No title

Applied Linguistics Table of Contents Volume 22, Issue 1: March 2001.

Editorial M Bygate, K Hyltenstam and C Kramsch pp. i-i Full-Text PDF (149 KB) Incidental vocabulary acquisition in a second language: the construct of taskinduced involvement B Laufer and J Hulstijn pp. 1-26 Abstract Full-Text PDF (149 KB) Task complexity, task difficulty, and task production: exploring interactions in a componential framework P Robinson pp. 27-57 Abstract Full-Text PDF (320 KB) Interaction in academic writing: learning to argue with the reader G Thompson pp. 58-78 Abstract Full-Text PDF (132 KB) Preposition entries in UK monolingual learners' dictionaries: problems and possible solutions S Lindstromberg pp. 79-103 Abstract Full-Text PDF (165 KB) Interpreting the discourse of HG Widdowson: a corpus-based critical discourse analysis R de Beaugrande pp. 104-121 Abstract Full-Text PDF (117 KB) Review. Corpus Linguistics. Investigating Language Structure and Use. D Biber, S Conrad, R Reppen G Barnbrook pp. 122-123 Details Full-Text PDF (37 KB) Review. Second Language Attrition in Japanese Contexts. L Hansen [ed] E Yukawa pp. 124-129 Details Full-Text PDF (37 KB) Review. Language Play, Language Learning. G Cook JA Belz pp. 129-134 Details Full-Text PDF (61 KB)

Review. Resisting Linguistic Imperialism in English Teaching. A Suresh Canagarajah D Johnson pp. 134-136 Details Full-Text PDF (46 KB) Review. Good to Talk? Living and Working in a Communication Culture. D Cameron G Cook pp. 136-140 Details Full-Text PDF (52 KB)

Applied Linguistics 22/1

# Oxford University Press 2001

Editorial From volume 22, Applied Linguistics introduces two small changes, ®rstly concerning policy on the publication of article-length responses to earlier articles, and secondly concerning the introduction of a new `Forum' section to the journal. Article-length responses to papers are subject to the normal refereeing process. In order to help to speed up the process, as far as possible the journal will publish article-length responses in the ®rst issue going to press following receipt of the ®nal version of the manuscript. In contrast, other papers are published in chronological order of receipt. We ask contributors to note, however, that normally no more than one such response will be published in any given issue. We hope that this will enable faster publication of articlelength responses. The ®rst paper published under this procedure is an article by Robert de Beaugrande, which appears in this issue. In this volume we also introduce a new `Forum' section to the journal. This section is intended to enable the publication of short contributions in a way which can allow a more rapid turn-around than do standard articles. We hope this might o€er a new interactive dimension to the journal. Pieces are welcome in the form of short responses to previous articles; notices about research in progress; items about the publication process itself; and pieces about research themes and research policies in applied linguistics, whether in the journal, our sponsoring associations (AAAL, BAAL and AILA), or the wider community. Items will be reviewed in-house and not usually sent to outside review, with the aim of enabling us to reduce the time lag between writing and publication to a minimum. Three copies of contributions, normally of up to 2000 words maximum, should be sent to either of the Editors or to the Reviews Editor. Martin Bygate, Kenneth Hyltenstam, Claire Kramsch

Applied Linguistics 22/1: 1±26

# Oxford University Press 2001

Incidental Vocabulary Acquisition in a Second Language: The Construct of Task-Induced Involvement BATIA LAUFER and JAN HULSTIJN 1

University of Haifa, Israel and 2University of Amsterdam, The Netherlands The paper makes an attempt to stimulate theoretical thinking and empirical research in the domain of L2 vocabulary learning by introducing a construct of involvement with motivational and cognitive dimensions: Need, Search, and Evaluation. Retention of hitherto unfamiliar words is claimed to be conditional upon the amount of involvement while processing these words. Involvement is operationalised by tasks designed to vary in the degree of need, search, and evaluation. The paper reviews a number of constructs that are currently debated and investigated in the literature on cognitive and motivational aspects of L2 learning. It also re-examines the existing empirical literature on task e€ect in the light of the proposed construct of task-induced involvement, stresses the need for deepening and broadening the construct, and discusses possibilities it o€ers for research on vocabulary learning.

Virtually all second language learners and their teachers are well aware of the fact that learning a second language (L2) involves the learning of large numbers of words. Not surprisingly, many learners are somewhat apprehensive when faced with such an enormous task and teachers as well as learners have always shown a keen interest in ®nding out how words can best be learned. For many years, instructional practice has been based on the view, unanimously supported by psychologists, that elaboration on features of new words promotes their retention (Anderson 1995: ch. 6; Baddeley 1997: ch. 7). This proposition is based on what learning psychologists have been repeating for many decades after William James (1890: 662), namely that `all improvement of the memory lies in the line of elaborating the associates'. This means that the more attention that is paid to the formal and semantic aspects of words and the richer the associations that are made with existing knowledge (e.g. in the form of establishing similarities and contrasts between old and new information), the higher are the chances that the new information will be retained. Likewise, all learners and teachers know that motivation promotes success and achievement in L2 learning and that students who experience high amounts of an external or intrinsic drive or need to learn, will achieve higher levels of pro®ciency than students with low levels of drive. In practice, this means that educationalists may attempt to increase students' motivation by

2 INCIDENTAL VOCABULARY ACQUISITION IN A SECOND LANGUAGE

providing tasks and materials which students may experience as interesting and appealing. We entirely agree with the views that elaboration and motivation are key factors in promoting vocabulary learning. However, we note with some concern that the theoretical and the empirical literature on concepts of cognition in the narrow, information-processing sense (such as elaboration and attention), as well as the literature on a€ective components of cognition (such as motivation and need), have not produced substantial progress in the domain of L2 vocabulary learning. In this respect, theory and research on L2 vocabulary learning is lagging behind developments in the domain of L2 grammar learning, where recent years have spurred new theoretical and empirical work on constructs such as `focus on form' (e.g. Long and Robinson 1998), `input processing' (VanPatten 1996), and task-based language teaching (Skehan 1998; Robinson 2000). In this paper, we make an attempt to stimulate both theoretical thinking and empirical research in the domain of L2 vocabulary learning by proposing a construct of task-induced involvement, with three motivational and cognitive dimensions: need, search, and evaluation. We begin with a review of the literature on a number of cognitive constructs (such as attention, implicit and explicit learning, and elaboration) and we brie¯y review the literature on motivation. This is followed by a consideration of the notion of incidental learning, which is crucial for a proper understanding of the literature of incidental L2 vocabulary learning. These three ground-clearing review sections are rounded o€ with a brief summary. Next, we examine the empirical literature on factors promoting incidental L2 vocabulary learning. Subsequently, we present our construct of task-induced involvement load, re-examine the empirical literature in the light of this construct and end our paper with a discussion of the needs to deepen and broaden the construct both theoretically and empirically.

COGNITIVE VIEWS ON THE REPRESENTATION AND ATTAINMENT OF L2 KNOWLEDGE In this section, we will ®rst look at the literature on L2 grammar acquisition (mostly referred to by `SLA', excluding vocabulary) and then review the literature on L2 vocabulary learning, examining which cognitive constructs have been invoked in these two domains. In trying to understand the SLA literature, it is helpful to distinguish between views on how grammatical knowledge is represented in the mind and views on how grammatical knowledge is acquired. With respect to the mental representation of grammatical knowledge, most researchers in the generative school conceptualise L2 grammatical knowledge as consisting of (1) a system of highly abstract principles and parameters dealing with mostly abstract grammatical phenomena such as word order, government, constraints on movement, agreement, tense, and co-reference, which are commonly referred to as Universal Grammar (UG) or as the `core',

BATIA LAUFER and JAN HULSTIJN

3

and (2) the remainder, or the `periphery' of grammatical knowledge, pertaining to phenomena such as regular and irregular forms of noun declension, verb conjugation, grammatical gender, etc. With regard to the core, the discussion centres around the question of the initial state, in particular the issue of how L2 learners, given the fact that they have already set the parameters of UG for their L1, will set, or reset, or not set at all, the parameters of L2. In contrast to the symbolic view, the connectionist view represents language knowledge as a neural network, consisting of a collection of units connected to each other by a set of pathways (Broeder and Plunkett 1994). The network is the product of a bottom-up process in which lower-level units have been grouped with each other to form higher-level units. In this view, it is not necessarily the case that the network (or parts of it) embodies grammatical notions such as CP, NP, AGR, etc. Nor does the network, in comprehending or producing language, apply rules. The fact that the outcome of comprehension or production processes appears to be regular does not imply that rules have been applied. Regular behaviour can also be generated by associative processes based on the relative strength of certain connections standing in competition with the strength of other connections. It should be noted that a generative theory of grammar is only concerned with the representation of knowledge and not with its acquisition or its transition from one stage to another (Gregg 1996, 2000). Connectionists, however, deal with representation and acquisition in a single theory. This will become apparent in a later section, concerned with implicit and explicit learning.

Noticing and attention With regard to the question of how grammatical knowledge is acquired, three positions are commonly distinguished (N. Ellis 1994a; R. Ellis 1993; Hulstijn and De Graa€ 1994), which are referred to as the No Interface, Strong Interface, and Weak Interface positions. According to these positions, explicit, metalinguistic knowledge, e.g. knowledge of the rules to be found in pedagogic grammars, does not a€ect the acquisition of implicit knowledge (No Interface), transfers into implicit knowledge (Strong Interface), or may indirectly a€ect the acquisition of implicit knowledge by focusing learners' attention on features in the input which are critical for the grammatical phenomenon to be acquired (Weak Interface). Schmidt (1994a, 1994b, 2000) further developed the Weak Interface position into his Noticing Hypothesis. The Noticing Hypothesis holds that, for L2 grammar acquisition to take place, it is not necessary to learn metalinguistic principles or rules, but that learners must `notice' critical features in utterances. In the case of subject±verb agreement, for instance, learners must notice the co-occurrence of certain subject and verb forms. Schmidt (2000) de®nes `noticing' as the subjective correlate of what

4 INCIDENTAL VOCABULARY ACQUISITION IN A SECOND LANGUAGE

psychologists call attention, roughly equivalent to `clear perception' and `detection within selective attention' (Tomlin and Villa 1994; see also Carr and Curran 1994). Reviewing the psychological literature on attention, Schmidt (2000) concludes that although there may be some forms of learning without attention (the evidence on unattended learning is highly contested in the psychological literature), it is known that preparatory attention and voluntary orienting vastly improve encoding: Since many features of L2 input are likely to be infrequent, non-salient, and communicatively redundant, intentionally focused attention may be a practical (though not theoretical) necessity for successful language learning. Language learners who take a totally passive approach to learning, waiting patiently and depending on involuntary attentional processes to trigger automatic noticing, are likely to be slow and unsuccessful learners (Schmidt 2000).

Schmidt argues that there is sucient ground to motivate an attentionally determined encoding-speci®city hypothesis for SLA, acknowledging that this is insucient to settle the issue of explaining the acquisition of many of the abstract features of UG. The growing interest in attention, `focus on form' (Long and Robinson 1998), `input processing' (VanPatten 1996) and `input enhancement' (Sharwood Smith 1993) has produced a still growing number of empirical studies designed to investigate which attention directing tasks may promote L2 grammar learning (e.g. Doughty 1991; Harley 1998; Hulstijn 1989; Lee 1998; Leow 1998; VanPatten and Oikkenon 1996; Williams 1999).

Implicit and explicit learning The discussion in the SLA literature on the question of how L2 grammatical knowledge is acquired has not only been in¯uenced by the issue of attention and noticing but also by the related distinction between implicit and explicit learning: Implicit learning is acquisition of knowledge about the underlying structure of a complex stimulus environment by a process which takes place naturally, simply and without conscious operations. Explicit learning is a more conscious operation where the individual makes and tests hypotheses in a search for structure. Knowledge attainment can thus take place implicitly (a nonconscious and automatic abstraction of the structural nature of the material arrived at from experience of instances), explicitly through selective learning (the learner searching for information and building then testing hypotheses), or, because we can communicate using language, explicitly via given rules (assimilation of a rule following explicit instruction) (N. Ellis 1994a: 1±2).

Schmidt (1994b) argues that attention to input is necessary for explicit learning and may be both necessary and sucient for implicit learning. According to N. Ellis (1994a: 3±4), many issues concerning implicit and

BATIA LAUFER and JAN HULSTIJN

5

explicit learning are as yet unresolved. What aspects of grammar and vocabulary can be learned implicitly? How modular and inaccessible is implicit learning? What are the various mechanisms of explicit learning? Are there sensitive periods for implicit language learning? What are the neural substrates of these processes? With regard to vocabulary learning, N. Ellis (1994b, 1994c) claims that the perceptual aspects of new words, i.e. acquiring their phonetic and phonological features, are learned implicitly as a result of frequent exposure. Similarly, the motor aspects of the articulation of word forms develop implicitly as a result of practice. However, the meaning of words is learned explicitly, requiring the conscious processing at the semantic and conceptual levels and paying attention to the form±meaning connections. Successful learners use sophisticated metacognitive learning strategies, such as inferring word meanings from context and semantic or imagery mediation, in this endeavour.

Depth of processing The notion of implicit learning, as de®ned by Ellis, and the distinction between implicit and explicit learning, have their origins in connectionism, which emerged as a new perspective in cognitive psychology in the 1980s (see Broeder and Plunkett 1994, for an introduction). In the early 1970s, however, an important breakthrough occurred in the study of learning and memory, with the seminal work of Craik and Lockhart (1972), to which we will now turn. Their depth of processing hypothesis, and the notions and hypotheses of other scholars that grew out of it in subsequent years, can, from the current viewpoint of connectionism, be categorised as dealing with explicit learning. Craik and Lockhart argued that the chance that some piece of new information will be stored in long-term memory is not determined by the length of time that it is held in short-term memory but rather by the shallowness or depth with which it is initially processed. They further postulated several levels of processing depth. For instance, processing the meaning of a new lexical item takes place at a rather deep level whereas processing the phonological form takes place at a rather shallow level. Craik and Lockhart were initially successful in providing evidence that semantic processing of lexical items resulted in higher retention than phonological or orthographical processing. The levels of processing theory, however, was challenged, re®ned and modi®ed, and eventually even abolished, in the following years. Two of the problems were: (1) What exactly constitutes a `level' of processing, and (2) How do we know that one level is `deeper' than another? For instance, can non-semantic processing tasks still be meaningful activities? In other words, to be meaningful, an orienting task need not involve thinking about the meaning of a word; it could just as well involve thinking about its pronunciation or spelling. Craik and Tulving (1975) suggested that what is critical to retention is not

6 INCIDENTAL VOCABULARY ACQUISITION IN A SECOND LANGUAGE

simply the presence or absence of semantic encoding, but the richness with which the material is encoded. A major obstacle facing all proposals resides in the diculty of providing an unambiguous, operationalisable de®nition of any notion proposed as a replacement for depth of processing, be it `encoding speci®city', `distinctiveness of encoding', `degree of elaboration', `cognitive e€ort', `degree of richness', etc. (Zechmeister and Nyberg 1982, ch. 12; Baddeley 1997). Yet, cognitive psychologists remained uni®ed in their view that `memory performance is determined far more by the nature of the processing activities engaged in by the learner than it is by the intention to learn per se' (Eysenck 1982: 203). Thus, researchers of knowledge representation, information encoding and retrieval, attention, and memory have not yet succeeded in providing adequate theoretical explanations of phenomena of human learning and memory in terms of quality (type) and quantity (duration and frequency) of information processing (Anderson 1995, ch. 6; Baddeley 1997, ch. 7). Yet they agree that processing new lexical information more elaborately (e.g. by paying careful attention to the word's pronunciation, orthography, grammatical category, meaning, and semantic relations to other words) will lead to higher retention than by processing new lexical information less elaborately (e.g. by paying attention to only one or two of these dimensions).

MOTIVATIONAL ASPECTS OF L2 LEARNING IN RELATION TO COGNITION Our discussion of learning and cognition in the previous section was restricted to the literature of cognition exclusively conceived of as information processing. Human beings, however, are, as Gray (1999) observes, not just information-processing devices but they also possess motives and emotions, and they are integrated in a socio-cultural environment. Motivation, emotion, and socio-cultural factors may a€ect the way in which humans process information. Thus, cognition, as the `ability to acquire, organise, remember, and use knowledge to guide behavior' (Gray 1999: 22), can be studied not only in a narrow sense, as in most of the literature on implicit and explicit memory and learning, but also in a broader sense, as in¯uenced by motivations, attitudes, and social and cultural environments. (For an attempt to develop an integrated view of a€ect and information processing in the domain of SLA, see Schumann 1994, and Young and Perkins 1995.) The acquisition of a second language normally requires the learning of many thousands of words, and most L2 learners, being just `human', may approach this heavy learning load with a great deal of apprehension. Therefore, we decided to include at least one aspect of motivation, namely `need' in our construct of task-induced involvement load. In this section, we brie¯y review the relevant SLA literature.

