, hypotheses resulting from the corpus study need to be formulated, so that they can subsequently be tested by means of an elicitation experiment. The number of different hypotheses that can be tested in an elicitation experiment is restricted, since the experiment is restricted in size. An experiment can only involve a limited number of different questions and take a limited amount of time, in order to rely on the full attention and co-operation of
3.5 - Conclusion
83
informants. For this reason I decided to restrict the elicitation experiment to the testing of the hypotheses for three types of variant NP: the fronted premodifier, the discontinuous AJP, and the floating deferred modifier. They represent three major types of mobility (fronting, discontinuity and floating). What is more, they are too infrequent to give a reliable basis for statistical processing, but frequent enough to provide a good basis for formulating hypotheses. Also, they all involve complex structures that cannot be retrieved easily from a (large) less sophisticatedly annotated corpus. The hypotheses that were chosen to be tested in the elicitation experiment are the following: 1.
An NP with a fronted premodifier cannot function as indirect object, object complement, or adverbial. 2. An NP with a discontinuous AJP cannot function as indirect object, object complement, or adverbial 3. An NP with a floating deferred modifier cannot function as indirect object, object complement, or adverbial. 4. In an NP with a fronted premodifier the determiner can be realized by an indefinite article only. 5. In an NP with a discontinuous AJP the determiner cannot be realized by a (cardinal) numeral or a demonstrative pronoun 6. In an NP with a discontinuous AJP the head cannot be realized by a proper noun 7. In an NP with a floating deferred modifier the head cannot be realized by a demonstrative pronoun 8. In a fronted AJP the premodifier slot can only be realized by an intensifying adverb. 9. A fronted AJP cannot contain an adjective phrase postmodifier 10. A floating deferred modifier cannot be realized by an AVP or an NP 11. A fronted premodifier is mutually exclusive with a limiter and another premodifier The way in which these hypotheses were tested is discussed in chapter 4, as well as the results that emerged from the elicitation experiment and the combination of the results with those of the corpus study.
Chapter 4 The elicitation experiment 4.1 Introduction In the discussion of the multi-method approach to descriptive linguistics in chapter 2, attention was paid to general issues concerning the design and presentation of elicitation experiments and to the way in which these issues were dealt with in the elicitation of variant NPs. In this chapter, the results of the elicitation experiment for the hypotheses which were formulated in the last section of the previous chapter are discussed. But before we present and interpret these results, the design and presentation of the experiment is given some more attention. The current elicitation experiment is based on the design which was used in the Survey of English Usage and consequently includes both performance and judgment tests. An illustration of the thirteen different types of test which were used is given in Appendix F. With regard to the presentation of the tests, several variables which might influence the performance of informants were taken into consideration. One of the measures which was taken to minimize the variability in judgments between and within informants is the introduction of test batteries to restrict the variation in the types of test offered to an informant. Table 4-1 gives the initial distribution of the thirteen types of test over three batteries. Table 4-1: Initial distribution of test types over batteries Tests in battery I Composition, rating Operation, selection Completion, supplem. Preference, ranking
Tests in battery II Tests in battery III Composition, select one Operation, compliance Completion, forced choice Completion, word placement Evaluation Preference, Ranking Similarity Frequency Normalcy
Before the actual experiment was carried out, a pilot was performed with twelve informants, all students at the University of Liverpool. The aim of the pilot was to test the general suitability of the design of the experiment. The results are not judged on their relevance for the study of variant NPs. In the pilot, the batteries of Table 4-1 were given to four informants each. The results of the pilot show that, except for the Operation Compliance test, all types of test were found to be comprehensible and that the computer program was easy to work with. The major problem with the Operation Compliance test was that it was hard to find a
86
Chapter 4 -The elicitation experiment
suitable operation to perform on sentences in order to achieve the relatively spontaneous production of variant NPs. Two operations were included in the pilot. The first asked to change the indefinite article into the definite article (“change into ) and the second to change an active sentence into the passive. The first operation induced the informants in all cases to completely rewrite the sentence into a structure which was uninteresting for the study of variant NPs. The second operation proved too difficult for informants, since most did not know the difference between an active and a passive sentence. On the basis of the results of the pilot study, I decided not to include the Operation Compliance test in the actual experiment. This led to a redistribution of the twelve remaining types of test over the three batteries. The final distribution is given in Table 4-2 as well as the number of questions which was included in a test type in the first and second round of elicitation. Table 4-2: Final distribution of test types over batteries battery I battery II battery III # questions # questions # questions Test type Per. 1 Per. 2 Test type Per. 1 Per. 2 Test type Per. 1 Per. 2 Compos. 5 5 Compos. 5 10 Operation, 10 10 rating select one selection Compl. 5 5 Compl. 10 5 Compl. 5 5 forced ch. word place. supplem. Evaluation 10 10 Preference, 10 10 Preference, 10 10 Rating ranking Similarity 10 10 Frequency 10 10 Normalcy 10 10 In the first round of elicitation, the batteries were offered in two different versions (A and B), while in the second round a third version (C) was also included.1 In all, the results of 116 informants were (successfully) obtained in the experiment.2 In the first round, 42 informants took part, 14 in each battery, 7 in each version. In the second round, an additional 74 informants participated, 25 in batteries I and II and 24 in battery III, varying from 7 to 9 informants for each version. The informants were mostly students at the University of Liverpool, while additionally some members of staff participated. All informants took part only once. In all, 76 women took part versus 40 men. The informants range in age from 18 to 57, with an average age of 21. They come from various study backgrounds (32 humanities, 42 social sciences, 35 physical sciences, and 7 unknown) and from various regions in the United Kingdom. Although not directly relevant to the present, more general descriptive study of variant noun phrases, it would have been interesting to find out whether there 1
For an explanation of the difference between the three versions of a battery see chapter 2, section 2.6.2. 2 Beside these 116, nine informants took part whose results were not included in the overall test results. Two informants handed in an empty diskette, without data. And the results of seven other students were disregarded because it was doubtful whether they were native speakers of British English, since they were born outside the UK.
4.2 - Results of the elicitation experiment
87
are differences in the results of informants over different variables, such as orderof-presentation of the tests, age, sex, etc. However, since the number of variables is high and the number of informants relatively low, it is hard to make reliable statistical observations on the interactions between factors as well as on the effects of individual factors. A study was conducted of the relation between the three variables ‘sex’, ‘age’, and ‘order-of-presentation’ by means of multivariate tests of the scores which were obtained in the judgment tests. The results indicate that there is a significant difference between the scores of men and women for the Preference Rating test in the group of informants who were presented with battery II in the second round of elicitation, while no other significant effects between variables were found.3 The group concerned consists of 17 women and 8 men. If we compare their average scores in the Preference Rating test, we find that the largest deviation in scores is 1.04 (on a seven point scale) with an average deviation of 0.4. Since no significant effect of ‘sex’ was found in the other batteries or even for the other judgment test in the battery in question, and since only a small deviation of average means between men and women occurs for the scores concerned, we can safely conclude that, although the multivariate test shows that there is a significant effect of the variable ‘sex’ in one case, there is no need to distinguish between men and women in the interpretation of the results in general. Overall, the results of the multivariate tests give no ground to assume that it is unreliable to generalize over the results of the different informants in the interpretation of the elicitation data. As was pointed out before, the specific use of the elicitation data is to supplement the corpus data. Thus while aware of the limitations of the elicitation experiment, I think we can safely proceed to discuss and interpret the results of the experiments for the testing of the hypotheses which were formulated on the basis of the corpus results as discussed in the previous chapter. 4.2 Results of the elicitation experiment For the elicitation experiment, two types of results can be distinguished. First of all, there are the general results about the adequacy of the design of the experiment and the informants’ performance. Secondly, there are the specific results for the separate questions in the tests. Before we proceed to discuss the results of the elicitation experiment for the testing of the hypotheses, we first take a look at the more general results.
3
I am grateful to dr. Frans van der Slik for his help with the statistics and his patience in trying to fathom all variables involved in the interpretation of the elicitation experiment. For the multivariate analysis, the GLM Multivariate procedure of SPPS 8.0 was used. The F statistic for sex for the group in question was significant (F=772.95; df=15; p < .05), using Pillai’s trace to approximate the F value.
88
Chapter 4 -The elicitation experiment
4.2.1
General results
In the interview which was conducted after the computerized part in the second round of experiments, informants were asked to answer the following questions about the elicitation experiment: 1. 2. 3.
What do you think the aim of this questionnaire is?4 Did you have any problems answering the questions? Did you feel it was difficult to judge on acceptability and/or similarity/ frequency/normalcy?5
In response to the first question most informants indicate that they thought the aim was to test their knowledge and (correct) use of English grammar in general. Almost 19 per cent of the informants mention word order differences or word placement as a topic, but when asked, no one thought that the test was aimed at a specific structure. Other topics which were mentioned were language change, the influence of education, social class, region, etc. on language use, and the difference between spoken and written language. We can conclude that informants were unable to identify the target structure. Whether this was due to the introduction of distractor sentences cannot be established. Over 40% of the informants said that they had no problems at all in answering the questions. Others mentioned that it was hard to construct sentences and that it was unclear whether they could change or introduce words or leave words out (25.7%). Almost 15% of the informants used the option to skip a question. In all, 76 questions were skipped (58 production and 18 judgment questions) out of a total of 3865 questions. The most important reason for skipping a production task was that informants felt they could not construct any sentence which made sense with the given words. The most important reason for skipping a judgment task (Preference Ranking) was that all sentences were equally unacceptable.6 More than half of the informants had no trouble in judging the sentences on their acceptability, similarity, frequency, and/or normalcy. In general, the seven-point scale proved adequate. For the acceptability tests only 4% of the informants indicated that the scale was too broad. For the Similarity test, however, the score was much higher. Here, 20% considered the scale too broad. For this test, and possibly also for Normalcy and Frequency, a five-point scale would have been sufficient and perhaps easier to work with for the informants. 4
To the informants the experiment was always referred to as ‘a questionnaire’. Most students were asked to participate by means of an e-mail message. The message contained the following sentences: “Your task will be to fill in a questionnaire with simple questions on English usage. I will use the results in my research on English language use.” 5 ‘Acceptability’ refers to the Evaluation, Preference Rating, and Preference Ranking tests. A choice was made between Frequency, Normalcy, and Similarity, depending on the type of test the informants had encountered in their experiment. 6 Whenever informants chose to skip a question, a new window popped up prompting them to give a reason for skipping the question. This feature of the program proved quite successful, since informants often gave very precise reasons for skipping a question.
4.2 - Results of the elicitation experiment
89
Another problem that informants had with most judgment tests was the lack of context. Up to 12% of the informants indicated that the appropriate score is dependent on the context in which the sentence would be embedded. Other factors which were said to be of influence on the judgments are variety (spoken or written language), genre, and emphasis. Informants also mention that it is hard to distinguish between what is ‘grammatically correct’ and what they use in everyday life. The time that students needed to complete the test varied considerably. The average time for completion of the test, calculated over all informants, was slightly over 19 minutes. The fastest informant needed only nine minutes, while the slowest informant took 40 minutes. Both the fastest and the slowest informant participated in the first round of experiments. In this round, the informants took part in groups and could have been more easily distracted, since they were not placed in separate rooms. In the second round of experiments, informants participated one by one and they were seated in a separate room to work with the computer program. However, for this group of informants the variation is still high. The fastest person in this group needed a bit less than ten minutes, while the slowest needed over 34 minutes. The main reason to record the time needed per question was to investigate whether sentences which include variant NPs took more time on average to receive a judgment than sentences which include prototypical NPs, the hypothesis being that variant NPs are more complex than prototypical NPs and thus need more time to be processed. The best task to test this hypothesis is the Evaluation task. However, there are some problems which prevent the reliable testing of this hypothesis. First of all, of course, there is the large variation between informants in the time needed to complete the experiment, which complicates averaging over the time needed by informants to give a judgment. Secondly, all tasks of a specific type are presented one after the other. The time records indicate that there is an important learning or familiarization process involved in the performance of informants. The time needed to judge a sentence decreases sharply through the Evaluation task (and other tasks). This effect is only partly cancelled through the introduction of versions of a battery in which the order of sentences is changed. In other words, the time needed to ‘learn’ a task interferes (too) heavily with the time needed to process a sentence. To successfully cancel this effect we would need to include more than ten sentences in an Evaluation task. Thirdly, the time needed to judge a sentence is clearly influenced by the score it gets on the sevenpoint scale. Sentences which are perfectly acceptable or extremely unacceptable are judged faster than less clear cases. And in the fourth place, the time needed to process a sentence is not only influenced by its complexity, but also by its length. Taking all this into account, I decided not to include the recorded times in the discussion of acceptability of variant NPs. The inclusion of the interview and the option to skip a question while stating the reason for doing so proved to be important features of the design of the experiment. These features enabled the informants to pinpoint the reasons for rejecting a sentence and thereby enabled me to better interpret the results of the tests and to detect the sentences which should be excluded from the discussion of
90
Chapter 4 -The elicitation experiment
the hypotheses, because their unacceptability was caused by a different variable from the one which was intended. An example is sentence (1).7 This sentence was offered in an Evaluation experiment to test the acceptability of an NP with floating deferred modifier functioning as indirect object. The sentence received an average score of 4.5 (‘unacceptable’). When it was offered in the interview, sentence (1) was rejected by nine informants and accepted by six. Some of the informants who rejected the sentence commented that the combination of bad with criticism was strange. It was only then that I realized that criticism is a ‘false friend’ of the Dutch word kritiek.8 To a different group the sentence was therefore presented with review instead of criticism. The sentence was now judged as acceptable by all informants, although five indicated to prefer the continuous version. It is clear that the interpretation of the elicitation data is not simply a question of analysing the average scores. The interviews and the skip option in which informants can explain their judgments form an important source of information for a careful interpretation of the results. (1)
She gave the play a bad criticism in which he plays a priest
[136]
To conclude, I think we can claim that the design and execution of the experiment were successful. The problems the informants encountered were few, while their nature was largely predictable (such as the lack of context which was needed to establish the acceptability of sentences). Also, many informants said that it had been fun to participate in the experiment because it really made them think about their language and because the computer program was easy to work with. The use of a computer instead of elicitation on paper had the advantage that informants could not retrieve their answers to previous questions, and could thus not compare them. Also, the data from the elicitation experiments was immediately electronically available, which simplified further processing of the data. 4.2.2
Hypothesis testing
The hypotheses which are tested in the elicitation experiment are based on the corpus data for three of the nine types of variant NP: the fronted premodifier, the discontinuous modifier, and the floating deferred modifier. In the following sections, the results of the elicitation experiment are discussed per type of NP. 9 The hypotheses that were formulated for these three types of NP are all based on the non-occurrence of constructs in the corpus. But before we present, interpret and discuss the results for these hypotheses, we first have to answer a more 7
The numbers between square brackets to the right of the examples correspond to the numbers given to the sentences in Appendix G. For a complete list of sentences offered in the interviews see Appendix J. 8 Such mistakes can be overcome by having all test sentences carefully proof-read by a native speaker. 9 Some of the results have been previously published in de Mönnink (1997b, 1999, and in press).
4.2 - Results of the elicitation experiment
91
elementary question: How do native speakers of British English judge the acceptability of the occurrences of variant NPs which we encountered in the corpus? It is important to answer this question for two reasons. In the first place, a corpus may contain unacceptable sentences, so it is not certain that the constructs we found are acceptable, and consequently whether they should be included in a description of the English noun phrase. Secondly, the acceptability scores on corpus examples will give us a clue as to how we can interpret the results for the other hypotheses which were based on non-occurrence in the corpus. To answer the question of acceptability of variant NPs, I studied the occurrences of variant NPs in the corpus and presented those NPs or NPs of the same type to the informants.10 The NPs were embedded in sentences which were kept as simple as possible in order to minimize the introduction of other elements in the sentence which could influence the judgments of informants. The sentences were offered in both judgment and performance tests. In doing so, I tried to test the acceptability, but also the productivity of the structures. I also tried to find out whether prototypical structures were (always) preferred over variant NPs. In general, we can conclude from the elicitation results that variant NPs are considered very common by native speakers of English. Also, informants had no problem constructing the target sentences themselves in the performance tests. In testing the acceptability of fronted premodification, it turned out that fronted AJPs introduced by too are considered more common than those introduced by so. When given the choice, informants chose structures with such over structures with so. In the interview, informants judged the structures with so as acceptable, but also mentioned that they sounded ‘old-fashioned’, ‘poetic’, and ‘formal’. When asked to rephrase the examples they either replaced so by such, or they deferred the AJP to post-head position. Take, for example, the sentences in (2) to (4).11 Both (3) and (4) are preferred over (2). When (2) and (3) are offered in a Similarity test the average scores vary from ‘exactly the same meaning’ to ‘similar meaning’.12 When offered in a Frequency or Normalcy test, structures with such are considered ‘extremely frequent’ and ‘perfectly normal’, while structures with so are judged as ‘(fairly) infrequent’ and ‘controversial’ with regard to their normalcy. This possibly suggests that the combination of so + adjective is disappearing from the English language and is being replaced by the combination such + adjective.
10
The sentences offered in the experiment have to be relatively simple in order to keep the tests manageable and to exclude other variable factors as much as possible. The sentences from the corpus which contain variable NPs are almost without exclusion relatively long and complex sentences. For this reason, these sentences could not be included in the experiment. Where possible I used the NPs from the corpus, but embedded them in relatively simple sentences. In other cases I invented the whole sentence, while sticking to the structures encountered in the corpus as closely as possible. 11 For a complete list of sentences used in the elicitation experiment, see Appendix G. For the performance tests this appendix only includes the target sentences which I wanted the informants to construct. In practice, many more sentences were created. 12 For the interpretation of the scores on the seven point scale see chapter 2, section 2.6.2.
92 (2) (3) (4)
Chapter 4 -The elicitation experiment He is so arrogant a person He is such an arrogant person He is a person so arrogant
[5] [47]
When informants have the choice between a discontinuous AJP introduced by too in which the first part stands in prototypical premodifier position or where the first part is fronted, we find that the fronted position is preferred over the prototypical position. Also, when the AJP is at the same time discontinuous, the discontinuous version is considered more acceptable than the continuous version in which the AJP occurs in post-head position. In other words, example (5) is judged to be more acceptable than example (6), while (6) is considered more acceptable than (7). In the performance tests, the discontinuous version in which the first part occurs in prototypical position does occur, but clearly less often than the fronted version and the version in which the whole adjective phrase occurs continuously in postmodifier position. Interestingly, while judged as (slightly) less acceptable, the continuous structure was created more often than the discontinuous one. This discrepancy between judgment and production of both fronted premodification and discontinuous modification was a recurrent pattern throughout the experiment. Also, the production of variant NPs would increase after informants had been confronted with these structures in a judgment test. This seems to indicate that, although they are judged as perfectly acceptable, variant NPs are not necessarily the most obvious construction to create. This may explain the relatively low frequency of variant NPs in the corpus. (5) (6) (7)
This is too short a distance to travel by car This is a distance too short to travel by car This is a too short distance to travel by car
[108] [83]
From the previous paragraph it can be concluded that discontinuous structures are judged as acceptable by native speakers of English when the AJP is introduced by too and when the first part is fronted. How about the acceptability of intensified discontinuous AJPs whose first part precedes the head of the NP, but follows the determiner, while the second part follows the head (exs. 8-10)? The results of the elicitation experiment indicate that such structures are clearly acceptable to the informants. What is more, they are always judged as more acceptable than the version in which the whole AJP is placed in post-head position. When the AJP occurs in the superlative degree, the continuous posthead version is even considered as unacceptable (exs. 9-10). While testing the acceptability of fronted premodification, we already concluded that discontinuous AJPs in which such precedes the determiner while the head of the AJP follows the determiner are judged as acceptable. The same is true for discontinuous AJPs in which the premodifier position is realized by a different adverb (ex. 11). (8) (9) (10)
It occurs with a much higher frequency than I expected He was the most arrogant person I have ever encountered This was the worst nightmare ever
[59] [51] [90]
4.2 - Results of the elicitation experiment (11)
This is quite a complicated question
93 [85]
From the scores in the judgment tests it can be concluded that, for most NPs with a floating deferred modifier, no real preference seems to exist for either the continuous version in which the modifier occurs adjacent to the head or other constituents of the NP, or the discontinuous (floating) version (exs. 12-13). An exception to this are instances in which a passive verb phrase separates the floating deferred modifier from the other constituents of the NP (exs. 14-15). In these cases, the discontinuous version was highly preferred. In general, informants constructed more NPs with floating deferred modifiers in the performance tests than NPs with adjacent postmodifiers. (12) (13) (14) (15)
Reports have come in all day about strange occurrences Everyone started asking questions except Irena A rumour was spread that the president was dead Plans were made to deal with the situation
[134] [118] [116] [133]
So, although they occur relatively infrequently in corpora, the variant NPs tested here are considered very common by the users of English. This indicates that a description of contemporary English should indeed account for the occurrence of fronted premodification, discontinuous modification and floating deferred modification. What conditions can be formulated on these structures is discussed in the following sections, where their acceptability is tested in contexts and situations in which they were not encountered in the corpus. 4.2.3
Results for fronted premodification
In the last section of chapter 3, five hypotheses were formulated with regard to the occurrence of fronted premodification. In this section, the results of the elicitation experiment for these hypotheses are discussed. The first hypothesis states that an NP with a fronted premodifier cannot function as indirect object, object complement, or adverbial. The hypothesis consists of three subhypotheses, which will be discussed separately. Þ
An NP with a fronted premodifier cannot function as indirect object
The following sentences were used to test this (sub-)hypothesis. The presented scores and percentages are averages over informants.13 The percentages for the performance tests indicate how many informants produced the sentence. These 13
Appendix G indicates exactly which tests were used to elicit which sentences. In the discussion here I only distinguish between performance and judgment tests and within the class of judgment tests, between Preference Ranking and the other tests which are rated on the seven point scale (here referred to with the collective term Rating, not to be confused with Preference Rating). The average scores over informants for all sentences offered in the judgment tests can be found in Appendix H.
94
Chapter 4 -The elicitation experiment
percentages can count up to over 100 for a group of related sentences, since in the Composite Rating test the informants were asked to produce as many sentences as possible. For Rating, the presented score is an average of the scores a particular sentence received on the seven point scale. For Ranking, it indicates the average position a particular sentence got on the five point scale. Note that the presentation of the scores below should not be mistaken for frequency tables.
[2] He asked too large a group of people to come He asked a too large group of people to come He asked a group of people too large to come
Perform. Judgment % Ranking Rating 68.0 1.1 12.0 2.9 16.0 3.0
[38] Why not give so hungry a child a little bread Why not give a child so hungry a little bread Why not give a so hungry child a little bread
36.0 40.0 24.0
2.2 2.4 4.4
2.6 4.9
Additionally, the sentence in (16) was offered to the informants in the interview. (16)
You should not tell so jealous a boy that you met his girlfriend in the pub.