BATIA LAUFER and JAN HULSTIJN

7

Motivation Since the ground breaking work of Gardner and Lambert (1972), the literature on L2 learning motivation distinguishes two types of motivation, in¯uenced by integrative and instrumental orientations (reviews are given by Crookes and Schmidt 1991; R. Ellis 1994: 508±517; Skehan 1989, ch. 4). Integrative orientation involves an interest in L2 learning because of an interest in the people that speak the L2 and their culture, while an instrumental orientation concerns the practical value and advantage of L2 learning. Whereas orientation refers to the underlying reasons of L2 learning, motivation refers to the directed e€ort L2 learners make. Both integrative and instrumental motivation have been shown to be strongly related to L2 achievement. The socio-educational model of Gardner (1985) explains, at a general level, how motivations (as well as self-con®dence and attitudes) are a€ected by the socio-educational environment. However, as Crookes and Schmidt (1991) observe, it does not specify how various educational settings and instructional tasks may a€ect L2 learning motivation and how motivation might in¯uence information processing by promoting learners' attention. Crookes and Schmidt therefore present an agenda for research on the causal relationship between motivation and information processing in SLA. Since the appearance of the Crookes and Schmidt paper we have found no such empirical studies, at least no studies which investigated experimentally how motivation could be fostered or directed to induce L2 grammar or vocabulary learning. Even Skehan (1998) and (Robinson 2000) summing up the many factors contributing to task e€ectiveness, do not explore the question of how tasks might a€ect learners' motivation or need to pay attention to the L2 input. Somewhat closer to the possible relationship between tasks and motivation or need comes DoÈrnyei (1994). In an attempt to gear motivation research speci®cally to L2 learning in classroom settings, DoÈrnyei proposed a threelevel conceptualisation of L2 motivation. At the language level, motivation refers to orientations toward the language, the people that speak it, and their culture. At the learner level, motivation is concerned with the need for achievement (see below) and self-con®dence. At the learning situation level, motivation is considered as a€ected by (1) the syllabus and learning materials, (2) teachers' attitudes and behaviour, and (3) cohesion and goal-orientedness of the learners group. Although DoÈrnyei explicitly acknowledges the importance of the perceived interest, relevance, and satisfaction with the syllabus and learning materials, he does not provide a classi®cation of learning tasks in terms of their potential to a€ect learners' need or motivation to pay attention to the L2 input. Most studies of motivation (and other socio-psychological variables) adopt a correlational approach, measuring motivation by means of self-report questionnaires. There are almost no studies adopting a causal approach by trying to experimentally manipulate learners' motivation. In an exceptional

8 INCIDENTAL VOCABULARY ACQUISITION IN A SECOND LANGUAGE

study, Gardner and MacIntyre (1991) compared vocabulary recall among two groups of psychology students who were given a list of 26 English±French word pairs. Students were native speakers of English in the Canadian province of Ontario; they were not students of French. One group was rewarded with $10 if they succeeded in learning 24 of the 26 words. Students in the other group were just told to do their best with memorising the words. The results showed that the ®nancially rewarded group performed signi®cantly better than the control group. Thus, this study provided some evidence that L2 vocabulary learning can be promoted by incentives. The incentive provided in this study, however, was extrinsic since it stemmed from the condition under which the learning task had to be performed and not from the task itself, or from the materials which were used to perform it. In a similar vein, Crookes and Schmidt (1991) refer to a study of cued recall of word list items (Eysenck and Eysenck 1980) in which an interaction was obtained between (1) absence or presence of monetary reward and (2) the strength of connections between the words and their corresponding retrieval cues. High incentive improved recall to weak retrieval cues but had no e€ect in the case of strong retrieval cues. This result suggests that the learning and retention of dicult words will bene®t more than the learning of easy words from need and motivation, whether they are induced by externally imposed incentives or by intrinsic drives. To our knowledge, no one has yet submitted this suggestion to an empirical test.

Need In the SLA literature, we have come across the term `need', in only three places (Crookes and Schmidt 1991; Oxford and Shearin 1994, and Skehan 1989). Crookes and Schmidt (1991) mention the construct of need as part of Keller's (1983) education-oriented theory of motivation. Keller distinguishes many components of motivation, one of which is called instrumental needs. Instrumental needs are met when the content of a lesson or course matches what students believe they need to learn. Keller observes that humans have needs for achievement, for aliation, and for power. In an exploration of theories of motivation, relevant to L2 learning, Oxford and Shearin (1994) mention two need theories, basing themselves on Landy (1985), who claims that need is primarily a creator of tension. The ®rst one is known as Maslow's hierarchy of needs, claiming that people regress from their lower-order needs, such as self-actualisation, when their higher-order needs, such as the need for food and sleep, are not met. The second theory, (proposed by McClelland et al. 1953 and mentioned by Skehan 1989 as well), is based on the need for achievement, induced by fear of failure. However, to our knowledge, none of the need theories mentioned give a detailed account of how learning tasks could be classi®ed in terms of the degree of need they may evoke in the learner.

BATIA LAUFER and JAN HULSTIJN

9

Summary From the literature reviewed in the preceding sections, we draw the following conclusions: 1 Attention, with noticing as its subjective correlate, appears to play a crucial role in both implicit and explicit language learning. 2 Whether grammatical knowledge can, cannot, or must be attained in an implicit or explicit way, is still an unresolved issue, complicated by di€erent views on how various domains of grammar (abstract principles of the grammar's core, or less abstract regularities and irregularities of its periphery), are represented in the mind (as mental rules, or as a collection of items associated with each other in a neural network). 3 Input and output modules of the lexicon, responsible for perceiving and articulating words, appear to be learned implicitly, whereas word meanings appear to be learned explicitly. Elaborative attention to a word's formal and semantic features is conducive to its retention, but the literature does not specify how `elaboration' is a€ected by tasks. 4 Motivation for learning an L2 is a€ected by the sociocultural as well as by the educational environment re¯ected in learners' perceived interest, relevance of the syllabus and learning materials and satisfaction with them. Although the importance of the educational setting is acknowledged, no speci®c suggestions have been made about the e€ect the various educational settings and instructional tasks may have on L2 learning motivation and about the e€ect motivation might have on information processing by promoting learners' attention. 5 Two studies appear to provide evidence that external incentives, in the form of ®nancial reward, increase performance in learning a list of vocabulary items. The results of one of these studies suggest that the learning of dicult words may bene®t more from the provision of incentives than the learning of easy words. 6 There is at least one theory which explicitly subsumes need under motivation. Need creates tension. We understand this assertion to mean that a mild degree of tension may positively a€ect information processing, and therefore may indirectly a€ect learning. 7 Instrumental need is ful®lled when the content of a lesson or course matches what students believe they need to learn. Again, as in the case of motivation, no speci®c proposals have been made for how instructional tasks might foster instrumental need. 8 In summary, the most important conclusion to be drawn, in the context of our objective to stimulate theory and research on L2 vocabulary learning, is that no attempts were found in the literature to operationalise general constructs such as noticing, elaboration, motivation, or need, at the micro level of learning tasks. The construct of task-induced involvement load, which we will propose, is meant to be a step in this direction.

10

INCIDENTAL VOCABULARY ACQUISITION IN A SECOND LANGUAGE

INCIDENTAL LEARNING Almost all of the many memory studies conducted in the tradition of Craik and Lockhart (1972) and Craik and Tulving (1975), adopted a so-called incidental learning design. In experiments investigating incidental vocabulary learning, learners are typically required to perform a task involving the processing of some information without being told in advance that they will be tested afterwards on their recall of that information. In contrast, participants in an intentional learning situation are told in advance that their recall will be tested afterwards. Incidental learning conditions allow researchers to investigate the e€ect of the particular kind of information processing they are interested in. One method is to expose subjects to the relevant material without instruction to learn (Type I design). This generally means that subjects must perform some sort of orienting task that leads them to experience the material to be tested but does not lead them to expect a later retention test. For example, subjects are given a list of words and asked to correct any spelling errors. Afterwards they are tested on the recall of all the words in the list. Another technique of investigating incidental learning is to ask subjects to learn something, but not the information targeted for subsequent testing (Type II design). For example, we give subjects a text to read and tell them they will be asked to recall the contents of the text. However, what the subjects are not told in advance, is that the text contains some unfamiliar words and that they will be tested afterwards on their recall of these words (see Hulstijn, in press, for an extensive treatment of incidental and intentional learning). The use of the terms incidental and intentional learning in the psychological literature goes back to the beginning of the twentieth century and has been used in experimental psychology for a long time. Hundreds of experiments on incidental and intentional learning have been conducted. Classical readings are Ebbinghaus (1964), Postman (1964), McLaughlin (1965) and Eysenck (1982). In operational terms, incidental and intentional learning can be distinguished simply in terms of prelearning instructions that either do, or do not, forewarn subjects about the existence of a subsequent retention test (Eysenck 1982: 198). Apart from the strictly methodological meaning of incidental learning as it is used in the experimental literature, the term incidental vocabulary learning has also been given a more general, educational meaning, referring to the learning without an intent to learn, or as the learning of one thing, e.g. vocabulary, when the learner's primary objective is to do something else, e.g. to communicate (Schmidt 1994a). Thus, in this more general de®nition, no mention is being made of the presence or absence of a forewarning of an upcoming retention test after the information-processing task. It is in this latter meaning, that incidental learning has become known in the ®eld of second and foreign language pedagogy, the most frequently quoted example being vocabulary learning as the by-product of reading (Krashen 1989). It should perhaps be pointed out that the notions of incidental and

BATIA LAUFER and JAN HULSTIJN 11

intentional learning should not be confused with the notions of implicit and explicit learning or memory, as discussed in the previous section. Although implicit learning can be incidental only (i.e. without learners' awareness of an upcoming retention test, or without learners' deliberate decision to commit information to memory), explicit learning can occur both intentionally and incidentally. The relevance of the above distinctions to L2 vocabulary learning is not dicult to see. Since linking word form to word meaning is an explicit learning activity requiring attention on the part of the learner, vocabulary can therefore be learnt intentionally as well as incidentally. With regard to the empirical study of intentional vocabulary learning, it is hard to ®nd out which factors are responsible for vocabulary retention as researchers have no control over what people do when they decide to commit words to memory. A researcher or a teacher may, for example, suggest the use of the key word method, yet the learner will choose another memorisation strategy with which s/he may feel more comfortable. Incidental learning, on the other hand, can be manipulated and therefore empirically investigated. Teachers and researchers have control over the tasks which they assign in the belief that they are conducive to vocabulary learning. Controlled experiments, aiming to contribute to pedagogy and to the understanding of how words are acquired, will therefore usually adopt the incidental design.

WHAT AFFECTS INCIDENTAL VOCABULARY LEARNING? EMPIRICAL EVIDENCE According to Hulstijn (1992), when learners were asked to infer the meaning of words from context by choosing the correct meaning from multiple choice options, subsequent retention of these words was better than when they were provided with a synonym of the words during a reading task. Luppescu and Day (1993), and Knight (1994) found that students who read a text and looked up unknown words in the dictionary remembered them better than students who read the text without a dictionary. We do not know whether in the latter case learners were trying to infer the meaning of unknown words, or were simply ignoring them. Hulstijn et al. (1996) found that the relatively few words that were looked up in the dictionary during a reading task yielded much higher retention scores on a subsequent test than the same words in the marginal gloss condition. In Paribakht and Wesche (1997), words that were practised in a series of exercises were retained better than words inferred from context. In Newton's (1995) study, items that were negotiated for meaning (by means of requests for clari®cations, or con®rmation checks) were better remembered than non-negotiated items, even by learners who were not using the words but simply observing the negotiation. Similarly, in R. Ellis (1994), interactionally modi®ed input, i.e. input that included clari®cations, resulted in more words that were learnt than pre-modi®ed input (input prepared on the basis of prior student interaction before the experimental task). In the latter case, however, more words were acquired per minute. Several studies

12

INCIDENTAL VOCABULARY ACQUISITION IN A SECOND LANGUAGE

point to the value of output in speaking tasks for retention of new vocabulary. In Ellis and He (1999), interactionally modi®ed output fared better than interactionally modi®ed input. In a story retelling task used by Joe (1995, 1998), items used in original contexts were remembered better than items used in non-original contexts. The only study which examined the e€ect of writing on the retention of new vocabulary is by Hulstijn and Trompetter (1998). They found that looking up the words in an L1±L2 dictionary and incorporating them in a composition was more e€ective than looking up the words in an L2±L1 dictionary for comprehension purposes. The study by Cho and Krashen (1994) is often cited as evidence for vocabulary acquisition through reading. Yet it is interesting to look closely at what the four subjects in the study were doing and how much vocabulary each of them acquired. One of them did not use a dictionary at all and she acquired seven words from a book. The second subject used the dictionary initially only and abandoned it at a later stage. She acquired eight words. The other two subjects used the dictionary consistently and even wrote the words with example sentences in a booklet (though they were not trying to memorise these words intentionally). These subjects acquired 17 and 34 words per book. Hence, the study shows that dictionary use, a self-imposed task of two of the subjects, resulted in better vocabulary learning than reading only. The Table 1 summarises the studies mentioned above. At ®rst glance, there seems to be no connection among all the studies surveyed above. The tasks in some studies are communicative, in others they are not; in some studies, output is required of students, in others, the understanding of input; some studies use spoken language, others written language; some studies use paper and pen; others use computers. And yet, in each study, one task is superior to the other in terms of incidental vocabulary learning. In explaining this di€erence, most authors suggest that the superior task required a deeper level of processing of the new words than the other task. Additional terms that have been used in connection with good retention induced by tasks are: greater depth of processing; better, more intense quality of information processing; degree of elaboration; quality of attention; richness of encoding. These explanations fall under the conclusion presented in the earlier review section that retention of information depends on the nature of the information processing. It should be noted that this explanation is usually a post-hoc explanation of the results. This is not surprising since the concepts of deep processing or elaboration defy simple formalisation and operationalisation. It is not dicult to decide that a totally meaning-oriented task requires deeper word processing than a totally form-oriented task, e.g. locate all words that end in a letter `e', or classify words by their part of speech. Yet in a normal language learning situation, teachers do not assign vocabulary tasks which do not include word meaning. Hence it is hard to decide in advance whether one instructional task requires deeper processing than another. Let us take, for example, three simple tasks learners can do with the word `skinny': (1) looking up its

BATIA LAUFER and JAN HULSTIJN 13

Table 1: Task e€ect on incidental vocabulary learning The more e€ective task

The less e€ective task

Reference

Meaning selected from several options Meaning looked up in a dictionary

Meaning explained by synonym Reading with/without guessing

Hulstijn 1992

Meaning looked up in a dictionary Reading and a series of vocabulary exercises Meaning negotiated Negotiated input Used in original sentences (oral task) Interactionally modi®ed output Used in a composition (L1±L2 look-up)

Meaning provided in a marginal gloss Reading only (and inferring meaning) Meaning not negotiated Premodi®ed input Used in non-original sentences Interactionally modi®ed input Encountered in a reading task (L2±L1 look-up) Reading only, words not looked up

Reading, words looked up in a dictionary (selfimposed)

Knight 1994; Luppescu and Day 1993 Hulstijn et al. 1996 Paribakht and Wesche 1997 Newton 1995 R. Ellis et al. 1994 Joe 1995, 1998 R. Ellis and He 1999 Hulstijn and Trompetter, 1998 Cho and Krashen 1994

meaning in a dictionary and writing a sentence with the word; (2) looking up its meaning and explaining the di€erence between `skinny', `thin', and `slim'; and (3) receiving a sentence with the word and trying to infer its meaning from four alternatives presented by the teacher. There are no de®nite criteria which would help us grade the three tasks in terms of the depth of processing they require. Hence, if our concern is e€ective learning and teaching, we do not know which task is more e€ective than which. And yet research on task e€ectiveness would require the identi®cation of criteria which could be observed, manipulated, and measured.

TASK-INDUCED INVOLVEMENT On the basis of the analysis of tasks surveyed earlier and on the basis of our conclusion drawn from the literature reviewed, we propose to identify the components of incidental tasks which we believe are conducive to the kind of elaborate processing crucial for learning. This proposal should be conceived as a ®rst attempt to stimulate researchers as well as practitioners to

14

INCIDENTAL VOCABULARY ACQUISITION IN A SECOND LANGUAGE

operationalise the general labels of `attention' and `elaboration' into concrete task-speci®c constructs. For now, three such components will be proposed which, taken together, constitute the construct of involvement.

Involvement Our ®rst assumption about determining factors in vocabulary retention is as follows: Assumption One: Retention of words when processed incidentally, is conditional upon the following factors in a task: need, search, and evaluation.

Taken together, these three factors combine into what will be referred to as involvement. Involvement is perceived as a motivational-cognitive construct which can explain and predict learners' success in the retention of hitherto unfamiliar words. We use the label cognitive in its narrow sense, i.e. referring to information processing only, with the exclusion of a€ective aspects of cognition, as explained in the preceding review. The need component is the motivational, non-cognitive dimension of involvement. It is concerned with the need to achieve. We interpret this notion not in its negative sense, based on fear of failure, but in its positive sense, based on a drive to comply with the task requirements, whereby the task requirements can be either externally imposed or self-imposed. If, for example, the learner is reading a text and an unknown word is absolutely necessary for comprehension, s/he will experience the need to understand it. Or, the need will arise during a writing or speaking task when the L2 learner wants to refer to a certain concept or object but the L2 word expressing it is unfamiliar. We propose to distinguish between `moderate' and `strong' need. Need is moderate when it is imposed by an external agent, e. g. the need to use a word in a sentence which the teacher has asked the learner to produce. Need is strong when imposed on the learner by him- or herself. A case in point is a decision to express a concept without knowing the appropriate word for it. In the case of need, moderate and strong subsume di€erent degrees of drive. Search and evaluation are the two cognitive (information processing) dimensions of involvement, contingent upon noticing and deliberately allocating attention to the form±meaning relationship (Schmidt 1994a, 2000). Search is the attempt to ®nd the meaning of an unknown L2 word or trying to ®nd the L2 word form expressing a concept (e.g. trying to ®nd the L2 translation of an L1 word) by consulting a dictionary or another authority (e.g. a teacher). Evaluation entails a comparison of a given word with other words, a speci®c meaning of a word with its other meanings, or combining the word with other words in order to assess whether a word (i.e. a form±meaning pair) does or does not ®t its context. If, for example, during a reading task, a word that is

BATIA LAUFER and JAN HULSTIJN 15

looked up is a homonym, a decision has to be made about its meaning by comparing all its meanings against the speci®c context and choosing the one that ®ts best. Another example is an L2 writing task in which an L1 word is looked up in a dictionary and three L2 alternatives are presented. The translations have to be evaluated against each other and the most suitable one has to be chosen for the speci®c meaning the L2 writer is trying to convey. But unlike in the preceding example, the evaluation in the writing task will involve additional syntagmatic decisions about the precise collocations of the word which the learner is trying to use. Evaluation, as illustrated by the two examples above, implies some kind of selective decision based on a criterion of semantic and formal appropriateness (®t) of the word and its context. If the evaluation entails recognising di€erences between words (as in a ®ll-in task with words provided), or di€erences between several senses of a word in a given context, we will refer to this kind of evaluation as `moderate'. If, on the other hand, evaluation requires making a decision about additional words which will combine with the new word in an original sentence or text, we will refer to it as `strong' evaluation.

Involvement load A real-life communicative situation, or a teacher-designed learning task can induce any one, two, or all three of the components of involvement for each word: need, search, and evaluation. A reading comprehension task which requires the learner to look up the meaning of a homonym in a dictionary, illustrates need (since knowing the word's meaning is necessary for the successful completion of the comprehension task), search (since the meaning of the word is looked up), and evaluation (since di€erent meanings of the word have to be compared and checked against the context before one is selected). If, however, the same task is simpli®ed for the learner by teacher's glosses for unknown words in the text margin, search and evaluation are no longer required. In the latter example, the task induces a weaker involvement in the word as only the need component is at work. The task of the former example, however, requiring need, search, and evaluation, induces a stronger involvement. Thus we may say that tasks di€er in the involvement load they generate. In conclusion, involvement load is de®ned here as the combination of the presence or absence of the involvement factors Need, Search, and Evaluation. The variability in involvement load that can be experienced for di€erent words leads us to formulate our second assumption. Assumption Two: Other factors being equal, words which are processed with higher involvement load will be retained better than words which are processed with lower involvement load [Note 1].

16

INCIDENTAL VOCABULARY ACQUISITION IN A SECOND LANGUAGE

Task-induced involvement load Before exploring the notion of involvement load in this subsection, we would like to point out that we adopt the general de®nition of task as provided by Richards et al. (1985: 289), as `an activity or action which is carried out as the result of processing or understanding language (i.e. as a response)'. In the socalled task-based approach, task is given a more speci®c meaning as `an activity in which: meaning is primary; there is some communication problem to solve; there is some sort of relationship to comparable real-world activities; task completion has some priority; the assessment of the task is in terms of outcome' (Skehan 1998: 95). We subscribe to the pedagogical aims of the task-based approach and thus adhere to Skehan's urge `to develop pedagogic interventions where learners focus on form naturally, rather than arti®cially' (Skehan 1998: 40). However, in the context of our aim to stimulate theory and empirical research (as opposed to sound pedagogical practice), it suces to adopt the more general de®nition even though that de®nition encompasses arti®cial noncommunicative tasks, such as ®lling in gaps or writing isolated and unconnected sentences with given words. Of course, in applying the notion of involvement to the language classroom, one would try to follow Skehan's principle. The examples in the previous subsection, show that, in a natural communicative task, di€erent words can induce di€erent involvement loads. In a teaching context, however, or for research purposes, tasks can be designed in such a way that the involvement load is (almost) identical for all the words targeted for teaching or research. We will refer to this involvement load as task-induced involvement load and illustrate it by analysing some learning tasks, widely practised in educational settings, in terms of need, search, and evaluation of newly met words (see Table 2). Let us ®rst consider several reading tasks. A reading comprehension task where unknown words are glossed for the student, but the comprehension questions can be answered without reference to these words (1 in Table 2) does not induce any need to focus on the glossed words (since they are irrelevant to the task), nor any search for their meaning (since they are glossed), nor any evaluation. A reading comprehension task with glossed words that are relevant to answering the questions (2 in Table 2) will induce a moderate need to look at the glosses (moderate because it is imposed by the task), but it will induce neither search nor evaluation. The same task with glosses removed (3 in Table 2), will not only induce need but also search (provided that the student has deemed the unknown words as relevant enough to look up). The presence or absence of evaluation may vary with type of word and context. If an unknown word has only one meaning and if the context allows a straightforward, literal interpretation of it, no decision has to be made about its contextual meaning. If, on the other hand, the word has several meanings, the reader has to select the meaning which makes sense in the context, a decision demanding moderate evaluationÐmoderate since the learner is not required to produce original language.

BATIA LAUFER and JAN HULSTIJN 17

Let us now imagine the same text, but the target words have been deleted from it. These words are listed at the bottom of the text with their translations or explanations and the task requires the learner to ®ll the text gaps with the correct words from the list (the task can be made more complex by adding to the list words that do not ®t the text at all) and to answer the comprehension questions (4 in Table 2). The ®ll-in task induces a moderate need, no search (the words are explained) and a moderate evaluation, since all the words in the list have to be evaluated against each other and the context of the gaps. The next three examples are of writing tasks. In the ®fth task the learner is asked to write original sentences with some new words. These words are translated or explained by the teacher. The task induces a moderate need, no search, and strong evaluation because the new words are evaluated against suitable collocations in a learner-generated context. Now suppose the learner is required to write a composition (as opposed to single sentences) and incorporate some L2 target words; the teacher has not provided these words in their L2 form, but by their L1 equivalent (6 in Table 2). The task will induce a moderate need and search since the L2 word forms have to be looked up, and again a strong evaluation as the words are used in learner-generated contexts. Finally, consider a case of a composition where the learner wants to use concepts for which s/he possesses no L2 form. S/he then decides to look up these L1 concepts for their L2 equivalence (in an L1±L2 dictionary) and use them in the composition (7 in Table 2). This task induces a strong need (selfimposed), search, and a strong evaluation. The above analysis is summed up in Table 2. A minus (±) indicates an absence of an involvement factor, a plus (+) indicates that the factor is present in its moderate version, a double plus (++) marks the strong version of an involvement factor. Table 2 clearly shows how di€erent tasks di€er in the involvement load they induce from the learner. This variability in task-induced involvement load, together with Assumption Two, lead us to formulate our third assumption. Assumption Three: Other factors being equal, teacher/researcherdesigned tasks with a higher involvement load will be more e€ective for vocabulary retention than tasks with a lower involvement load.

In summary, we propose a motivational-cognitive construct of involvement, consisting of three basic components: need, search, and evaluation. Each of these three factors can be absent or present when processing a word in a natural or arti®cially designed task. The combination of factors with their degrees of prominence constitutes involvement load. Our basic assumption regarding vocabulary retention is that retention of hitherto unfamiliar words is conditional, in general, upon the degree of involvement in processing these words. The concept of involvement can be operationalised by devising tasks with various degrees of need, search, and evaluation and can therefore be submitted to empirical investigation.

18

INCIDENTAL VOCABULARY ACQUISITION IN A SECOND LANGUAGE

Table 2: Task-induced involvement load Task

Status of target words

Need

Search

Evaluation

1. Reading and comprehension questions 2. Reading and comprehension questions 3. Reading and comprehension questions

Glossed in text but irrelevant to task

±

±

±

Glossed in text and relevant to task

+

±

±

Not glossed but relevant to task

+

+

4. Reading and comprehension questions and ®lling gaps 5. Writing original sentences 6. Writing a composition

Relevant to reading + comprehension. Listed with glosses at the end of text Listed with glosses +

±

±/+ (depending on word and context) +

±

++

Concepts selected + by the teacher (and provided in L1). The L2 learner± writer must look up the L2 form Concepts selected ++ (and looked up) by L2 learner±writer

+

++

+

++

7. Writing a composition

THE INVOLVEMENT LOAD HYPOTHESIS: EMPIRICAL AND THEORETICAL PERSPECTIVES The Involvement Load Hypothesis: empirical support Let us revisit the Task Table (Table 1) presented earlier, comparing each pair of tasks in terms of their involvement load. The table is reproduced as Table 3, this time showing which involvement component is present in one task of the pair and absent in the other. Three examples will be singled out for detailed illustration of the di€erences in the involvement load of the tasks. Hulstijn (1992) showed that when meanings of words had to be inferred they were retained better than words with given meanings. If we compare the

BATIA LAUFER and JAN HULSTIJN 19

Table 3: Previous research re-visited The more e€ective task

The less e€ective task

Reference

Meaning selected from several options +evaluation Meaning looked up in a dictionary +search Meaning looked up in a dictionary +search Meaning negotiated ++ need, +search Negotiated input +search Used in original sentences ++evaluation Used in a composition (L1±L2 look up) ++evaluation Interactionally modi®ed output ++evaluation Reading and a series of vocabulary exercises +evaluation/++evaluation Reading, words looked up in a dictionary +search

Meaning explained by synonym

Hulstijn 1992

Reading with/without guessing +/±search Meaning provided in a marginal gloss Meaning not negotiated

Knight 1994; Luppescu and Day 1993 Hulstijn et al. 1996 Newton 1995

Premodi®ed input

R. Ellis et al. 1994 Joe 1995, 1998

Used in non-original sentences Encountered in a reading task (L2±L1 look up) ±/+evaluation Interactionally modi®ed input Reading only (and inferring meaning) ±/+evaluation Reading only, words not looked up

Hulstijn and Trompetter 1998 R. Ellis and He 1999 Paribakht and Wesche 1997 Cho and Krashen 1994

two tasks in terms of involvement load, we can see that the di€erence lies in the absence of evaluation in the synonym-condition and presence of evaluation in the multiple-choice condition. Learners had to evaluate all the alternative meanings against the text context. (In both conditions there was a moderate need, induced by the researcher, and no search). Newton (1995) found that words which were negotiated for meaning in an interactive task were retained better than words that were not negotiated. Put di€erently, the choice to negotiate a word implies a need, induced by the learner, and also a search for meaning. (Search for meaning does not have to be in a dictionary only. The learner can search the text context, ask a teacher, or peers.) When unknown words are not negotiated, it means the learner has no need for them and therefore performs no search.

20

INCIDENTAL VOCABULARY ACQUISITION IN A SECOND LANGUAGE

Joe (1995) shows that words with a `high degree of generation' are better retained than words with a low degree of generation, or no generation at all. What this means is that when words are used in a learner-generated, original context, they are better retained than if they are used in a non-original context (for example, memorised from text), or not used at all. The di€erence probably lies in all three factors of involvement, but a fair comparison cannot be made since memorisation belongs to the realm of intentional learning whereas our construct of involvement applies to incidental learning. Thus, the results of the study conducted by Newton and, perhaps in part, the results of Joe's study can also be explained in terms of involvement. Words which were retained better underwent a higher involvement load than words which were retained less well.

The Involvement Load Hypothesis: a theoretical perspective As mentioned in the introductory section, psychologists have associated better learning with depth of processing, or degree of elaboration, or quality of attention to information. Our construct of involvement decomposes the concept of processing into two cognitive components (search and evaluation) and adds a motivational component (need). This proposal should be conceived of as a ®rst attempt to stimulate researchers as well as practitioners to operationalise traditional general labels such as noticing, attention, elaboration, and motivation, into task-speci®c components. Thus, although involvement will vary with words, involvement can also be in¯uenced by task factors. It is on the task factors that we direct our focus. All three factors can be manipulated separately and in di€erent combinations by researchers or teachers as shown in Table 2. As a research paradigm, therefore, our proposal o€ers multiple possibilities to explore the relationship between retention and various aspects of deep processing. This can be done by designing experimental tasks with varying involvement loads. How does the construct of involvement load relate to the Input Hypothesis (Krashen 1985) and the Output Hypothesis in SLA (Swain 1985, 1996)? The Involvement Load Hypothesis does not predict that any output task will lead to better results than any input task. It predicts that higher involvement in a word induced by the task (natural or arti®cial) will result in better retention regardless of whether it is an input or an output task. To illustrate this point, let us consider three tasks: one input and two output tasks. The input task is to read a text for comprehension. During the reading, the learner decides to look up certain words in a dictionary. Since it was the learner's decision, the need is characterised as strong. Looking up the meaning of a word implies a search. Let us assume the looked-up words have one unequivocal meaning and the context does not lead to interpretation diculties: hence there is no evaluation in the task. The involvement load of the task is therefore ++need, +search, ±evaluation. Now consider an output task: an exercise where words have to be ®lled in

BATIA LAUFER and JAN HULSTIJN 21

into sentence blanks. The words to be ®lled in are written out in a list with their translations or explanations. Since the use of words is imposed by the teacher, need is moderate. There is no search as the words are glossed. In order to ®ll in the correct word in each blank, the candidate words provided by the teacher have to be evaluated against each other in the context of each blank. Therefore the task requires a moderate amount of evaluation. The involvement load of this output task is +need, ±search, +evaluation. Finally, consider another output task: writing original sentences with words which are provided with their translations or explanations. The need and search of this task are the same as in the previous output task. Yet the amount of evaluation is higher as the words are used in an original context. The involvement load of this task is + need, ±search, ++evaluation. The Involvement Load Hypothesis predicts that, among the ®rst two tasks, it is the input task, reading with look-ups, that will yield better retention since it is more involving than the ®ll-in output task. It also predicts that the input task and the second output task, sentence writing, will be equally bene®cial for retention since the two are identical in their involvement load. Similarly, the model predicts that task e€ect does not depend on the mode as such, i.e. whether the task is aural, oral, or visual. What matters is the motivationalcognitive dimensions of the task, i.e. its involvement load. The above predictions can be tested, as will be suggested in the next section.

SUGGESTIONS FOR FURTHER RESEARCH Further investigation of the involvement load e€ect Our basic contention is that the e€ectiveness of a task is determined by the involvement load it induces. Research should be conducted to test this contention. Tasks with di€erent involvement loads could be compared with regard to their e€ect on incidental vocabulary learning. A particularly interesting comparison would involve conditions where the input and output tasks have identical involvement loads. If involvement load is the determining factor in task e€ectiveness, irrespective of whether the task is input or output oriented, the two conditions should yield similar retention results.

Establishing the relative importance of need, search, and evaluation So far the amount of the involvement load has been conceived of as the sum of the plusses (of need, search, and evaluation). For example, a task consisting of +need, +search, ±evaluation has the same involvement load as a task consisting of +need, ±search, +evaluation. Yet all three factors may not be equally important for vocabulary learning. Earlier we suggested that the weight (impact on incidental learning) of search might be lower than that of need and evaluation. Empirical research should compare tasks with the same

22

INCIDENTAL VOCABULARY ACQUISITION IN A SECOND LANGUAGE

number of components (two), but a di€erent distribution of the components involved.

Comparing the quality and quantity of exposure to words A crucial question in understanding vocabulary learning is whether retention depends on what one does with the word rather than how often one meets it. In pedagogy, the question is whether task type is just as important, more so, or less so than the number of tasks in which a new word appears. Put di€erently, we would like to ®nd out whether the quality of exposure to new vocabulary during `incidental' encounters can compensate for the relatively limited amount of exposure which is characteristic of learning a second language in a non-language speaking environment. Research could compare vocabulary retention along two dimensions: varying task involvement loads and varying the number of exposures to the investigated words.

CONCLUDING REMARKS: A PLEA FOR THEORY BUILDING The generality of the constructs in the theoretical literature and the paucity of empirical research testing the noticing and elaboration hypotheses in the domain of L2 vocabulary learning led us to adopt `a bottom-up approach' and start our investigation at the task end of the issue. We examined a number of studies on incidental vocabulary learning and inferred from them, in an inductive fashion, three constructs: need, search, and evaluation. We are fully aware of the possible criticisms of such a bottom-up approach since it draws post hoc conclusions from existing instructional tasks rather than from results of controlled experiments of the kind psychologists conduct in laboratories. Yet this is not an unacceptable scholarly procedure. Advancement in scholarly inquiry, both in terms of theory building and in terms of providing empirical data, can proceed in a number of ways (Kuhn 1974). One way is to begin at the higher end of the hermeneutic cycle: developing theory, deducing hypotheses from it, testing them empirically, and applying the theory's constructs and theses to practice. Another way is to begin at the lower end of the hermeneutic cycle: looking at empirical data, trying to discern general patterns, and establishing links with existing theories or developing new theories. Whichever route is taken, full circle must be made eventually. Given the fact that the last decade has not shown much progress in applying cognitive notions to the domain of L2 vocabulary learning (in comparison to the domain of L2 grammar learning, where research on noticing, pushed output, recasts, and other types of feedback is thriving) we felt justi®ed in adopting the latter, `bottom-up' approach. As mentioned in the introduction, we hope that our Involvement Load Hypothesis will be a ®rst step in stimulating theoretical and empirical work and will be followed by other proposals of greater scope and depth. Greater depth could be re¯ected in more precise de®nitions of the involvement

BATIA LAUFER and JAN HULSTIJN 23

components and a more thorough theoretical link between them and theories of information processing and a€ective components of cognition. Broader scope could take the form of adding new components to the three proposed here and, eventually, providing a theoretical underpinning for each of them. (Revised version received March 2000)

ACKNOWLEDGEMENTS We are grateful to our colleagues for their helpful comments on an earlier version of this paper: Nanda Poulisse, Rob Schoonen, and Elisabeth van der Linden from the University of Amsterdam, Paul Meara from the University of Swansea, Paul Nation from the Victoria University of Wellington, and Norbert Schmitt from the University of Nottingham.

NOTE 1 Examples of `other ' factors are phonological, morphological and semantic complexity, in relation to learners' knowledge, i.e. L2 pro®ciency and conceptual development (Laufer 1990a, 1990b, 1997; Schmitt and Meara, 1997).

REFERENCES Anderson, J. R. 1995. Cognitive Psychology and its Implications (4th edn.). New York: Freeman. Baddeley, A. 1997. Human Memory: Theory and Practice (rev edn.). Hove (UK): Psychology Press. Broeder, P. and K. Plunkett. 1994. `Connectionism and second language acquisition' in N. Ellis (ed.): Implicit and Explicit Learning of Languages. London: Academic Press. 421±53. Carr, T. H. and T. Curran. 1994. `Cognitive factors in learning about structured sequences.' Studies in Second Language Acquisition 16: 205±30. Cho, K.-S. and S. Krashen. 1994. `Acquisition of vocabulary from the Sweet Valley Kids Series: adult ESL acquisition.' Journal of Reading 37: 662±7. Craik, F. I. M. and R. S. Lockhart. 1972. `Levels of processing: A framework for memory research.' Journal of Verbal Learning and Verbal Behavior 11: 671±84. Craik, F. I. M. and E. Tulving. 1975. `Depth of processing and the retention of words in episodic memory.' Journal of Experimental Psychology: General, 104: 268±94. Crookes, G. and R. W. Schmidt. 1991. `Motivation: Reopening the research agenda.' Language Learning 41: 469±512.