The results indicate that it is acceptable to have an NP with fronted premodifier in the function of indirect object. The somewhat lower score for [38] is influenced by the status of so + adjective, as discussed in the previous section. This is also clear from the comments of informants on sentence (16) in the interview. Of the 15 informants to whom this sentence was presented, 14 said that they found it acceptable, but eight remarked that it was ‘long-winded’, ‘formal’, ‘not common’, ‘posh’, and/or ‘old-fashioned’. Þ
An NP with a fronted premodifier cannot function as object complement
To test this (sub-)hypothesis the following sentences were used:
I consider Bill/John far too arrogant a person I consider Bill/John a far too arrogant person I consider Bill/John a person far too arrogant [29] The teacher called Bill so eloquent a speaker [101] He regarded it too difficult a question to answer He regarded it a question too difficult to answer [8]
Perform. % 32.0 36.0 16.0 42.9 42.9
Judgment Rating 1.56 4.95 1.85 1.85
Again, the fronted AJP introduced by so stands out by its low acceptability score. When [29] was offered in the interview, the reactions were similar to the other examples with so. Twelve out of fifteen informants accepted the sentence but indicated that they considered it ‘long-winded,’ ‘flowery’, ‘poetic’, ‘not used in
4.2 - Results of the elicitation experiment
95
speech’, etc. The three that rejected the sentence indicated that they objected to the use of so. From the results of examples [8] and [101] it is clear that it is otherwise perfectly acceptable to have an NP with fronted premodifier functioning as object complement in the sentence. Þ
An NP with a fronted premodifier cannot function as adverbial
This (sub-)hypothesis was tested by the following sentences:
[20] It goes back too long a time [102] It goes back too long a time for you to remember It goes back a too long time for you to remember [24] Last year he stayed out far longer a time [103] Last year he stayed out far longer a time than today [61] Last year he stayed out a far longer time than today
Perform. Judgment % Rating 3.2 56.0 1.8 0.0 5.0 4.2 44.0 3.5 12.0 2.5
Only the results of example [102] and the performance results of [103] clearly suggest that an NP with fronted premodifier can function as adverbial in the sentence. Sentences [20] and [24], however, score relatively low in the Evaluation test, while the Normalcy score for [103] is also ‘controversial’. The low scores for [24] and [103] may be explained by the fact that in these examples the premodifier in the AJP is realized by far while the whole AJP occurs in the comparative degree. Such a structure was suggested as a possible realization of a fronted premodifier by Aarts and Aarts (1988: 110), but it may well be less acceptable as fronted premodifier than AJPs introduced by so, too, or as. It is certainly less frequent, which may explain the relatively low score in the Normalcy test. Unfortunately, no other examples of far + comparative adjective were included in the experiment with which to compare the results for [24] and [103]. Overall, I think we can conclude that the function of adverbial can be realized by an NP with fronted premodifier. While the results for the test sentences do not lead to a straightforward rejection of the hypothesis, they certainly do not indicate that the examples are totally unacceptable and that the hypothesis should thus be accepted. When we take the results for the three sub-hypotheses into account, the first hypothesis can be rejected. An NP with a fronted premodifier can function as an indirect object, object complement, and/or adverbial in the sentence. The nonoccurrence of such instances in the corpus should presumably be ascribed to the low frequency of fronted premodification in combination with the low frequency of NPs realizing these functions in general. The function which an NP realizes in the sentence or clause thus poses no restriction on the potential mobility of the premodifier. The second (main) hypothesis for fronted premodification states that in an NP with a fronted premodifier the determiner can only be realized by an indefinite
96
Chapter 4 -The elicitation experiment
article. This implies that other common realizations of the determiner, such as the definite article or a demonstrative pronoun, are impossible when the determiner is preceded by an AJP. Two sentences were used to test this hypothesis. (17) (18)
He has too firm this grip on us It depends on how strong the person you are
[4] [19]
The first sentence was elicited by means of a Completion Forced Choice test. The informants had to fill in the determiner function of the noun phrases in two related sentences. In one sentence the determiner followed a fronted premodifier, while in the other sentence the determiner preceded the premodifier. The choice was between an indefinite article and a demonstrative pronoun. This resulted in one of the following pairs of sentences: (a) (b)
He has too firm a grip on us He has too firm this grip on us
He has this very firm grip on us He has a very firm grip on us
If both the (a) and the (b) versions are correct, we would expect to find great variety in the answers of informants. As it is, only one out of 14 informants chose the second pair. Sentence (18) was elicited by means of a Completion Word placement test. The informants were asked to rewrite sentence (19) using the phrase ‘how strong’ in it. If the second hypothesis is correct, informants would have to change the definite article into an indefinite article in order to be able to use the AJP how strong as a fronted premodifier to person. The instructions for the test did not tell informants that they were allowed to do so. However, out of fifteen informants four actually violated the rules and changed the article, resulting in the construction of sentence (19a). Two informants skipped the question, of which one stated that there was ‘no possible well formed sentence unless the is changed to a’. Three informants constructed sentence (19b)/(18). Possibly, they preferred to strictly follow the rules of the assignment, taking the construction of an unacceptable sentence into the bargain. The other five informants created sentence (19c), thereby avoiding the construction of an NP with a fronted premodifier. That sentence (18) is indeed unacceptable becomes clear from the remarks of informants when it was presented to them in the interview. Eleven out of fifteen informants immediately rejected the sentence and stated that the definite article should be replaced by the indefinite article. When sentence (19a) was offered to informants in an Evaluation test, it was judged as totally acceptable (with an average score of 1.4). (19)
It depends on the person you are a. It depends on how strong a person you are b. It depends on how strong the person you are c. It depends on the person how strong you are
4.2 - Results of the elicitation experiment
97
On the basis of these results the second hypothesis can be accepted. Although we have not tested every possible realization of the determiner, we can safely accept the hypothesis based on the fact that two otherwise common realizations of the determiner are judged to be unacceptable when a premodifier precedes the determiner. The obligatory realization of the determiner by an indefinite article poses a strong restriction on the overall structure of an NP with a fronted premodifier, in that it also restricts the realization of the head of the NP. An indefinite article does not (normally) co-occur, for example, with a head that is realized by a proper noun, a nominal adjective, a numeral, or a pronoun. The third hypothesis for fronted premodification mentioned in chapter 2 states that the premodifier slot in a fronted AJP can only be realized by an intensifying adverb (phrase). In the corpus we only found AJPs premodified by too, as, and so in fronted position as well as occurrences of such, while Aarts and Aarts (1988) mention that a fronted AJP can also contain one of the intensifying adverbs how(ever), ever so, that, this, enough, more, less and far (the latter only when preceding an adverb in the comparative degree or one of the intensifying adverbs too, more, and less).14 In the discussion of the previous hypothesis, we already saw that fronting of an AJP in which the premodifier is realized by how is acceptable (ex. 19a). To test the current hypothesis the following sentences were included in the experiment.
[3] [6] [15] [30] [31] [37] [68] [104] [98]
He did exceptionally brave a deed (yesterday) He did an exceptionally brave deed (yesterday) He told truly astonishing a story I never travelled this short a distance by car This is far more difficult a question This is a far more difficult question This is perfectly boring a film This is a perfectly boring film Tom, very convincing a person, told her to go She told a more convincing story than yesterday She told more convincing a story than yesterday Fred is clever enough a man to look through that Fred is a clever enough man to look through that Fred is a man clever enough to look through that
Perfor. Judgment % Rank. Rat. 4.0 4.1 44.0 1.6 0.0 2.1 0.0 71.8 0.0 83.3 4.2 1.3 1.6 2.4 8.0 2.3 3.1 56.0 1.6 1.5 48.0 2.7
The results of the elicitation experiment clearly indicate that it is impossible to have a non-intensified AJP in fronted position. In the Completion test (the scrambled parts of) sentence [3] were offered with the article a instead of an. The idea was that this would prompt the informants to construct an NP with a fronted 14
In a discussion of sentences [24] and [103] above we saw that a fronted AJP in the comparative degree in which far functions as AJP premodifier is considered as less acceptable than a fronted AJP introduced by so, as, or too.
98
Chapter 4 -The elicitation experiment
premodifier, if possible. However, four informants changed the article to an, while six informants skipped the question, indicating that the article should be an. This result indicates that the prototypical position is clearly preferred over fronted position for a non-intensified AJP. The same result is obtained from the judgment test where the fronted version is judged as unacceptable, while the prototypical version is judged as perfectly acceptable. Sentences [6] and [31] are not constructed with a fronted premodifier at all, and sentence [37] is judged as unacceptable in an Evaluation test. Next to premodification by so, as, too, and how, a fronted AJP can be intensified by this (ex. [15]). For more, however, the results are less clear. Sentence [30] was never constructed with a fronted premodifier, while sentence [104] was ranked third in a Preference, Ranking test. Although a fronted AJP introduced by more is perhaps not totally unacceptable, the version in which the AJP occurs in prototypical premodifying postition is clearly more productive and preferred by native speakers of English. Another way to intensify an AJP is through postmodification by means of the adverb enough. Sentence [98] was used to test whether an adjective that is postmodified by enough can occur as a fronted premodifier in the NP. Again, we see that the fronted version is not very productive in the Composition test and the prototypical premodifying position is clearly the preferred position. To conclude, then, the results of the elicitation experiment not only give reason to accept the third hypothesis (the premodifier slot in a fronted AJP can only be filled by an intensified adverb), but they also show that not every intensifying adverb is equally acceptable in this position. The fourth hypothesis which was tested by means of the elicitation experiment claims that a fronted adjective phrase cannot contain a postmodifier. Adjective phrases that contain a postmodifier tend to be deferred to post-head position in the noun phrase. In the discussion of the previous hypothesis, we saw that, when the AJP is postmodified by the adverb enough, prototypical premodifier position is preferred. In fronted position the AJP receives a ‘controversial’ judgment; it is neither acceptable, nor is it entirely unacceptable. Two other sentences were used to test this hypothesis. In both sentences the adjective phrase postmodifier is realized by a (short) non-finite clause. Sentence (20) was judged as unacceptable in a Preference Rating test (average score is 4.5). Sentence (21) was never constructed in an Operation Selection test and was judged as highly unacceptable in a Preference Rating test (average score = 5.6). To conclude, the fourth hypothesis can be accepted. In general, an AJP with postmodifier is preferred in either post-head position or prototypical premodifier position. When the AJP postmodifier is realized by an intensifying adverb, fronting is more acceptable than when the postmodifier is realized by a (non-finite) clause, but even then fronting is unproductive and the acceptability is doubtful. (20) (21)
I think this is too difficult to answer a question This is too easy to do a job
[18] [36]
4.2 - Results of the elicitation experiment
99
The fifth and last hypothesis states that a fronted premodifier is mutually exclusive with a premodifier in prototypical position. In the elicitation test only one sentence was used to test this hypothesis. In an acceptability test, sentence (22) scored 6.46 on the seven point scale, indicating that it is judged as extremely unacceptable. However, the unacceptability can partly be explained by the perceived status of so + adjective in English. That the extreme unacceptability must indeed be attributed to the presence of both a fronted and a prototypical premodifier was established in the interviews in which 15 informants were asked to comment on sentence (22) and sentence (24), while another 15 were asked to comment on sentence (23). Sentence (23) was accepted by 11 informants (although some commented on the use of so), while sentences (22) and (24) were rejected by all informants. On the basis of these results, the fifth hypothesis can be accepted. (22) (23) (24) 4.2.4
I had so romantic a long evening yesterday I had so romantic an evening yesterday John is far too arrogant a cold person
[9]
Results for discontinuous modification
For NPs with a discontinuous modifier only three hypotheses were tested in the elicitation experiment. As was the case for fronted premodification, the first hypothesis deals with the distribution of NPs over sentence/clause functions and can be divided into three sub-hypotheses. Þ
An NP with a discontinuous modifier cannot function as indirect object
Perfor. Judgment % Rank. Rat. [94] We gave a completely different department from 60.0 2.8 2.4 ours the assignment We gave a department completely different from 20.0 3.9 2.4 ours the assignment We gave the assignment to a completely different 1.4 department from ours [95] You asked a less intelligent student than before to 32.0 2.4 do the job You asked a student less intelligent than before to 44.0 2.7 do the job (25) (26)
He asked an even larger party than yesterday to come along He asked a party even larger than yesterday to come along
Sentences [94] and [95] were presented in both performance and judgment tests. Additionally sentences (25) and (26) were presented to informants in the interview. Out of 15 informants 12 accepted sentence (25), while only eight accepted sentence (26). Of the seven informants that rejected sentence (25), three
100
Chapter 4 -The elicitation experiment
suggested to split the AJP into a discontinuous modifier of the NP. All the results show that it is acceptable to have an NP with a discontinuous modifier in the function of indirect object in the sentence. The hypothesis can thus be rejected. Þ
An NP with a discontinuous modifier cannot function as object complement
The results of sentence [101] for NPs with a fronted premodifier as discussed in the previous section already indicate that it is possible to have an NP with discontinuous modifier in object complement function in the sentence. The two other sentences which were used to test this hypothesis also suggest that this subhypothesis can be rejected. Sentences [52] and [54] were judged as perfectly acceptable.
[52] His illness made him a very different person from his friends His illness made him a person very different from his friends [54] I consider John an easy man to persuade I consider John a man easy to persuade [101] He regarded it too difficult a question to answer He regarded it a question too difficult to answer Þ
Perfor. Judgment % Rank. Rat. 1.8 4.4 1.4 2.2 42.9 42.9
1.4 1.85 1.85
An NP with a discontinuous modifier cannot function as adverbial
The sentences used to test this hypothesis were already discussed in section 4.2.3, when the possibility to have an NP with fronted premodifier in adverbial function in the sentence was tested. There we concluded that an NP with fronted premodifier was even more acceptable in adverbial position when the AJP was at the same time discontinuous. Additionally, sentence [61] received an average score of 2.5 in a Normalcy test, indicating that it is also possible to have an NP with a discontinuous AJP of which the first part is not fronted functioning as adverbial in the sentence. In other words, the third sub-hypothesis can also be rejected. To conclude, the first hypothesis which was defined in chapter 3 can be rejected. NPs with a discontinuous modifier can function as indirect object, object complement, and adverbial in the sentence/clause. No restriction can thus be defined as to the function which an NP with discontinuous modifier can take in the sentence/clause. The second hypothesis which was tested in the elicitation experiment states that the determiner in an NP with a discontinuous modifier cannot be realized by a (cardinal) numeral or a demonstrative pronoun.
4.2 - Results of the elicitation experiment
[73] These are two particularly complicated sums to calculate These are two sums particularly complicated to calculate [91] Today we ate three even bigger apples than yesterday Today we ate three apples even bigger than yesterday [97] You have this more positive attitude than me/Sue You have this attitude more positive than me/Sue
101 Perfor. Judgment % Rank. Rat. 1.1
(27)
a. They have two sufficient incomes to buy the house b. They have two incomes sufficient to buy the house
(28)
a. You have this more positive attitude than me You have an attitude more positive than me b. You have a more positive attitude than me You have this attitude more positive than me
2.2 56.0
3.1
44.0
3.0
52.0 36.0
2.7 5.2
The results for these sentences give no straightforward evidence for either accepting or rejecting the hypothesis. Only sentence [73] seems perfectly acceptable. Unfortunately this sentence was only presented in a Preference Ranking test. The result can thus not be compared to that of another test. The other sentences were mostly judged as ‘controversial’, although the sentences are constructed by informants in the performance tests. Sentence [97] was offered in a Completion Forced-choice test. The informants could either produce the pair of sentences in (28a) or the one in (28b). It may well be that the choice for (28a) was a negative choice, because the pair in (28b) was judged as even less acceptable. Indeed, two informants skipped the question with the remark that neither choice would lead to an acceptable pair of sentences. Sentences (27a) and (27b) were offered in the interview. Sentence (27a) was accepted by six out of fifteen informants, while sentence (27b) was accepted by ten. Informants would clearly prefer a relative clause here in which the adjective phrase is embedded. Although the sentences used to test the second hypothesis were not judged as straightforwardly acceptable by the informants, they were not totally rejected either. Given the right context, it may well be possible to have an NP with a discontinuous modifier in which the determiner is realized by a (cardinal) numeral or a demonstrative pronoun. In testing this hypothesis, one of the disadvantages of using an elicitation experiment was encountered. The sentences are presented (virtually) without context, which may influence the acceptability of structures, especially of the more complex ones. The third hypothesis states that an NP with a discontinuous AJP cannot be headed by a proper noun. Normally, proper nouns are not modified. However, if the
102
Chapter 4 -The elicitation experiment
proper noun has lost its unique reference, it acts syntactically like a common noun in that it can co-occur with a determiner and modifiers. We would thus expect a proper noun without unique reference to be able to have a discontinuous modifier as well. Three sentences were used to test this hypothesis. (29) (30) (31)
Frank is probably the poorest Pendleton of them all He is a better Romeo than Tom ever was He plays a more convincing Romeo than Tom
[41] [44] [48]
Sentences (29) and (31) scored 1.64 and 1.54 in an Evaluation and Normalcy test respectively. Sentence (30) was produced by 56% of the informants in a Completion, Forced-choice test. From these results it is obvious that the third hypothesis can be rejected. It is perfectly acceptable to have a discontinuous AJP as modifier to a (non-unique) proper noun. 4.2.5
Results for floating deferred modification
For NPs with a floating deferred modifier three hypotheses were formulated in chapter 3. The first hypothesis consists of three sub-hypotheses again. The first sub-hypothesis states that an indirect object cannot be realized by an NP with a floating deferred modifier. Three sentences were used to test this sub-hypothesis. The results of the judgment tests for these three sentences suggest that the hypothesis can be accepted. What is striking, however, is that both the continuous and the discontinuous version of all sentences score low in the judgement test, while in the performance tests the continuous version is clearly constructed more often and apparently without problem. Perfor. Judgment % Rank. Rating [121] He gave the matter no thought on how to win the 92.9 3.7 game 3.1 He gave the matter on how to win the game no 7.1 3.9 thought 3.0 [128] I forgave the man everything who stole my bike 14.3 1.8 5.7 I forgave the man who stole my bike everything 64.3 2.1 5.3 (32)
She gave the play a bad review in which he plays a priest
To get more information on the reason for the unacceptability, the sentences were also presented in an interview. Sentence [121] was accepted by the majority of informants, although many commented that they would not use it because it was too complex and seemed ‘jumbled up’. They clearly prefer the indirect object to be included in a prepositional phrase (He gave no thought to the matter…). These comments explain the relatively low scores in the Frequency and Normalcy tests for both the discontinuous and the continuous version. Sentence [128] was rejected by the majority of informants. In all cases the reason for rejection was the use of the word everything in the sentence, which was considered strange in
4.2 - Results of the elicitation experiment
103
the given context. Everything should preferably be preceded by for. Again, these results explain the low scores for the Preference Rating test. Since the judgments are clearly not based on the occurrence of a floating deferred modifier, it is best to disregard this sentence in the discussion of the first sub-hypothesis. Additionally, sentence (32) was presented in the interview. The sentence was judged as acceptable by all informants, although five indicated to prefer the continuous version. The previous discussion illustrates again that the interpretation of the elicitation data is not simply a straightforward question of analysing the average scores. Considering all the data, we can conclude that the first sub-hypothesis can be rejected. Given the right context, an NP with a floating deferred modifier can function as indirect object in a sentence. When both the indirect and the direct object functions are realized by a noun phrase, a floating deferred modifier can cause (structural) ambiguity: it can either be analysed as a floating modifier to the indirect object or as a normal postmodifier of the direct object. Native speakers of English will be prompted to try to connect the modifier to the nearest NP. This can cause confusion in the interpretation of a sentence and may explain the comments made by informants in the interview that the sentence seems “jumbled up”. Floating deferred modifiers of indirect object NPs are thus restricted to clear cases where the floating deferred modifier does not cause ambiguity in the interpretation of the NP. The restriction is then not of a syntactic, but of a semantic nature. The second sub-hypothesis states that an NP with a floating deferred modifier cannot function as object complement. The hypothesis was tested by means of three sentences. The test results for each sentence and each type of test indicate that the discontinuous version is not possible, or at least highly irregular. Other than with the previous sub-hypothesis, here the discontinuous versions were hardly ever constructed in the performance tests. However, here again the interpretation of the data is not as straightforward as it may appear at first sight. All three sentences are structured in the same way. The modifier is separated from its head by means of an adverbial of time. Perfor. Judgment % Rank. Rat. [135] She called him a dog yesterday with a bad heart 7.0 She sold him a dog with a bad heart yesterday 93.0 [144] They elected him president today of the company 0.0 3.6 5.3 They elected him president of the company today 93.0 1.3 2.4 [145] They will appoint her Head tomorrow of the 4.9 school A possible reason for informants to reject the discontinuous versions would be that a time adverbial is not preferred in sentence medial position. This is supported by data collected in the interviews. When confronted with sentence [135] or [144] only three out of thirty informants totally rejected the sentences.
104
Chapter 4 -The elicitation experiment
The others indicated that the word order seemed strange but was certainly not wrong. When asked to rephrase the sentence they placed the time adverbial in sentence initial or final position. The variable ‘preferred placement of the adverbial’ may thus interfere here and skew the results toward a preference for the continuous version. Even if the second sub-hypothesis cannot be accepted or rejected unequivocally on the basis of the elicitation data, the data show that the occurrence of an NP with floating deferred modifier as object complement is very restricted, both in the possible realization of the intervening material and in the frequency of occurrence. The continuous version is highly preferred. The third sub-hypothesis states that an adverbial cannot be realized by an NP with a floating deferred modifier. Three explanations can be offered for the nonoccurrence of adverbial NPs with a floating deferred modifier:15 1. 2. 3.
It is (syntactically) impossible to defer a modifier outside the boundaries of an adverbial NP. The non-occurrence in the corpus is the result of the combined low frequencies of an adverbial realized by an NP and floating deferred modification in general. A modifier that occurs outside the boundaries of the adverbial NP it modifies will not be recognized as modifier and receives a different interpretation at sentence or clause level.
Put differently, the third explanation states that a floating deferred modifier of an NP that functions as adverbial will receive a different interpretation than when it occurs adjacent to the noun it modifies, thereby causing a difference in meaning of the sentence. This difference in meaning can be elicited by means of a Similarity test. Sentences (33) and (34) were presented in both the continuous and the discontinuous versions in a Similarity test. (33) (34)
a. The minute he went home everybody started laughing b. The minute everybody started laughing he went home a. The moment they entered the room the situation changed b. The moment the situation changed they entered the room
[138] [139]
If the third explanation is the most likely one, the scores resulting from the Similarity test should be high, indicating that the continuous version means something entirely different from the discontinuous version. For sentences (33) and (34) similarity scores are indeed found to be extremely high, 6.7 and 6.2 respectively. It can thus be concluded that the clauses have a different function in the a. sentences than in the b. sentences. In the continuous version they function as modifier to the noun. In the discontinuous version, the original subject and verb of the sentence are now interpreted as a postmodifying clause, while the constituents of the original postmodifying clause now function at the level of the 15
These three explanations for the non-occurrence of adverbial NPs with a floating deferred modifier are not necessarily mutually exclusive.
4.2 - Results of the elicitation experiment
105
main sentence. If this analysis is correct, extraction of the postmodifying clause would not be possible when the clause is introduced by a subordinator, since a subordinate clause cannot function independently as the main clause in a sentence. This is indeed the case (cf. sentences 33’ and 34’). (33’) a. The minute that he went home everybody started laughing *b. The minute everybody started laughing that he went home (34’) a. The moment that they entered the room the situation changed *b. The moment the situation changed that they entered the room The explanation of a difference in the interpretation of the sentence is also supported by the results of the Composition test. Here the answers were highly inconsistent, indicating no clear preference for one version over the other. This result would indeed be expected if the different interpretations are completely independent. It can thus be concluded that the non-occurrence in the corpus of a floating deferred modifier to an NP that functions as adverbial can be attributed to the fact that these constructions cannot be obtained from the corpus simply because they receive a different analysis. At the same time, this explanation means that the postmodifier in an adverbial NP cannot be moved without severe consequences for the interpretation of the sentence. This poses a heavy restriction on the mobility of the (post)modifier. In this light the third sub-hypothesis can be accepted. The second hypothesis formulated in chapter 3 states that a demonstrative pronoun functioning as head cannot co-occur with a floating deferred modifier. Three sentences were used to test this hypothesis. The results from the judgment tests show that informants have a clear preference for the continuous version. However, informants did produce floating deferred modifiers in the performance tests. In one case, the Forced Choice Selection task ‘forced’ informants to produce a noun phrase with a floating deferred modifier of which the head was realized by a demonstrative pronoun. If this construction is impossible in English, one would expect informants to skip the question. However, none of the informants did.
[123] His figure was that still of a young man/athlete His figure was still that of a young man/athlete [148] We invited those to our party we know from London We invited those we know from London to our party [145] We invited those to our party we love We invited those we love to our party (35)
His figure was still that, I feel, of a young man
Perfor. Judgment % Rank. Rat. 35.7 5.0 64.3 1.6 35.7 64.3 4.2 1.3
106
Chapter 4 -The elicitation experiment
In the interviews, both the continuous and the discontinuous version of sentence [148] was presented to informants, as well as sentence (35). Only three informants accepted the discontinuous version of sentence [148]. The others indicated a preference for the continuous version and commented that the discontinuous structure was perhaps not wrong in formal written language. All informants accepted the continuous structure, but again some commented that it sounded formal and that they would not use it in speech. Sentence (35) evoked the same remarks.16 Taking all the data into account, the second hypothesis can be rejected. Although an adjacent modifier is clearly preferred, it is possible to have a floating deferred modifier to a demonstrative pronoun. The third hypothesis for NPs with a floating deferred modifier consists of two sub-hypotheses. The first claims that a modifying adverb phrase cannot occur outside the boundaries of the NP as a floating deferred modifier. Basically, the same three explanations can be suggested here as for the non-occurrence of adverbial NPs with a floating deferred modifier above. Here, the third explanation implies that, if a modifying adverb phrase is separated from the head it modifies, it is interpreted as an adverbial at sentence/clause level. Again, any difference in interpretation can be brought to light by means of a Similarity test. The following pairs of sentences were offered to informants in a Similarity test: (36) (37) (38) (39)
a. Everything around was quite still b. Everything was quite still around a. It is where evenings here usually end b. It is where evenings usually end here a. Mary here was a tower of strength to me b. Mary was a tower of strength to me here a. The space below has been found empty b. The space has been found empty below
[119] [129] [130] [140]
The similarity scores range between 3.3 (controversial) and 4.5 (fairly different). A difference in meaning does seem to exist, although not so clearly as for the NPs functioning as adverbial. The difference can be ascribed to the different interpretation of the adverb phrase, either as modifier to the NP or as adverbial at sentence level. The second sub-hypothesis states that a modifying noun phrase cannot occur outside NP boundaries as a floating deferred modifier. Again, the same line 16
In the corpus we find that a demonstrative pronoun followed by an adjacent postmodifier is indeed the most frequent in the more formal genres non-fiction and scripted speech.
4.3 - The combined results
107
of reasoning can be followed as for the previous sub-hypothesis. The result is different, however. Sentence pair (40) was offered in a Similarity test and received an average score of 1.7, indicating that the two sentences have exactly the same meaning and that the modifying NP is not interpreted differently when it occurs non-adjacent to the noun it modifies. In an Evaluation test, sentence (41) received an average score of 3.8, indicating that its acceptability is controversial. Sententences (42) and (43) were offered both with and without the preposition of in the interviews. With of the sentences were judged as acceptable by all informants, although some mentioned that they would prefer the prepositional phrase to immediately follow the head. Without the preposition, about half of the informants accepted the sentence. The other half rephrased the sentence by including the modifying NP in a prepositional phrase or relative clause. To conclude, the second sub-hypothesis can be rejected. A floating deferred modifier can be realized by a noun phrase, although realization by either prepositional phrase or clause is clearly preferred. (40) (41) (42) (43)
I discovered a hole this morning the size of an apple I discovered a hole the size of an apple this morning A car was stolen yesterday that colour A girl appeared (of) your age, smoking a cigarette A rock was found (of) that same shape
[127] [111] [114]
4.3 The combined results We have now collected both corpus data and experimental data on the occurrence of fronted premodification, discontinuous modification, and floating deferred modification in British English. In this section, the results of both studies are combined in order to reach an integral description of the NP structures in question. 4.3.1
Combined results for fronted premodification
The combined results of the corpus study and the elicitation experiment give rise to the description of an NP with a fronted premodifier as shown in Figure 4-1 (p. 108). In the corpus, no examples were found in which a fronted premodifier cooccurs with a limiter. This possible restriction on the structure was not tested in the experiment. We can thus not exclude the (optional) limiter slot from the structure in Figure 4-1. The fronted premodifier is always realized by an intensifying adjective phrase. This is either (identifying) such or an adjective modified by an intensifying adverb phrase. The modifying adverb phrase may consist of a head only, or a premodifier plus head. The head of the adverb phrase is preferably realized by one of the adverbs as, so, too, how, or this, although realization by more or enough is also marginally acceptable, while the acceptability of the realization by less, how(ever), ever so, and that has yet to be established.