DoÈrnyei, Z. 1994. Motivation and motivating in the foreign language classroom. Modern Language Journal 78: 273±84. Doughty, C. 1991. `Second language instruction does make a difference: Evidence from an empirical study of SL relativization.' Studies in Second Language Acquisition 13: 431±69. Ebbinghaus, H. 1964. Memory: A Contribution to Experimental Psychology. New York: Dover. Ellis, N. C. 1994a. `Implicit and explicit language learningÐAn overview' in N. C. Ellis (ed.): Implicit and Explicit Learning of Languages. London: Academic Press. 1±32. Ellis, N. C. 1994b. `Consciousness in second language learning: Psychological perspectives on the role of conscious processes in vocabulary acquisition' in J. H. Hulstijn and R. Schmidt (eds.): Consciousness in Second Language Learning. AILA Review 11: 37±56. Ellis, N. 1994c. `Vocabulary acquisition: The implicit ins and outs of explicit cognitive mediation' in N. C. Ellis (ed.): Implicit and Explicit Learning of Languages. London: Academic Press. 211±82. Ellis, R. 1993. `The structural syllabus and second language acquisition.' TESOL Quarterly 27: 91±113.

24

INCIDENTAL VOCABULARY ACQUISITION IN A SECOND LANGUAGE

Ellis, R. 1994. The Study of Second Language Acquisition. Oxford: Oxford University Press. Ellis, R. and X. He. 1999. `The roles of modified input and output in the incidental acquisition of word meanings.' Studies in Second Language Acquisition 21: 285±301. Ellis, R., Y. Tanaka and A. Yamazaki. 1994. `Classroom interaction, comprehension and the acquisition of L2 word meanings.' Language Learning 44: 449±91. Eysenck, M. W. 1982. `Incidental learning and orienting tasks' in C. R. Puff (ed.): Handbook of Research Methods in Human Memory and Cognition. New York: Academic Press. 197± 228. Eysenck, M. W. and M. C. Eysenck. 1980. `Memory scanning, introversion-extraversion and levels of processing.' Journal of Research into Personality 13: 305±15. Gardner, R. C. 1985. Social Psychology and Second Language Learning: The Role of Attitudes and Motivation. London: Edward Arnold. Gardner, R. C. and W. E. Lambert. 1972. Attitudes and Motivation in Second-Language Learning. Rowley, MA: Newbury House. Gardner, R. C. and P. D. MacIntyre. 1991. `An instrumental motivation in language study: Who says it isn't effective?' Studies in Second Language Acquisition 13: 57±72. Gray, P. 1999. Psychology (3rd edn). New York: Worth Publishers. Gregg, K. R. 1996. `The logical and developmental problems of second language acquisition' in W. C. Ritchie and T. K. Bhatia (eds.): Handbook of Second Language Acquisition. San Diego: Academic Press. 49±81. Gregg, K. R. 2000. `Learnability, linguistic theory, and second language acquisition' in P. Robinson (ed.): Cognition and Second Language Instruction. Cambridge: Cambridge University Press. Harley, B. 1998. `The role and focus-on-form tasks in promoting child L2 acquisition' in C. Doughty and J. Williams (eds.): Focus on Form in Classroom Second Language Acquisition. Cambridge: Cambridge University Press. 156±74. Hulstijn, J. H. 1989. `Implicit and incidental second language learning: Experiments in the processing of natural and partly artifical input' in H. W. Dechert and M. Raupach (eds.): Interlingual Processing. Tubingen: Gunter Narr. 49±73. Hulstijn, J. H. 1992. `Retention of inferred and

given word meanings: Experiments in incidental vocabulary learning' in P. J. Arnaud and H. BeÂjoint (eds.): Vocabulary and Applied Linguistics. London: Macmillan. 113±25. Hulstijn, J. H. in press. `Intentional and incidental second-language vocabulary learning: A reappraisal of elaboration, rehearsal and automaticity' in P. Robinson (ed.): Cognition and Second Language Instruction. Cambridge: Cambridge University Press. Hulstijn, J. H. and R. De Graaff. 1994. `Under what conditions does explicit knowledge of a second language facilitate the acquisition of implicit knowledge? A research proposal' in J. H. Hulstijn and R. Schmidt (eds.): Consciousness in Second Language Learning. AILA Review 11: 97±112. Hulstijn, J. H. and P. Trompetter. 1998. `Incidental learning of second language vocabulary in computer-assisted reading and writing tasks' in D. Albrechtsen, B. Henriksen, I. M. Mees, and E. Poulsen (eds.): Perspectives on Foreign and Second Language Pedagogy. Odense, Denmark: Odense University Press. 191±200. Hulstijn, J. H., M. Hollander, and T. Greidanus. 1996. `Incidental vocabulary learning by advanced foreign language students: The influence of marginal glosses, dictionary use, and reoccurrence of unknown words.' Modern Language Journal 80: 327±39. James, W. 1890. Principles of Psychology. Vol. 1. New York: Holt. Joe, A. 1995. `Text-based tasks and incidental vocabulary learning: A case study.' Second Language Research 11: 149±58. Joe, A. 1998. `What effects do text-based tasks promoting generation have on incidental vocabulary acquisition?' Applied Linguistics 19: 357±77. Keller, J. 1983. `Motivational design of instruction' in C. M. Reigeluth (ed.): Instructional Design Theories and Models. Hillsdale, NJ: Erlbaum. 289±320. Knight, S. 1994. `Dictionary: The tool of last resort in foreign language reading? A new perspective.' Modern Language Journal 78: 285±99. Krashen, S. D. 1985. The Input Hypothesis. Issues and Implications. Harlow, England: Longman. Krashen, S. 1989. `We acquire vocabulary and spelling by reading: Additional evidence for

BATIA LAUFER and JAN HULSTIJN 25

the input hypothesis.' The Modern Language Journal 73: 440±64. Kuhn, T. S. 1974. The Structure of Scientific Revolutions (2nd edn.). Chicago, IL: The University of Chicago Press. Landy, F. 1985. Psychology of Work Behavior. Homewood, IL: Dorsey Press. Laufer, B. 1990a. `Words you know: how they affect the words you learn' in J. Fisiak (ed.): Further Insights into Contrastive Linguistics. Amsterdam, Netherlands: Benjamin's. 573± 93. Laufer, B. 1990b. `Ease and difficulty in vocabulary learning: some teaching implications.' Foreign Language Annals 23: 147±56. Laufer, B. 1997. `What's in a word that makes it hard or easy? Intralexicalfactors affecting the difficulty of vocabulary acquisition' in N. Schmitt and M. McCarthy (eds.): Vocabulary Description, Acquisition and Pedagogy. Cambridge: Cambridge University Press. 140±55. Lee, J. F. 1998. `The relationship of verb morphology to second language reading comprehension and input processing.' Modern Language Journal 82: 33±48. Leow, R. P. 1998. `The effects of amount and type of exposure on adult learners' L2 development in SLA.' Modern Language Journal 82: 49±68. Long, M. H. and P. Robinson. 1998. `Focus of form: Theory, research and practice' in C. Doughty and J. Williams (eds.): Focus on Form in Classroom Second Language Acquisition. Cambridge: Cambridge University Press. 15± 63. Luppescu, S. and R. R. Day. 1993. `Reading, dictionaries and vocabulary learning.' Language Learning 43: 263±87. MacIntyre, P. D. and R. C. Gardner. 1989. `Anxiety and second-language learning: Toward a theoretical clarification.' Language Learning 39: 251±75. McClelland, D. C., J. W. Atkinson, R. W. Clark, and E. L. Lowell. 1953. Achievement Motive. New York: Appleton-Century-Crofts. McLaughlin, B. 1965. ` ``Intentional'' and ``incidental'' learning in human subjects: The role of instructions to learn and motivation.' Psychological Bulletin 63: 359±76. Newton, J. 1995. `Task-based interaction and incidental vocabulary learning: a case study.' Second Language Research 11: 159±77. Oxford, R. and J. Shearin. 1994. `Language

learning motivation: Expanding the theoretical framework.' Modern Language Journal 78: 12±28. Paribakht, T. S and M. Wesche. 1997. `Vocabulary enhancement activities and reading for meaning in second language vocabulary acquisition' in J. Coady and T. Huckin (eds.): Second Language Vocabulary Acquisition. A Rationale for Pedagogy, Cambridge: Cambridge University Press. 174± 200. Postman, L. 1964. `Short-term memory and incidental learning' in A. W. Melton (ed.): Categories of Human Learning. New York: Academic Press. 145±201. Richards, J., J. Platt, and H. Weber. 1985. Longman Dictionary of Applied Linguistics. Longman, UK. Robinson, P. 2000. `Task complexity and second language syllabus design' in P. Robinson (ed.): Cognition and Second Language Instruction. Cambridge: Cambridge University Press. Schmidt, R. 1994a. `Deconstructing consciousness in search of useful definitions for applied linguistics' in J. H. Hulstijn and R. Schmidt (eds.): Consciousness in Second Language Learning. AILA Review 11: 11±26. Schmidt, R. 1994b. `Implicit learning and the cognitive unconscious: Of artificial grammars and SLA.' in N. C. Ellis (ed.): Implicit and Explicit Learning of Languages. London: Academic Press. 165±210. Schmidt, R. 2000. `Attention' in P. Robinson (ed.): Cognition and Second Language Instruction. Cambridge: Cambridge University Press. Schmitt, N. and P. Meara. 1997. `Researching vocabulary through a word knowledge framework: Word associations and verbal suffixes.' Studies in Second Language Acquisition 19: 17±36. Schumann, J. H. 1994. `Where is cognition? Emotion and cognition in second language acquisition.' Studies in Second Language Acquisition 16: 231±42. Sharwood Smith, M. 1993. `Input enhancement in instructed SLA.' Studies in Second Language Acquisition 15: 165±79. Skehan, P. 1989. Individual Differences in Second-Language Learning. London: Edward Arnold. Skehan, P. 1998. A Cognitive Approach to Language Learning. Oxford: Oxford University Press.

26

INCIDENTAL VOCABULARY ACQUISITION IN A SECOND LANGUAGE

Swain, M. 1985. `Communicative competence: Some roles of comprehensible input and comprehensible output in its development' in S. Gass and C. Madden (eds.): Input and Second Language Acquisition. Rowley, MA: Newbury House. 235±53. Swain, M. 1996. `Three functions of output in second language learning' in G. Cook and B. Seidlhover (eds.): Principles and Practice in the Study of Language. Oxford: Oxford University Press. Tomlin, R. S. and V. Villa. 1994. `Attention in cognitive science and second language acquisition.' Studies in Second Language Acquisition 16: 183±204.

VanPatten, B. 1996. Input Processing and Grammar Instruction. Norwood, NJ: Ablex. VanPatten, B. and S. Oikkenon. 1996. `Explanation versus structured input in processing instruction.' Studies in Second Language Acquisition 18: 495±510. Williams, J. N. 1999. `Memory, attention, and inductive learning.' Studies in Second Language Acquisition 21: 1±48. Young, R. and K. Perkins. 1995. `Cognition and conation in second language acquisition theory.' International Review of Applied Linguistics 33: 142±64. Zechmeister, E. B. and S. E. Nyberg. 1982. Human Memory: An Introduction to Research and Memory. Monterey, CA: Brook/Cole.

Applied Linguistics 22/1: 27±57

# Oxford University Press 2001

Task Complexity, Task Diculty, and Task Production: Exploring Interactions in a Componential Framework PETER ROBINSON Aoyama Gakuin University, Tokyo, Japan This paper describes a framework for examining the e€ects of the cognitive complexity of tasks on language production and learner perceptions of task diculty, and for motivating sequencing decisions in task-based syllabuses. Results of a study of the relationship between task complexity, diculty, and production show that increasing the cognitive complexity of a direction-giving map task signi®cantly a€ects speaker±information-giver production (more lexical variety on a complex version and greater ¯uency on a simple version) and hearer±information-receiver interaction (more con®rmation checks on a complex version). Cognitive complexity also signi®cantly a€ects learner perceptions of diculty (e.g. a complex version is rated signi®cantly more stressful than a simple version). Task role signi®cantly a€ects ratings of diculty, though task sequencing (simple to complex versus the reverse sequence) does not. However, sequencing does a€ect the accuracy and ¯uency of speaker production. Implications of the ®ndings for task-based syllabus design and further research into task complexity, diculty, and production interactions are discussed.

The development of theoretically motivated, empirically substantiable, and pedagogically feasible sequencing criteria has long been acknowledged as a major goal of research aimed at operationalizing task-based approaches to syllabus design (Candlin 1987; Long 1985; Long and Crookes 1992; Nunan 1989; Robinson et al. 1995; Skehan 1996). The present paper describes a framework for distinguishing the relative complexity of tasks with this ultimate goal in mind. My aims are to propose task sequencing criteria which are feasible and researchable, and to examine the e€ects of sequencing decisions motivated by the framework on measures of learner a€ect, and production. To these ends I propose distinctions between cognitively de®ned task complexity, learner perceptions of task diculty, and the interactive conditions under which tasks are performed. Drawing extensively on previous research and speculation I then identify the cognitive components of task complexity in terms of a series of dimensions and report ®ndings of previous research into the e€ects of these on task performance. Following this, I describe results of a study exploring relationships between the cognitive complexity of tasks, learner ratings of task diculty and task production.

28

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

THE GRADING AND SEQUENCING ISSUE: PREVIOUS PROPOSALS In task-based approaches to pedagogy and syllabus design such as those of Long (1985, 1997) and Skehan (1996) pedagogic tasks are not sequenced on the basis of their linguistic content but according to variously de®ned notions of increases in task complexity or diculty. However, there is no current consensus on how the scope of these two terms di€ers, and necessarily, therefore, on the components of each. Long (1985) speci®cally argued that a task-based syllabus should adopt pedagogic tasks as the unit of analysis, sequenced from simple to complex, and gradually approximating the demands of real world target tasks identi®ed via needs analyses, nominating such factors as the number of steps involved in performing the task, and remoteness of the information drawn on in time and space as factors possibly contributing to `complexity'. Skehan (1996, 1998), drawing on earlier proposals by Candlin (1987) distinguishes three factors, contributing to the `diculty' of tasks; (1) code complexity, i.e. the syntactic and lexical diculty of language input; (2) cognitive complexity, e.g. the processing demands of tasks, and the availability of relevant schematic knowledge; and (3) communicative stress, e.g. the result of di€erentials in time pressure, the modality of task performance, and the scale, or number of participants involved. Brindley (1987) and Nunan (1989) proposed essentially identical classi®cations of factors in their proposals for sequencing decisions. Nunan distinguished learner, activity, and input factors while Brindley distinguished learner, task, and text factors. Brindley's learner factors included con®dence and motivation; ability to learn at the pace required and prior learning experience; and possession of necessary language skills and relevant cultural knowledge. Brindley claimed all of these make tasks easier. Brindley also identi®ed a number of separate task factors such as degree of cognitive complexity, number of steps, amount of context support, and amount of time provided. Along with these Brindley proposed that text factors, including the length, clarity, and familiarity of texts used on tasks also a€ect their diculty for the learner. The issue of the extent to which linguistic content of texts should in¯uence sequencing decisions in task-based approaches is a problematic one, over which researchers have disagreed. I suggest that the textual input to tasks, adjusted to such factors as pro®ciency, is a decision about task `content', and this is in principle separable from decisions about task sequencing which should be based on increases in cognitively de®ned complexity. One reason is that using linguistic text factors to grade and sequence tasks may result in those `structure-trapping' tasks described by Skehan (1996, 1998) which use tasks to implement a linguistic syllabus. Skehan argues against this role for tasks, as does Long (1985, 1997). However, Nunan (1989) proposed that task designers should `create an interesting/relevant text or task at the appropriate level of diculty', and then see what items on the linguistic syllabus can be

PETER ROBINSON 29

taught through it. In Widdowson's (1990) terms, Nunan argued that tasks should be used to `realize' or implement a linguistic syllabus. Identifying this and other problems with the proposals put forward for grading and sequencing tasks, Widdowson wrote: But the use of problem solving tasks as units of syllabus content, as distinct from activities for syllabus realization, encounters a number of diculties. By what criteria, for example, are such tasks to be sequentially arranged? If the sequence is to be in accord with natural learning . . . information is needed about cognitive development at di€erent stages of maturation, about the conditions, psychological and social, which attend the emergence in the mind of general problem solving capabilities. Armed with such information, we could perhaps relate these capabilities to certain task types, analysed into their constituent features, and then given token realizations and arranged in an order of increasing complexity. Such information is not, to my mind, currently available (Widdowson 1990: 147±8).1

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK CONDITIONS In this paper I report research with the goal Long, Skehan, and Widdowson have identi®ed, of motivating decisions about sequencing tasks `in an order of increasing complexity'. However, as illustrated by the brief review of previous research and speculation, in many discussions of task sequencing the terms complexity and diculty are used interchangeably and the scope of potential in¯uences on them is argued to be very wide, including, cognitive, a€ective, linguistic, interactional, experiential, and many other factors. With this in mind, following Spilsbury et al. (1990)2 and others engaged in psychological enquiry into task e€ects, throughout this paper I use the terms `complexity' and `diculty' to describe two di€erent in¯uences on task performance. I distinguish both from task conditions, the interactive demands of tasks, as Figure 1 shows. Interactions between complexity, condition, and diculty are bound to occur during task performance, and examining some of these interactions was one of the aims of the present paper. I will ®rst describe each factor separately, then brie¯y discuss likely interactions between them.