108
Chapter 4 -The elicitation experiment
In a noun phrase with a fronted premodifier the determiner is an obligatory constituent which is realized by an indefinite article. A premodifier slot in prototypical position is missing, since the results of the experiment indicated that such occurrences are judged as unacceptable. In the corpus, the head was either realized by a common noun or by the proform one. Other possible realizations of the head were not tested by means of the elicitation experiment, but are not very likely, given the realization of the determiner. NP (LIM)
PRFR
DT
NPHD
AVP
AJP(int)
DTP
N(com) PRO(one) …
(AJPR) AVP(int)
AJHD (AJPO)
DTCE
(NPPO)* PP CL
ADJ AVP(int) ART(indef)
Figure 4-1: NP with fronted premodifier No restrictions can be formulated as to the distribution of NPs with a fronted premodifier. The results of the elicitation experiment show that these NPs are also acceptable in indirect object, object complement and adverbial position, although such occurrences were not encountered in the corpus. Frequencies of occurrence in the corpus show that NPs with a fronted premodifier have a clear preference for the direct object, subject complement and prepositional complement positions, while occurring less frequently as subject. Additionally, the results from the elicitation experiment, and especially from the interviews, show that the combination of the intensifying adverb so with an adjective is associated with formal, written language and is considered poetic and archaic usage. Although informants used the construction in the experiment, they claim not to use the it in their normal language use. Instead, informants prefer the construction with such (a) plus adjective, which has exactly the same meaning according to the results of the Similarity test. 4.3.2
Combined results for discontinuous modification
All hypotheses which were tested for NPs with a discontinuous modifier were rejected. This implies that no clear-cut restrictions apply to such NP structures. From the corpus study, we concluded that four different types of discontinuous modification can be distinguished. The first and most frequent type is represented in Figure 4-2 (p. 109). Although the NP structure can in theory become very complex, it should be noted that in most NPs with a discontinuous modifier the only optional slot which is realized is the determiner slot, since there exists a clear preference for less complex structures. The determiner function is most commonly realized by an indefinite or definite article and the head by a common
4.3 - The combined results
109
noun, although other realizations are also possible. The realization of the postmodifier slot in the first part of the discontinuous AJP is restricted to occurrences of enough. Also, realization of this slot blocks the realization of the AJP premodifier. In other words, the premodifier and postmodifier in the first part of the discontinuous AJP are mutually exclusive. NP (LIM) (DT)
MODI
AVP
AJPD1
DTP (AJPR)
(NPPR)* NPHD (NPPO)* MODI (NPPO)* AJP NP
AJHD (AJPO)
N(com) PP N(prop) CL PRO(one)
AJPD2
PP
AJPO*
Figure 4-2: Most common type of NP with a discontinuous modifier Instead of following the head (or postmodifier) of the NP immediately, the second part of the discontinuous modifier, the AJP postmodifier, can also occur outside the boundaries of the NP towards the end of the sentence. In that case, the same restrictions hold as for NPs with a floating deferred modifier as discussed in the next section. The second type of NP with a discontinuous modifier is represented in Figure 4-3. Here, the first part of the discontinuous modifier occurs before the determiner in fronted position. This type is subject to the restriction such as formulated for fronted premodification in the previous section. NP (LIM)
MODI
DT
AVP
AJPD1
DTP
(AJPR) AJHD (AJPO)
NPHD
(NPPO)*
N(com) PRO(one) DTCE …
PP CL
MODI
NPPO)*
AJPD2
PP
AJPO*
AVP(int) ADJ AVP(int) ART(indef) Figure 4-3: NP with a discontinuous and fronted modifier The last two types of NP with a discontinuous modifier are combined in Figure 44 (p. 110). The third type occurs when the part of the AJP following the head of the NP is not realized. In the fourth type, the discontinuous modifier consists of three parts. An NP with a discontinuous modifier essentially has a similar distribution over sentence/clause functions as a prototypical NP. However, it shows a clear preference for the functions of direct object and subject complement.
110
Chapter 4 -The elicitation experiment NP
(LIM) MODI
DT
MODI (NPPR) NPHD (NPPO)* (MODI) (NPPO)*
AVP AJPD1
DTP
AJPD2
AJPR
DTCE
AJHD
AVP
ART
AJP N(com) PP NP N(prop) PRO(one)
AJPD3
PP CL
AJPO*
ADJ
Figure 4-4: NP with a discontinuous modifier with two parts ocurring before the head 4.3.3
Combined results for floating deferred modification
As was the case for NPs with a discontinuous modifier, the structure of an NP which is modified by a floating deferred modifier is fairly unrestricted (cf. Figure 4-5). The only difference in realization between an NP with a floating deferred modifier and an NP with an adjacent postmodifier is that the floating deferred modifier cannot be realized by an adverb phrase. When a modifying adverb phrase is postponed, the sentence receives a different interpretation and analysis. The adverb phrase is then no longer analysed as a postmodifier of the NP but, instead, functions as adverbial at sentence or clause level. In Figure 4-5, the floating deferred modifier is not attached to the rest of the NP structure, but is separated by interrupting sentence or clause material. The realization of this interrupting material can range from small verb particles to full clauses or sequences of different categories. NP (LIM)
(DT)
AVP
DTP
(NPPR)* NPHD AJP NP
(NPPO)*
N(com) AVP N(prop) PP NUM CL PN(dem) NADJ Figure 4-5: NP with a floating deferred modifier
FNPPO AJP NP PP CL
Again, it should be noted that although the structure of a noun phrase with a floating deferred modifier can in theory be very complex, most of the occurrences found in actual language use are found to consist of only a (determiner plus) head plus floating modifier. With regard to the distribution over sentence/clause functions, the only clear difference with a prototypical NP is that an NP with a floating deferred modifier cannot realize the function of adverbial. If the modifier of an NP in adverbial function is postponed, the sentence will receive a different
4.4 - The data cycle continued
111
interpretation and thus a different analysis. Additionally, the occurrence of an NP with a floating deferred modifier in the function of object complement or indirect object is restricted with regard to the possible realization of the interrupting material. 4.4
The data cycle continued
Combining the corpus data with the experimental data has so far led to a fairly complete description of NPs with a fronted premodifier, a discontinuous modifier, and/or a floating deferred modifier. The data cycle as described in chapter 2 has been completed once and the research can satisfactorily end here. However, it can also be continued by pursuing the data cycle further. There are several reasons for doing so. First of all, the results of the elicitation experiment may induce a more specific corpus search. One of the problems with a qualitative study of the mobility of NP constituents is that we need corpora which have been enriched with detailed syntactic information, since it is impossible to find the occurrences automatically in a raw corpus or in a corpus which has been enriched with word class information only. As a consequence, the corpora that can be used are but few, the size of the corpus is restricted, and the number of occurrences of variant NPs with a mobile constituent which are encountered is small. If, through the results of the elicitation experiment, a clear (lexical) restriction on the realization of variant NPs can be formulated, this enables a more specific search for occurrences using other, larger corpora which have been annotated in a less sophisticated manner. In the current study, only the results for NPs with a fronted premodifier were such that they enable a further, more restricted corpus study. The results indicate, among other things, that the realization of the determiner in NPs with a fronted premodifier is restricted to the indefinite article and that the adjective is modified by an intensifying adverb, most notably so, as and too. We can use this information to do a lexical search in a corpus which has not been enriched with detailed syntactic (or even word-class) information. A search was made in the British National Corpus (BNC) containing 100 million words of present-day British English. The BNC has been tagged automatically. Unfortunately, however, using the client/server software tool SARA, tags can only be used in searches in combination with a specific word.17 First of all, I tried to find occurrences of fronted AJPs in which the premodifier was realized by that, however, less, and ever so, since their possibility was suggested by Aarts and Aarts (1988) but they were not found in the corpus, nor tested in the experiment. Additionally, I searched for more + adjective in fronted position, since the results of the experiment for this combination were not straightforwardly interpretable. Of all combinations occurrences were found in the BNC, except for ever so + adjective. Of that + adjective 34 occurrences were 17
For more information on the British National Corpus and SARA see Burnard (1995), and Aston and Burnard (1998). I am also grateful to Peter Schneider, Sebastian Hofmann, and Hans Martin Lehmann for letting me use the web-interface to SARA (BNCweb) they developed at Zürich University.
112
Chapter 4 -The elicitation experiment
found as fronted modifier in the NP, 51 occurrences of however, 9 of less, and 36 of more. Clearly, all these intensifying adverbs are possible modifiers of fronted adjectives. The fact that the combination of ever so + adjective was not found as fronted modifier does not necessarily mean that it is impossible, but it does indicate that it is extremely infrequent. Examples from the BNC of the combinations found in fronted position are given in (44) to (47). (44) (45) (46) (47)
And she’s not that good an actress. [G0A 3041] Parents should not try to monopolize their teenagers, with however good a motive. [BLW 1262] The plasma membrane is also thought to be far less rigid a structure than originally proposed. [HSB 113] Otto was speaking - spluttering would be more apt a description - with his mouth full. [FAT 1608]
Secondly, I searched for as...a, so...a and too...a in a span of 5 words, specifying that as, so and too are adverbs (tagged AVO) and a is an article (tagged ATO).18 This yielded 1157 occurrences of a fronted premodifier realized by as+adjective of which 1075 form the first part of a discontinuous modifier, 764 occurrences of so+adjective of which 189 introduce a discontinuous modifier, and 1167 occurrences of too+adjective of which 299 function as the first part of a discontinuous modifier. The frequencies for so, as, and too + adjective in fronted position is clearly much higher than for the other intensifying adverbs. It is therefore not strange that we found only these occurrences in the corpus study, and not the others. Having now collected a total of 3218 extra occurrences of NPs with a fronted premodifier, we can use these occurrences to check whether the results of the elicitation experiment regarding, for example, the distribution of NPs with fronted premodification (and discontinuous modification) are not disproved, or to find actual examples of NPs which are possible according to the results of the elicitation experiment, but which were not found in the first corpus study. In the corpus no occurrences were found of an NP with a fronted premodifier or a discontinuous modifier functioning as indirect object, object complement, or adverbial. The results of the elicitation experiment, on the other hand, indicated that such NP structures are perfectly acceptable in all sentence functions tested. A study of the examples in the BNC yields 35 occurrences where an NP with a fronted premodifier functions as object complement (exs. 48-50), of which 24 are also discontinuous; five NPs functioning as adverbial (exs. 51-52); and no NPs with fronted premodifier functioning as indirect object. (48) (49)
18
It therefore gives little hint of what makes its author so extraordinary a figure. [ABF 561] … you will find motherhood not quite as difficult a task as I suspect you were anticipating [CDE 520]
Query = [(((“so”=AVO)*(“a”=ATO))/5)]
4.4 - The data cycle continued (50) (51) (52)
113
… the generic use of data and methods should not make this too surprising an observation. [HPU 406] I knew her so short a time. [HGV 2918] As fine a singer as he was, it was Robinson’s writing and producing talents that would be valuable to Berry Gordy, boss of the fledging Motown operation. [CAT 1025]
Contrary to the expectation based on the results of the elicitation experiment, the BNC material also includes examples in which a fronted premodifier co-occurs with a modifier in prototypical position. The premodifier is either realized by a (one-word) noun phrase (53-55) or realized by a (one-word) adjective phrase (5659). The premodifier can also have multiple realizations or be coordinated (6061). (53) (54) (55) (56) (57) (58) (59) (60) (61)
On the other hand you may be expecting too fast a weight loss. [AD0 1187] ... in putting together so complex a weapon system. [ABA 914] Never before had Ramsey run up against so formidable a Christian opponent,... [A68 997] Never before had there been so savage a fiscal squeeze; [A66 691] The mother was as perfect a physical specimen as we could find. [AN8 2164] ... it was strange for so lovely a young woman to be leading a life devoid of romance. [FS1 608] So obvious a structural fault must be easy to diagnose and put right. [H0E 213] ... ‘The World Is Turning On’ is as dandy a catchy guitar pop record as anything the Teenage Fannies have concocted. [CK4 1111] ..., handbells provide ‘as good a practical and possible evangelistic tool as a church can have’. [FPY 1096]
On the basis of these examples we can conclude that the fifth hypothesis for NPs with a fronted premodifier was wrongly accepted. Still, the fact remains that sentence (22) was judged as unacceptable by the informants, indicating that not just any adjective can co-occur with a fronted premodifier. Probably, a further restriction applies to the adjectives and nouns that can co-occur. What is striking is that more than fifty per cent of the premodifiers that co-occur with a fronted premodifier is realized by a noun phrase, whereas normally less than thirty per cent of the premodifiers is realized by a noun phrase. Premodifying noun phrases are classifying; that is, they create “a subclass of the class denoted by the head of the noun phrase” which they modify (Aarts and Aarts, 1988: 63). Adjective phrases are often descriptive; they describe “the referent of the noun phrase in terms of a particular quality of the referent”. However, if we look at the examples above (and indeed all the other examples found in the BNC), we find that the underlined adjective phrases are here classifying. Exceptions may be young in sentence (58) and practical and possible in sentence (61). Young is typically a
114
Chapter 4 -The elicitation experiment
descriptive modifier (cf. ex. 62), but in combination with words like lady, man, woman it forms a lexicalized expression and is more like a classifying than like a descriptive modifier. Descriptive adjectives can themselves usually be premodified by an intensifying adverb, but classifying adjectives cannot. Young in (62) is thus clearly descriptive, since it can be modified by very. For young in (58), however, premodification is unacceptable (cf. ex. 58’). The same is true for practical and possible in sentence (61). The description of NPs with a fronted premodifier (cf. Figure 4-1, p. 108 and Figure 4-3, p. 109) should thus be extended to include a prototypical premodifier position which can be realized by a classifying NP or AJP. (22) I had so romantic a long evening yesterday. [9] (62) I met a very young boy yesterday. (58’) *... it was strange for so lovely a very young woman to be leading a life devoid of romance. A second reason to continue the data cycle is to test new hypotheses which resulted from the elicitation experiment. The data from the interviews, for example, indicate that fronted premodification, and especially the AJP introduced by so, is associated with written, rather than spoken English. It is felt to be formal and even poetic usage. We would thus expect to find the majority of the occurrences of the fronted premodifier in written material, in those genres in which the author wants to evoke a formal atmosphere or a poetic, romantic one. Is this expectation justified by the distribution found in the BNC? The BNC includes a spoken component of approximately 10 million words and a written component of approximately 90 million words. Both components comprise different styles and genres. In the design of the BNC, the written component is subdivided into imaginative texts and informative texts. The imaginative texts are fictional, literary and/or creative texts. This text type corresponds to my ‘fiction’ genre. All other texts belong to the class of informative texts, which corresponds roughly to my ‘non-fiction’ genre. Table 4-3 gives the distribution of the occurrences of fronted premodification over the different genres in the BNC. Table 4-3: Distribution of NPs with fronted premodifier in the BNC
Written - Fiction - Non-fiction Spoken Total
BNC % 21.91 67.74 10.35 100.00
as...a % 21.95 69.32 8.73 100.00
so...a % 27.09 70.16 2.75 100.00
too...a % 22.28 67.78 9.94 100.00
We see that the distributions of the occurrences of fronted premodification introduced by as and too carefully follow the general distribution of texts in the BNC. The distribution of fronted premodification introduced by so, however, diverges from this distribution in that significantly fewer occurrences are found in the spoken component and more in fiction (imaginative) texts. This finding
4.5 - Conclusion
115
reflects the judgments of native speakers on the status of so + adjective in contemporary British English. This possibly suggests that the combination so + adjective is disappearing from the English language and is being replaced by the combination such (a) + adjective. A third reason to continue the research by further following the data cycle is that we do not want to stop at providing a description of variant NPs alone, but additionally want to provide an explanation for the mobility of constituents. One of the conclusions that can be drawn from the elicitation experiment is that native speakers of English often have no preference for either the prototypical or the variant version of an NP. However, in some contexts or situations, there exists a clear preference for the variant NP. How can this be explained? A thorough investigation of the syntactic characteristics of NPs with a mobile constituent can contribute to studies of word order variation in theoretical linguistics and in the fields of psycholinguistics and discourse analysis. Previous studies in these areas have already suggested explanations for, and restrictions on the mobility of constituents in general. A detailed corpus study of variant NPs can give further insight into the appropriateness of such theories for the mobility of NP constituents. This is the topic of the next chapter. 4.5
Conclusion
The present chapter illustrates the usefulness of the multi-method approach in trying to obtain a complete description of variant NPs. First of all, the results of the elicitation experiment were discussed for NPs with a fronted premodifier, a discontinuous modifier, and/or a floating deferred modifier. Secondly, the experimental data were combined with the data obtained from the corpus study as discussed in chapter 3. As a result, the description of the variant NPs which were included in the experiment could be further specified. Thirdly, it was shown that the description could be even further completed by continuing the research along the data cycle which was described in chapter 2. In the next chapter, a selection of existing descriptions of the mobility of constituents in natural languages will be discussed as well as their appropriateness for the description of variant NPs as it has resulted from the research presented so far.
Chapter 5 The description of mobility 5.1 Introduction In the preceding chapters the mobility of constituents in the English noun phrase was discussed. ‘Mobility’ was defined as the situation in which an immediate constituent (or part of an immediate constituent) of the NP occurs in a different position from the one it typically occupies. The typical position of a constituent is determined by the descriptive model (or theory) that is adhered to. In this study, the traditional description of the noun phrase is taken as a starting point, or, more specifically, the prototypical NP structure as it was defined in chapter 1, Figure 16 (p. 19). In the prototypical structure, the NP is a contiguous sequence of (immediate) constituents, while an immediate constituent (IC) itself is also a contiguous string. In many cases, the mobility of a constituent results in a discontinuous structure, either a discontinuous NP, or a discontinuous immediate constituent of the NP. Discontinuity is here defined as the situation in which the words of an (immediate) constituent (at some level in the analysis of the sentence) are not adjacent to each other. Obviously, whether something is treated as mobility or ‘movement’ of a constituent and whether a structure is considered discontinuous is very much dependent on the prototypical structure that is presupposed. The question whether the phenomenon of discontinuous constituency is a theory-dependent notion is addressed by Bunt (1996). He claims that every description of language must include some kind of division into constituents (whether implicitly or explicitly), since the semantic interpretation of language requires decisions on which elements in the sentence are combined. In that sense, the phenomenon of discontinuity is theory-independent, since non-adjacent parts that semantically seem to belong together do occur in natural languages and hence should be described as such in descriptive theories of language. However, the way in which the phenomenon of discontinuity manifests itself in the various theories or models may differ. Still, most grammar formalisms are explicitly or implicitly based on the idea of a constituent as a continuous sequence of words, being primarily designed to describe continuous constituents and having to take recourse to special operations for dealing with discontinuities. This is because concatenation, and therefore the notion of adjacency, plays a central role in most grammar formalisms. The description of
Chapter 5 - The description of mobility
118
structures made up of non-adjacent elements therefore in general presents difficulties. (Bunt, 1996: 4) Almost the same point can be made about the incorporation of some notion of mobility (or movement) of constituents in the various descriptions of language. The order in which words can occur in a sequence is often fixed to a certain extent. Most descriptions of language include a facility to account for fixed word order, and are therefore also forced to account for deviations on the fixed order of words, or even the occurrence of free word order.1 While a description in terms of fixed word order and immediate constituents makes the phenomenon of discontinuous constituency explicit, it is in fact the interpretation of the linear production of language that causes the difficulties. In this chapter, I discuss the treatment of mobility and discontinuity in various descriptive models. More specifically, I examine the ways in which variation in word order is described in the formal and the functional approaches to language (sections 5.3 and 5.4 respectively) and evaluate those approaches based on the occurrences of mobility in the NP. In section 5.5, I discuss the various approaches in the light of their explanatory function for the occurrence of mobility in the NP. I start by discussing the different choices that linguistic models make with respect to meta-theoretical issues which influence their general approach to the description of mobility (section 5.2). 5.2 Meta-theoretical issues which influence the description of mobility While the phenomenon of mobility/discontinuity is theory-independent, the way in which it is described or accounted for is theory- or description-dependent. Linguistic models differ in the approach they take to constituency, in their focus of description (performance or competence), and in the number of different levels of description they incorporate in their model (deep structure vs. surface structure; syntax, semantics, and pragmatics). In this section, I discuss some of the major approaches that can be found regarding these distinctions and their influence on the various treatments of mobility. 5.2.1
Constituency
Descriptions differ in what exactly they consider to constitute a ‘constituent’ (or unit of description) and thus in their conception of movement and discontinuity. Most grammars that adopt a description in terms of ICs also require them to be continuous sequences. Some models (such as Hudson’s Word Grammar) base syntactic structure on grammatical relations between individual words, rather than on constituent structure.2 However, even in Word Grammar words (‘dependents’) 1
Interestingly, discontinuity is also recognized in (relatively) free word order languages such as Finnish. See for example Dowty (1996). 2 See Hudson (1984, 1990).
5.2 - Meta-theoretical issues
119
are required to be adjacent to their heads, thereby forcing a partly fixed ordering of words.3 The problem of discontinuity for IC analysis was recognized almost as soon as IC analysis was introduced by Bloomfield and it has since received a great deal of attention. Huck and Ojeda (1987) distinguish three major strategies that have been adopted by linguistic models where discontinuity is concerned: 1. 2. 3.
allow for discontinuous structures in IC analysis or phrase structure rules represent constituency and non-contiguity at different syntactic levels (multilevel syntax) disallow discontinuity at the syntactic level, but allow non-contiguous elements to be mapped onto one semantic representation.4
One of the first to adopt a definition of discontinuous constituency in IC analysis was Wells (1947).5 Wells extended rigid binary IC analysis by allowing for multiple (three of more) ICs as well as by allowing for discontinuous ICs. However, in order not to make the analysis of sentences into immediate constituents a tremendously intricate affair, he proposed a restriction to discontinuity that requires that a discontinuous structure has a continuous counterpart with virtually the same meaning: A discontinuous sequence is a constituent if in some environment the corresponding continuous sequence occurs as a constituent in a construction semantically harmonious with the constructions in which the given discontinuous sequence occurs. (Wells, 1947: 104) Interestingly, roughly the same assumption, that a discontinuous constituent has an (underlying) continuous counterpart, motivates the rejection of discontinuity in phrase structure (deep structure) analysis in Transformational Grammar. The phrase structure rules in transformational grammar create only continuous sequences in deep structure (DS). The final surface structure (SS) of a sentence is derived from DS through (one or more) transformation(s) (Figure 5-1, p.120). While the order of constituents in SS may be different from that in DS, SS is also always a continuous sequence. In this model, the mobility of constituents is accounted for by means of movement transformations. Transformational Grammar is thus an example of a linguistic model that follows the multi-level syntax strategy to account for discontinuity.
3
The definition of adjacency is rather flexible however: “D is adjacent to H provided that every word between D and H is a subordinate either of H, or of a mutual head of D and H” (Hudson, 1990: 117). 4 As an example of this approach Huck and Ojeda give Gazdar’s (1981) view on extraposed relative clauses. I will not discuss this approach here, but see section 5.2.3 for a discussion of levels of description other than the syntactic level. 5 Before him, Bloomfield and Pike had also discussed the topic.
Chapter 5 - The description of mobility
120
In early generative grammar there were essentially no constraints on the distance that the deep structure of a sentence might be from its surface structure, nor was there any reason to expect that the two levels would share any particular principles of organization. (Newmeyer, 1996: 162) However, over the years, other models have developed out of the generative framework that generate SS directly. Examples of such models are Lexical Functional Grammar (LFG; Kaplan and Bresnan, 1982) and Generalized Phrase Structure Grammar (GPSG; Gazdar et al., 1985). What is more, the generative framework itself has also been developing gradually in the direction of a more surface-structure approach. With the introduction of traces in SS, deep structure was no longer needed to derive the semantic interpretation of a sentence. In Government and Binding (GB; Chomsky, 1981) and other theoretical approaches that adopt similar theoretical principles, deep structure phrase markers are related to surface structure phrase markers by application of transformations (move a). The movement transformations leave behind traces which are co-indexed with their antecedents and the movement operations result in phonetic gaps at the movement sites. In theory, it is possible to base generate the SS with traces without having to explicitly postulate DS. In the Minimalist theory (Chomsky, 1995), the levels of DS and SS no longer have grammatical significance. They now form the path of ‘overt syntax’. Overt syntax constitutes those parts of the system that are relevant to both the PF (Phonetic Form) and the LF (Logical Form) of language. At ‘Spell-Out’ the derivation of a particular linguistic expression splits into a track leading to PF and a track leading to LF. PF and LF are now the only recognized linguistic levels (Figure 5-2). DS SS PF
{ {
overt syntax
Spell-Out
covert syntax
LF
sem. interpret.
Figure 5-1: GB Theory
PF
LF
sem. interpr.
Figure 5-2: Minimalist Theory
The elimination of DS and SS as levels over which grammatical principles can be defined has some implications for the treatment of mobility in the generative framework.6 Prior to Minimalism, the mobility of constituents was (determined and) restricted by locality conditions, such as the subjacency principle, and island conditions. These conditions apply on the SS representation. However, in the minimalist theory, conditions on SS phrase markers are no longer possible. “This requires reanalysing such cases in terms of conditions on derivations, (rather than 6
See also Hornstein (1995).