Task complexity It is important to distinguish between di€erentials in the processing demands of tasks, which are a consequence of task structure and design, and di€erentials in the resources learners bring to tasks which are attributable to a range of individual di€erence variables. Task complexity, I argue, is the result of the attentional, memory, reasoning, and other information processing demands imposed by the structure of the task on the language learner. These di€erences in information processing demand, resulting from design characteristics, are relatively ®xed and invariant (Sternberg 1977;

30

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

Figure 1: Task complexity, condition and diculty Suedfeld and Coren 1992). So, for any given learner, a simple task (adding six and seven, or giving directions from A to B using a simpli®ed map of a small area with which the speaker is familiar) will always be less demanding than a complex task (doing calculus, or giving directions from A to B and then C using a detailed map of a large area the speaker is unfamiliar with) since the di€erential in the extent of attentional focus, working memory, reasoning, and other cognitive demands is a ®xed and invariant feature of the tasks. Task complexity will help explain within learner variance when performing any two tasks. The e€ects of complexity di€erentials should be revealed by the fact that the cognitively simpler, less resource-demanding task will involve a lower error rate, and/or be completed faster, and be less susceptible to interference from competing tasks than the more complex task. It is true that variance in task performance should also result from repeating the same task, whether simple or complex, since task practice and automatization reduce resource demands. I argue this is a separate factor again to the complexity issue (see Bygate 1996; DeKeyser in press). Task complexity, then, involves a number of previously identi®ed task factors including those suggested by Long (1985), Skehan's cognitive complexity, and Brindley's task factors. Figure 1 describes task complexity as consisting of a number of dimensions which can be manipulated during task design. The dimensions are represented by +/± a component which may be present or absent (though they may also be thought of as continua, along which there is relatively more, versus relatively less of a component such as planning time, or prior knowledge, etc.). These are divided into two categories, resource-directing dimensions and resource-depleting dimensions. The two categories identify an important di€erence in the way these dimensions a€ect resource allocation during L2

PETER ROBINSON 31

task performance (Robinson in pressb). Increasing task complexity along resource-directing dimensions (e.g. by requiring reasoning, in addition to simple information transmission) makes greater resource demands which can be met by using speci®c features of the language code. In this case those features might include use of logical connectors (if±then, therefore, because) along with the syntactic permutations necessary to embedding and subordination of clauses. In contrast, complex tasks along resource-depleting dimensions make greater demands on attention and working memory, but do not direct resources to features of language code that can be used in completing the task. This happens when, for example, a secondary task is added to a primary task, draining resources from primary task completion, or when planning time for task performance is removed. As discussed below, these two categories of task complexity dimension are likely to interact and a€ect task production in measurable ways. For example, making a pedagogic task simple along a resource-depleting dimension (e.g. by allowing planning time) and complex along a resource-directing dimension (e.g. by requiring reasoning) allows optimum resource allocation to satisfy the linguistic demands of the task, compared to making the task complex on both dimensions simultaneously. Thus, with these dimensions and distinctions in mind, simpler pedagogic versions of a target real world task, which support learners in channelling their attention to needed aspects of language code, can be designed and practised. These would be followed by progressively more complex versions which approach the demands of the real world (e.g. thinking up and giving answers to unexpected questions about a job application during an interview).

Task difficulty I have argued that the cognitive factors contributing to complexity are a consequence of the structure of the task which imposes resource demands. Satisfying those demands in order to successfully complete the task is dependent on the extent of the resources a learner brings to the task. Consequently, the cognitive factors contributing to task complexity must be distinguished from learner factors which may make a task more or less dicult (as opposed to complex). These are a consequence of di€erentials between learners in their available attentional, memory, and reasoning resource pools i.e. di€erentials in the resources they draw on in responding to the demands of tasks (see Gopher 1992). Complexity and diculty cannot be assumed to be in a ®xed relationship to each other for two reasons (see note 2). First, two learners di€ering in intelligence or aptitude may ®nd the same maths or language learning task more or less dicult than each other as a result of inherent ability di€erentials between them. That is, di€erences in the limits or extent of the attentional, memory, and reasoning resource pools two learners bring to a learning task may contribute to di€erences in performance, and their perceptions of the task's diculty. Secondly, resource pools can also be

32

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

a€ected by such temporarily limiting factors as motivation. Greater motivation for one learner can result in a temporary expansion of resource pools currently available to meet the demands of a particular task (e.g. heightened attention to and rehearsal of input in working memory) relative to another learner equivalent or superior in intelligence or aptitude but having less motivation, and so fewer temporarily available resources. Task diculty, then, will help explain between learner variance in performance on any one task because it is determined by those factors contributing to di€erences between them in the extent of available cognitive resources. The learner factors contributing to between learner di€erences in the size of resource pools available to meet task demands, and so to task diculty, are of two kinds: (1) a€ective variables, such as con®dence, motivation, and anxiety, which may change and so a€ect the size of resource pool availability on a temporary basis; and (2) ability variables, such as intelligence, aptitude, and cognitive style which are more permanent and ®xed determinants of resource pools, and more stable over a course of instruction than changeable (and methodologically susceptible) a€ective variables. Although it may often be dicult for materials and syllabus designers to predict the e€ects of tasks on a€ective variables, given their temporal nature and the large potential range of in¯uences upon them, they are extremely important to on-line methodological decisions about pairing and grouping students when implementing the syllabus (see DoÈrnyei and Malderez 1997). In contrast to a€ective variables, ability variables such as aptitude, intelligence, and cognitive style can often be diagnosed ahead of syllabus implementation but here again there is as yet no clear research evidence of stable and predictable interactions between them and L2 task complexity and performance (for research into this in other educational domains see Ackerman 1988; Fleishman 1978; Snow 1994, and Sawyer and Ranta in press for a review of SLA studies). A fourth ability variable, pro®ciency, is often the basis of placement into di€erent levels of a programme. Therefore pro®ciency is not likely to be a factor that is available for on-line methodological manipulation. However, pro®ciency level will, of course, in¯uence decisions about task content and the level of language input to a task. Task diculty, then, also involves a number of factors previously identi®ed by research and speculation about in¯uences on task sequencing, such as Candlin and Skehan's communicative stress, and Brindley and Nunan's learner factors.

Task conditions Task conditions, or the interactive demands of tasks, involve neither task factors nor learner factors alone, but rather participation factors such as the direction of information ¯ow (one-way or two-way) and the communicative goals (one or many solutions) of task performance (see Long 1989). Participant factors include, e.g. gender, familiarity with each other and with task role, and relative status (see Plough and Gass 1993; Yule and MacDonald 1990), as well

PETER ROBINSON 33

as task goal and task interpretation (Coughlan and Du€ 1994). Context of task performance, which is another aspect of the conditions under which the task is performed, will also clearly a€ect learners' perceptions of role and status, and thus in¯uence collaboration and production during interactive tasks (see Douglas 1998; Selinker and Douglas 1985; Tarone and Liu 1995).

Interactions Interactions between all three in¯uences on task performance and learningÐ complexity, diculty, and conditionÐmay be expected. For example complexity along some dimensions (e.g. tasks with no planning time, or with dual task demands) may a€ect task diculty (e.g. as measured by their e€ects on learner anxiety) more than complexity along other dimensions (e.g. tasks with no prior knowledge, or requiring reference to many elements). Similarly, participant factors such as role (e.g. information-giver versus receiver in a one-way closed task) may a€ect task diculty (such that the information-giver role induces greater anxiety than the more passive information-receiver role). However, of the three, I suggest it is di€erences in task complexity which are the logical basis for prospective decision making about task-based syllabus design and the sequencing of pedagogic tasks. It is also possible that stable relationships may exist between increases in task complexity (the cognitive demands of tasks) and learner perceptions of diculty, assessed via a€ective variables, with more complex tasks also being judged to be more dicult. Investigating this was one aim of the research described later in this paper.

COGNITIVE DIMENSIONS OF TASK COMPLEXITY E€ects of task complexity on performance, production, and development The proposal for sequencing pedagogic tasks on the basis of their cognitive complexity is motivated by its potential positive impact in three interrelated domains, as Figure 2 shows. The focus of the study reported later in this paper is on the second of these. Before considering in detail the possible impact of task complexity on production, however, brief mention is made here of its relationship to performance and development. The desired outcome of task-based instruction is the ability to achieve real world target task goals (see Long 1985; Long and Crookes 1992; Skehan 1998) as measured by an estimate of successful performance. The rationale for developing target task ability for L2 learners by gradually increasing the complexity of the tasks used to prepare them for this is similar to that adopted in other ®elds, such as mathematics education or pilot training, where simpler problems and ¯ight simulations are practised prior to more complex versions (see, e.g., Case 1985; Hancock et al. 1995; Snow 1994). Testing whether a desired outcome has been achieved can be done directly,

34

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

Figure 2: In¯uences of task complexity on L2 development, production and performance. Source: Robinson in pressa through performance-referenced tests in which the criterion is whether or not the learner succeeds on the task, (e.g. making a bank transfer, ®nding a book in the library, or making notes on a lecture which enable the learner to adequately identify subsequently needed information). Testing can also be done indirectly, via system-referenced tests which assess the learners' knowledge of the language system. In this case the criterion is often a score, for example 80 per cent or above, on a test of grammar, or reading and listening comprehension (see Bachman and Palmer 1996; Baker 1990; McNamara 1996; Norris et al. 1998; Robinson and Ross 1996 for discussion). A further, and fundamentally important reason for using pedagogic tasks, sequenced in order of increasing cognitive complexity, as the basis of syllabus design is that such a sequencing decision should e€ectively facilitate L2 development: the acquisition of new L2 knowledge, and restructuring of existing L2 representations. I am in broad agreement with the acquisitional arguments for using tasks in language pedagogy put forward by Long (1985), Long and Robinson (1998), Skehan (1998), Skehan and Foster (in press), and Doughty (in press). Brie¯y, I argue that increasing the cognitive demands of tasks increasingly engages cognitive resources, leading to more attention to, and incorporation and rehearsal of (oral/written) task input in working memory (see in particular, Doughty's speci®cation of the mechanisms involved in this process). It also leads to greater attention to and modi®cation of output, facilitating the `noticing' and `pushing' of speaker production that Schmidt (1990) and Swain (1995) argue is important in promoting interlanguage change (for further discussion see Robinson 1995b, 1996, in preparation, in pressa). Most of the empirical work on task complexity there is to date has focused on its e€ects on learner language, using measures of accuracy, ¯uency, and complexity of production, the intermediate stage in Figure 2. As Skehan (1998)

PETER ROBINSON 35

and Bygate (1996) have observed, this research has in part been motivated by the prospect that modulating the design and information processing demands of tasks can lead to language practice and consolidation in each of these important areas. For example, some tasks may lead learners to prioritize ¯uency, others to prioritize complexity or accuracy of production. To make predictions about the e€ects of task complexity on these aspects of task production it is necessary to distinguish between monologic, non interactive tasks such as narratives and interactive tasks such as discussion or information transfer. Complex monologic tasks, I would argue, should elicit less ¯uent, but more accurate and complex production, relative to simpler tasks when complexity is manipulated along the resource-directing dimensions identi®ed in Figure 1 which make increasing functional demands on the language user (i.e. +/± here-and-now, +/± few elements, +/± no reasoning demands). This is because, as described previously, increasing complexity along these dimensions is potentially a means of directing resources to a wider range of functional and linguistic requirements. These predicted performance e€ects are motivated by the following considerations: (1) structural complexity tends to accompany functional complexity in discourse (Givon 1985) and therefore complexity of oral production will develop in response to the increasingly complex functional demands the learner has to meet during communication; (2) greater attention to L2 speech, induced by the communicative demands of speci®c tasks, sometimes results in higher levels of accuracy on task relevant, communicatively non-redundant language (Loschky and Bley-Vroman 1993; Tarone and Parrish 1988);3 and (3) high communicative and cognitive task demands can lead learners to `push production' (Swain 1995), `stretch interlanguage' (Long 1989) and expend the additional `mental e€ort' on production that leads to destabilization of fossilized forms (Schmidt 1983). In contrast, increasing monologic task complexity along +/± planning time, +/± prior knowledge, or +/± single task dimensions leads to a depletion of attentional and memory resources. This should have the e€ects of reducing ¯uency, accuracy, and complexity on the more complex tasks, and e€ects, reported in the following section, have been found to be in line with these predictions for the di€erence between planned (simpler) and unplanned (more complex) task performance by Skehan (1996) and others. Synergistic e€ects of simultaneously increasing complexity along these resource-directing and resource-depleting dimensions can be expected and research is needed to investigate these. For example, the e€ects of increasing complexity along resource-directing dimensions should be greater when these are simultaneously simpler along the resource-depleting dimensions. That is, giving planning time and relevant background knowledge will increase the attentional resources available to meet the functional and linguistic demands of tasks requiring reasoning, or reference to many elements, and should lead to greater accuracy and complexity of production on these tasks than when planning time and/or relevant prior knowledge are not available. The clearest prediction that can be made about the e€ects of task

36

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

complexity on language production during interactive tasks is that complex versions of tasks should result in more negotiation, and consequently more con®rmation checks and clari®cation requests than simpler versions (cf. Long 1983). This should not a€ect the predictions for accuracy made above, but it may have consequences for the predictions for complexity, since more interaction and turntaking may mitigate speakers' attempts to produce complex syntax and subordination, resulting instead in greater numbers of elliptical yes/no or single clause answers to clari®cation requests and con®rmation checks relative to performance on less interactively negotiated simple versions of a task.

Dimensions of task complexity: research findings The predictions described above for the e€ects of task complexity on monologic task performance have received some support from previous studies, as the following brief review demonstrates. However, very few studies have been done on the e€ects of task complexity on interactive task productionÐa further motivation for the present study. It is important to note, also, that in many cases of actual classroom pedagogic task performance tasks will be made progressively more complex for learners along a number of dimensions simultaneously. The dimensions adopted for manipulation will be based on an analysis of the performance demands of the target task (see Norris et al. 1998 for examples of this). However, experimental and quasiexperimental SLA research in the componential framework reported here has often involved studying the e€ects of manipulating complexity along one dimension independently of others (cf. Sternberg 1977).4 The facilitating e€ect of + prior knowledge on task performance receives support from research outside the ®eld of SLA (e.g. Anderson 1981; Britten and Tresser 1982; Joseph and Dwyer 1984) as well as from within it for its e€ects on L2 reading and listening comprehension (e.g. Barry and Lazarte 1998; Carrell and Wise 1998; Urwin 1999). Good and Butterworth (1980) found prior knowledge (of a familiar route) resulted in signi®cantly more ¯uent ®rst language (L1) speech production on a route description task than no prior knowledge (describing an unfamiliar route). Similarly, Chang (1999) in a small-scale study of six Taiwanese learners of English on a monologic one-way task found topic familiarity led to signi®cantly greater ¯uency (words per error free T-unit, and words per minute), but had no e€ect on accuracy (error rate per T-unit). There is some evidence, however, that prior knowledge may interact with pro®ciency level in facilitating task performance. Clapham (1996) found prior knowledge of a domain did not facilitate performance for subjects taking reading tests who were at a low level of pro®ciency, but increasingly facilitated performance for subjects at higher levels of pro®ciency. In general, ®ndings for the e€ects of manipulating +/± planning time show planning leads to gains in accuracy, ¯uency, and complexity. Ellis (1987)

PETER ROBINSON 37

found more accurate production of past tense morphemes on planned versus unplanned narratives, and Crookes (1989) found a trend for more complex syntax on planned versus unplanned information gap tasks. Foster and Skehan (1996) found planning time led to increases in ¯uency, accuracy, and complexity, though Skehan and Foster (1997) found gains in ¯uency but more limited evidence of gains in accuracy (Skehan and Foster in press). Undoubtedly, giving planning time, and focusing attention on relevant aspects of task structure (Skehan 1996) reduces cognitive demands. The e€ects for greater accuracy on simpler tasks along this dimension are also possibly due in part to learner avoidance of problematic forms and to a narrowing of their productive repertoire to tried and trusted forms during the planning phase. The third resource-depleting dimension of L2 task complexity identi®ed in Figure 1, +/± single task demand, is motivated by ®ndings from research in other educational domains where performance on a primary task is examined to see to what extent decrements occur when a secondary task is added (e.g. Gopher 1992; Sanders 1998; Sarno and Wickens 1995; Wickens 1989, 1992). Robinson and Lim (1993) operationalized the dimension using a one-way interactive direction-giving task requiring speakers to give directions from point A to point B on a map to a partner. In the single task condition the route was marked on the map for the speaker, while in the dual task condition the route was not marked, following the thinking that in this latter condition the speaker would have to both think up the route and describe it (two tasks) compared to simply describing an identi®ed route (one task). Production on the route-not-marked map task was less ¯uent than on the route-marked task, with no di€erences for accuracy and complexity, using the measures adopted in Robinson (1995a). This dimension of complexity falls within the area Skehan (1996) has called the `degree of structure' task materials impose on learner performance, and relates to those task activity factors which Nunan (1989) claims contribute to their `diculty', e.g. listening for comprehension in order to simply tick a multiple-choice solution, versus listening to simultaneously writing the answer out while listening. Robinson (1995a) and Rahimpour (1997) operationalized the resourcedirecting dimension of +/± here-and-now as a distinction between narratives performed when learners describe a series of events in the present tense while looking at pictures illustrating them (here-and-now), versus narratives performed from memory without looking at the pictures, and delivered in the past tense (there-and-then). These studies were motivated by ®ndings from L1 development which show there-and-then reference to emerge later than present tense, context supported reference (e.g. Sachs 1983), and by similar ®ndings from SLA research (Meisel 1987). Robinson (1995a) found the complex there-and-then condition elicited signi®cantly more lexical variety, with a strong trend to greater accuracy and dys¯uency, but with no signi®cant ®ndings for complexity. Accuracy was measured via target like use (TLU) of articles, ¯uency by numbers of pauses and words per utterance, and

38

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

Figure 3: Increasingly cognitively complex versions of a map task complexity by multipropositional utterances, S-nodes per T-unit and a measure of lexical complexity, the ratio of lexical to grammatical words. Similarly Rahimpour (1997) found there-and-then narratives were more accurate (in error-free T-units), and more dys¯uent. These ®ndings are in line with the predictions for the e€ects of increasing cognitive complexity on monologic task production made above. Both Brown et al. (1984) and Prabhu (1987) claim that tasks requiring selective information transmission + reasoning to establish causality, and justi®cation of beliefs are more complex than tasks requiring non-selective information transmission, without these demands. They also claim that tasks requiring a few clearly di€erent elements to be distinguished from each other (e.g. apples, trees, and clouds), are easier than tasks requiring many similar elements to be distinguished (e.g. cars in a trac jam, buildings and streets on a map). Complex tasks on both dimensions (i.e. +/± reasoning demands and +/± many elements) are also likely to require a wider range of language than simpler tasks, e.g. greater use of logical connectors, subordination, complex noun phrases, and a wider variety of attributive adjectives (Brown 1995: 138± 9; Brown and Yule 1983: 135). Some consequences can be drawn from the dimensions described above for task sequencing and syllabus design. Since the dimensions are separable, tasks can be made more complex on an increasing number of them. In the sequence illustrated in Figure 3, learners are initially presented with a map of a small area they are familiar with, with the route marked on, and are given planning time to decide how to describe the route to a partner. Progressively more complex versions can be practised by removing planning time, then removing the route, using a map of an unfamiliar area, then ®nally increasing

PETER ROBINSON 39

the size of the map. In this way tasks increase in complexity and authenticity, gradually approximating the demands of real-world target tasks. Practice on increasingly complex versions presumably avoids the decrements in performance that result from going from a simple to an authentic version of the map in one step, and reduces perceptions of the diculty of the most complex version. Investigating the validity of this assumption is another aim of the research reported below.