5.2 - Meta-theoretical issues
121
representations), showing that these principles actually apply at LF and/or eliminating large classes of movement rules” (Hornstein, 1995: 63-64). This last option is very much in line with Chomsky’s basic argument that the model needs to be as simple as possible. It is thus to be expected that the view on mobility in the generative tradition in the coming years will change drastically.7 In (a monolevel) IC syntax, the mobility and especially the discontinuity of constituents requires some kind of adaptation or technique to handle this deviation in the syntactic organization of the sentence. In practice, models often adopt one or more of the following strategies: 1. 2. 3. 4.
express relations between constituents adopt (co-indexed) ‘traces’ in the tree structure allow for crossing branches in the tree structure change/relax the notion of constituency
The dislocation of constituents can partly be made visible by explicating the relations constituents have to each other. If, for example, the basic word order in a language is Subject-Verb-Object, but the object is ‘moved’ to the beginning of the sentence (and the subject to the end), the mobility is clear from the fact that the order of two constituents between which a certain relation holds (here e.g. the verb and its object) has changed, while the mobility is not apparent from the order of the categories (NP-VP-NP), which remains the same. If a constituent has moved to another structural level, for example an object is moved outside the clause which contains the constituents with which the object is in direct relation, the position of the moved constituent will also have to be explicated, for example by using a term like raised object. This approach was adopted in this study where, for example, a modifier which prototypically follows the head, but which has moved outside the boundaries of the noun phrase towards the end of the sentence, is still considered to hold the same relation to the head (and other constituents of the NP), while the name indicates its non-prototypical positioning (floating deferred noun phrase postmodifier). The adoption of traces to indicate the prototypical position of moved constituents (and to control movement) is used in many generative theories and monolevel theories that have developed out of the generative framework. Additionally, some suggestions have been made within the generative framework to allow for discontinuous trees. McCawley (1982) suggests distinguishing two different types of movement operations: relation-changing operations and orderchanging operations. Relation-changing operations involve a change in constituent structure, but not necessarily constituent order, while order-changing operations change the order of constituents without changing constituency itself, thereby giving rise to the possibility of discontinuous structures. For GPSG, Ojeda (1987) argues that the recognition of phrase markers exhibiting 7
Indeed, at the moment, generative linguists differ in opinion on whether both overt and covert movement exists or only one of the two, and on whether there is both leftward movement and rightward movement, or only leftward movement. See also section 5.3.
Chapter 5 - The description of mobility
122
discontinuity (and multidominance) would avoid some of the problems which have arisen in the GPSG account of unbounded dependencies. One problem that arises with the adoption of discontinuous constituency in IC analysis is that the representation is not straightforward. Discontinuous trees cannot (easily) be translated into labelled bracketings. In tree representations, some kind of crossing-branches representation has to be introduced (cf. Figures 53 and 5-4).8
a better solution
than we had expected
Figure 5-3: Discontinuous representation McCawley, 1982)
P:VP
Su
Wells
(taken
from
Det.:Indef.art......................................... A
Su
S
following
Prim.aux............ had Prim.aux............ been LV..................... found
Dim Su:NP
H:N.......... solution
Dim: Adj.P
Dim
better than we had expected
Figure 5-4: Discontinuous presentation following Aarts and Aarts (1988) of the sentence A better solution had been found than we had expected. The formalization of linguistic theories brings to light the problem of achieving an explicit description of constituent structure, which is flexible enough to deal with discontinuity, while at the same time restrictive enough to control the ambiguity of the analyses. Phrase structure rules generally abstract over (immediate) constituents which are continuous sequences. That is, they determine the immediate dominance and linear precedence relations of continuous (sub-) trees. A node x can only precede a node y if all daughters of x precede the daughters of y. Also, a node is required to either precede or dominate another node, but it cannot do both. This excludes discontinuity in a tree. However, over the past two decades, some attempts have been made to create formal descriptions of discontinuous constituency structures. Most notable is Bunt’s Discontinuous Phrase Structure Grammar (Bunt, 1991, 1996). An important development in the formalization of discontinuity and mobility is the separation of phrase structure rules into immediate dominance (ID) rules and linear precedence (LP) statements. The ID/LP format was first introduced for GPSG, and has later been adopted in other generative theories. Instead of trying to include discontinuity in rigid IC analysis, one can also try to get rid of phrasal constituents in the description of language altogether. This 8
See also Figure 1-3 (p. 13) for Nida’s representation of discontinuity.
5.2 - Meta-theoretical issues
123
approach is adopted for example in Dependency Grammar (Mel’ uk, 1988) and Word Grammar. Instead of breaking the sentence into phrases, these frameworks establish links between individual words. Most dependency theories still require that a word plus all its dependents form a contiguous substring, thereby disallowing discontinuity. However, Covington (1990) presents a dependency parser that merely prefers words and their dependents to be continuous, but does not require them to be so. Covington’s approach to discontinuity closely resembles Hudson’s definition of the adjacency principle (cf. footnote 3). 5.2.2
Descriptive aims
Descriptions differ in their descriptive aims. They either want to model the competence or the performance of speakers. In a strict performance description it is sufficient to account for the structures that (can) occur in language use. In other words, the grammar is observationally adequate.9 However, if the aim is to describe the competence of the idealized native speaker, the grammar should also possess descriptive and explanatory adequacy.10 In other words, the grammar not only describes the well-formed sentences, it also explains why one structure is well-formed while another is considered to be deviant. In a performance description of mobility, knowledge about conditions on the mobility of constituents can serve the purpose of restricting the ambiguity of the grammar, that is the number of possible analyses suggested for a sentence. In a competence description, conditions are an integral part of the grammar, since they can explain why a structure is ill-formed. Also, in a competence description it is important to explicate in some way the relation between different word orders, for example, the relation between continuous and discontinuous structures, while in a description that aims to be observationally adequate only, the relation between variants is in principle uninteresting as long as all variants can be accounted for. If the aim is to achieve a universal description of language, the conditions that apply to the mobility of constituents will have to be as general as possible. The description of the NP in this study is in essence a description of language use which aims to be both observationally and descriptively adequate. The objective is to describe all possible NP structures in English, without necessarily having to explain why other structures are impossible (or nonoccurring). Conditions on the mobility of constituents in this context have the primary function of restricting the ambiguity of the description of the NP and not so much of explaining the occurrence and non-occurrence of constructions. It is, however, not unimaginable that an explanation of why certain structures are used can give insight into formal restrictions on NP structures.
9
In practice, grammars that aim to describe language use try to be both observationally and descriptively adequate. In other words, they produce analyses that harmonize with linguists’ intuitions about the structure of sentences. 10 For an explanation of the different types of adequacy, see Chomsky (1965).
Chapter 5 - The description of mobility
124 5.2.3
Level(s) of description
A third characteristic in which linguistic models differ is the levels of description that they assume to constitute part of ‘grammar’ and the context in which they seek to find an explanation for the structuring of sentences (and thus for the occurrence of mobility). The levels of description that models have at their disposal are the phonological level, the syntactic level, the semantic level, and/or the pragmatic level.11 So far in this study we have been mainly concerned with the syntactic aspects of the mobility of constituents in the English NP. The definition provided for discontinuity is also purely syntactic.12 However, we have also seen that others have opted for a more semantically motivated definition. Indeed, for a description and/or explanation of mobility, the semantics and the information structure of the sentence seem to play an important role. The semantic interpretation determines what we consider to belong together in a sentence, in other words what we consider to constitute a constituent and thus what we consider to be a discontinuous constituent. The information structure of a sentence largely determines the position and ordering of the constituents. One of the main questions in pragmatic studies is: why can one and the same proposition be expressed by two or more sentence structures? For this reason, word order flexibility has received a fair amount of attention in discourse pragmatic (or functional) literature. Unfortunately, syntax and discourse pragmatics are often studied independently of each other. A notable exception is the work by Lambrecht (1994), and I subscribe to his view on the combination of syntax and discourse pragmatics: Syntax may be autonomous in its own domain, but by its nature it must provide the resources for expressing the communicative needs of speakers. Therefore its nature cannot be fully understood unless we explain the principles which determine its function in discourse. In my view, the most promising but perhaps also the most difficult approach to grammatical analysis is one in which the different components of grammar are seen not as hierarchically organized independent subsystems but as interdependent forces competing with each other for the limited coding possibilities offered by the structure of the sentence. (Lambrecht, 1994: 11-12) 11
The pragmatic level is sometimes also referred to as the functional level, the discourse level, or the information level. 12 Note that, while the definition of discontinuity is syntactic, the notion of constituency in itself presupposes that both syntactic and semantic notions apply.
5.3 - Word order variation in the formal approach
125
Examples of linguistic models which attempt at an integrated description of various grammatical domains are Lexical Functional Grammar (Kaplan and Bresnan, 1982), Functional Grammar (Dik, 1989), Role and Reference Grammar (Van Valin, 1993a), and Givón’s functional-typological approach (Givón, 1984, 1990). Related, but taking a slightly different perspective, is Hawkins’ processing approach to syntax. His approach is also functional, but from a more psycholinguistic point of view. It explores the effect on sentence structure of the necessity to process sentences in real time. In the next sections, I discuss some of the treatments of mobility that are found in the literature. I will distinguish between the formal approach and the functional approach.13 The overview of the literature is by no means exhaustive, but the discussion tries to give a broad view of the types of structures that are generally ranged under the mobility phenomenon and the nature of the conditions that have been formulated to restrict their description or explain their occurrence. 5.3 Word order variation in the formal approach In the linear structure of the sentence, words or constituents can, in principle, move out of their basic or prototypical position in two ways: they can move anywhere to the right of their prototypical position, or they can move anywhere to the left. Rightward movement is associated with right extraposition, right dislocation, postposing, deferral, and/or floating. Leftward movement is associated with left extraposition, fronting, raising, left dislocation, preposing, and/or topicalization. Movement can be either optional or obligatory. In generative theories movement phenomena have most often been explained by the adoption of adjunction. Constituents which are moved to the left are left-adjoined to some projection, while constituents which are moved to the right are rightadjoined. Constituents can either be base-adjoined, or move to that position as a result of move a. Figure 5-5 gives an example of right-adjunction. Figure 5-6 (p. 126) gives an example of (multiple) left-adjunction where two adverb phrases (AdvP) are adjoined to the verb phrase (VP).14 VP V’ V
AdvP NP
read this book very often Figure 5-5: Example of right-adjunction
13
The formal approach is mainly determined by discussions of mobility in the (broad) generative framework, since it is there that the phenomenon of mobility has received the most attention. 14 The examples are taken from Haegeman and Guéron (1999: 80-81).
Chapter 5 - The description of mobility
126 VP VP
AdvP
VP
AdvP
V’ V
NP
probably always remember this story Figure 5-6: Example of (multiple) left-adjunction With the introduction of ‘The Antisymmetry of Syntax’ (Kayne, 1994) and the Minimalist Program, the discussion of movement has regained in vigour. Kayne proposes a restrictive theory of word order in which linear order is always completely determined by phrase structure. He argues that (prototypically) “complements must always follow their associated head and that specifiers and adjoined elements must always precede the phrase that they are sister to” (Kayne, 1994: 3). In Minimalism, a constituent moves in order to satisfy its morphological properties. An element is extracted from the lexicon containing certain morphological features. These features have to be satisfied (‘checked’) somewhere in the process of derivation. Languages can differ in the point at which features are checked, before or after Spell-Out.15 All features need to be checked at LF for the output to be valid. If they are not all checked, the derivation ‘crashes’. A constituent can move to more than one landing site during a derivation. Spell-Out position of the moved constituent is the highest position with strong features. At Spell-Out the order of constituents is as we perceive it in language use. In other words, the constituent order does not change after SpellOut. As was said above, a constituent can move to the right or to the left of its prototypical position. However, in Antisymmetry and Minimalism the direction of movement is restricted to the left. By excluding the possibility of rightadjunction in Antisymmetry, movement of constituents to the right is impossible. In the Minimalist theory, movement is always leftward, since constituents (‘trees’) can only move to head and specifier positions to the left in the tree.16 The removal of rightward movement from a model implies that all structures that were previously assumed to be the result of rightward movement have to be reanalysed in terms of stranding and/or leftward movement. There exists some disagreement 15
To be more precise, languages differ in whether a feature is weak or strong. Strong features have to be checked before Spell-Out, weak features are checked after Spell-Out. 16 See Veenstra (1998) for a discussion and a formalization of leftward movement in the Minimalist Program.
5.3 - Word order variation in the formal approach
127
on this point, however. Alphonce and Davis (1997), for example, argue that movement is inherently non-directional. They further argue that linear order is not determined by syntactic principles. Instead, it is the processing mechanism that determines the order of constituents. In other words, linear order is an issue of performance, not of competence. In their approach, LF structures are strictly hierarchical and linearity constraints are derived from morphological, phonological and/or processing considerations. A strong constraint on all types of rightward movement is that they are ‘upward bounded’, i.e. a rightward moved element cannot move beyond the first sentence (or clause) up. Upward boundedness was recognized by Ross (1967). The constraint is known as the ‘Right Roof Constraint’ and is considered to be an important constraint on rightward movement in general. Another important constraint on movement suggested by Ross is the Complex NP Constraint which prohibits extraction out of an S’ dominated by an NP that contains a lexical head noun. As was argued by Rizzi (1982), languages differ in the distance a moved constituent can cover. Wh-movement in English, for example, can cross one IP at most, while in Italian the distance of wh-movement is at most one CP. This difference in the distance of movement is known as the Subjacency Condition. The subjacency condition seems to hold for leftward movement in general, but is less easily generalized over rightward movement. The removal of rightward movement simplifies the treatment of movement. As we shall see in section 5.5.3 for example, analysing extraposition out of NP as ‘stranding’ instead of movement to the right allows for a simple account of the restrictions imposed on this construction by the Right Roof Constraint. In language typology and traditional descriptive linguistics a mainly morpho-syntactic approach is followed in finding explanations for word-order variation. Following Greenberg’s (1963a) pioneering work, languages have often been classified as having a fixed basic order of subject, verb and object in simple clauses.17 What is more, the order of verb and object in the simple clause is also claimed to determine the order of modifier and noun in the NP: OV-order: GenN/AdjN/RelN/DemN/NumN VO-order: NGen/NAdj/NRel/NDem/NNum English, for example, has a basic word order of subject-verb-object (SVO). Since the verb precedes the object, the basic order in the NP should be noun-modifier. Clearly, if we take this order as the starting point of a study of mobility in the noun phrase instead of the prototypical structure defined in chapter 1, the result would be quite different. For instance, every modifier occurring before the head noun would now be the result of the mobility of that modifier to the left. In other
17
For an extensive discussion of (the different interpretations of) the term ‘basic order’ see Siewierska (1988).
Chapter 5 - The description of mobility
128
words, the basic word order that is adopted in a linguistic theory is essential for that theory’s treatment of the mobility or ‘movement’ of constituents. Since most languages do not conform fully to the harmonic structure suggested by Greenberg, other motivations had to be found to account for the exceptions. One of the more important explanations for both phrase and clause constituents is the heaviness-hierarchy introduced by Hawkins (1983): Rel < Gen < Adj < (Dem, Num) According to this hierarchy relative clauses are the ‘heaviest’ constituents, while demonstratives and numerals are the ‘lightest’. A relatively light modifier prefers to precede the head, while a relatively heavy modifier follows it. In traditional descriptive linguistics the preference of a heavy constituent to occur at the end of a clause is indicated by the term ‘end-weight’ (Quirk et al., 1972; 1985).18 Neither in the typological nor in the descriptive approach is it clear exactly what it is that makes a constituent ‘heavy’. Hawkins suggests that the heaviness of a constituent is dependent on: a. b.
the number of syllables/words the structural complexity
However, which of these factors is the most important or how they interact is not explicated. 5.4 Word order variation in the functional approach Beside the morpho-syntactic explanation of end-weight, traditional grammars give a more discourse-oriented explanation of ‘end-focus’. This corresponds to the approach that is adopted in functional descriptions of language which concentrate on the functional notions of ‘topic’, ‘comment’ and ‘focus’. These studies depart from the idea that a sentence is prototypically constructed from constituents with low information value at the beginning, to those with high information value at the end. The portion of the clause which contains the old information (‘topic’, ‘theme’) precedes the portion which contains the new information (‘comment’, ‘focus’). Birner and Ward (1998), for example, argue that “constructions involving some preposed constituent require that constituent to represent information that is old in some sense (either to the hearer or to the discourse, and either relatively or absolutely), while constructions involving a postposed constituent require that constituent to represent information that is new in some sense” (Birner and Ward, 1998: 16). Two constructions that do not conform to this general rule are Left Dislocation (LD) and Right Dislocation (RD). Left-dislocated constituents (usually) represent new information and rightdislocated constituents represent information that is given in some sense. Givón 18
At sentence level the principle of end-weight has its analogue in the principle of resolution (Quirk et al., 1972: 791).
5.5 - Mobility in the NP
129
(1990: 757) states that “left dislocation is typically a device to mark topics - most commonly definite - that have been out of the focus of attention for a while, and are being brought back.” Right dislocation is often looked upon as an ‘afterthought’ or ‘repair’ device.19 The speaker uses a pronoun, but realizes that the referent may be unclear and thus he/she provides an explicit referent. This interpretation of right dislocation is supported by the fact that the distance between the pronoun and the dislocated element is small (Givón, 1990: 761) and that the dislocated element is (mostly) preceded by an intonation pause. LD and RD are not only functionally distinct from other types of movement but they are also syntactically distinct in that they have a pronoun in their prototypical position with which the left or right dislocated constituent is co-referential. Sentence (1) is an example of left dislocation and sentence (2) of right dislocation. John, I never saw him there He used to own a house in London, my brother.
(1) (2)
Over the past few years, a more cognitively-based understanding of word order variation has emerged from the fields of psycholinguistics and functional linguistics. Explanations are sought in the various ‘principles of information processing’. These include general psycholinguistic processes with regard to the production and interpretation of an utterance in real time and the capacity of our short-term memory. When we produce an utterance, it is customary that we surround the new information that we want to communicate with old, alreadystored information. This enables the hearer to interpret the new information correctly and to efficiently store it in memory (‘grounding’). 5.5 Mobility in the noun phrase In this section, the findings of the research reported on in this study are discussed in the light of the treatment of movement as found in the formal and the functional approaches discussed above. It is not attempted to give complete descriptions of the various types of mobility in the existing theories. 5.5.1
Fronted premodification
Fronted premodification has received little attention in the literature. While some of the traditional and structural descriptive studies give examples of adjective phrases preceding an indefinite article (cf. Kruisinga, 1909-1932; Nida, 1966; Quirk et al., 1972, 1985; Aarts and Aarts, 1982), most descriptions lack an elaborate description of fronted premodification. Quirk et al., for example, restrict their discussion to one occurrence of so plus adjective of which they say:
19
For example Tomlin (1986), and Geluykens (1987).
Chapter 5 - The description of mobility
130
However, in rather formal contexts, so plus adjective can be placed before the indefinite article [3b]: so beautiful a daughter
[3b] (Quirk et al., 1985: 1323)
The most elaborate discussions are found in Aarts and Aarts (1982; see chapter 3, section 3.3.3) and Kruisinga (1909-1932, part 3): When an attributive is qualified by so, as, how, however, too, the adverbs with the adjectives have emphatic preposition, the noun following with an indefinite article. ... Occasionally, the adjective has mid-position. An only too vivid memory. Malet, Calmedy But his face and voice made a so deep impression that during the next few minutes I ordered many pairs. Galsworthy, Caravan p. 183 (Kruisinga, part 3: 227-228) In chapter 4, we established that the fronted premodifier is always realized by an intensifying adjective phrase (either (identifying) such or an adjective modified by an intensifying adverb phrase) and that the fronted modifier is indeed followed by an indefinite article. The head of the modifying adverb phrase is preferably realized by one of the adverbs as, so, too, how, or this, although realization by more, less, enough, how(ever), and that are also found. Additionally, the AVP head can be preceded by another premodifying AVP. In the generative framework, the occurrence of fronted premodification has received attention from Bresnan (1973) and Abney (1987). A fundamental difference in these two studies is that Bresnan takes the lexical head hypothesis as a starting point, while Abney opts for the functional head hypothesis. In essence this means that Bresnan presumes that the adjective is the head of the AJP (or AP) and that the intensifying adverb phrase or, to be more precise, the degree element (DegP) occupies the specifier position of the lexical projection AP (Figure 5-7, p. 131). Abney, on the other hand, assumes that the AJP is a projection of the degree element (Figure 5-8, p. 131).20
20
In the same way that an NP is seen as a projection of the determiner (see also chapter 1, section 1.2.3). Arguments in support of the DegP-hypothesis were given by Corver (1990, 1991) and in the rest of the discussion I will adopt this analysis.
131
5.5 - Mobility in the NP AP
DegP
DegP Spec
A’ Deg’
A
Spec XP
Deg Figure 5-7: Lexical head hypothesis
Deg’ Deg
AP Spec
A’
A XP Figure 5-8: Functional head hypothesis
Before we go on to discuss the treatment of fronted premodification in the generative framework, let us first recapitulate the different positions an intensifying adverb + adjective can take in the structure of the NP: it can occur before the indefinite article (ex. 3); too + adjective can follow the determiner (ex. 4)21; adverb + adjective can follow the head of the NP if the adjective phrase includes a postmodifier (ex. 5); occasionally, the entire adjective phrase including the postmodifier is found preceding the article (ex. 6); the adjective phrase can occur discontinuously (exs. 7-8) and the adjective phrase postmodifier can float outside NP boundaries towards the end of sentence (ex. 9). (3) (4) (5) (6) (7) (8) (9)
And with your toys, Capitano Trent, how small a target can you hit? Again, however, we must be careful not to reach a too hasty conclusion. We visited our farming cousins and enjoyed the delights of a life so different from our own. ... it covers one’s stance to as wide as possible a range of phenomena. The plasma membrane is also thought to be far less rigid a structure than originally proposed. Is that too small a task for you to do? So crucial a part was Gen. Robertson’s signal to play in the events which are the subject of this report that it will be considered in a separate chapter.
So how can we account for the variation of the position of the DegP in the structure of the NP? The first question we have to answer is: where is the DegP base generated, to the left or to the right of the noun? If we follow Kayne, the DegP must be adjoined to noun, since it is a modifier of the noun, and adjunction is always to the left. If we take the NP to be a lexical projection, fronting can be explained by left-adjoining the DegP to the NP, while the determiner is in [Spec,NP] (Figure 5-9, p. 132). However, this does not explain that the DegP can also follow the determiner in some cases (e.g. a too narrow definition). A more straightforward account is reached if we adopt the DP analysis (Figure 5-1). In 21
Kruisinga also gives an example of so + adjective following the determiner, but no occurrences of this were found in the corpus study (including a search of the BNC) and in the elicitation experiment such examples were judged as relatively unacceptable.
Chapter 5 - The description of mobility
132
this analysis, the DegP can move to [Spec,DP], or possibly be left-adjoined to the DP. NP2 NP1
DegP DP
N’
N too vivid an image Figure 5-9: Fronting in the NP analysis DP Spec
D’ D
NP2 DegP
NP1 N’
N a too vivid image Figure 5-10: Fronting in the DP analysis So far, the analysis is in line with Kayne’s ‘Linear Correspondence Axiom’ in that both adjunction and movement are to the left. But how can we explain a discontinuous DegP in this analysis? The only way to achieve discontinuity would be to move (the head of) NP1 to the left, either by means of XP movement or by head-to-head movement. The NP cannot move inside the DegP, since that would violate the ban on movement to a non-c-commanding position. The only other position available is [Spec,DP], but that would place the noun in front of the determiner. In the same way, head-to-head raising can only move the N to D. The next step should move Deg and A to the left, leaving the DegP internal XP behind. However, if we take the DegP structure to be as in Figure 5-8 (p. 131), and if we assume that only constituents can move, it is hard to explain how the Deg and the A can move out of the DegP together, without the XP. It seems to me that we have to postulate either right adjunction or rightward movement or both to explain the facts. Rightward movement can move the DegP internal XP to a position after the head of the NP. But where? Additionally, Deg and A still have to move to the left. It seems uneconomical to adopt an analysis that needs two movements, when a derivation with only one movement is available. Let us therefore investigate an alternative approach that uses the possibility of right-adjunction, but not necessarily of rightward
133
5.5 - Mobility in the NP
movement. In this alternative approach, the analysis of DegP is as given in Figure 5-11, while in the structure of the DP DegP is right-adjoined to NP. Taking the XP to be a complement of Deg seems expedient, since it is the adverb that determines the realization of the XP. In case of as, for example, the XP is a PP or a CP (clause) in which the preposition or the conjunct is realized by as; in case of so it is a PP introduced by as, or a CP introduced by as or that; in case of too it is a CP in which the subordinator is realized by to; etc. DegP2 DegP1 Spec
XP Deg’
Deg
AP Spec
A’ A
XP
Figure 5-11: Alternative DegP analysis Now consider the structure in (10): (10)
It was so different a procedure from the rest of the operation that they often dreaded it.