THE PRESENT STUDY Research questions and hypotheses The present study examined the e€ects of increasing task complexity on measures of learner production on two versions of the map task described above (version 3 and version 5). I also examined the e€ects of increasing task complexity on learner ratings of task diculty, as assessed by responses to a questionnaire assessing a€ective factors completed immediately after performance on each version. I was also interested in the e€ects of sequencing decisions (from simple to complex versus the reverse sequence) on measures of production and ratings of diculty. There were three research questions.

Research Question 1 `Is language production qualitatively di€erent on simple and complex versions?' In line with the theoretical motivation described above, I hypothesized that the cognitively complex version would push speaker (information-giver) output more than the simpler version by making additional functional demands (reference to many versus few elements) and therefore increases in lexical variety and grammatical accuracy would be expected for speaker production relative to the simpler version. Since I expected that increasing the cognitive complexity of the task would cause additional communication problems, I also hypothesized that the hearer (information-receiver) would be forced to make more clari®cation requests and con®rmation checks on the more complex version, compared to the simpler version. Choice of these functional units was motivated, in part, by Long's (1983) ®nding that greater problematicity of communication between native speakers (NS) and non-native speakers (NNS), compared to NS±NS communication, is also accompanied by greater use of clari®cation requests and con®rmation checks. Consequently, I expected the greater interaction on the complex version would lead to more frequent speaker interruption, and turn-taking, resulting in less ¯uent and complex speaker production, relative to the simple version.

40

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

Research Question 2 `Is the di€erence in cognitive complexity of the tasks matched by learner perceptions of di€erences in their diculty?' I hypothesized that it would be matched in three of the areas covered by the questionnaire: the complex version would be rated more dicult overall, more stressful, and ability to perform the task would be rated lower than for the simple version. I had no directional hypotheses for the remaining two areas, interest in and motivation to perform the simple and complex versions.

Research Question 3 `Does sequencing tasks from simple to complex versus complex to simple a€ect learner production and ratings of diculty?' There is some evidence from applied psychological research that perceptions of diculty are a€ected by the order in which simple and complex tasks are performed (Hancock et al. 1995) but there has been no L2 task research into this issue. Following the ®ndings of Hancock et al., that performing simple versions of tasks before complex versions reduces perceptions of the diculty of the complex version, I hypothesized the complex version would be judged to be less dicult and stressful, and perceived ability on it would be higher, when it was performed in the sequence simple±complex versus the sequence complex±simple. Some recent SLA research has examined the e€ects of task repetition on measures of production (Bygate 1996) ®nding increases in accuracy, ¯uency, and complexity as a result of practice. However, in the present study, though versions of the task were repeated, these di€ered in complexity. I assumed complexity di€erentials would be a more powerful in¯uence on production than repetition of task versions, and therefore that production on the simple version (and on the complex version) would be the same whether performed in the sequence simple±complex, or in the sequence complex±simple.

Participants and research design The participants in the study were 44 Japanese university undergraduates, aged between 20 and 24 at Aoyama Gakuin University in Tokyo. All had completed six years of high school English education, and two years of English classes at Aoyama Gakuin, prior to taking specialized content courses in their third and fourth years of study. There were two between-participants factors (Sequence and Role) and one within-participants factor (Task Complexity) in this design. Participants were randomly assigned to the role of speaker (information-giver) or hearer (information-receiver) on the two map tasks. The speaker was asked to give directions from point A to point B, marked on both their maps. The hearer had the same maps with only point A marked on. The hearer was asked to listen and draw the routes described by the speaker and to identify point B on each map. Hearers were told they could ask questions if they wanted to. Participants were seated opposite each other, but

PETER ROBINSON 41

were asked to hold their maps at an angle so they could not be seen by the other person. One map task (the simpler version) required the speaker to give directions from A to B using a map covering a small area which they had prior knowledge of, Aoyama Gakuin University campus. The complex version required them to give directions from A to B using a map of a larger area they were likely to be unfamiliar with, taken from an authentic street map of the Nihombashi area in central Tokyo. The two dimensions of complexity distinguishing the cognitive demands of these tasks, amount of information and availability or not of prior knowledge, are related to other dimensions in Figure 3 above and correspond to version 3 and version 5. Task sequencing was controlled for. Half the dyads were randomly assigned to perform the simpler (version 3) task then the complex (version 5) task, and half to perform the reverse sequence.

Measures taken Speaker (information-giver) production was assessed for accuracy, ¯uency, and complexity. To measure accuracy a general measure, error free C-units (EFC) was used.5 Syntactic complexity was assessed by a measure of subordination, clauses per C-unit (CPC), and lexical complexity or variety by token type ratio (TTR). Fluency was assessed by a measure of the number of words per C-unit (WPC). Hearer (information-receiver) production was assessed using two interactional units, clari®cation requests (CR) and con®rmation checks (CC).6 Immediately following performance of each task both participants completed a brief questionnaire. The questionnaire was brief to ensure minimal disruption between performance of the two tasks, and therefore to facilitate comparison of their relative demands. Each item was rated on a 9-point Likert scale. These items were worded in the following way, from left 0, to right 9: 1 2 3 4 5

I thought this task was easy/I thought this task was hard; I felt relaxed doing this task/I felt frustrated doing this task; I didn't do well on this task/I did well on this task; This task was not interesting/This task was interesting; I don't want to do more tasks like this/I want to do more tasks like this.

Responses to these ®ve items assessing overall perception of task diculty, ratings of stress, perceived ability to complete the task, interest in task content, and motivation to complete these and other tasks like them, were used to assess learner perception of task diculty.

42

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

Table 1: Descriptive statistics for speaker and hearer production on simple and complex versions Speaker production

Simple Complex Probability

Hearer production

TTR M/SD

%EFC M/SD

WPC M/SD

CPC M/SD

CC M/SD

CR M/SD

4.3/1.4 3.6/1.1 p = .01

58/17 62/13 p = .13

6.6/1.4 5.9/1.3 p = .03

1.05/0.8 1.05/0.8 n.s.

1.9/2.3 4.5/3.5 p < .01

0.8/1.0 1.5/1.4 n.s.

Key:

TTR = Token type ratio CPC = Clauses per C-unit

EFC = % Error free C-units CC = Con®rmation checks

WPC = Words per C-unit CR = Clari®cation requests

Table 2: Descriptive statistics for questionnaire responses on simple and complex versions Diculty M/SD Simple 3.5/2.1 Complex 5.4/2.3 Probability p < .01

Stress M/SD

Ability M/SD

Interest M/SD

Motivation M/SD

3.7/2.3 4.8/2.4 p < .01

5.0/2.1 4.4/2.0 p = .07

5.0/2.1 5.7/2.1 p = .08

5.3/2.2 5.2/2.3 n.s.

RESULTS Main analyses Research Question 1. Task complexity and task production As in a previous study of a similar direction giving task (Foster and Skehan 1996), both simple and complex versions resulted in language of low complexity (CPC), on average little more than one clause per C-unit. Due to the non-normal distribution of this data a Wilcoxon matched pairs signed ranks test was performed for production on the two versions and showed no signi®cant di€erence. All other data were normally distributed. Consequently parametric analyses were performed. Results of a repeated measure MANOVA, using three measures of speaker production as dependent variables (TTR, EFC, and WPC), shows a signi®cant interaction of Task Complexity and Production measure (F (2/20) = 6.039 p < .01). Planned a priori means comparisons show this is attributable to the signi®cantly lower token type ratio for production on the complex version, indicating greater lexical variety

PETER ROBINSON 43

in production on this task (F = 6.081, p = .01) and to signi®cantly greater numbers of words per C-unit on the simple version (F = 5.007, p = .03). There is also a trend to greater accuracy on the complex version in line with the research hypotheses (F = 2.271, p = .13) (see Figure 4 and Table 1). A repeated measure MANOVA, using measures of hearer interaction (CC and CR) as dependent variables, shows a signi®cant main e€ect for Task Complexity and a signi®cant interaction of Task Complexity and Interaction measure (F (1/21) = 6.745 p < .01). Planned comparisons show this is attributable to greater use of con®rmation checks on the complex version (F = 24.254, p < .01) (see Table 2). The results for clari®cation requests, while in the predicted direction, are not signi®cant. Therefore the hypotheses for the ®rst research question are con®rmed for the lexical variety and ¯uency of speaker production, and for use of con®rmation checks.

Research Question 2. Task complexity and task diculty A repeated measure MANOVA, using responses to the task diculty questionnaire items as dependent variables, shows a signi®cant main e€ect for the within-participants factor Task Complexity (F (1/41) = 18.723, p < .01) and a signi®cant interaction of Diculty measure and Task Complexity (F (4/ 38) = 4.024, p < .01). A priori planned comparisons show ratings of overall diculty (F = 27.810, p < .01) and stress (F = 9.599, p < .01) are signi®cantly higher on the complex version. Ratings of ability are not signi®cant, (F = 3.168, p = .07) but in the direction predicted by the hypotheses, since con®dence in ability is judged lower on the complex version. Ratings of interest and motivation do not di€er signi®cantly (see Figure 5 and Table 2). The hypotheses for the second research question are therefore largely con®rmed.

Research Question 3. Task sequencing, task diculty, and task production To examine the e€ect of sequence on production this factor was entered into the two repeated measure MANOVA used to address the ®rst research question. For speaker (information-giver) production there is a signi®cant interaction of Sequence and Production measure (F (2/19) = 11.675 p < .01), but no signi®cant interaction of Sequence, Task Complexity, and Production measure. As Table 3 shows, the interaction of Sequence and Production measure is attributable to greater accuracy on both versions when performed in the simple±complex sequence, and greater ¯uency on both versions when performed in the sequence complex±simple. For hearer (informationreceiver) interaction there is no signi®cant main e€ect for Sequence, or signi®cant interaction of Sequence and Interaction measure. The hypothesis that sequence would not a€ect production is therefore con®rmed only for the amount of interaction.

44

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

Figure 4: Speaker production on simple and complex versions of the map task

Figure 5: Diculty ratings for simple and complex versions of the map task

PETER ROBINSON 45

To examine the e€ect of sequencing on ratings of diculty the factor Sequence was entered into the repeated measure MANOVA used to answer the second research question. There is no signi®cant main e€ect for Sequence, or signi®cant interaction of Sequence with Diculty measure, or Task Complexity. The hypotheses for the e€ect of sequencing on ratings of complex task diculty are therefore not con®rmed.

Summary of the main analyses 1 2

3 4

Task complexity a€ects speaker and hearer production. Task complexity signi®cantly a€ects the lexical variety and ¯uency of speaker production, and amount of hearer interaction in the predicted direction. Cognitive demands of tasks and ratings of their diculty are related. Task complexity signi®cantly a€ects ratings of overall diculty, and stress in the predicted direction. There is a trend to less con®dence in ability on, and more interest in, the complex version. Sequencing and ratings of diculty are unrelated. Sequencing from simple to complex versus the reverse sequence does not signi®cantly a€ect ratings of the diculty of the complex version. Sequencing a€ects speaker production but not interaction. Sequencing tasks (from simple to complex versus the reverse sequence) has signi®cant e€ects on accuracy and ¯uency of speaker production, but has no e€ect on the amount of interaction.

Further analyses Task diculty and production The main analyses reveal a substantial in¯uence of task complexity on production. An interesting further question is whether learner factors contributing to Diculty, such as anxiety and motivation are also related to the quality of production. To further examine the production data, correlations of the Diculty ratings and speaker and hearer production were examined. No correlations of hearer interaction and the Diculty measures are signi®cant. However, Diculty is signi®cantly correlated with aspects of speaker (information-giver) production. On the simple version there is a signi®cant positive correlation of perceived ability and ¯uency (WPC) (r = .410, p = .03). Those speakers who produced more words per Cunit were more con®dent that they had been able to successfully complete the task than those who produced fewer words per C-unit. There is also a signi®cant negative correlation of motivation and token type ratio (TTR) (r = ±.451, p = .03). In other words, speakers whose production had relatively low token type ratios (and was therefore more lexically varied) rated their motivation to complete the task higher than those who produced less lexically varied language (with consequently higher token type ratios). On the complex version, only perceived ability correlates signi®cantly, and positively

46

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

with ¯uency (WPC) (r = .551, p < .01), as it does on the simple version. These ®ndings for the e€ects of learner ratings of Diculty on task production are interesting, suggesting that a€ective responses to tasks and the ¯uency and lexical variety of production are positively related. However, in general the ®ndings are weaker than those reported earlier for the e€ects of cognitively de®ned Complexity on production. Diculty is not related to hearer production on either version. Only ratings of ability to successfully complete the task are positively and signi®cantly related to speaker ¯uency on both simple and complex versions.

Task diculty and role As mentioned previously it is likely that task condition factors, such as Role may a€ect diculty ratings in a one-way information transfer task such as the one used in the present study, since the speaker (information-giver), has a greater onus of responsibility if the dyad is to successfully complete the task. To examine the e€ect of Role on ratings of Diculty, this was entered as a factor in the repeated measure MANOVA used to address the second research question. There is a signi®cant interaction of Role and Diculty measure (F (4/37) = 3.466 p < .01) but no interaction of these with Task Complexity which, as Figure 6 shows, is attributable to the fact that hearers, on both simple and complex versions, judge tasks to be less dicult overall, less stressful, and their ability to be higher, than speakers do. There is no di€erence in speaker and hearer ratings of interest and motivation. It is clear then, that Role does a€ect ratings of Diculty di€erentially, in line with the earlier speculations.

Summary of the further analyses 1

2 3

Diculty and speaker production are related. High ratings of con®dence in ability to complete the task are accompanied by more ¯uent production on both versions. Greater lexical variety is also positively related to motivation on the simple version. Diculty and hearer interaction are unrelated. Role a€ects ratings of diculty. Hearers judge simple and complex versions to be less dicult overall, less stressful, and are more con®dent of their ability on them than speakers.

DISCUSSION Task complexity and production In line with the theoretical motivation given earlier in this paper for the e€ects of task complexity on interactive task production, I hypothesized that the greater cognitive demands of the complex version would push the speaker (information-giver) to produce more lexically varied and more accurate

PETER ROBINSON 47

Figure 6: Diculty ratings for speaker and hearer across simple and complex versions language. I also hypothesized that there would be more interaction on the complex version, leading to more con®rmation checks and clari®cation requests on the part of the hearer (information-receiver). This greater amount of interaction generated by the complex version would also have the e€ect of reducing the ¯uency and complexity of speaker (informationgiver) production, relative to performance on the simple version. These hypotheses were largely con®rmed. There is clear evidence that increasing task complexity qualitatively changes the nature of speaker production on task. There is greater lexical variety on the complex version and greater ¯uency on the simple version (see Figure 4 and Table 1). The ®nding for greater lexical variety (lower token type ratios) on the complex version is no doubt in part a re¯ection of its additional functional demands, referring to and distinguishing between a greater number of elements, such as streets, buildings, and other landmarks. It is also possibly the result of the di€erentials in assumed shared prior knowledge, since less complex and explicit noun phrases could be used as referring expressions on the simple version where such knowledge was available (Bongaerts et al. 1987 report similar ®ndings for the e€ect of shared knowledge on the complexity of referring expressions). The signi®cant di€erence in average words per C-unit is also in line with the prediction made earlier in this paper, that production on simple versions will be more ¯uent than on complex versions of tasks. This result was also no

48

Speaker production Simple±complex sequence

Mean SD

Complex±simple sequence

Simple version

Complex version

Simple version

Complex version

TTR

EFC

WPC CPC

TTR

EFC

WPC CPC

TTR

EFC

WPC CPC

TTR

EFC

WPC CPC

4.1 1.6

65 18

6.0 1.3

3.1 1.2

68 13

5.7 1.4

4.6 1.1

47 14

7.3 1.4

4.2 0.8

54 11

6.1 1.1

1.06 1.00

1.06 1.10

1.02 0.50

Key: TTR = Token type ratio, EFC = % Error free C-units, WPC = Words per C-unit, CPC = Clauses per C-unit

1.03 0.40

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

Table 3: Descriptive statistics for speaker production on tasks performed in the simple±complex, and complex±simple sequences.

PETER ROBINSON 49

doubt partially a€ected by the relatively high amount of interaction generated by the complex version where hearer con®rmation checks often resulted in brief phrasal, or elliptical yes/no answers. Since elliptical yes/no answers are considered to be C-units (see note 5) then a preponderance of these on the complex version reduced the average numbers of words per C-unit. For the same reason, I hypothesized that more complex interactive tasks, encouraging more negotiation would also likely result in less complex production (measured in clauses per C-unit) than simpler counterparts. This was not con®rmed in the present study since production on both versions was at a low level of complexity. Task complexity did not signi®cantly a€ect speaker (information-giver) accuracy either, although there is a trend in the direction of the research hypotheses towards greater accuracy on the complex version, suggesting that increasing the cognitive demands of tasks may direct greater attentional resources to e€ective `microplanning' (Levelt 1989), or syntactic preparation of production units, compared to less cognitively demanding tasks. Given the non-signi®cant ®nding for accuracy di€erences, however, this speculation is in need of con®rmation from future research. Finally, as predicted, there was clear evidence of an e€ect for task complexity on measures of negotiation and interaction. Hearers made extensive use of con®rmation checks on the complex version, with a trend towards more use of clari®cation requests. This con®rms the relationship between communicative demand and cognitive demand and is an interesting ®nding in itself for those concerned to present tasks to learners which optimize the amount of negotiation on task.

Task complexity and task difficulty Di€erences in the cognitively de®ned complexity of tasks were also re¯ected in learner perceptions of task diculty. These ®ndings are promising for those concerned to implement task-based syllabuses, for two reasons. First, the results show that the complex version is rated signi®cantly more stressful and dicult overall, with a trend to less con®dence in ability on the complex version. To this extent task sequencing decisions based on complexity are also legitimate for learners in terms of their a€ective demands (see Figure 5 and Table 2). Additionally, and encouragingly, di€erences in task complexity are unrelated to ratings of interest, and motivation, suggesting that as tasks increase in cognitive complexity, approaching the authenticity of target task demands, there is no loss of interest in or motivation to complete the task. Task role also had an e€ect on ratings of diculty (see Figure 6). The results described above are clearest for speakers. Hearers are considerably more con®dent of their ability, and perceive less overall diculty and stress on both versions than speakers. This is understandable, given that the onus for successful task performance was di€erentially distributed in this one-way task. Di€erent ®ndings would no doubt emerge from studies adopting two-way task conditions.

50

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

Task complexity and task sequencing What e€ect do decisions about the order in which simple and complex versions are performed have on ratings of diculty and task production? Contrary to expectations sequencing tasks from simple to complex versus the reverse sequence did not signi®cantly a€ect ratings of the diculty of the complex version, which were equivalent whether it was performed ®rst or second. It is possible that ®ndings in the predicted direction would have emerged if the simplest version in Figure 3 had been contrasted with the complex version 5 used in the present study, although this remains an issue for future research, and an important one for decisions about task sequencing and their e€ects on learners. As predicted, task sequencing did not a€ect the di€erentials in hearer (information-receiver) interaction on simple and complex versions. However, interestingly, and unexpectedly, there was a signi®cant interaction of speaker (information-giver) production and sequence. There was no interaction of speaker production, sequence, and task complexity, since as Table 3 shows, performance on simple and complex versions in either sequence still reveals the e€ects described above (i.e. greater lexical variety on the complex version, and greater ¯uency on the simple version). However, these results also reveal the equally powerful in¯uence of sequencing decisions on speaker (information-giver) production. Sequencing tasks from complex to simple led to increases in ¯uency of speaker production on both versions, compared to the reverse sequence, while sequencing from simple to complex led to increases in accuracy on both versions compared to the reverse sequence. It is not clear how this ®nding can be explained. One possibility, with regard to both ¯uency and accuracy, is that performing the simple version ®rst (with its relatively lower attentional resource demands) led to more ecient scheduling and coordinatingÐor `time-sharing'Ðof the two main task components (thinking up the route and describing it). Ecient time-sharing has positive e€ects on the quality of attentional allocation to each task component (see Gopher 1992; Wickens 1992). With regard to speaker production this therefore led to more ecient and economical speech on both versions in this sequence, resulting in fewer words per C-unit relative to tasks in the reverse sequence. Consequently more attention was also available for focusing on accuracy of production on both versions performed in this sequence. Sequencing tasks from complex to simple did not have this advantage. This less than optimum sequence for scheduling and co-ordinating task components (doing the complex, more attentional resource demanding version ®rst) resulted in production which was neither so economical, nor so accurate.

PETER ROBINSON 51

Limitations of the study and issues for future research The present study has a number of limitations and these should be acknowledged. For example, the method I chose to identify ratings of task diculty, while adopted for good reason, i.e. its minimal intrusion into task performance and on-line re¯ection on task demands, needs to be supplemented by studies adopting more extensive and detailed instrumentation, such as modi®ed versions of Horwitz et al.'s (1986) classroom anxiety scale (see e.g. Holthouse 1995), and motivated by comprehensive models of classroom and task motivation such as the one developed by DoÈrnyei and Otto (1998). Additionally, studies of subjective responses to task demands, using methodologies such as interviews, and protocol analyses, are to be encouraged since they will likely enable clearer examination and conclusions to be drawn regarding the speci®c causes of task diculty for learners (see Jourdenais in press). They will also provide insight into the extent to which learners di€er in their interpretation of task demands (see Coughlan and Du€ 1994). Within the framework described in this paper it will be important to examine the extent to which learner interpretations of the demands of the task match the designer's intentions in manipulating the dimensions of complexity I have described and the e€ects of di€ering interpretations on production. Finally, the generalizability of the present ®ndings must be demonstrated, both with regard to the participants and the content of tasks.

SUMMARY AND CONCLUSION Each of the three broad factors I have identi®ed in this framework for studying the e€ects of pedagogic tasks on second language learner production is multidimensional, as Figure 1 shows. Clearly, the full extent of variation in task production cannot be reduced to any one factor or dimension of that factor alone. However, I have attempted to make a case for basing sequencing decisions in task-based approaches to syllabus design on distinctions between the cognitive demands of tasks which contribute to their relative complexity. The present study has shown that the complexity of tasks does exert a considerable in¯uence on learner production. For this, and two other reasons I would argue that sequencing tasks on the basis of their cognitive complexity is to be preferred over sequencing decisions based on task diculty or task conditions. First, as discussed earlier in this paper, factors a€ecting perceptions of diculty (such as motivation and anxiety) are hard or impossible to diagnose in advance of pedagogic task performance, and so are problematic as a basis for a priori prospective decisions about sequencing tasks. They are, of course, important to assess on-line, and adjustments to syllabus decisions may often be made based on themÐand this is fundamental to on-line and retrospective approaches to syllabus design (e.g. Breen 1984). Decisions about task conditions, I also suggest, will largely be motivated by the needs analysis,

52

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

and by ®delity to the target task performance the course is attempting to develop. They will therefore play little role in sequencing decisions since speci®c task conditions will have been determined as appropriate for target task performance from the outset and will be replicated in each version of pedagogic tasks approximating those target task demands. In support of this proposal the present study has provided some evidence that cognitive complexity is a robust, and manipulable in¯uence on learner production, and is therefore a feasible basis for design and sequencing decisions which operationalize a task-based syllabus. Adjusting the cognitive demands of a one-way direction-giving map task has signi®cant e€ects on the quality of speaker (information-giver) production, on the amount of hearer (information-receiver) interaction, and on learner ratings of task diculty. The ®nding that increases in task complexity a€ect some aspects of learner ratings of diculty, e.g. stress, but not others, such as interest in the task or motivation to complete it, are of clear relevance to theoretical proposals for, and practical implementations of task-based approaches to syllabus design. Powerful e€ects of task complexity were also found on measures of interaction, with more complex tasks resulting in greater numbers of con®rmation checks, and on the quality of speaker production, with the complex version resulting in less ¯uent, but more lexically varied language. Both of these ®ndings are also of clear relevance to decisions about task-based materials design and methodological options in task selection. Sequencing tasks from simple to complex versus the reverse sequence did not a€ect ratings of the complex version's diculty, or the extent of interaction, but it did result in shorter, more accurate and arguably therefore more ecient Cunits on both versions. This ®nding has implications for task-based syllabus design, and I have interpreted it as support for sequencing versions of tasks from simple to complex, modulated along the dimensions of task complexity described in this paper, some of which are illustrated in Figure 3. Many other studies of this area are needed, not least with respect to the in¯uence of other diculty variables on language production and task performance not operationalized in this study, such as aptitude. Con®dence in pedagogic decision making about the implications of task complexity for sequencing decisions and its e€ects on production can only be increased by cumulative ®ndings emerging from a range of studies, adopting a variety of measures and methodologies. What I hope to have achieved in this paper is a sense of the pedagogic importance and relevance of the questions asked (a number of them not so far addressed by research to date), and to establish a conceptual and methodological framework for examining them which is also pedagogically feasible. (Revised version received April 2000)

PETER ROBINSON 53

NOTES 1 One reviewer of this paper suggested that this quotation implies that Widdowson agrees with Nunan, i.e. that tasks be used to implement a linguistic syllabus, in the absence of any criteria for independently grading and sequencing the problem solving tasks themselves. Another reviewer also points out that this quotation has major implications for any kind of syllabus, and no current proposals for syllabus design could meet the stringent criterion Widdowson puts forward here. 2 Discussing the relationship between intelligence and measures of working memory load in a verbal reasoning task Spilsbury et al. write; `Task diculty and task complexity need to be distinguished carefully. Occasionally these terms are used as synonyms. That this is not justi®ed is demonstrated by instances when the diculty of a task is increased without increasing the task complexity by, for example, making perception of a cognitive task more dicult by using smaller print. To some extent the confusion between these terms turns on the lack of an adequate and acceptable de®nition for complexity. The de®nitions do, however, generally include some reference to notions of intricacy or of interconnectedness either in terms of the number of elements in a task or steps needed to solve a problem' (Spilsbury et al. 1990: 1069). That is, size of print di€erentially a€ects perceptions of diculty, depending on di€erences in eyesightÐa physical ability variable. Intelligence is also an ability variable, as is language learning aptitude, albeit mental not physical abilities. Spilsbury et al.'s main point, then, is that intelligence a€ects the perceptions of the diculty of cognitive tasks, but task complexity is logically independent of it, depending instead on task factors such as the working memory load it imposes. 3 One caveat to the claim made here about accuracy is that where speci®c language features are assessed, as in target-like use analysis of e.g. articles, these should be chosen with the learner's developmental level in mind (Bley-Vroman 1983). If learners are not developmentally ready to pro-

duce features of syntax or morphology, it is unlikely that complex tasks will induce them with a high degree of accuracy, though they may `push' learners to make incursions into the next level they are ready to pass through. 4 Throughout this paper I have termed the approach I have taken to describing task complexity as a `framework'. My research strategy has involved examining the e€ects of manipulating dimensions of tasks that have been claimedÐwithin SLA research, as well as outside itÐto contribute to cognitive complexity. The approach is similar to Sternberg's (1977) early attempts to identify the cognitive components of information processing on tasks used to measure human mental abilities. However, the approach is subtly di€erent. Sternberg (1977) identi®ed components of information processing used in response to tasks; i.e. `performance components' used in the execution of certain processing strategies, and `metacomponents' used to monitor and select performance components. That is because he was concerned to develop a theory of intelligence. I identify components of tasks as a series of dimensions which make di€erential information processing demands. That is because I am concerned to develop a theory of task complexity. A full theory of task complexity would relate the components of tasks contributing to their relative complexity to the components of information processing (see Robinson 1999). 5 Communication units or C-units are similar to T-units, which Chaudron (1988) de®nes as any syntactic main clause and its associated subordinate clauses. As Chaudron (1988), Brock (1986), and Skehan and Foster (1997) point out, C-units di€er in that elliptical answers to questions can also be considered complete predications, and are therefore more suitable for analysing interactive discourse of the type required during performance on these tasks. In coding Cunits I based criteria on de®nitions o€ered by Chaudron, Skehan and Foster, and Brock. The basic criteria are as follows: 1. A main, or independent clause, plus

54

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

whatever other subordinate clauses or nonclauses are attached to, or embedded within it. (Chaudron) . . . or grammatically independent predications (Brock), e.g. Go straight to the rotary (1 C-unit) and go along the rotary (1 C-unit) and turn left (1 C-unit) you go right the rotary along the rotary (1 C-unit) 2. Where the speaker's turn is interrupted by listener backchannelling, and repetition, count both halves of the speaker's turn as one C-unit if they are one clause, e.g. S yeah go and turn left at the headquarters (1 C-unit) L turn left 3. Where the speakers appear to abandon a clause and reformulate it, do not count the abandoned portion in the C-unit, e.g. er you can go er (no C-unit) you can go right the rotary along the rotary (1 C-unit)

4. Elliptical answers to questions count as a C-unit, i.e. `yes' can be a whole unit of communication when it is an answer to a clausal question, e.g. S yes (1 C-unit) L I reach Headquarters S on your right yes (not a C-unit) L right? S yeah yes to rotary L go straight the main street (1 C-unit) 6 Three native English speaking raters separately coded all the data for C-units, then compared analyses and reached consensus. Subsequently one rater analysed all the data for the C-unit based measures and for token type ratio. There was between 90 per cent and 100 per cent agreement between this rater and the other two raters for all measures based on a randomly selected sample of eight of the transcripts.

REFERENCES Ackerman, P. 1988. `Determinants of individual differences during skill acquisition: Cognitive abilities and information processing.' Journal of Experimental Psychology: General 117: 288±318. Anderson, J. R. 1981. `Effects of prior knowledge on memory for new information.' Memory and Cognition 9: 237±46. Bachman, L. and A. S. Palmer. 1996. Language Testing in Practice: Designing and Developing Useful Language Tests. Oxford: Oxford University Press. Baker, D. 1990. A Guide to Language Testing. London: Edward Arnold. Barry, S. and A. A. Lazarte. 1998. `Evidence for mental models: How do prior knowledge, syntactic complexity, and reading topic affect inference generation in a recall task for nonnative readers of Spanish?' Modern Language Journal 82: 176±93. Bley-Vroman, R. 1983. `The comparative fallacy in interlanguage studies: The case of systematicity.' Language Learning 33: 1±17. Bongaerts, T., E. Kellerman, and A. Bentlage. 1987. `Perspective and proficiency in L2 referential communication.' Studies in Second Language Acquisition 9: 171±200. Breen, M. 1984. Process syllabuses for the language classroom. In C. J. Brumfit (ed.),

General English Syllabus Design, Oxford: Pergamon. 47±60. Brindley, G. 1987. `Factors affecting task difficulty' in D. Nunan (ed.): Guidelines for The Development of Curriculum Resources. Adelaide National Curriculum Resource Centre. 45±56. Britten, B. and A. Tresser. 1982. `Effects of prior knowledge on the use of cognitive capacity in three complex tasks.' Journal of Verbal Learning and Verbal Behaviour 21: 421±36. Brock, C. 1986. `The effects of referential questions on ESL classroom discourse.' TESOL Quarterly 20/1: 47±59. Brown, G. 1995. Speakers, Listeners and Communication. Cambridge: Cambridge University Press. Brown, G., and G. Yule. 1983. Teaching the Spoken Language. Cambridge: Cambridge University Press. Brown, G., A. Anderson, R. Shillcock, and G. Yule. 1984. Teaching Talk: Strategies for Production and Assessment. Cambridge: Cambridge University Press. Bygate, M. 1996. `Effects of task repetition: Appraising the developing language of learners' in J. Willis and D. Willis (eds.): Challenge and Change in Language Teaching. Oxford: Heinemann. 136±46.

PETER ROBINSON 55

Candlin, C. 1987. `Towards task-based language learning' in C. Candlin and D. Murphy (eds.): Language Learning Tasks. London: Prentice Hall. 5±22. Carrell, P. and T. Wise. 1998. `Prior knowledge, topic interest and L2 reading' Studies in Second Language Acquisition 20. Case, R. 1985. Intellectual Development: Birth to Adulthood. Norwood, NJ: Ablex. Chang, Y. F. 1999. `Discourse topics and interlanguage variation' in P. Robinson (ed.): Representation and Process: Proceedings of the 3rd Pacific Second Language Research Forum, Vol. 1. Tokyo: PacSLRF. 235±41. Chaudron, C. 1988. Second Language Classrooms. New York: Cambridge University Press. Clapham, C. 1996. The Development of IELTS: A Study of the Effect of Background Knowledge on Reading Comprehension. Cambridge: Cambridge University Press. Coughlan, P. and P. Duff. 1994. `Same task, different activities: Analysis of an SLA task from an activity theory perspective' in J. Lantolf and G. Appel (eds.): Vygotskian Approaches to Second Language Research. Norwood, NJ: Ablex. 157±72. Crookes, G. 1989. `Planning and interlanguage variation.' Studies in Second Language Acquisition 11: 183±99. DeKeyser, R. In press. `Automaticity and automatization' in P. Robinson (ed.): Cognition and Second Language Instruction. New York: Cambridge University Press. DoÈrnyei, Z. and A. Malderez. 1997. `Group dynamics and foreign language teaching.' System 25: 65±81. DoÈrnyei, Z. and I. Otto. 1998. `Motivation in action: A process model of L2 motivation.' Thames Valley University Working Papers in Applied Linguistics 4: 43±69. Doughty, C. In press. `Cognitive underpinnings of focus on form' in P. Robinson (ed.): Cognition and Second Language Instruction. New York: Cambridge University Press. Douglas, D. 1998. `Testing methods in contextbased second language research' in L. Bachman and A. Cohen (eds.): Interfaces between Second Language Acquisition and Language Testing. New York: Cambridge University Press. 141±55. Ellis, R. 1987. `Interlanguage variability in narrative discourse: Style shifting and use of the past tense.' Studies in Second Language Acquisition 9: 1±20.