The PP from the rest of the operation is clearly a complement of the adjective different and thus occupies the AP internal XP position. As a result, movement of Deg and A cannot be an instance of XP movement, since Deg and A do not form a separate constituent in this analysis. Alternatively, movement can be explained by head-to-head movement. A0 first raises to Deg0 , then both raise to D0. Why then does raising take place? Clearly, it is the degree element that triggers raising to D, because without such an element an adjective cannot be fronted. On its own, it seems that the adjective can be raised as far up as N0. Apparently, A-to-Deg occurs before Deg-to-D, since the degree element cannot be raised independently (*so a different procedure that...). There are some exceptions to this, however. Consider the examples in (11) to (14): (11) (12) (13) (14)
quite a good idea such a rude boy rather a better mood than usual quite a pleasant horse to ride
Adverbs like such, rather, and quite differ from adverbs like so, too, as, more, and how(ever) in that they can move independently from the adjective but are similar in the fact that they can be fronted. In other words, all these adverbs have
134
Chapter 5 - The description of mobility
a feature in common, which they can select and which triggers movement to D, but only as, so, too, how, this, more, less, enough, how(ever), and that also trigger the A to move to Deg. In conclusion, I argue that fronting can only be explained adequately in a generative framework if right-adjunction is allowed. In other words, Kayne’s antisymmetry theory is too restrictive. The minimalist restriction of only allowing for leftward movement, on the other hand, can be retained. By adopting the analysis as suggested above, an integrated account can be given of fronted premodification and occurrences of discontinuous modification. Exactly what triggers the adverb and the adjective to move to the left is not clear.22 Neither is it explained why the determiner can only be realized by an indefinite article.23 To my knowledge, fronted premodification has never received attention in a functional description of language. A general principle for constituent ordering that seems to apply in this structure was introduced by Dik (1989): The Principle op Pragmatic Highlighting Constituents with special pragmatic functionality (New Topic, Given Topic, Completive Focus, Contrastive Focus) are preferably placed in “special positions”, including, at least, the clause-initial position. (Dik, 1989: 343) Intuitively I would say that in case of fronted premodification, fronting serves the purpose of giving extra emphasis to the adjective. A first method to give the adjective extra emphasis is by modifying it by means of an intensifying adverb. Fronting seems to reinforce this process. However, it is hard to say what the special pragmatic functionality of fronting is exactly. In general, fronting is often said to concern old or given information. In case of fronted premodification this does not seem to apply. Consider the next lines from The Bloody Wood: ‘I’ve had a damnable day - and you can guess I’m not happy about Mrs Martineau. And now I must be off. Another couple of calls, as a matter of fact.’ ‘I wouldn’t like your job - although I wish I did as useful a one.’ In this example, useful is clearly new information. The whole NP (as useful a one) can be referred to as the focus of the clause. This is supported by the indefiniteness of the NP. In functional approaches, indefinite NPs are said to be typically used to introduce a new referent into the discourse. This would also explain the preference of NPs with a fronted premodifier for the direct object, 22
Movement of the adverb to the left of the article seems to give it extra emphasis. Perhaps the explanation can be found in introducing an (optional) feature <elative> that can be checked by movement of Deg to D. 23 Perhaps head movement to D is restricted to indefinite determiners in general.
5.5 - Mobility in the NP
135
subject complement and prepositional complement positions, and their less frequent occurrence as subject and indirect object. Givón (1979) has studied the distribution of definite and indefinite NPs over sentence functions and has established that indefinite NPs hardly ever occur in the functions of subject and indirect object. The properties of NPs with a fronted premodifier discussed so far seem to suggest that we are dealing with (focus) NPs that introduce new information into the discourse. What is more, the new information may well be the information contained in the adjective, instead of the information in the head noun (cf. the example from The Bloody Wood above where the head one cannot represent a new referent). The fact that it is the adjective which contains the focal information can then explain its non-prototypical position in the NP in terms of the Principle of Pragmatic Highlighting. The first position in the NP may fulfil the same role as the first position in the clause. While these are interesting observations, it would need a more in-depth study of the pragmatic properties of NPs with a fronted premodifier to establish whether these arguments turn out to be valid. It is clear, however, that the general assumption that preposed constituents represent old information does not apply to fronting in the noun phrase. 5.5.2
Discontinuous adjective phrases
While the phenomenon of discontinuity has received a fair amount of attention in the literature, discontinuous adjective phrases are discussed surprisingly little. Surprisingly, since the discontinuous AJP is mentioned by many grammarians as a possible modifier in the noun phrase, and seems a perfect example of discontinuity. This fact is recognized by Baltin (1987), who discusses a related structure, namely ‘degree complements’, which he calls: “symptoms par excellence of discontinuous constituency” (Baltin, 1987: 11). However, he gives a slightly different definition of discontinuity in that he describes it “as a situation in which an element is selected by another element with which it is not contiguous and the intervening material precludes an establishment of continuous constituency between the selector and the selectee” (Baltin, ibid.). In most cases of discontinuity, including most occurrences of discontinuous AJPs, the intervening material does not preclude a continuous structure. Why discontinuity does occur is not clear. It is generally recognized that discontinuous constituents are universally disfavoured. In functional approaches we find that many ordering principles are an elaboration of a more general iconic principle: what belongs together semantically goes together syntactically.24 The occurrence of a discontinuous modifier in the NP is a violation of this general principle and has to be explained in a different way. A possible explanation is offered by Hawkins’ Early Immediate Constituents (EIC) principle. In essence, this principle says that the most favoured order of constituents is that order in which the lowest number of dominated elements has to be scanned for IC recognition, “making this process faster and also more efficient, since the same 24
This principle is also known as Behagel’s First Law. See also Rijkhoff (1990: 5).
136
Chapter 5 - The description of mobility
amount of information is extracted from less input” (Hawkins, 1994: 58). The EIC predicts that, in general, discontinuities of ICs are not preferred. However, if discontinuity of an IC in the NP would shorten the processing of the NP (or even of the clause that contains the NP) the EIC predicts that it can be advantageous in some cases to have discontinuity. Hawkins believes that the major determinant of word-order variation is syntactic weight (or length) and that pragmatic considerations of information structure play no role. If we apply the argument of syntactic weight to discontinuous modification, we would have to argue that in the basic word order of the NP in English the AJP occurs before the head noun. If the AJP includes a (heavy) postmodifier, the whole AJP is placed after the head, because of its relative weight, or the AJP is split in two, in which case the lighter part remains in place, while the heavier part, the adjective phrase postmodifier(s), is moved to the right. What then is the processing gain of discontinuity of the AJP in terms of EIC? The results from the elicitation experiment have shown that the discontinuous structure is often preferred over the version in which the whole AJP is placed in post-head position, while an NP with a postmodified adjective in premodifying position is (usually) judged as unacceptable. The EIC is expressed in terms of IC-to-non-IC ratios. EIC (Expanded) The human parser prefers linear orders that maximize the IC-to-non-IC ratios of constituent recognition domains. Orders with the most optimal ratios will be preferred over their non-optimal counterparts in the unmarked case; orders with non-optimal ratios will be more or equally preferred in direct proportion to the magnitude of their ratios. For finer discriminations, IC-to-non-IC ratios can be measured left-toright. (Hawkins, 1994: 78-79) We would thus expect (a constituent containing) an NP with a discontinuous AJP to receive a higher IC-to-non-IC ratio than its counterparts with a continuous AJP. Now consider the following examples of discontinuous modification: (15) (16) (17) (18) (19)
This is a particularly complicated benefit to calculate. They are a different order of object from the objects they relate when treated as a relationship. But just under an hour ago, the most alarming news of all came through. From the talk then, the violence he advocates is on a somewhat grander scale than merely cutting the throat of a single Tory ex-minister. The flattery was hardly subtle and Lampart was too clever a man to miss it.
What are the IC-to-non-IC ratios for the constituent recognition domains (CRDs) of the sentences in (15) to (19)? The CRD of a node is the set of terminal and
137
5.5 - Mobility in the NP
non-terminal nodes that must be parsed to recognize that node, proceeding from the terminal node that constructs the first IC on the left, to the terminal node that constructs the last IC on the right, and including all intervening terminal and nonterminal nodes that they construct. Thus the CRD of S in (15), for example, starts at This and ends at a, since the NP is constructed (recognized) by its left periphery (following Hawkins).25 Clearly, under these definitions, none of the ICto-non-IC ratios for the CRDs of S are influenced by the discontinuity of the AJP. The same goes for the ratios for VP. The gain should thus be found in the processing of the NP itself. The NP in (15), for example, can in theory have the structures as represented in Figure 5-12 (p. 139).26 If we assume that discontinuity adds an extra IC to the NP (Figure 5-12c), we see that the EIC correctly explains the preferences for the three possible structures.27 EIC can even explain why the fronted and discontinuous structure in (19) is preferred over any of the other possible variants. According to Hawkins’ theory, an adjective phrase can be recognized by its right periphery (the adjective) if no modifier follows. The EIC for an NP is thus higher when the first part of the discontinuous AJP is fronted (Figure 5-13). Following the same line of reasoning, we would also expect the phrase particularly complicated to be preferred in fronted position. However, in that case fronting is impossible. It seems to me that an explanation for this cannot be found in terms of EIC (or weight), but should instead be found in a pragmatic context. NP AdjP
Det
AdvP Adj Adv
N
VP VP V
NP
IC-to-non-IC ratio = 4/5=80%
N too
clever a
man to miss
it
Figure 5-13: IC-to-non-IC ratio for too clever a man to miss it.
25
If the realization of the left periphery can be the intensifying adverb of a fronted adjective phrase, the structure in (19) would get the same score as its counterpart where the adjective is in normal premodifying position. In other words, fronting does not effect ICto-non-IC ratios negatively. 26 Obviously, the IC-to-non-IC ratios depend heavily on the structure that is assigned to a sentence/clause or phrase. In Figure 5-12, I have tried to draw the trees as they would presumably be drawn by Hawkins. 27 The issue whether discontinuity adds an extra IC to the NP or not will not be further pursued here. I would like to remark, however, that from a processing point of view it seems expedient to (try to) start a new IC after having recognized the head of the NP.
138
Chapter 5 - The description of mobility
The way in which discontinuous AJPs can be described in a formal (generative) framework was already discussed in section 5.5.1. So far, we have ignored those discontinuous AJPs whose postmodifier is non-adjacent to the rest of the noun phrase, and occurs further towards the end of the sentence. These structures will be included in the discussion of floating deferred postmodification in the next section. 5.5.3
Floating deferred postmodification
Floating deferred postmodification, or ‘extraposition from NP’ as it is commonly referred to, has received a fair amount of attention in (generative) literature. The treatment of extraposition out of the noun phrase is often related to extraposition of the clause at sentence level (it/there sentences), and combined with extraposition out of the determiner phrase (floating deferred DTP postmodifiers, sometimes also referred to as ‘result clause extraposition’) and occasionally with extraposition out of a (modifying) AJP. For all those constructions a general rule of extraposition is formulated. The phenomenon of extraposition is especially interesting in that it seems to present the strongest case in English for rightward movement.28 Now that the generative model has moved on to a theory that does not allow rightward movement, the phenomenon of extraposition has to be reconsidered. A strong restriction on extraposition is that it is rightward bounded, i.e. an embedded sentence (or PP) can only move to the end of the next sentence up. This restriction is known as the Right Roof Constraint (RRC) and was introduced by Ross (1967). In Ross’ description an (NP postmodifying) clause or PP would be extraposed/adjoined to the end of the S containing the NP, but could never go further. Ross defined the RRC in terms of ‘command’: the constituent to be extraposed has to command the node to which it is right-adjoined. The occurrences in the corpus of floating deferred postmodification, floating deferred DTP postmodification and floating deferred AJP postmodification, all observe the RRC. Akmajian (1975) showed that Ross’ rightward boundedness condition was not strong enough since it could not explain why a PP cannot be extracted from an NP that is embedded in an NP postmodifying PP. (20)
a. A review of a new book about French cooking came out yesterday. b. A review came out yesterday of a new book about French cooking. c. *A review of a new book came out yesterday about French cooking.
Akmajian argued that this fact can be explained by application of the notion of ‘cycle’ to the NP (as well as to the S) and by adopting the restriction that a constituent can only be extraposed one cycle up from the cycle containing it. Whether this restriction holds for all occurrences in the corpus is less clear. The corpus contains a few occurrences of extraposition out of an NP-embedded NP
28
See also Rochemont and Culicover (1997).
139
5.5 - Mobility in the NP (a)
NP Det
AdjP
N
AdvP
AdjP
Adv
Adj
VP
IC-to-non-IC ratio = 3/12=25%
V a particularly complicated to calculate benefit
(b)
NP AdjP
Det N AdvP
AdjP
Adv
Adj
VP
IC-to-non-IC ratio = 3/8=37%
V a benefit particularly complicated to calculate
(c)
NP Det
AdjP AdvP
N Adj
VP V
Adv
IC-to-non-IC ratio = 4/8=50%
a particularly complicated benefit to calculate
Figure 5-12: IC-to-non-IC ratios for the variants of a particularly complicated benifit to calculate.29
29
The broken lines indicate the boundaries of the CRD of the NP.
140
Chapter 5 - The description of mobility
(exs. 21-22). However, in none of the examples is the postmodifier separated from the NP by a VP (as is the case in example 20c) and in none of the examples does the highest NP function as a (non-existential) subject of the clause containing it. The NP cycle of Akmaijan does seem to hold for all such instances of extraposition out of NP. Perhaps then the examples in (21) and (22) are not caused by the movement of the postmodifier, but by the movement of the interrupting adverbial. (21) (22)
And there are a number of things there that she would like to show you. So for instance you’re taking a photograph of a line of people with a flash gun right going away from you
Baltin (1981) shows that a PP which is extraposed from a subject is attached to S, whereas a PP which is extraposed from an object is attached to VP. Along the same lines, Reinhart (1980) argues that a distinction should be made between extraposition of clauses at the level of the sentence (e.g. expletive it constructions), and extraction of clauses out of the NP. While extraposition of a clause at sentence level attaches it to VP, extraposition of a clause out of the NP attaches the clause to the matrix sentence. Baltin (1981) adapts the Rule of Extraposition to the Government and Binding framework and formulates the Principle of Generalized Subjacency to account for restrictions on rightward movement. The principle states that all major category nodes are bounding nodes for rightward movement, while only a subset of these nodes are bounding nodes for leftward movement.30 Guéron (1980) is to my knowledge the first to investigate whether extraposition out of NP actually involves movement. She wonders whether PPs may not be simply base-generated in extraposed position and linked to the NP by interpretive rules. After all, PP extraposition does not leave a visible “gap”. Guéron proposes a constraint on the government of extraposed complements that applies at the level of logical form. McCawley (1982) is the first to consider extraposition out of NP as an example of an order-changing transformation. In other words, extraposition out of NP does not change constituency and the extraposed PP or clause is still dominated by the NP from which it is moved. Culicover and Rochemont (1990) differentiate extraposition out of NP from other types of movement such as wh-movement and topicalization. In the first place, a wh-phrase or a topicalized element may not be related to a gap within a subject NP, whereas an extraposed constituent can. In other words, rightward movement out of a subject is possible, but not leftward movement. Secondly, an extraposed phrase is strictly bounded relative to its antecedent, whereas the others can occur unboundedly far from the gap to which they are related. Furthermore, Culicover and Rochemont propose that phrases extraposed from subjects are adjoined not to IP but to VP and, following Guéron (1980) and Guéron and May (1984), they suggest that extraposition should not be analysed in 30
Obviously, AP cannot be a bounding node since a floating deferred adjective phrase postmodifiers can cross both the AP node containing the postmodifier and the NP containing AP.
5.5 - Mobility in the NP
141
terms of movement at all. Instead, extraposed phrases are base-generated in their extraposed positions and restrictions on extraposition are due to restrictions on the relation between the extraposed phrase and its antecedent. Kayne (1994) suggests a stranding analysis for extraposition which is in line with his general ban on rightward movement. He cannot adopt a base generation approach of the extraposed constituent, since right-adjunction of the constituent to VP or IP is also prohibited in his Antisymmetry theory (nor is a ternary branching structure possible). In Kayne’s approach the whole NP (or more accurately the whole DP) is base-generated to the right of the verb. During the derivation, the NP or part of the NP is moved to the left in order to receive case. The advantage of this approach is that it provides an explanation for movement (namely case assignment), while in older accounts extraposition was a purely optional rule. Additional advantages are that stranding provides a better account of the Right Roof Constraint and that it explains why the relative clause occurs to the right of its head, rather than to the left. The RRC can now be explained by the fact that the part of the NP that is moved cannot move to a non c-commanding position. A position outside the first S up would be a non ccommanding position. In Kayne’s analysis the relative clause is a complement of D and the noun has moved to Spec-CP. An example analysis is given in Figure 5-14 (p. 142). A problem with this analysis is that the article and the noun together do not form one constituent and thus cannot be moved. Kayne uses this aspect to explain the fact that extraposition is impossible (or awkward) with the definite article the (as Ziv and Cole (1974) have observed). In Kayne’s analysis the is positioned in D, while a co-occurs with the noun in Spec-CP. The indefinite article + noun can thus be moved since they form one constituent, whereas the definite article + noun cannot. While extraposition is indeed found more with NPs in which the determiner is realized by an indefinite article (28.3%) or not realized at all (40.8%), 12.8% of the floating deferred postmodifiers found in the corpus are part of an NP in which the determiner is realized by a definite article (exs. 24-26).31 It is hard to see how such examples can be described in Kayne’s theory. (24) (25) (26)
The question arises whether it is possible to combine the two theories. ... until the empire gives the order to the section commanders to withdraw. The word is coming down the line, in a number of ways, that the Soviet Union must kick-start itself into action.
31
Also, extraposition out of a definite noun phrase is possible in other languages. An example is Dutch. Consider for example the following sentences which are taken from Shannon (1993): (i) Dit roept de vraag op waardoor deze categorie gewettigd wordt. ‘This calls the question up what the category is licensed by.’ (ii) Hierboven is de vraag gesteld hoe sociale klassen kunnen worden onderscheiden. ‘Above the question was asked how social classes can be distinguished.’
Chapter 5 - The description of mobility
142
The discussion about the proper treatment of extraposition from NP in the current generative theory is in full swing, and it would go too far to try and provide a conclusive minimalist analysis in this book. However, it is clear from the discussion above that the stranding analysis as suggested by Kayne, cannot in its present form explain the facts reported here. Undeniably, an analysis of extraposition in terms of stranding has some clear advantages in terms of simplifying the account of movement. Also, our findings do not render such an analysis impossible. However, a stranding analysis would need a different NP structure from the one represented in Figure 5-14. Possibly, a solution can be found in adopting right-adjoined structures again, as was argued above for the treatment of fronted premodification. DP Spec
D’ D
CP Spec
C’ C
IP NP
I’ I
VP VP
PP
V’ the man i who
V we knew
NP ti in high school
Figure 5-14: DP analysis following Kayne (1994) From a communicative point of view, there are two important factors that are said to influence extraposition: weight and information value. As was said in section 5.3, it is not clear exactly how the weight of a constituent can be calculated. One determinant of weight seems to be the length of the constituent in terms of the number of (orthographic) words. If we compare the average number of words of floating deferred postmodifiers in the corpus with the average number of words for adjacent postmodifiers, we find that floating deferred postmodifiers are indeed longer. For the whole corpus, the average number of words for floating deferred postmodifiers is 9.32, while adjacent postmodifiers consist of 5.57 words on
5.5 - Mobility in the NP
143
average.32 Another determinant of the weight of a constituent is its structural complexity. A (perhaps somewhat simplistic) way to calculate structural complexity is to count the number of ICs which a constituent contains. If we compare the average number of ICs of floating deferred postmodifiers with those of adjacent postmodifiers, we find that the number is indeed higher: 2.84 for floating deferred postmodifiers vs. 2.31 for adjacent postmodifiers.33 These findings for weight correspond to Hawkins’ Early Immediate Constituents principle.34 By extraposing the heavy postmodifier of a noun phrase which is followed by additional sentence material, the IC-to-non-IC ratio for that sentence will become higher and the order will thus be preferred over the adjacent ordering. In terms of information value, an extraposed element is expected to contain new (or important) information. Birner and Ward (1998) show that the postposed NPs in existential and presentational there-sentences represent information that is unfamiliar in some sense, either new to the hearer or new to the discourse. Shannon (1993) claims that in Dutch extraposition is only possible out of an NP which is the focus of the sentence/clause. And Quirk et al. (1985) mention that the postponement of part of the NP is used to achieve an information climax with end-focus.35 To be able to systematically test the applicability of these claims on the occurrences of floating deferred postmodification in the corpus, would require the corpus to be enriched with a discourse annotation. Unfortunately, such an annotation does not exist for the corpus at hand and a manual analysis would be too time-consuming for what is only a minor aspect of the present study. Without carrying out a proper discourse analysis, however, some observations can be made on the basis of the corpus material and the elicitation results that seem to support the ‘new information/focus’ claims. In the first place, floating deferred postmodifiers are found the most with NPs in subject position (almost 51% of all occurrences). This is unexpected, since the subject does not typically contain the focus of the sentence. However, if we take a closer look at these examples we see that the verb in these cases is often either an intransitive verb or a passivized transitive. In contrast to subjects of transitive verbs, subjects of intransitive verbs frequently appear as focus of the sentence. By extraposing the postmodifier, i.e. the part of the NP containing the focus, end-focus is achieved. In case of transitive verbs, the object often contains the focus of the 32
The average number of words is higher for floating deferred postmodifiers in all samples in the corpus. The difference is the most extreme for The Bloody Wood: 13.36 vs. 5.44 words. 33 The average number of ICs is higher for floating deferred postmodifiers in all but one of the samples in the corpus. The only exception is found in Nil Carborundum. However, in this sample only 2 examples of floating deferred postmodification are found, and the figure has no real statistical significance. The difference in the number of ICs is the most extreme for The Bloody Wood: 3.64 vs. 2.45 ICs. 34 See also section 5.5.2. 35 According to Quirk et al. (1985: 1399) the factor of end-focus often coincides with the factor of end-weight, which aims to achieve a stylistically well-balanced sentence.
144
Chapter 5 - The description of mobility
sentence. When the sentence is passivized, the object becomes the focuscontaining subject of the passive sentence. Again, end-focus can be achieved by extraposing the postmodifier. Secondly, in almost 32% of all instances of floating deferred postmodification the NP functions as direct object or subject complement of the sentence. These functions typically contain the focus. In these cases, the postmodifier is separated from the rest of the NP by an adverbial. 5.5.4
(Floating) deferred determiners
While the discussion of floating deferred postmodifiers as being either a case of rightward movement or a case of stranding is still in full swing, floating deferred determiners are now commonly considered as instances of stranding in generative theories. Sportiche (1988) is the first to analyse quantifier floating as quantifier stranding. In her analysis, the quantifier (Q) is treated as an adjunct to the NP. In other approaches (e.g. Abney, 1987), Q is considered as a modifier of the NP on a par with adjectives. In more recent generative theories, Q is considered as the head of the functional quantifier phrase (QP) and either positioned lower or higher than DP (Figure 5-15, p. 145). For the analysis of a structure like all the three children both possibilities are combined resulting in a tree structure as represented in Figure 5-16 (p. 145). The discontinuous structure now derives from moving the DP to the left, while the QP is stranded in its base-generated position. This explains why only some, namely QP external, QPs can be found stranded (all, both, each, enough). It does not explain, however, why quantifier stranding in English is obligatory when the NP is realized by a personal pronoun,36 why the quantifier is in most cases directly adjacent to the head of the NP (in my definition: deferred, but not floating), and why a non-adjacent quantifier is never found far from the head of the NP.37 It seems to me that these facts can be related to the need for (nominative) case assignment in the generative model and when excepting a basegenerated position of the subject to the left of the VP. After the DP has moved to receive case, only heads out of the VP can move up to a position lower than DP. As far as I know, floating deferred determiners have received no special attention in the functional literature. In general, quantification is treated in terms of scope, related to logical treatments of quantification. It is hard to see how quantifier floating can be explained from general principles proposed for postponed structures such as weight and information value.
36
Except, of course, in a partitive construction such as all of us. In the corpus study, non-adjacent quantifiers were only found with subject NPs and they were separated from the head of the NP by a verb (either an auxiliary or a full verb phrase). 37
145
5.6 - Conclusion (a)
DP
Spec
(b) Spec
D’ D
QP
QP Spec
Q’ Q
Q’ Q
DP Spec
NP
D’ D
NP
Figure 5-15: Possible QP positions QP Spec
Q’ Q
DP Spec
D’ D
QP Spec
Q’ Q
NP
all the three children Figure 5-16: Analysis of all the three children 5.6 Conclusion In this chapter, I have focused on the various treatments of mobility found in two of the main approaches in linguistics, namely the formal and the functional approach. The most important conclusion of this chapter must be that, while the phenomena of mobility and discontinuity have received a lot of attention in both formal and functional approaches to language description, so far no theory exists which can account for all the aspects of mobility in the noun phrase as reported on in this study. The wide interest in movement and discontinuity comes from the fact that these are language-independent notions. The extent to which languages allow mobility of constituents differs, but in general movement and discontinuous structures are recognized in all natural languages. It is for this reason that linguistic theories have to include some facility to account for mobility and discontinuity. The way in which this is implemented is theory-dependent. From a meta-theoretical point of view, the treatment of these phenomena is dependent on
146
Chapter 5 - The description of mobility
the approaches a theory takes towards constituency, descriptive aims and level(s) of description. The discussion of mobility in the formal approach has been largely determined by studies carried out in the generative framework. Generative theories have changed immensely over the years where the treatment of movement is concerned. They have developed from a purely transformational approach into the direction of a more surface-structure approach, they have greatly reduced the number of movement rules to one general movement rule (move a), and they have limited the directionality of move a to movement to the left. The most important principles proposed to explain mobility in the functional approach are weight and information value. In the discussion of the treatments of mobility in the NP, I have focussed on those constructions that have received the most attention in the formal and functional approach, i.e. discontinuous AJPs, floating deferred postmodification (extraposition out of NP), and (floating) deferred determiners (quantifier floating/stranding). Additionally, I have paid attention to fronted premodification, since that is the only NP construction that clearly seems to involve leftward movement of an IC. Some of the most important results of the discussion are that: -
-
a (generative) model should allow for both left and right adjunction in order to describe mobility in the NP properly; a description in terms of only leftward movement is not rendered impossible by the NP structures discussed here; fronted premodification forms an exception to the general assumption that a constituent out of which something is fronted represents old or given information - instead, NPs with a fronted premodifier seem to introduce new information into the discourse; the occurrence of discontinuity can be explained to a large extent by Hawkins’ Early Immediate Constituents principle, which is principally based on considerations of syntactic weight; Ross’ Right Roof Constraint applies to all examples of mobility in the NP encountered in the corpus; Akmaijan’s proposal for an NP cycle is not observed by all occurrences of mobility in the corpus, but does seem to be a helpful notion for distinguishing between different types of mobility; the stranding analysis of extraposition out of NP as suggested by Kayne has to be reconsidered for occurrences of extraposition out of definite NPs; the average syntactic weight when expressed in number of words and/or the number of ICs is indeed higher for floating postmodifiers in the corpus than for adjacent postmodifiers.
To conclude, I would like emphasize again that a proper analysis of mobility should entail a combination of syntactic and pragmatic principles.
Chapter 6 Summary and conclusion 6.1 Summary In this book, I have reported on the attempt to achieve two goals. The first goal was to investigate the mobility of the constituents in the English noun phrase and gain insight into the nature and frequency of variant NPs. The second goal, which follows from the first, was to develop a multi-method approach to descriptive studies that combines corpus data and experimental data. The main reason to study the mobility of NP constituents was that the phenomenon of word order variation in general and of phrasal constituents in particular has so far received little attention in English descriptive grammars. Where such variation is mentioned, the description often lacks, or provides only tentative, information on the frequency of occurrence and the conditions under which variation occurs. A second reason for this study is found in the field of corpus linguistics. For the analysis of corpora, corpus linguistics aims at the exhaustive description of a language. The formal grammar, which plays a central role in the analysis of corpora, should describe all possible regular structures as well as their variants. Insight into the mobility of constituents can help formulate conditions on word order variation in the formal grammar, thereby restricting the ambiguity of the analyses. In this chapter, I sum up the findings of the present study and discuss them in relation to these goals and reasons. Chapter 1 provides a description of the structure of the English noun phrase and its possible realizations. It gives an overview of the descriptions of the English NP as they can be found in grammatical handbooks of English. These different descriptions are then combined to constitute a consensus description of the NP: the prototypical NP structure. This integral structure is characterized by an immediate constituent analysis representing a rank-hierarchy of the units of description and the linear order of constituents. Within this structure, all categories can be assigned a ‘typical’ function. All variations on the prototypical structure are considered to yield ‘variant NPs’. Where the variation is caused by the mobility of a constituent or part of a constituent, the structure is subject of study in this research. An immediate constituent of the NP can occur to the left (fronted) or to the right (deferred) of its prototypical position, and it can occur inside or outside NP boundaries (floating). If only part of an immediate constituent occurs outside its prototypical position, the constituent as a whole is referred to as ‘discontinuous’. The present study of the mobility of constituents in the NP is restricted to occurrences in contemporary British English and aims to be a description of language use. The primary source of data used for the study was corpus data.
148
Chapter 6 - Summary and conclusion
However, every corpus study is finite and cannot guarantee a complete picture. For this reason, the corpus study was supplemented by means of experimental data, an approach here referred to as ‘the multi-method approach’. The design of the multi-method approach was discussed in Chapter 2. In general, the design is independent of the type of descriptive study that is conducted and has some clear advantages for descriptive studies that aim for a comprehensive description, especially studies that involve infrequent phenomena. The multi-method approach combines quantitative and qualitative data obtained from a corpus with judgments of native speakers. In chapter 2, it is argued that a methodologically sound descriptive study should follow a cyclic process (cf. Figure 2-1, p. 34). The cycle includes four stages: the (re)formulation of hypotheses, the testing of hypotheses on corpus data, the (re)formulation of hypotheses, and the testing of hypotheses on intuitive data. A descriptive study can start and end at any point on the data cycle as long as the whole round is completed at least once. Following the data cycle is in itself no guarantee for a sound empirical approach. The way in which the intuitive data and the corpus data are obtained and analysed plays an important role in this respect. While the introspection by one person can offer a valid contribution at the starting point of the description, it should not be the final point in the process, since it would interfere with the empirical grounding of the research. To guarantee the collection of objective intuitive data, a carefully designed experiment should be carried out with a representative group of native speakers. The design of such an experiment for the study of variant NPs was one of the focus points of this study (section 2.6). The design is based on that used in the Survey of English Usage (Quirk and Svartvik, 1966, 1979; Greenbaum and Quirk, 1970) and comprises 7 performance tests and 6 judgment tests. For the presentation of the experiment, a Windows-based computer program was developed. The program and other issues regarding the presentation of the experiment are discussed in section 2.6.2. The experiment was carried out with mostly students at the University of Liverpool over two rounds. In total, 120 people took part. In the second round, the actual experiment was carried out by means of the computer and followed by a face-to-face interview. The starting point of the current research was the following hypothesis: the prototypical description of the NP accounts for all possible occurrences of the NP in British English. This hypothesis was tested on corpus data. Ideally, the design of the corpus used for the present research should be such that the chosen samples are maximally representative of contemporary British English, and that the sample sizes, and thus the corpus as a whole, are sufficiently large to fully represent the mobility of NP constituents. Although discussions on how to achieve and measure optimal representativeness are in full swing, it is clear that the corpus used for the present research is far from ideal as far as its representativeness of contemporary British English is concerned. Allowing for practical restrictions on the availability of annotated corpus material, I have tried to make the corpus as representative as possible by including samples of various text categories and (transcribed) spoken samples. All samples have been annotated with detailed syntactic information and are stored in a syntactic
6.2 - Evaluation of the multi-method approach
149
database, which enabled further quantitative and qualitative research. The design of the corpus was discussed in section 2.5. Chapter 2 ends with a discussion of the combined use of experimental and corpus data. It is argued that, although the two types of data may appear heterogeneous at first sight, they are in fact very similar in that they both reflect language use and provide evidence which can help the linguist decide on which structures to incorporate in the description of a language. Whereas the first two chapters were concerned with general and methodological issues, Chapters 3 and 4 discuss the results of the application of the multi-method approach to the study of mobility in the English noun phrase. Chapter 3 gives the results of the corpus study, and Chapter 4 of the elicitation experiment. The most important results of both studies are reviewed in the next sections. Section 6.2 evaluates the multi-method approach and section 6.3 sums up the findings on the mobility of NP constituents and relates them to descriptions of mobility found in formal and functional approaches such as discussed in Chapter 5. 6.2 Evaluation of the multi-method approach The multi-method approach combines corpus data with experimental data. The present study was the first attempt to motivate and implement the existing idea of combining the two types of data into a practical and empirical approach to descriptive linguistics. In the present research, a corpus study preceded an elicitation experiment. The aim of the corpus study was to find all variant NPs resulting from the mobility of (part of) an immediate constituent of the NP and examine their nature and frequency. The corpus which was compiled for the present study comprises 170,000 words and includes five distinct text categories: fiction, non-fiction, drama, scripted speech, and spontaneous speech. This corpus is referred to as Corpus A. Because the number of variant NPs in Corpus A was relatively low, some additional textual material was used to supplement the data of Corpus A. This material, referred to as Corpus B, includes only fiction and non-fiction samples. Since only Corpus A was designed to be representative of contemporary British English, it was only this corpus that was used for the quantitative part of the corpus study. All samples in the corpus have been enriched with detailed syntactic annotation which enabled an automatic search for NP structures. In general, the corpus study provided a good overview of NP structures in British English. As far as the design of the corpus is concerned, the inclusion of different genres proved important in that the results pointed to a preference for mobility in the least formal genres. However, it is hard to establish statistical significance for these findings, because of the great divergence in the number of prototypical NPs and the number of variant NPs. The problem of obtaining statistical support is even more apparent for those situations in which the frequency of occurrence of the number of variant NPs is so low that the expected cell frequencies drop below five, in which case statistical techniques such as the chi-square test are no longer reliable. This problem of obtaining statistical
150
Chapter 6 - Summary and conclusion
significance for findings in the corpus can only partly be solved by increasing the size of the corpus, since the divergence in frequency between variant and prototypical NPs will remain. Another way to test the reliability of the results from the corpus study is to formulate hypotheses on the basis of these results and to test those hypotheses by means of an elicitation experiment. In this research, the elicitation experiment was restricted to the testing of hypotheses for three types of non-variant NP: the fronted premodifier, the discontinuous AJP, and the floating deferred modifier. The actual experiment was carried out with 120 people over two rounds. This number proved too small for a reliable study of some methodological aspects, such as the influence on the test results of variables such as the order of presentation and the sex and/or age of informants. Fortunately, the design of the experiment aimed to restrict the influence of such variables as much as possible. So even if a certain variable had a skewing effect, the effect will have been levelled out in the total test results. In general, the design and the execution of the experiment were successful. The computer program was easy to work with for the students and it simplified the processing of the data afterwards. However, there are a few things that I would do differently, if I were to repeat the experiment. I would -
also include an interview in the first round of elicitation, because the interview proved to be a valuable source of information for the interpretation of the results afterwards; include a little more instruction for the performance tests and a few extra examples; use a five-point scale for the similarity, frequency and normalcy judgments, because a seven-point scale proved too broad for both the scoring and the interpretation; include several native speaker linguists in the pilot to comment on the choice of test sentences, so that non-native influences in the sentences that distract the informant from the essence of the sentence are spotted at an early stage; include the same sentence in at least four types of test (two performance and two acceptability test) to simplify the interpretation of the scores; try to get more data by increasing the number of sentences in an experiment. Especially the Evaluation test proved to be valuable for testing the hypotheses. try to get a larger group of informants to be able to test the influence of the different design variables on the test results.
On the basis of the elicitation experiment, all hypotheses could be either accepted or rejected. Combination of these findings with the data obtained from the corpus study resulted in an extensive description of the variant NPs which were included in the experiment. The results of the multi-method approach were further justified when the data cycle was continued for NPs with a fronted premodifier. To conclude, the multi-method approach has proven to be a powerful method for a study that aims to provide a comprehensive description of a structure, especially a qualitative study of an infrequent phenomenon. However,
6.3 - Discussion of the results
151
the success of the multi-method approach for the study of mobility in the NP, and for future research, is bound by certain prior conditions. In the first place, the research depends heavily on existing knowledge about the structure of the NP as described in the literature. Without this basis, the study would have been too unrestricted for a successful corpus study and the number of hypotheses that would have to be tested in the elicitation experiment would have been too large to be feasible. In the second place, the availability of a corpus of which the annotation is tailored to the specific need is indispensable. Thirdly, while the multi-method approach proved very successful for the investigation of an infrequent phenomenon, the frequency of mobility in the NP could not have been much lower than observed, since this would have heavily impaired the formulation of the hypotheses to be tested in the elicitation experiment. In the fourth place, while the multi-method approach is an important step in further developing linguistic methodology toward an empirical approach, the method still depends to a great extent on the linguist’s interpretation of the corpus data and the experimental data. 6.3 Discussion of the results Corpus A comprises a total of 54,517 NPs of which 24,274 are complex, i.e. consist of more than a head only. Of the complex NPs, 5.3% cannot be accounted for by the prototypical NP structure. The majority of these variant NPs (82.8%) can be explained by the mobility of constituents. Taking the prototypical NP structure as a starting point, we can, in theory, expect the mobility of constituents to result in variant NPs with (a combination of): a. b. c. d. e. f. g. h. i. j. k. l. m.
a deferred modifier a floating deferred modifier a fronted modifier a floating fronted modifier a deferred determiner a floating deferred determiner a fronted determiner a floating fronted determiner a deferred limiter a floating deferred limiter a floating fronted limiter a discontinuous modifier a discontinuous determiner
In practice, eight of these thirteen types of variant NP were found. No occurrences were found of a floating fronted modifier (d), a (floating) fronted determiner (g-h), a floating deferred limiter (j), and a floating fronted limiter (k). Additionally, one non-prototypical constituent of the NP was found to be mobile, namely the emphasizer, resulting in a floating deferred emphasizer (but not a fronted emphasizer). The nine types of variant NP which were found in the
152
Chapter 6 - Summary and conclusion
corpus are discussed in detail in section 3.3. Section 3.4 discusses the distribution of variant NPs. As far as the distribution over text types is concerned, variant NPs with a mobile constituent occur significantly more in the least formal genres and significantly less in the formal genres. However, according to the contingency coefficient for these two variables, the relation is weak. As far as the distribution over sentence/clause and phrase functions is concerned, variant NPs have in general almost the same distribution as prototypical NPs. There are some differences, however. NPs with a floating constituent, for example, have a clear preference for the subject position, whereas NPs with a fronted modifier show a preference for the direct object position, while they are hardly ever found in subject position. No occurrences were found of NPs with a fronted, floating, and/or discontinuous constituent functioning as indirect object, object complement, or subject attribute. The results of the corpus study gave rise to several hypotheses on conditions on the mobility of constituents. Eleven hypotheses, for the fronted premodifier, the discontinuous AJP, and the floating deferred modifier, were tested in the elicitation experiment. These hypotheses were all based on the nonoccurrence of constructs in the corpus. Additionally, the variant structures that did occur in the corpus were also included in the experiment to test their acceptability. All variant NPs which were encountered in the corpus were judged as acceptable by the informants in the elicitation experiment. This indicates that, although they occur relatively infrequently in the corpus, they are considered as common constructs by the users of English. Also, the elicitation of these structures helped to shed some light on general ideas that native speakers hold about these structures, which facilitates the interpretation of the rest of the data: -
-
fronted AJPs introduced by too are considered to be more frequent than those introduced by so. The structures with so are judged to be ‘old-fashioned’, ‘poetic’ and ‘formal’. Informants prefer constructions with such instead of so; in general there exists a discrepancy between judgment and performance. Even if a variant structure is judged as more acceptable than its prototypical counterpart, the prototypical version is often produced more frequently; if a sentence containing a variant NP is compared with a prototypical structure which is unacceptable, the acceptability score for the variant version may be relatively high in a judgment test.
Other results from this part of the experiment are: -
If a complex AJP in the superlative degree occurs as a modifier, the continuous post-head version is unacceptable. For NPs with a floating deferred modifier, no preference exists for the continuous or the floating version, except in those instances in which a passive verb separates the first part of the NP from its postmodifier. In these latter cases, the discontinuous version is highly preferred.
6.3 - Discussion of the results
153
The results of the elicitation experiment are discussed in detail in Section 4.2. The results were then combined with those already obtained in the corpus study in order to reach integral descriptions for the structure of NPs with a fronted premodifier, a discontinuous AJP, and/or a floating deferred modifier. The combined results are discussed in Section 4.3. The most important results are summarized here: 1. 2. 3. 4.
5. 6.
7. 8.
Fronting is restricted to intensifying AJPs, either such or an adjective modified by an intensifying adverb phrase. In an NP with a fronted premodifier the determiner is an obligatory constituent which is realized by an indefinite article. An NP with a fronted premodifier cannot also contain a premodifier in prototypical position. NPs with a fronted premodifier have a clear preference for the direct object, subject complement, and prepositional complement positions, but are also acceptable in other typical NP functions. The realization of the postmodifier slot in the first part of the discontinuous AJP is restricted to occurrences of enough. NPs with a discontinuous modifier have a clear preference for the direct object and subject complement positions, but are also acceptable in other typical NP functions. A floating deferred modifier cannot be realized by an adverb phrase. NPs with a floating deferred modifier have a clear preference for the subject position. They cannot occur in adverbial function, but are acceptable in all other typical NP functions.
The results for fronted premodification enabled a more specific corpus search for such constructs in the British National Corpus. The results of this continuation of the data cycle validate and strengthen all findings of the research so far, except one: a fronted premodifier can co-occur with a modifier in prototypical position (point 3 above). However, the realization of the premodifier in prototypical position does seem to be restricted to classifying NPs and AJPs. 6.3.1
Results for descriptive linguistics
In this study the emphasis has been on the observational task of linguistics. This is necessary because so far this task has received comparatively little attention from mainstream linguistics, and every scientific discipline has to start from reliable findings obtained through a well-designed observational methodology. As a well-grounded observational study, the present research forms a contribution to descriptive linguistics in that it provides a comprehensive description of the English NP, which includes both qualitative and quantitative information. More specifically, it provides a description of the mobility of constituents in the NP. In general, the description provided here supplements existing NP descriptions. Furthermore, the research relates the mobility of NP
154
Chapter 6 - Summary and conclusion
constituents to existing studies on word order variation as they are found in formal and functional approaches to linguistics (Chapter 5). One reason for this is to find an explanation for the occurrence of variant NP structures. Another reason is to test the appropriateness of existing theories for the mobility of NP constituents in order to arrive at an integral description of the mobility phenomenon. An important result of the discussion in Chapter 5 is that the mobility of constituents in the NP, and more specifically the fronting of a premodifier, can only be explained adequately in a (generative) framework which allows both left and right adjunction. This renders Kayne’s antisymmetry theory too restrictive (Kayne, 1994). Also, while Kayne’s stranding analysis for floating deferred modification (or extrapostion) cannot, in its present form, explain all occurrences of that structure reported in this book, the examples of mobility in the NP do not disprove the possibility of a framework that allows for leftward movement only. As far as finding explanations for the occurrence of variant NP structures is concerned, explanations can be found in the syntactic weight and the information value of constituents. Occurrences of discontinuity in the NP can be explained by Hawkins’ Early Immediate Constituents (EIC) principle (Hawkins, 1994), which says that the most favoured order of constituents is that order in which the lowest number of dominated elements has to be scanned for IC recognition. This principle is motivated by considerations of efficiency and ease of processing and is based on the notion of syntactic weight. The findings in the corpus also support an explanation of floating deferred modification in terms of weight. Besides syntactic weight, the information value of a constituent seems to play an important role in explaining the occurrence of mobility. A first study of the discourse aspects of NPs with a floating deferred modifier indicates that such NPs are the focus of the sentence and that the floating deferred modifier contains the new information. Discontinuity in these structures has the function of achieving an information climax with end-focus. NPs containing a fronted premodifier are also the focus of the sentence. In this case the modifying AJP contains the new information and is placed in NP-initial position to receive extra pragmatic highlighting. Whether these explanations are valid should be investigated by means of an in-depth study of the pragmatic properties of variant NPs. Besides a pragmatic study of variant NPs, it would be interesting to further investigate the status of modifying AJPs consisting of so + adjective. The present study aimed to provide a synchronic description of mobility. It nevertheless pointed to a diachronic development which is possibly taking place, namely the disappearance of fronted premodification by means of so + adjective and its replacement by such + adjective. A diachronic (corpus) study of this phenomenon could shed some light on the validity of this indication of language change. 6.3.2
Results for corpus linguistics
The present research has contributed to the field of corpus linguistics in two ways. In the first place, it has given insight into conditions on the mobility of constituents in the English NP. Knowledge of such conditions helps to create
6.3 - Discussion of the results
155
formal grammars for corpus analysis in which the ambiguity of the descriptions is controlled as much as possible. However, for the phenomenon at hand the gain is found more in the theoretical evidence that the multi-method approach can provide such conditions, than in the practical want to formalize them. Considering the low frequency of the types of variant NP and thus the relatively small improvement in parsing accuracy, the lack of strong conditions for most types of variant NP, and the increase in ambiguity of the parser when adding the extra NP structures, it would be unwise for most parser purposes to add the extra rules. A possible exception is inclusion in a parser for the purpose of analysing unrestricted input with the aim of providing at least the contextually appropriate analysis among the analyses produced by the parser. Secondly, and perhaps more importantly, the present study has contributed to the further development of the methodology of corpus-based studies. It has pointed out some of the weaknesses of using only corpus data, especially for a qualitative study of an infrequent phenomenon, and has stressed the importance of collecting judgments through the use of a carefully designed experiment. The multi-method approach was illustrated by means of the study of the mobility of constituents in the English noun phrase, but can and should in fact be used for any study that aims to provide a comprehensive description of a structure in a language.
References cited Aarts, F. (1971): “On the distribution of noun-phrase types in English clausestructure”, in: Lingua 26: 281-293. Aarts, F. (1975): “The Great Tradition or grammars and Quirk’s grammar”, in: Dutch Quarterly Review of Anglo-American Letters 5: 98-126. Aarts, F. (1976): “The description of linguistic variation in English: from Firth till the present day”, in: English Studies 57: 239-251. Aarts, F. and J. Aarts (1982): English syntactic structures. Functions and categories in sentence analysis. Oxford: Pergamon Press. 2nd edition (1988). New York: Prentice Hall. Aarts, J. (1991): “Intuition-based and observation-based grammars”, in: K. Aijmer and B. Altenberg (eds.), 44-62. Aarts, J. (1995): “Corpus analysis”, in: J. Verschueren, J. Östman, and J. Blommaert (eds.), 565-570. Aarts, J., H. van Halteren, and N. Oostdijk (1998): “The linguistic annotation of corpora: the TOSCA analysis system”, in: International Journal of Corpus Linguistics 3(2): 189-210. Aarts, J. and T. van den Heuvel (1985): “Computational tools for the syntactic analysis of corpora”, in: Linguistics 23: 303-335. Aarts, J. and W. Meijs (eds.) (1984): Corpus linguistics. Recent developments in the use of computer corpora in English language research. Amsterdam: Rodopi. Aarts, J. and W. Meijs (eds.) (1986): Corpus linguistics II. New studies in the analysis and exploitation of computer corpora. Amsterdam: Rodopi. Aarts, J., I. de Mönnink, and H. Wekker (eds.) (1997): Studies in English language and teaching. Amsterdam: Rodopi. Abney, S. (1987): The English noun phrase in its sentential aspect. PhD thesis. Cambridge, Mass.: Massachusetts Institute of Technology. Aijmer, K. and B. Altenberg (eds.) (1991): English corpus linguistics: Studies in honour of Jan Svartvik. London: Longman. Akmajian, A. (1975): “More evidence for an NP cycle”, in: Linguistic Inquiry 6: 115-129. Alphonce, C. and H. Davis (1997): “Motivating non-directional movement”, in: D. Beerman et al. (eds.), 7-35. Altenberg, B. (1994): “On the functions of such in spoken and written English”, in: N. Oostdijk and P. de Haan (eds.), 223-240. Anagnostopoulou A., H. van Riemsdijk, and F. Zwarts (eds.) (1997): Materials on left dislocation. Amsterdam: John Benjamins. Aston, G. and L. Burnard (1998): The BNC handbook: Exploring the British National Corpus with SARA. Edinburgh: Edinburgh University Press. Baker C. and J. McCarthy (eds.) (1981): The logical problem of language acquisition. Cambridge Mass.: MIT Press. Baltin, M. (1981): “Strict bounding”, in: C. Baker and J. McCarthy (eds.), 257295.
158
References
Baltin, M. (1987): “Degree complements”, in: G. Huck and A. Ojeda (eds.), 1126. Beerman, D., D. LeBlanc, and H. van Riemsdijk (eds.) (1997): Rightward movement. Amsterdam: John Benjamins. Bergenholtz, H. and B. Schaeder (eds.) (1979): Empirische Textwissenschaft. Aufbau und Auswertung von Textcorpora. Koenigstein: Scriptor. Biber, D. (1990): “Methodological issues regarding corpus-based analysis of linguistic variation”, in: Literary and Linguistic Computing 5: 257-69. Biber, D. (1993): “Representativeness in corpus design”, in: Literary and Linguistic Computing 8: 243-57. Biber, D. and E. Finegan (1986): “An initial typology of English text types”, in: J. Aarts and W. Meijs (eds.), 19-46. Birner, B. and G. Ward (1998): Information status and noncanonical word order in English. Amsterdam: John Benjamins. Bloomfield, L. (1933): Language. New York: Holt, Rinehart and Winston Inc. Bolinger, D. (1972): Degree words. The Hague: Mouton. Bresnan, J. (1973): “Syntax of the comparative clause construction in English”, in: Linguistic Inquiry 4: 275-343. Bresnan, J. (ed.) (1982): The mental representation of grammatical relations. Cambridge Mass.: MIT Press. Brown, G., K. Malmkjær, and J. Williams (eds.) (1996): Performance and competence in second language acquisition. Cambridge: Cambridge University Press. Brown, K. and J. Miller (1996): Concise encyclopedia of syntactic theories. Cambridge: Cambridge University Press. Bunt, H. (1991): “Parsing with discontinuous phrase structure grammar”, in: M. Tomita (ed.), 49-63. Bunt, H. (1996): “Discontinuous constituency: Introduction”, in: H. Bunt and A. van Horck (eds.), 1-10. Bunt, H. and A. van Horck (eds.) (1996): Discontinuous constituency. Berlin/New York: Mouton de Gruyter. Burnard, L. (1995): Users reference guide for the British National Corpus. Version 1.0. Oxford: Oxford University Computing Services. Chaudron, C. (1983): “Research on metalinguistic judgments: A review of theory, methods, and results”, in: Language Learning 33: 343-377. Chomsky, N. (1957): Syntactic structures. The Hague: Mouton. Chomsky, N. (1965): Aspects of the theory of syntax. Cambridge, Mass.: MIT Press. Chomsky, N. (1981): Lectures on Government and Binding. Dordrecht: Foris Publications. Chomsky, N. (1995): The Minimalist Program. Cambridge Mass.: MIT Press. Corver, N. (1990): The syntax of left branch extractions. PhD dissertation, Tilburg University. Corver, N. (1991): “Evidence for DegP”, in: NELS 21: 33-47. Covington, M. (1990): “Parsing discontinuous constituents in dependency grammar”, in: Computational Linguistics 16-4: 234-236.
References
159
Culicover, P. and M. Rochemont (1990): “Extraposition and the complement principle”, in: Linguistic Inquiry 21: 23-47. Dik, S. (1989): The theory of functional grammar. Part 1: The structure of the clause. Dordrecht: Foris. Dixon, R. (1982): Where have all the adjectives gone? Berlin: Mouton. Dowty, D. (1996): “Toward a minimalist theory of syntactic structure”, in: H. Bunt and A. van Horck (eds.), 11-62. Drijkoningen, F. and K. Hengeveld (eds.) (1993): Linguistics in the Netherlands 1993. Amsterdam: John Benjamins. Ellis, R. (1991): “Grammaticality judgments and second language acquisition”, in: Studies in Second Language Acquisition 13: 161-186. Fillmore, C. (1992): “‘Corpus linguistics’ or ‘Computer-aided armchair linguistics’”, in: J. Svartvik (ed.), 35-60. Fries, C. (1952): The structure of English. An introduction to the construction of English sentences. New York: Harcourt, Brace and Company. Fries, P. H. (1972): “Problems in the tagmemic description of the English noun phrase”, in: L. Heilmann (ed.), 222-229. Garside, R., G. Leech, and G. Sampson (eds.) (1987): The computational analysis of English. A corpus-based approach. London: Longman. Gazdar, G. (1981): “Unbounded dependencies and coordinate structures”, Linguistic Inquiry 12: 155-184. Gazdar, G., E. Klein, G. Pullum, and I. Sag (1985): Generalized Phrase Structure Grammar. Cambridge, Mass.: Harvard University Press. Geluykens, R. (1987): “Tails (right dislocations) as repair mechanism in English conversations”, in: J. Nuyts and G. De Schutter (eds.), 119-130. Giorgi, A. and G. Longobardi (1991): The syntax of noun phrases. Configuration, parameters and empty categories. Cambridge: Cambridge University Press. Givón, T. (1979): On understanding grammar. New York: Academic Press. Givón, T. (1984): Syntax: A functional-typological introduction, vol. I. Amsterdam: John Benjamins. Givón, T. (1990): Syntax: A functional-typological introduction, vol. II. Amsterdam: John Benjamins. Gleason, H. (1965): Linguistics and English grammar. New York: Holt, Rinehart and Winston Inc. Greenbaum, S. (1973): “Informant elicitation of data on syntactic variation”, in: Lingua 31: 201-212. Greenbaum, S. (1976): “Contextual influence on acceptability judgements”, in: Linguistics 187: 5-11. Greenbaum, S. (1977): Acceptability in language. The Hague: Mouton. Greenbaum, S. (1984): “Corpus analysis and elicitation tests”, in: J. Aarts and W. Meijs (eds.), 193-201. Greenbaum, S. (1988): “A proposal for an international corpus of English”, in: World Englishes 7: 315. Greenbaum, S. (1992): “A new corpus of English: ICE”, in: J. Svartvik (ed.), 171-179.
160
References
Greenbaum, S. and R. Quirk (1970): Elicitation experiments in English: linguistic studies in use and attitude. London: Longman. Greenbaum, S. and R. Quirk (1990): A student’s grammar of the English language. London: Longman. Greenberg, J. (1963a): “Some universals of grammar with particular reference to the order of meaningful elements”, in: J. Greenberg (ed.), 73-113. Greenberg, J. (ed.) (1963b): Universals of language. Cambridge, Mass.: MIT Press. Guéron, J. (1980): “On the syntax and semantics of PP extraposition”, in: Linguistic Inquiry 11: 637-677. Guéron, J. and R. May (1984): “Extraposition and logical form”, in: Linguistic Inquiry 15:1-31. de Haan, P. (1989): Postmodifying clauses in the English noun phrase. A corpusbased study. Amsterdam: Rodopi. de Haan, P. (1992): “The optimum corpus sample size?”, in: G. Leitner (ed.), 319. Haegeman, L. and J. Guéron (1999): English grammar. A generative perspective. Oxford: Blackwell. Halliday, M.A.K. (1985): An introduction to functional grammar. London: Edward Arnold. van Halteren, H. (1997): Excursions into syntactic databases. Amsterdam: Rodopi. van Halteren, H. and T. van den Heuvel (1990): Linguistic exploitation of syntactic databases. Amsterdam: Rodopi. Harman, G. (1963): “Generative grammars without transformation rules: a defense of phrase structure”, in: Language 39: 597-616. Harman, G. (1966): “The adequacy of context-free phrase structure grammars”, in: Word 22: 276-293. Harris, Z. (1951): Methods in structural linguistics. Chicago: University of Chicago Press. Hawkins, J. (1983): Word order universals. New York: Academic Press. Hawkins, J. (1994): A performance theory of order and constituency. Cambridge: Cambridge University Press. Heilmann, L. (ed.) (1972): Proceedings of the eleventh international congress of linguistics. Bologna: Il Mulino. Hetzron, R. (1978): “On the relative order of adjectives”, in: H. Seiler (ed.), 165184. Hornstein, N. (1995): Logical form. From GB to Minimalism. Oxford/Cambridge MA: Blackwell. Huck, G. and A. Ojeda (1987a): “Introduction”, in: G. Huck and A. Ojeda (eds.), 1-10. Huck, G. and A. Ojeda (eds.) (1987b): Discontinuous constituency. Syntax and Semantics 20. London/Orlando: Academic Press Inc. Hudson, R. (1984): Word Grammar. Oxford: Basil Blackwell. Hudson, R. (1990): English Word Grammar. Oxford: Basil Blackwell.
References
161
Jacobson, P. (1996): “Constituent structure”, in: K. Brown and J. Miller (eds.), 54-67. Jespersen, O. (1909-1949): A modern English grammar on historical principles. Copenhagen: Munksgaard. Johansson, S., G. Leech, and H. Goodluck (1978): Manual of information to accompany the Lancaster-Oslo/Bergen Corpus of British English, for use with digital computers. Oslo: Dept. of English, University of Oslo. Kaplan, R. and J. Bresnan (1982): “Lexical-functional grammar: A formal system for grammatical representation”, in: J. Bresnan (ed.), 173-281. Kayne, R. (1994): The antisymmetry of syntax. Cambridge Mass.: MIT Press. Kirk, J. (ed.) (in press): Corpora galore. Analyses and techniques in describing English. Papers from the 19th International Conference on English Language Research on Computerized Corpora, Newcastle, 1998. Amsterdam: Rodopi. Kruisinga, E. (1909-1932): A handbook of present-day English. Groningen. Ku era, H. and W. Francis (1967): Computational analysis of present-day American English. Providence, Rhode Island: Brown University Press. La Galy, M. et al. (eds.) (1974): Papers from the Tenth Regional Meeting. Chicago Linguistics Society. Chicago: University of Chicago. Lambrecht, K. (1994): Information structure and sentence form. Topic, focus and the mental representation of discourse referents. Cambridge: Cambridge University Press. Lancashire, I., C. Meyer, and C. Percy (eds.) (1996): Synchronic corpus linguistics. Papers from the 16th International Conference on English Language Research on Computerized Corpora, Toronto, 1995. Amsterdam: Rodopi. Leech, G. (1987): “The computational analysis of English. General introduction”, in: R. Garside, G. Leech, and G. Sampson (eds.), 1-15. Lees, R. (1964): “On the so-called ‘substitution in frames’ technique”, in: General Linguistics 6: 11-20. Leitner, G. (ed.) (1992): New directions in English language corpora. Berlin: Mouton de Gruyter. Ljung, M. (ed.) (1997): Corpus-based studies in English. Papers from the 17th International Conference on English Language Research on Computerized Corpora, Stockholm, 1996. Amsterdam: Rodopi. Longacre, R. (1960): “String constituent analysis”, in: Language 36: 63-88. McCawley, J. (1982): “Parentheticals and discontinuous constituent structure”, in: Linguistic Inquiry 13: 91-106. Mel’ uk, I. (1988): Dependency syntax: theory and practice. Albany: State University of New York Press. de Mönnink, I. (1996): “A first approach to the mobility of noun phrase constituents”, in: I. Lancashire, C. Meyer, and C. Percy (eds.), 265-79. de Mönnink, I. (1997a): “Using corpus and experimental data: a multi-method approach”, in: M. Ljung (ed.), 227-244. de Mönnink, I. (1997b): “Eliciting floating postmodifiers”, in: H. Strik, N. Oostdijk, C. Cucchiarini, and P.A. Coppen (eds.), 65-79.
162
References
de Mönnink, I. (1999): “Combining corpus and experimental data”, in: International Journal of Corpus Linguistics 4(1): 77-111. de Mönnink, I. (in press): “A moving phrase. A multi-method approach to the mobility of constituents in the English noun phrase”, in: J. Kirk (ed.), 133147. Nagata, H. (1989): “Judgments of sentence grammaticality and field-dependence of subjects”, in: Perceptual and Motor Skills 69: 739-747. Newmeyer, F. (1983): Grammatical theory. Its limits and its possibilities. Chicago: The University of Chicago Press. Newmeyer, F. (1996): Generative linguistics. London/New York: Routledge. Nida, E. A. (1966): A synopsis of English syntax. The Hague: Mouton. Nuyts J. and G. De Schutter (eds.) (1987): Getting one’s words into line: On word order and Functional Grammar. Dordrecht: Foris. Ojeda, A. (1987): “Discontinuity, multidominance, and unbounded dependency in Generalized Phrase Structure Grammar: some preliminaries”, in: G. Huck and A. Ojeda (eds.), 257-282. Oostdijk, N. (1988a): “A corpus linguistic approach to linguistic variation”, in: Literary and Linguistic Computing 3: 12-25. Oostdijk, N. (1988b): “A corpus for studying linguistic variation”, in: ICAME Journal 12: 3-14. Oostdijk, N. (1991): Corpus linguistics and the automatic analysis of English. Amsterdam: Rodopi. Oostdijk, N. (1993): Unpublished ICE manual: DOCREF = ICE-PHRASES 0.2.930215. Oostdijk, N. (in press): “Linguistic delicacy in explorative studies”, in: J. Kirk (ed.), 281-294. Oostdijk, N. and J. Aarts (1997): “Multiple postmodification in the English noun phrase”, in: J. Aarts, I. de Mönnink, and H. Wekker (eds.), 107-121. Oostdijk, N. and P. de Haan (eds.) (1994): Corpus-based research into language. Amsterdam: Rodopi. Ouhalla, J. (1994): Introducing transformational grammar: from rules to principles and parameters. London: Edward Arnold. Pike, K. (1943): “Taxemes and immediate constituents”, in: Language 19, 65-82. Postal, P. (1967): Constituent structure: a study of contemporary models of syntactic description. Bloomington: Indiana University. Poutsma, H. (1904-1926): A grammar of late modern English. Groningen: P. Noordhoff. Quirk, R. (1968): Essays on the English language. Medieval and modern. London: Longman. Quirk, R., S. Greenbaum, G. Leech, and J. Svartvik (1972): A grammar of contemporary English. London: Longman. Quirk, R., S. Greenbaum, G. Leech, and J. Svartvik (1985): A comprehensive grammar of the English language. London: Longman. Quirk, R. and J. Svartvik (1966): Investigating linguistic acceptability. The Hague: Mouton.
References
163
Quirk, R. and J. Svartvik (1979): “A corpus of modern English”, in: H. Bergenholtz and B. Schaeder (eds.), 204-218. Radford, A. (1988): Transformational Grammar. Cambridge: Cambridge University Press. Reinhart, T. (1980): “On the position of extraposed clauses”, in: Linguistic Inquiry 11: 621-624. van Riemsdijk, H. and F. Zwarts (1997): “Left dislocation in Dutch and the status of copying rules”, in: A. Anagnostopoulou et al. (eds.), 13-29. Rijkhoff, J. (1990): “Explaining word order in the noun phrase”, in: Linguistics 28: 5-42. Rizzi, L. (1982): Issues in Italian syntax. Dordrecht: Foris. Rochemont M. and P. Culicover (1997): “Deriving dependent right adjuncts in English”, in: D. Beerman et al. (eds.), 279-300. Rodman, R. (1997): “On Left Dislocation”, in A. Anagnostopoulou et al. (eds.), 31-54. Ross, J. (1967): Constraints on variables in syntax. PhD dissertation, MIT. Schütze, C. (1996): The empirical base of linguistics. Chicago: The University of Chicago Press. Seiler, H. (ed.) (1978): Language universals. Tübingen: Narr. Shannon, T. (1993): “Focus and the extraposition of noun phrase complement clauses in Dutch”, in: F. Drijkoningen and K. Hengeveld (eds.), 117-128. Shohamy, E. (1996): “Competence and performance in language testing”, in: G. Brown, K. Malmkjær, and J. Williams (eds.), 138-151. Siewierska, A. (1988): Word order rules. London: Croom Helm. Snow, C. and G. Meijer (1977): “On the secondary nature of syntactic intuitions”, in: S. Greenbaum (ed.), 163-177. Sportiche, D. (1988): “A theory of floating quantifiers and its corollaries for constituent structure”, in: Linguistic Inquiry 19: 425-449. Stageberg, N. C. (1965): An introductory English grammar. New York: Holt, Rinehart and Winston. Strik, H., N. Oostdijk, C. Cucchiarini, and P.A. Coppen (eds.) (1997): Proceedings of the Department of Language and Speech, 20 (1996). Nijmegen: Dept. of Language and Speech. Svartvik, J. (ed.) (1992): Directions in corpus linguistics. Proceedings of Nobel Symposium 82. Stockholm, 4-8 August 1991. Berlin/New York: Mouton de Gruyter. Tomita, M. (ed.) (1991): Current issues in parsing technology. Dordrecht: Kluwer Academic Publishers. Tomlin, R. (1986): Basic word order: Functional principles. London: Croom Helm. Van Valin, R. (1993a): “A synopsis of Role and Reference Grammar”, in: R. Van Valin (ed.): 1-164. Van Valin, R. (ed.) (1993b): Advances in Role and Reference Grammar. Amsterdam: John Benjamins. Veenstra, M. (1998): Formalizing the Minimalist Program. Groningen: Groningen Dissertations in Linguistics 24.
164
References
Verschueren, J., J. Östman, and J. Blommaert (eds.) (1995): Handbook of pragmatics. Amsterdam/Philadelphia: John Benjamins Publishing Company. Wells, R. (1947). “Immediate constituents”, in: Language 23: 81-117. Ziv, Y. and P. Cole (1974): “Relative extraposition and the scope of definite descriptions in Hebrew and English”, in: M. La Galy et al. (eds.): 772-786.
APPENDIX A: Design of Corpus A Written: Prose: Fiction Allingham, M.: The Mind Readers. 1965 (edition used: Penguin Books 1968, pp. 46-103). Innes, M.: The Bloody Wood. 1966 (edition used: Penguin Books 1968, pp. 27-89). Innes, M.: Carson’s Conspiracy. A Sir John Appleby Mystery. 1984 (edition used: Penguin Books 1986, pp. 79-147, Br. English with Am. spelling). Prose: Non-fiction Paul, J.: Cell Biology. London, 1965 (edition used: 1967, pp. 102-178) From the IPSM Test Corpus, a corpus of computer manuals: Lotus (Am. Eng. with Am. spelling) (2,952) Dynix (Am. Eng. with Am. spelling) (3,408) Trados (Br. Eng. with Am. spelling) (4,221)
Words: 20,266 21,558 20,011 61,835
19,368 10,581
29,949 Drama: Livings, H.: Stop It, Whoever You Are. Harmondsworth, 1962 (edition used: 1967, pp. 15-79). Livings, H.: Nil Carborundum. Harmondsworth, 1963 (edition used: 1967, pp. 214-239).
14,022 5,642 19,664
Spoken: Spontaneous speech: Direct conversation: between 2 persons S1A-009: Mother and son, 2 July 1991 [m,f] S1A-025: Brother and sister, March 1991 [m,f] S1A-039: Flatmates, Oct. 1991 [2f] S1A-059: Counselling interview, Feb. 1991 [2m] S1A-061: Colleagues’ lunchtime conversation, Feb. 1992 [2m] S1A-067: Friends, 8 Nov. 1991 [2f] S1A-083: Tennis coaches, April 1992 [2f] S1A-085: Friends, March 1992 [m,f]
1,974 1,079 2,035 1,878 1,988 1,960 1,975 2,030 14,919
166 Direct conversation: more than 2 persons S1A-007: Family conversation, 8 June 1991 [3m,2f] S1A-016: Marketing discussion, April 1991 [2f,3m] S1A-022: Family conversation, 10 June 1991 [m,3f] S1A-028: Birthday party conversation, July 1991 [2m,3f] S1A-046: Family conversation, Nov. 1991 [2m,2f] S1A-056: Mealtime conversation (4 subtexts), 7 Feb. 1992 [2m,f + 1 extra-corpus] S1A-063: Colleagues’ conversation [2m,2f] S1A-068: Students’ Union Office conversation, 6 March 1992 [2m,f] Prepared/scripted speech: SECA Commentary SECB News broadcast SECD Lecture type 1 - restricted audience SECF Magazine-style reporting SECJ Dialogue
Appendix A
1,991 2,112 1,815 1,973 2,171 2,151 1,836 1,889 15,938 9,060 5,169 7,409 4,710 6,885 33,233
APPENDIX B: Design of Corpus B Written: Prose: fiction Barker, C.: The Damnation Game. 1985 (edition used: Sphere Books Ltd. 1986, pp. 144-193). Herbert, J.: The Fog. 1975 (edition used: New English Library 1981, pp. 120-176). Hughes, D.: The Joke of the Century. 1986 (Edition used: Taplinger Publishing Co., Inc. 1986, pp. 97-157). James, P.D.: A Taste for Death. 1986 (edition used: Faber and Faber 1986, pp. 175-223). Roberts, P.: Tender Prey. 1983 (edition used: Pan Books Ltd. 1985, pp. 137-203). Sacks, O.: The Man Who Mistook His Wife for a Hat & Other Clinical Tales. 1986 (edition used: Simon and Schuster Inc. 1986, pp. 7-41, 77-80, 87-96, 103-110, 125). Stevens, R.T.: Flight from Bucharest. 1977 (edition used: Fontana Books 1978, pp. 93--144). Prose: non-fiction Brazier, M.: Medicine, Patients and the Law. An Up-to-Date and Informative guide. 1987 (edition used: Penguin Books Ltd. 1987, pp. 157-205). Hyland, M.: Introduction to Theoretical Psychology. 1981 (edition used: The Macmillan Press Ltd. 1981, pp. 32-90). Richards, J. Radcliffe: The Sceptical Feminist: A philosophical enquiry. 1980 (edition used: Routledge and Kegan Paul Ltd. 1980, pp. 90-138). Swann, D.: Competition and Consumer Protection. 1979 (edition used: Penguin Books Ltd. 1979, pp. 141-192).
20,012 20,000 20,007 20,007 20,016 20,011 20,005 140,058
20,008 20,022 20,015 20,010 80,055
APPENDIX C: Explanation of the abbreviations used in the example analyses Functions: A AJHD AJPO AJPR AVHD AVPR CS Dim DT/Det DTCE DTDE DTPE DTPO DTPR DTPS EMPH FEMPH FDTDE FNPPO H LIM MODI MVB NOSU NPHD NPPO NPPR OD OI P PRFR PRSU PUNC SBHD SU/Su SUB TO UTT V
adverbial adjective phrase head adjective phrase postmodifier adjective phrase premodifier adverb phrase head adverb phrase premodifer subject complement discontinuous modifier determiner central determiner deferred determiner predeterminer determiner phrase postmodifier determiner phrase premodifier postdeterminer emphasizer floating emphasizer floating deferred determiner floating postmodifier head limiter discontinuous modifier main verb notional subject noun phrase head noun phrase postmodifier noun phrase premodifier direct object indirect object predicator fronted premodifier provisional subject punctuation subordinator phrase head subject subordinator to-infinitive marker utterance verb
NOFU
signifies absence of a function
Appendix C
170 Categories: ADJ ADV AVP/AdvP AJP/Adj.P AJPD1 AJPD2 AJPD3 ART AVP CL CONJN DTP LV N NP NUM PM PN PP PRIT PRTCL RPDU S SUBP TXTU VP
adjective adverb adverb phrase adjective phrase 1st part of discontinuous adjective phrase 2nd part of discontinuous adjective phrase 3rd part of discontinuous adjective phrase article adverb phrase clause conjunction determiner phrase lexical verb noun noun phrase numeral punctuation mark pronoun prepositional phrase provisional it particle reported utterance sentence subordinator phrase textual unit verb phrase
Attributes: act attru com decl indef indic motr past per prop sing unm
active attributive common declarative indefinite indicative monotransitive past tense period proper singular unmarked word order
APPENDIX D: Explanation of the abbreviations used in the text and figures AP A/Adj BNC BW CB CC CM CP CRD D/DET Deg DegP Dem DP DS EIC fcri03 fhum03 fpsy02 Gen GPSG I IC ICE ID IP LF LP MR N NC neco02 Num nwom02 PF POM PREM Q QP Rel RRC SEU SI
Adjective Phrase Adjective British National Corpus The Bloody Wood Cell Biology Carson’s Conspiracy. A Sir John Appleby Mystery Computer Manuals, IPSM Test Corpus Complementizer Phrase Constituent Recognition Domain Determiner Degree element Degree Phrase Demonstrative Determiner Phrase Deep Structure Early Immediate Constituents A Taste for Death The Joke of the Century Tender Prey Genitive Generalized Phrase Structure Grammar Inflection Immediate Constituent International Corpus of English Immediate Dominance Inflection Phrase Logical Form Linear Precedence The Mind Readers Noun Nil Carborundum Competition and Consumer Protection Numeral The Sceptical Feminist: A philosophical enquiry Phonetic Form Postmodifier Premodifier Quantifier Quantifier Phrase Relative Right Roof Constraint Survey of English Usage Stop It, Whoever You Are
172 SP1 SP2 SP3 Spec SS TOSCA XP
APPENDIX Spontaneous conversation between 2 persons Spontaneous conversation between more than 2 persons Prepared/scripted speech Specifier Surface Structure TOols for Syntactic Corpus Analysis any Phrase
APPENDIX E: Distribution of prototypical NPs in Corpus A Table E-1: Absolute frequencies of prototypical NPs in Corpus A Fiction DT NPHD
Non- Drama Spon. Scrip. Total fiction speech speech # 3,539 1,304 1,054 1,403 2,077 9,377 1,149 113 7
918 143 7
235 18 2
377 24 3
805 116 8
3,484 414 27
1,021 49 0
717 81 5
196 4 0
359 10 1
862 78 8
3,155 222 14
420 42 3 26 0 3
364 39 0 21 0 1
45 2 0 1 0 0
89 8 0 1 1 0
346 45 4 42 0 6
1,264 136 7 91 1 10
LIM DT NPHD
47
5
19
20
14
105
LIM DT NPPR NPHD
14 6
1 0
5 0
5 0
3 1
28 7
15 0
5 1
5 0
7 1
11 2
43 4
9 1 1
6 0 0
0 0 0
1 0 0
1 0 0
17 1 1
44
2
9
29
15
99
LIM NPPR NPPR NPHD
1 1
3 0
1 0
6 0
0 0
11 1
LIM NPHD NPPO
9
6
4
2
6
27
LIM NPPR NPHD NPPO
2
1
0
0
0
3
429 32 3
905 90 7
135 11 1
247 26 3
458 48 1
2,174 207 15
433 8 0
381 27 2
112 1 0
220 6 0
417 27 0
1,563 69 2
71 7 0 4 0 7,509
161 14 1 12 1 5,231
12 0 0 0 0 1,872
21 4 2 3 0 2,879
DT NPPR NPHD DT NPPR NPPR NPHD DT NPPR NPPR NPPR NPHD DT NPHD NPPO DT NPHD NPPO NPPO DT NPHD NPPO NPPO NPPO DT NPPR NPHD NPPO DT NPPR NPPR NPHD NPPO DT NPPR NPPR NPPR NPHD NPPO DT NPPR NPHD NPPO NPPO DT NPPR NPHD NPPO NPPO NPPO DT NPPR NPPR NPHD NPPO NPPO
LIM DT NPPR NPPR NPHD LIM DT NPHD NPPO LIM DT NPHD NPPO NPPO LIM DT NPPR NPHD NPPO LIM DT NPPR NPPR NPPR NPHD NPPO LIM DT NPPR NPHD NPPO NPPO LIM NPHD LIM NPPR NPHD
NPPR NPHD NPPR NPPR NPHD NPPR NPPR NPPR NPHD NPHD NPPO NPHD NPPO NPPO NPHD NPPO NPPO NPPO NPPR NPHD NPPO NPPR NPPR NPHD NPPO NPPR NPPR NPPR NPHD NPPO NPPR NPHD NPPO NPPO NPPR NPHD NPPO NPPO NPPO
346 81 35 10 3 0 23 4 1 0 5,496 22,987
Appendix E
174 Table E-2: Relative frequencies of prototypical NPs in Corpus A
Non- Drama Spon. Scrip. Total % fiction speech speech 18.80 14.39 16.23 14.43 19.94 17.20
Fiction DT NPHD
6.10 0.60 0.04
10.13 1.58 0.08
3.62 0.28 0.03
3.88 0.25 0.03
7.73 1.11 0.08
6.39 0.76 0.05
5.42 0.26 0.00
7.91 0.89 0.06
3.02 0.06 0.00
3.69 0.10 0.01
8.28 0.75 0.08
5.79 0.41 0.03
DT NPPR NPPR NPHD NPPO NPPO
2.23 0.22 0.02 0.14 0.00 0.02
4.02 0.43 0.00 0.23 0.00 0.01
0.69 0.03 0.00 0.02 0.00 0.00
0.92 0.08 0.00 0.01 0.01 0.00
3.32 0.43 0.04 0.40 0.00 0.06
2.32 0.25 0.01 0.17 0.00 0.02
LIM DT NPHD
0.25
0.06
0.29
0.21
0.13
0.19
LIM DT NPPR NPHD
0.07 0.03
0.01 0.00
0.08 0.00
0.05 0.00
0.03 0.01
0.05 0.01
0.08 0.00
0.06 0.01
0.08 0.00
0.07 0.01
0.11 0.02
0.08 0.01
LIM DT NPPR NPHD NPPO NPPO
0.05 0.01 0.01
0.07 0.00 0.00
0.00 0.00 0.00
0.01 0.00 0.00
0.01 0.00 0.00
0.03 0.00 0.00
LIM NPHD
0.23
0.02
0.14
0.30
0.14
0.18
LIM NPPR NPHD LIM NPPR NPPR NPHD
0.01 0.01
0.03 0.00
0.02 0.00
0.06 0.00
0.00 0.00
0.02 0.00
LIM NPHD NPPO
0.05
0.07
0.06
0.02
0.06
0.05
LIM NPPR NPHD NPPO
0.01
0.01
0.00
0.00
0.00
0.01
NPPR NPHD
2.28 0.17 0.02
9.98 0.99 0.08
2.09 0.17 0.02
2.54 0.27 0.03
4.40 0.46 0.01
3.99 0.38 0.03
2.30 0.04 0.00
4.20 0.30 0.02
1.72 0.02 0.00
2.26 0.06 0.00
4.00 0.26 0.00
2.87 0.13 0.00
0.38 0.04 0.00 0.02 0.00 39.89
1.78 0.15 0.01 0.13 0.01 57.70
0.18 0.00 0.00 0.00 0.00 28.83
0.22 0.04 0.02 0.03 0.00 29.62
0.78 0.10 0.00 0.04 0.00 52.78
0.63 0.06 0.01 0.04 0.00 42.16
DT NPPR NPHD DT NPPR NPPR NPHD DT NPPR NPPR NPPR NPHD DT NPHD NPPO DT NPHD NPPO NPPO DT NPHD NPPO NPPO NPPO DT NPPR NPHD NPPO DT NPPR NPPR NPHD NPPO DT NPPR NPPR NPPR NPHD NPPO DT NPPR NPHD NPPO NPPO DT NPPR NPHD NPPO NPPO NPPO
LIM DT NPPR NPPR NPHD LIM DT NPHD NPPO LIM DT NPHD NPPO NPPO LIM DT NPPR NPHD NPPO LIM DT NPPR NPPR NPPR NPHD NPPO
NPPR NPPR NPHD NPPR NPPR NPPR NPHD NPHD NPPO NPHD NPPO NPPO NPHD NPPO NPPO NPPO NPPR NPHD NPPO NPPR NPPR NPHD NPPO NPPR NPPR NPPR NPHD NPPO NPPR NPHD NPPO NPPO NPPR NPHD NPPO NPPO NPPO
Total %
APPENDIX F: Illustration of the types of test in the original test design 1. Composition Rating Give all possible sentences that can be constructed using all of the words or word groups once. (At least one sentence is possible.) on how to win the game, the matter, no thought, he gave 2. Composition Select one Construct a sentence using all of the words or word groups once. Fill in the sentence that comes to mind first. was dead, a rumour, that the president, was spread 3. Operation Compliance Change the active sentence into a passive sentence. They made plans to go home 4. Operation Selection Rewrite the sentence, starting with: This is a ... This house is larger than yours 5. Completion Forced Choice Fill one blank with ‘sold’ and the other with ‘called’. She < > him a dog yesterday with a bad heart She < > him a dog with a bad heart yesterday 6. Completion Word Placement Rewrite the sentence using ‘a few days ago’ with it. Tom had a telegram saying Mary was ill 7. Completion Supplementation Complete the sentence in any way you like. Something in his eyes told me I couldn’t trust him. There was …
176
Appendix F
8. Evaluation Judge acceptability on a seven-point scale (one = perfectly acceptable, seven = totally unacceptable). She gave the play a bad criticism in which he plays a priest 9. Preference Rating Read both sentences, then judge the acceptability for each of them on a sevenpoint scale (one = perfectly acceptable, seven = totally unacceptable). No more was heard of him No more of him was heard 10. Preference Ranking Mark the sentences with order of preference (1 = most preferred). Today they elected him president of the company They elected him president today of the company Of the company they elected him president today They elected him president of the company today President of the company they elected him today 11. Similarity Read both sentences, then judge their similarity on a seven-point scale (one = completely similar in meaning, seven = completely different in meaning). The space below has been found empty The space has been found empty below 12. Frequency Read both sentences, then judge for each of them their frequency of occurrence on a seven-point scale (one = very frequent, seven = very rare). On your PC a screen appears listing the titles On your PC a screen listing the titles appears 13. Normalcy Read both sentences, then judge the normalcy for each sentence on a seven-point scale (one = very normal, seven = very strange). Everyone except Irena started asking questions Everyone started asking questions except Irena
APPENDIX G Sentences used in the elicitation experiment The following tables give an overview of the sentences which were used in the experiment to elicit the different types of variant NP. The first number in the code divided by the dashes indicates the period in which the test was presented to informants (1=Nov/Dec 96; 2=May 97). The roman number indicates the battery in which the sentence was included and the third number indicates the type of test by means of which it was elicited (according to the numbers in Table 1 of Chapter 2). Thus, 1-II-2 means that the sentence was elicited by means of a Composition, Select One test in the second battery of the first round of elicitation, 2-I-8 means that the sentence was elicited by means of an Evaluation test in the first battery of the second round of elicitation, etc. Fronted premodification: Construction production 1. (I think) He has much too firm a grip on us 1-II-2 1-III-7 2. He asked too large a group of people to come 3. He did exceptionally brave a deed (yesterday) 2-II-2 4. He has too firm this grip on us 1-I-5 5. He is so arrogant a person 6. He told truly astonishing a story 2-I-5 7. How could you ask so unnecessary a question 1-III-7 8. I consider Bill/John far too arrogant a person 2-II-2 9. I had so romantic a long evening yesterday! 10. I have never met quite so arrogant a person 1-II-2 before 11. I have never seen so powerful a machine before 12. I never saw so beautiful a house 13. I never saw so beautiful a person 14. I never travelled so short a distance by car 15. I never travelled this short a distance by car 16. I rather think this is too small a scale 17. I think this is rather too small a scale 18. I think this is too difficult to answer a question 19. It depends on how strong the/a person you are 1-II-6 20. It goes back too long a time 21. It goes back too long for you to remember a time 2-II-2 22. Kim thought it was true, so naive a child 2-II-2 23. Kim, so naive a child, thought it was true 24. Last year he stayed out far longer a time 25. Linda is so elegant a dancer! 26. Paul and I had so romantic an evening 1-III-7 27. She is so proud a mother 28. The man asked too large a group of people to 2-II-6 come with him 29. The teacher called Bill so eloquent a speaker 30. This is far more difficult a question 1-III-4 2-III-4
judgment 2-III-10 2-III-10 2-I-11
2-I-8 1-I-8 1-II-12 1-II-9 1-III-10 1-II-9 1-III-13 2-I-11 2-I-11 2-II-9 1-I-8 2-I-8 2-III-13 2-III-13 2-I-8 1-III-13 2-I-11 1-I-8
2-II-12
Appendix G
178 31. 32. 33. 34. 35. 36. 37. 38.
This is perfectly boring a film This is quite complicated a question This is rather too small a scale This/Tom is so eloquent a speaker This is too easy a job This is too easy to do a job Tom, very convincing a person, told her to go Why not give so hungry a child a little bread?
2-III-4 1-II-6 2-III-4 1-III-4 2-III-4 2-III-4
2-I-1
1-II-12 2-II-9 2-I-8 2-II-9 2-III-10
Discontinuous modification: Construction 39. (Friday) He told me of such things as war and hate 40. Bill is a more capable man of murder than his brother 41. Frank is probably the poorest Pendleton of them all 42. Fred is a clever enough man to look through that 43. Hate isn’t an easy emotion to hide 44. He is a better Romeo than Tom ever was 45. He is a brilliant man in many ways 46. He is a too clever man to miss the train 47. He is such an arrogant person 48. He plays a more convincing Romeo than Tom 49. He travelled to such exotic places as Andorra 50. He was speaking in the same tone as before 51. He was the most arrogant person I have ever encountered 52. His illness made him a very different person from his friends 53. How could you ask such an unnecessary question 54. I consider John an easy man to persuade 55. I have never seen such a powerful machine before 56. I have seen some (much) bigger disasters than this one 57. It goes back a too long time for you to remember 58. It is of much greater importance than I thought 59. It occurs with a much higher frequency than I expected 60. It was of much greater importance than I thought at first 61. Last year he stayed out a far longer time than today 62. Linda is such an elegant dancer! 63. Mandy wants the same things as you do 64. Mary left, a prouder person than before
production
judgment 1-II-12 1-III-10
2-II-6
2-I-8
2-III-7
2-I-8
2-I-1 1-III-7 2-I-5
2-II-9 2-III-10
1-I-11 1-III-13 1-III-13 2-I-11 2-II-12 2-III-13 1-III-13 1-II-12 2-II-12 1-II-12 1-III-7 1-I-8 1-III-10 1-II-12 2-III-7 2-III-13 1-II-2 2-II-12 2-II-12 2-III-13 1-III-13 1-II-12 2-I-8
Appendix G 65. She gave more convincing evidence than yesterday 66. She is such a proud mother 67. She is the coolest lady I have ever met 68. She told a more convincing story than yesterday 69. That was the strongest gust yet 70. The book was of greater value than she expected 71. There are now many easier things to lay hands on 72. There is something more exciting going on than that 73. These are two particularly complicated sums to calculate 74. They asked the same questions as before 75. This class contains a too noisy thirty for me to handle 76. This is a (much) higher frequency than I expected 77. This is a (much) larger house than yours 78. This is a different object from/to/than the one you saw before 79. This is a different question/sentence from the one you saw before 80. This is a larger house than yours 81. This is a particularly complicated sum to calculate 82. This is a too easy job to do 83. This is a too short distance to travel by car 84. This is an easy job to do 85. This is quite a complicated question 86. This is rather a small scale 87. This is such an eloquent speaker 88. This is the same sweater (that/as) you wore/were wearing last year 89. This was my worst nightmare ever/of all 90. This was the/my worst nightmare ever 91. Today we ate three even bigger apples than yesterday 92. Tom is such an eloquent speaker 93. Tom was announced as the best player ever 94. We gave a completely different department from ours the assignment 95. You asked a less intelligent/dumber student than before (to do the job) 96. You have a much more macabre mind than I have 97. You have this more positive attitude than me/Sue
179 2-II-9 2-II-2
2-I-11 2-II-9 2-III-13 2-II-12 2-III-10
2-III-7 2-II-9 2-III-10 1-II-12 2-III-10 2-I-11 2-II-2 1-III-4 2-III-4 1-III-4 1-I-11
1-II-9
1-III-4 1-III-4 2-III-4 1-III-10 2-III-4 1-II-6 2-III-4 2-III-4 1-III-4 2-III-4
2-I-11 2-II-12 2-III-13
2-I-5 2-I-1
2-I-1 2-II-2
2-II-9 1-II-12 2-II-12 2-II-9 2-III-10 2-III-13 1-II-9
2-I-5
2-II-12
Appendix G
180 Fronted + discontinuous: Construction 98. Fred is clever enough a man to look through that 99. He does hardly as useful a job as mine/as I do 100. He is too clever a man to miss the train 101. He regarded it too difficult a question to answer 102. It goes back too long a time for you to remember 103. Last year he stayed out far longer a time than today 104. She told more convincing a story than yesterday 105. This class contains too noisy a thirty for me to handle 106. This is quite as heavy a bag as mine 107. This is too easy a job to do 108. This is too short a distance to travel by car 109. This was too difficult a problem for Tom
production
judgment 2-II-9 2-III-10
2-III-7 1-II-2 2-II-2 2-II-6
1-III-13 1-III-13 2-III-13 2-III-13
2-II-2
2-III-10 2-I-8
1-III-4 2-III-4 2-III-4
1-I-11 2-II-9 1-III-10
1-II-6
Floating deferred modification: Construction 110. A car was stolen that colour yesterday 111. A car was stolen yesterday that colour 112. A girl appeared of your age smoking a cigarette 113. A girl appeared your age smoking a cigarette 114. A girl appeared your age, smoking a cigarette 115. A girl your age appeared, smoking a cigarette 116. A/The rumour was spread that the president was dead 117. Every time a new screen appears when you click that button 118. Everyone started asking questions except Irena 119. Everything was quite still around 120. Green Summer, the house, as I see it, of my childhood, was sold at last 121. He gave the matter no thought on how to win the game 122. He gave the order to his commanders to withdraw 123. His figure was that still of a young athlete 124. His figure was that still of a young man 125. I bought a book last week on trains 126. I bought an interesting book yesterday on trains 127. I discovered a hole this morning the size of an apple 128. I forgave the man everything who stole my bike 129. It is where evenings usually end here 130. Mary was a tower of strength to me here 131. No more was heard of him 132. On your PC a screen appears listing the titles 133. Plans were made to deal with the situation 134. Reports have come in all day about strange occurrences 135. She called him a dog yesterday with a bad heart 136. She gave the play a bad criticism in which he plays a
prod. 2-III-7
1-II-2 1-I-1 1-II-2
1-I-1
judgment 2-I-8 2-III-13 2-III-13 2-I-11 2-I-11 1-III-13 1-I-11 1-III-13 2-I-11 2-I-8 1-II-12 1-III-13 1-III-10 1-II-9
1-I-5 1-I-11 1-III-10 1-II-6 1-I-1
1-I-11 1-II-9 1-III-10 2-I-11 2-I-11 1-I-8 1-II-9 1-II-12
1-II-2 1-II-9 1-I-5 1-I-8
Appendix G
181
priest 137. Ten days, however, after they married they had their first fight 138. The minute he went home everybody started laughing 139. The moment they entered the room the situation changed 140. The space has been found empty below 141. There is good evidence, however, that he is guilty 142. There was no doubt in my mind that I had won 143. There was something in his eyes I couldn’t trust 144. They elected him president today of the company 145. They will appoint her Head tomorrow of the school 146. Those we invited to our party we love 147. Tom had a telegram a few days ago saying Mary was ill 148. We invited those to our party we know from London 149. We invited those to our party we love
1-I-8 1-I-11 1-I-11 1-I-11 1-II-2 1-II-6 1-III-7 1-I-1
1-II-9 1-III-10 1-I-8 1-III-10
1-II-2 1-I-5 1-III-10
Distractor sentences: Construction 150. Cruel Tom hit the dog yesterday 151. He was at a complete loss 152. I saw a very strange looking man in the park yesterday 153. I saw a very strange man in the park yesterday 154. I think we should/ought to go now 155. John went running to the park with Mary 156. John went with Mary to the park yesterday 157. Last night I thought we would win 158. Similarities between the two have been found 159. The dog should not be kept unleashed 160. This is a boy who wrote me a note 161. This is the man (who/that) I talked to a few days ago 162. Tom sold Irena a book the other day
production 2-II-2 1-I-5 2-I-5
judgment
1-I-8 1-II-6 2-II-6 2-III13 2-II-2 1-I- 1-II- 1-III- 2-I- 2-II- 2-III11 12 10 11 12 10 2-III10 1-I-8 2-II-2 1-III-4 2-III4 1-III-4 2-III4 1-III- 2-III13 13
APPENDIX H: Average scores of the judgment tests The tables below give the average scores of informants for the judgment tests. The numbers in the ‘Construction’ column correspond to the numbers of the sentences in Appendix F. If the number is in normal lay-out, the elicited sentence corresponds exactly to the Construction in Appendix F. If the number is in italics, a different (non-variant) version of the Construction was elicited. In the Similarity test both the Construction in F and its non-variant counterpart were offered to the informants.
Construct. 136 137 131 158 145 152 19 54 9 29
First period, Battery I, 14 informants Evaluation Similarity Mean Std. Dev. Construct. Mean 4.50 1.09 125 1.71 2.86 1.23 106 2.50 2.64 1.08 138 6.71 1.36 0.63 156 1.57 4.79 1.48 45 3.43 1.21 0.43 118 2.14 1.43 1.34 139 6.21 1.36 0.50 140 3.71 6.50 0.65 127 1.79 4.86 1.23 79 2.71
Std. Dev. 0.83 1.61 0.76 0.85 1.95 2.11 1.31 1.54 0.89 1.54
Construct. 20 64 41 105 111 24 37 8 40 120
Second period, Battery I, 25 informants Evaluation Similarity Mean Std. Dev. Construct. Mean 3.24 1.67 129 3.32 1.68 0.75 114 4.28 1.64 0.95 17 2.56 5.56 1.47 5/47 2.04 3.80 1.53 156 1.40 4.20 1.66 119 3.40 4.16 1.95 74 5.20 1.56 0.92 27/66 2.64 3.20 1.44 130 4.48 4.40 1.78 88 3.60
Std. Dev. 1.68 1.88 1.29 1.37 0.58 2.10 2.00 1.68 1.87 1.98
184
Appendix H
First period, Battery II, 14 informants Preference Rating Frequency Construct. Mean Std. Dev. Construct. Mean 12 3.36 1.69 72 3.29 12 72 1.86 0.95 2.21 163 2.21 1.42 34 4.36 163 1.43 0.65 92 1.50 131 1.21 0.58 132 2.14 131 132 4.36 1.55 2.36 134 2.14 1.61 52 1.79 134 52 2.07 1.27 4.43 156 96 2.14 1.17 2.00 96 2.93 1.38 156 3.14 14 1.64 0.84 55 1.43 14 5.57 0.94 11 3.93 128 5.71 1.20 50 1.36 128 50 5.29 2.05 4.71 39 79 1.50 1.09 2.79 79 3.86 1.23 39 3.14 144 5.36 1.78 121 3.71 144 121 2.43 2.06 3.93 63 123 5.07 1.38 3.36 123 1.64 1.15 63 2.21
Std. Dev. 1.77 1.12 1.34 0.65 1.66 1.15 0.80 1.22 1.30 1.75 0.76 1.07 0.63 1.86 1.48 1.46 1.77 1.38 1.82 1.89
Second period, Battery II, 25 informants Preference Rating Frequency Construct. Mean Std. Dev. Construct. Mean 38 2.60 1.26 97 2.72 38 97 4.92 1.78 5.24 6 91 3.12 1.76 1.56 91 6 3.04 1.70 4.72 70 1.16 0.37 89 1.36 70 3.12 1.64 89 2.72 67 1.28 0.54 47 1.08 67 6.56 0.82 5 4.12 68 94 2.44 1.66 3.84 94 2.40 1.04 68 1.56 156 42 1.48 0.77 1.32 98 3.08 1.26 156 2.88 59 163 1.24 0.52 2.84 163 1.56 0.96 59 1.64 93 107 1.48 0.82 6.44
Std. Dev. 1.43 1.33 0.96 1.59 0.91 1.43 0.28 1.56 1.34 0.58 0.48 1.72 1.31 0.76 0.96
Appendix H 36 65 65 18 18
185 5.56 1.52 2.28 3.96 4.48
1.56 0.92 1.59 2.07 1.76
93 51 51 60 60
1.36 1.16 5.88 1.68 5.36
0.76 0.37 1.24 0.75 1.29
First period, Battery III, 14 informants Preference Ranking Normalcy Constr. Mean Std. Constr. Mean Std. Constr. Mean Dev. Dev. 149 122 1.71 1.07 3.00 1.11 45 1.29 122 2.71 0.61 149 4.21 0.89 25 3.21 122 149 1.71 0.61 2.93 1.00 62 1.00 122 149 45 4.00 0.39 3.50 1.51 3.29 122 149 4.86 0.53 1.36 0.84 15 2.07 144 3.57 0.85 128 1.75 0.75 121 3.07 144 128 15 1.71 0.47 2.33 0.78 3.21 144 128 121 4.29 0.73 4.67 0.49 3.00 144 128 1.29 0.47 2.08 1.24 162 1.14 144 128 162 4.14 0.77 4.17 0.58 1.36 39 118 13 1.57 0.51 4.00 0.94 1.36 13 3.36 0.50 39 2.60 1.51 49 2.21 13 39 49 3.79 0.58 3.20 1.40 1.29 13 39 4.86 0.53 2.90 1.20 118 1.93 13 39 1.43 0.51 1.60 0.84 101 1.86 125 101 108 1.14 0.36 1.79 0.70 1.86 108 125 2.14 0.66 1.57 0.65 116 1.43 108 116 3.14 0.77 125 2.64 0.74 3.43 108 125 3.64 0.74 4.57 0.51 100 1.36 108 125 4.93 0.27 4.43 0.51 46 4.64 54 156 3.00 0.88 3.64 0.74 156 2.71 1.33 54 1.43 0.85 156 54 1.86 0.95 2.21 1.05 156 54 5.00 0.00 4.79 0.58 156 54 2.43 1.09 2.93 0.92
Std. Dev. 0.61 1.05 0.00 0.99 1.54 1.38 1.67 1.71 0.36 0.50 0.63 1.67 0.61 1.14 1.23 0.95 0.65 1.70 0.63 1.60
186
Appendix H
Second period, Battery III, 24 informants Preference Ranking Normalcy Constr. Mean Std. Constr. Mean Std. Constr. Mean Dev. Dev. 2 38 2.21 1.22 2.88 1.08 112 3.67 38 2 2.37 0.92 2.96 1.27 48 1.54 38 2 48 3.96 0.86 3.92 1.02 3.25 38 4.42 1.02 2 1.13 0.34 113 4.00 38 2 2.04 1.12 4.13 0.80 162 1.29 42 1.62 0.77 73 1.13 0.34 90 1.50 42 73 4.00 0.88 4.46 0.78 102 1.75 73 98 2.25 1.03 2.21 0.72 90 2.04 98 73 162 4.46 0.78 3.58 0.88 1.50 42/98 73 2.67 1.20 3.62 1.06 57 5.04 156 3 4.08 0.97 2.79 1.22 22 3.09 3 1.67 0.64 156 2.71 1.23 23 2.00 3 156 3.37 1.01 1.87 0.99 154 1.13 3 156 1.58 0.65 4.92 0.28 154 2.29 3 156 4.29 0.75 2.71 1.00 95 2.42 71 95 157 2.00 0.59 3.32 1.49 2.75 157 71 2.88 0.45 3.50 1.47 67 1.25 157 71 67 1.21 0.51 2.50 1.26 6.00 157 71 4.50 0.88 2.95 1.36 61 2.50 157 4.37 0.49 71 2.73 1.39 103 3.50 94 2.79 0.88 68 1.29 0.75 94 1.92 1.02 104 2.37 0.65 94 68 4.67 0.76 4.17 0.48 94 1.83 1.05 68/104 2.54 0.88 94 3.79 0.83 68/104 4.63 0.77
Std. Dev. 1.46 0.93 1.59 1.44 0.55 0.78 0.99 1.04 0.59 1.08 2.19 1.13 0.45 1.68 1.25 1.59 0.53 1.10 1.22 1.14
APPENDIX J: Questions which were asked in the interviews 1. 2. 3. 4.
What do you think the aim of this questionnaire is? Did you have any problems answering the questions? Did you feel it was difficult to give judgments on acceptability and frequency/normalcy/similarity? Would you judge the following sentences as acceptable or unacceptable? Why?
Group I: a. I forgave the man everything who stole my bike. b. The teacher called Bill so eloquent a speaker. c. She gave the play a bad criticism in which he plays a priest. d. I never saw a so beautiful house. e. It was of importance much greater than I thought at first. f. Bill is a more capable man of murder than his brother. g. This class contains a too noisy thirty for me to handle. h. I consider John far too arrogant a person. Group II: a. This class contains a too noisy thirty. b. You should not tell so jealous a boy that you met his girlfriend in the pub. c. It succeeded as expected, so very cunning a plan. d. They have two sufficient incomes to buy the house. e. A rock was found that same shape. f. He asked an even larger party than yesterday to come along. g. Tom is a better second of the match than John was. h. He got through some very difficult questions to answer. i. Today she offered evidence more convincing than yesterday. j. A girl appeared your age, smoking a cigarette. Group III: a. Which particularly complicated sums to calculate did he solve? b. You should not tell a jealous boy that you met his girlfriend in the pub. c. Jane won the game, a better player than all the others. d. They have two incomes sufficient to buy the house. e. A rock was found of that same shape. f. He asked a party even larger than yesterday to come along. g. Tom is a better second than John. h. He got through some questions very difficult to answer. i. Today she offered more convincing evidence than yesterday. j. A girl appeared of your age, smoking a cigarette.
Appendix H
188 Group IV: a. They elected him president today of the company. b. We invited those to our party we know from London. c. She gave the play a bad review in which Tom plays a priest. d. They will appoint her Head of the school tomorrow. e. I had so romantic an evening yesterday. f. John is an arrogant, cold person. g. It depends on how strong the person you are. h. I never travelled so short a distance by car. i. This is a question different from the ones you heard before. j. There is something more exciting going on than that. Group V: a. He gave the matter no thought on how to win the game. b. We invited those we know from London to our party. c. His figure was still that, I feel, of a young man. d. They will appoint her Head tomorrow of the school. e. I had so romantic a long evening yesterday. f. John is far too arrogant a cold person. g. I have never met so arrogant a person before. h. I never travelled a so short distance by car. i. This is a different question from the one you heard before. j. There is something going on more exciting than that. 5.
Other comments?