Fleishman, E. A. 1978. `Relating individual differences to the dimensions of human tasks.' Ergonomics 21: 1007±19. Foster, P. and P. Skehan. 1996. `The influence of planning and task type on second language performance.' Studies in Second Language Acquisition 18: 299±324. Givon, T. 1985. `Function, structure, and language acquisition' in D. Slobin (ed.): The Crosslinguistic Study of Language Acquisition: Vol. 1. Hillsdale, NJ: Lawrence Erlbaum. 1008±25. Good, D. A. and B. Butterworth. 1980. `Hesitancy as a conversational resource: Some methodological implications' in H. Dechert and M. Raupach (eds.): Temporal Variables in Speech Production. The Hague: Mouton. Gopher, D. 1992. `Analysis and measurement of mental workload' in G. d'Ydewalle, P. Eelen, and P. Bertelson (eds.): International Perspectives on Psychological Science: Vol. 2, State of the Art. Hillsdale, NJ: Lawrence Erlbaum. 265±91. Hancock, P., G. Williams, and C. Manning. 1995. `Influence of task demand characteristics on workload and performance.' The International Journal of Aviation Psychology. Special Issue on Pilot Workload: Contemporary Issues 5/1: 63±86. Holthouse, J. 1995. Anxiety and Second Language Learning Task Type. Unpublished MA dissertation, University of Queensland, Australia. Horwitz, E. K., M. B. Horwitz, and J. Cope. 1986. `Foreign language classroom anxiety.' Modern Language Journal 70: 125±32. Joseph, J. H. and F. M. Dwyer. 1984. `The effects of prior knowledge, presentation mode, and visual realism on student achievement.' Journal of Experimental Education 52: 110±21. Jourdenais, R. In press. `Cognition, instruction, and protocol analysis' in P. Robinson (ed.): Cognition and Second Language Instruction. New York: Cambridge University Press. Levelt, W. J. M. 1989. Speaking: From Intention to Articulation. Cambridge, MA: MIT Press. Long, M. H. 1983. `Native speaker non native speaker interaction and the negotiation of comprehensible input.' Applied Linguistics 4: 126±41. Long, M. H. 1985. `A role for instruction in

56

TASK COMPLEXITY, TASK DIFFICULTY, AND TASK PRODUCTION

second language acquisition: Task-based language teaching' in K. Hyltenstam and M. Pienemann (eds.): Modelling and Assessing Second Language Acquisition. Clevedon, Avon: Multilingual Matters. 77±99. Long, M. H. 1989. `Task, group, and taskgroup interactions.' University of Hawai'i Working Papers in ESL 8: 1±25. Long, M. H. 1997. Focus on Form in Taskbased Language Learning. Paper presented in the McGraw-Hill Teleconference on approaches to grammar instruction in communicative teaching, California State University, Long Beach, USA. Long, M. H. and G. Crookes. 1992. `Three approaches to task-based syllabus design.' TESOL Quarterly 26: 55±98. Long, M. H. and P. Robinson. 1998. `Focus on form: Theory, research, practice' in C. Doughty and J. Williams (eds.): Focus on Form in Classroom Second Language Acquisition. New York: Cambridge University Press. 15± 41. Loschky, L. and R. Bley-Vroman. 1993. `Grammar and task-based methodology' in G. Crookes and S. Gass (eds.): Tasks in Language Learning: Integrating Theory and Practice. Clevedon, Avon: Multilingual Matters. 123±67. McNamara, T. 1996. Second Language Performance Testing. New York: Longman. Meisel, J. 1987. `Reference to past events and actions in the development of natural second language acquisition' in C. Pfaff (ed.): First and Second Language Acquisition Processes. Rowley, MA: Newbury House. 206±25. Norris, J., J. D. Brown, T. Hudson, and J. Yoshioka. 1998. Developing Second Language Performance Tests. University of Hawaii Second Language Teaching and Curriculum Center Technical Report 19: University of Hawaii Press. Nunan, D. 1989. Designing Tasks for the Communicative Classroom. Cambridge: Cambridge University Press. Plough, I. and Gass, S. 1993. `Interlocutor and task familiarity: Effects on interactional structure' in G. Crookes and S. Gass (eds.): Tasks and Language Learning: Integrating Theory and Practice. Clevedon, Avon: Multilingual Matters. 95±122. Prabhu, N. 1987. Second Language Pedagogy. Oxford: Oxford University Press.

Rahimpour, M. 1997. Task Complexity, Task Condition, and Variation in L2 Oral Discourse. Unpublished Ph.D dissertation, University of Queensland, Australia. Robinson, P. 1995a. `Task complexity and second language narrative discourse.' Language Learning 45: 99±140. Robinson, P. 1995b. `Attention, memory and the ``noticing'' hypothesis.' Language Learning 45: 283±331. Robinson, P. 1996. `Connecting tasks, cognition and syllabus design' in P. Robinson (ed.): Task Complexity and Second Language Syllabus Design: Data-based Studies and Speculations. Brisbane: University of Queensland Working Papers in Applied Linguistics (Special Issue). 1±16. Robinson, P. In preparation. Task Complexity, Second Language Development, and Syllabus Design. Cambridge University Press. Robinson, P. In pressa. `Task complexity, cognitive resources and second language syllabus design: A triadic framework for investigating task influences on SLA' in P. Robinson (ed.): Cognition and Second Language Instruction. New York: Cambridge University Press. Robinson, P. In pressb. `Attention and memory during SLA' in C. Doughty and M. Long (eds.): Handbook of Research in Second Language Acquisition. Oxford: Blackwell. Robinson, P. and J. J. Lim. 1993. Cognitive Load and the Route-marked not-marked Map Task. Unpublished data, University of Hawai'i at Manoa, Department of ESL, Honolulu, USA. Robinson, P. and S. Ross. 1996. `The development of task-based testing in English for Academic Purposes programs.' Applied Linguistics 17: 523±49. Robinson, P., S. Ting, and J. Urwin. 1995. `Investigating second language task complexity.' RELC Journal 25: 62±79. Sachs, J. 1983. `Talking about the there-andthen: The emergence of displaced reference in parent-child discourse' in K. Nelson (ed.), Children's Language Vol. 4. New York: Gardner Press. 1±28. Sanders, A. 1998. Elements of Human Performance. Mahwah, NJ: Lawrence Erlbaum. Sarno, K. and C. Wickens. 1995. `Role of multiple resources in predicting time-sharing efficiency: Evaluation of three workload models in a multiple task setting.' The

PETER ROBINSON 57

International Journal of Aviation Psychology. Special Issue on Pilot Workload: Contemporary Issues 5/1: 107±30. Sawyer, M. and L. Ranta. In press. `Aptitude, individual differences and program design' in P. Robinson (ed.): Cognition and Second Language Instruction. New York: Cambridge University Press. Schmidt, R. 1983. `Input, acculturation and the acquisition of communicative competence' in N. Wolfson and E. Judd (eds.): Sociolinguistics and Second Language Acquisition. Rowley, MA: Newbury House. 137±74. Schmidt, R. 1990. `The role of consciousness in second language learning.' Applied Linguistics 11: 129±58. Selinker, L. and D. Douglas. 1985. `Wrestling with ``context'' in interlanguage theory.' Applied Linguistics 6: 190±204. Skehan, P. 1996. `A framework for the implementation of task-based instruction.' Applied Linguistics 17: 38±62. Skehan, P. 1998. A Cognitive Approach to Language Learning. Oxford: Oxford University Press. Skehan, P. and P. Foster. 1997. `Task type and task processing conditions as influences on foreign language performance.' Language Teaching Research 1: 185±212. Skehan, P. and P. Foster. In press. `Cognition and tasks' in P. Robinson (ed.): Cognition and Second Language Instruction. New York: Cambridge University Press. Snow, R. E. 1994. `Abilities in academic tasks' in R. Sternberg and R. Wagner (eds.): Mind in Context: Interactionist Perspectives on Human Intelligence. New York: Cambridge University Press. 3±37. Spilsbury, G., L. Stankov, and R. Roberts. 1990. `The effects of a task's difficulty on its

correlation with intelligence.' Personality and Individual Differences 11: 1069±77. Sternberg, R. 1977. Intelligence, Information Processing and Analogical Reasoning: The Componential Analysis of Human Abilities. Hillsdale, NJ: Lawrence Erlbaum. Suedfeld, P. and S. Coren. 1992. `Cognitive correlates of conceptual complexity.' Personality and Individual Differences 13: 1193±9. Swain, M. 1995. `Three functions of output in second language learning' in G. Cook and B. Seidlhofer (eds.): Principle and Practice in Applied Linguistics: Studies in Honour of H.G. Widdowson. Oxford: Oxford University Press. 125±44. Tarone, E. and G. Liu. 1995. `Situational context, variation, and second language acquisition theory' in G. Cook and B. Seidlhofer (eds.): Principle and Practice in Applied Linguistics: Studies in Honour of H. G. Widdowson. Oxford: Oxford University Press. 107±24. Tarone, E. and B. Parrish. 1988. `Task-based variation in interlanguage: The case of articles.' Language Learning 38: 21±44. Urwin, J. 1999. Second Language Listening Task Complexity. Unpublished Ph.D dissertation, Monash University, Australia. Wickens, C. 1989. `Attention and skilled performance' in D. Holding (ed.): Human Skills. New York: John Wiley. 71±105. Wickens, C. 1992. Engineering Psychology and Human Performance. 2nd edn New York: Harper Collins. Widdowson, H. G. 1990. Aspects of Language Teaching. Oxford: Oxford University Press. Yule, G. and M. MacDonald. 1990. `Resolving referential conflicts in L2 interaction: The effect of proficiency and interactive role.' Language Learning 40: 539±56.

Applied Linguistics 22/1: 58±78

# Oxford University Press 2001

Interaction in Academic Writing: Learning to Argue with the Reader GEOFF THOMPSON University of Liverpool, UK The view of written texts as embodying interaction between the writer and reader is now well established, and underlies many aspects that may be focused on in the training of novice writers of academic text. In this paper, I argue that interaction can draw on both interactive and interactional resources: interactive resources help to guide the reader through the text, while interactional resources involve the reader collaboratively in the development of the text. I use the concept of the `reader-in-the-text' (Thompson and Thetela 1995) to explore a central form of interactional resource: the inclusion in the text of a voice that is intended to be attributable to the reader. I identify a particular set of discourse contexts in which this happensÐwhere the writer brings in the reader's view in order to contradict itÐand outline the lexico-grammatical features which signal the other voice in those contexts; and I place these in a broader perspective on written text as a stage-managed form of dialogue. The impetus for the study comes from working with novice writers; and I discuss a number of examples where written drafts were improved by exploiting the interactional resources described, and argue for the value of raising students' awareness of these resources.

It has long been accepted that developing the skill of e€ective writing involves, amongst other things, developing an awareness of the audience and an ability to re¯ect and exploit that awareness in the way the text is written (see e.g. Nystrand 1986; Kirsch and Roen 1990; Grabe and Kaplan 1996; Johns 1997). One area directly a€ected by audience awareness is the way in which the text is organized and the organization is signalled. Any text can in principle be seen as a record of a dialogue between writer and reader in which, as Widdowson (1984: 59) argues, `the writer has to conduct his interaction by enacting the roles of both participants'. That is, pro®cient writers attempt to second-guess the kind of information that readers might want or expect to ®nd at each point in the unfolding text, and proceed by anticipating their questions about, or reactions to, what is written. The text is built up as a series of writer responses to these anticipated reactions. Hoey (e.g. 1988) highlights the various kinds of textual clues that writers give to readers as to the way each move in the interaction ®ts inÐthat is, clues as to the reaction that the co-operative reader is assumed to have had. Amongst these clues at the sentence-to-sentence level are conjuncts and conjunctions, such as `therefore' and `so' (both of which can be seen as assuming a question on the lines: `What is the consequence of what you have just told me?'). On a

GEOFF THOMPSON 59

larger scale, writers also exploit predictable text patterns that their readers will ®nd easy to identify, such as Problem±Solution, with their associated lexical signals, as a way of guiding the readers' expectations of how the text is going to develop. However, to explore interaction only in terms of writers' implicit assumptions about the reactions of readers gives an incomplete picture. In this paper I wish to explore a di€erent though complementary perspective that highlights the ways in which writers can bring the underlying dialogue to the surface. This involves them in interacting overtly with their readers (or, more accurately, the `readers' that they themselves enact), by including their questions and reactions in the text and thus assigning to them roles in a stagemanaged form of exchange. The initial impetus for the study came from working with novice writers: it was through trying to pinpoint the source of particular problems in their drafts and to suggest ways of improving their handling of argumentation that I was led to examine the forms of interaction focused on here. In the next part of the paper, therefore, I present examples to illustrate the kinds of problems that arose, to outline the steps in the redrafting process, and to show how making the textual interaction overt resulted in more satisfying versions. On the basis of this, I argue that students training to improve their pro®ciency as writers can bene®t from explicit attention to the ways in which interaction can be performed in text.

PERFORMING INTERACTION IN WRITTEN TEXT In Thompson and Thetela (1995), we distinguish two main types of interaction in written text. The aspects mentioned in the opening paragraph above, which are related to awareness of the audience's likely reactions and needs, may be termed interactive: these primarily involve the management of the ¯ow of information and thus serve to guide readers through the content of the text (in using the term `interactive' in this way, we are following Widdowson 1984 and Hoey 1988). The other type may be termed interactional: these are aspects which aim to involve readers in the argument or ethos of the text. The resources drawn on for this purpose are those which allow writers to conduct more or less overt interaction with their audience, by appearing in the text to comment on and evaluate the content through the use of modality and evaluation (see Hunston and Thompson 2000), and by assigning speech roles to themselves and the readers. For example, an interrogative mood choice typically constructs the role of questioner. This role may be assigned to the readerÐ the question is projected as being asked by the readerÐin which case the writer is most likely to be assigned the complementary role of answerer. In the following extract from an advertisement (taken from Thompson and Thetela 1995: 114), this distribution of roles is signalled explicitly:

60

INTERACTION IN ACADEMIC WRITING

[1] After all, are not all these things exactly what makes a car worth driving? To which we answer: yes In many cases, the assignment of roles is less obviously performed; but in approaching a text from the interactional perspective we constantly focus on whether the source of each statement, question, command, etc. is intended to be the writer or the reader. We also focus on how the complementary role (answering, obeying, etc.) is accommodated in the text. Questions and commands are usually communicatively unsuccessful if they do not elicit the response that they demand; and yet in written text it is clearly not a straightforward matter to include a response from the reader. One way of getting round this problem is to continue the text as if the required response has been given. In the following extract from another advertisement, the question and statement are both addressed by the writer to the reader, but the text is fully coherent only on the assumption that the question is followed by a positive response from the reader: [2] Do you want strength in times of crisis? A reinforced passenger safety cell with a ring of steel and front and rear crumple zones will help. The normal assumption in writing is that the expected response occurs: a `dispreferred second' (Pomerantz 1984) or `discretionary alternative' (Halliday 1994)Ðsuch as answering `no' to the question in [2]Ðis not a live option. In other words, when the initiation comes from the writer, the reader is construed as having provided the expected response. A recipe, for instance, can be read without carrying out the instructions (much less giving a verbal response undertaking to do so), but the text itself is constructed as if each of the commands were successfully obeyed: reference to `the mixture' projects a situation in which the preceding instructions for mixing ingredients have been followed. I have talked in terms of speech roles being assigned to the reader. However, I have indicated above that this is the reader as enacted by the writer: there is clearly no guarantee that the real-world readers will in fact provide the response that the text constructs for them. It is therefore more accurate and more useful to talk of the `reader-in-the-text' (Thompson and Thetela 1995). This term is preferred to other possible labels such as `ideal reader', because it highlights the central importance of evidence from the text. For example, the second advertisement extract above construes a reader-in-the-text who provides the required response, irrespective of how any real-world reader responds, by the way in which the text constructs a `gap' where the response must come. Examination of actual readers' responses to the text can be put to one side, at least provisionally, in the kind of analysis presented in this paper. In principle, it can be taken that writers assume a match, or aim to promote convergence, between reader and

GEOFF THOMPSON 61

reader-in-the-textÐthat at some level readers ®t or accept the roles assigned to them by proxy. Whether this is in fact the case depends on a range of factors beyond the text. It is worth noting that the two aspects of interaction, the interactive and the interactional, are essentially the two sides of the same coin. For example, a question `from the reader', which would be categorized as having primarily an interactional purpose, frequently serves to signal where the text is going next (Tadros 1985), and thus also has an interactive function. Indeed, one way of viewing the relationship, following Widdowson (1984: 61), is that the interactional represents the overt performance in the text of the interactive. Rather than simply moulding the text interactively to ®t the readers, writers may choose at any point to bring their management of the unfolding of the text to the surface and to engage themselves and the readers explicitly in the process: in these cases, the text acts out the organizing interactionally instead of just embodying it. Since writers make assumptions about the questions that might plausibly be asked by the reader and construct the text to provide answers, there is obviously the option of expressing the questions explicitly in the voice of the reader-in-the-text. The reasons why this option might be selected are very varied but typically re¯ect an attempt to involve the reader in some way. This attempt can be seen from two complementary angles. On the one hand, the writer o€ers a token of solidarity by overtly demonstrating understanding of, and concern for, the reader's processing of the text. On the other hand, more manipulatively, the writer spells out the question that the co-operative reader ought to be expecting to be answered at that point and thus encourages the reader to accept the direction the text is taking. For instance, in one of his essays discussing interaction in writing, Widdowson (1984: 71) writes: [3] I have mentioned Halliday's ideational and interpersonal functions. He postulates a third function: the textual. According to Halliday this provides the means whereby language makes links with itself so that individual sentences are fused into texts. This reads perfectly smoothly: it is clear that the third sentence is interactively designed to answer a reader question along the lines `What is the textual function?'. In fact, however, [3] is a version from which I have omitted one sentence: in the original, Widdowson takes the interactional option and includes an explicit question: [3'] I have mentioned Halliday's ideational and interpersonal functions. He postulates a third function: the textual. Where does this come from? According to Halliday it provides the means whereby language makes links with itself so that individual sentences are fused into texts. One of the main reasons why the writer chooses to enact the interaction overtly here is perhaps indicated by the surprised, even faintly querulous, wording of the question. This projects a reader-in-the-text who is already

62

INTERACTION IN ACADEMIC WRITING

dubious about the concept, and prepares the way for a critical evaluation of Halliday's view of the textual function in what follows. The interactional wording thus functions to involve the reader more persuasively in the developing argument. The importance of the interactional perspective is that it highlights the possibility of seeing the text not just as constructed with the readers' needs in mind, but as jointly constructed, with communicative space being left for the readers to contribute to the achievement of the text's goals. The readers' views are politely and collaboratively taken into account; but collaboration is a twoway process, and the readers are therefore encouraged to take part in the interaction and to collaborate back, by accepting, even if only provisionally, the roles, stances and arguments that are attributed to them. Achieving involvement, through a convergence of the reader with the reader-in-the text, is a crucial step in most types of argumentative, persuasive text, including academic papers and assignments;1 and collaboration is a central form of involvement. This perspective clearly accords well with Bakhtin's (1986) emphasis on the dialogic nature of language use: the insight that writers, just as much as speakers, are constantly responding to actual or imagined utterances from others, and that every utterance has `dialogic overtones' (1986: 93). One way in which these overtones can become more overt is by the incorporation of the other utterance in some way in the text. For example, certain genres, such as instructional lea¯ets (Al-Sharief 1996) and direct sales letters (Frank 1989), exploit this option fairly frequently in the form of questions `from the reader' which the writer then answers. In academic written text, interactional resources are less salient, but they still play an essential role, as is increasingly recognized in discourse studies (see, e.g., Webber 1994, on questions; and Swales et al. 1998, on imperatives). In the present paper, I wish to explore one particular way of incorporating the other utterance which has so far received little attention and which appears to be an important resource in academic text as well as other registers: introducing a declarative proposition that is assignable to the reader-in-thetext, in order to argue against it.

ARGUING AGAINST THE READER A sample text It will be useful to start by analysing an example of a text that can be taken as successfully deploying this resource. The following extract is from the introduction to Bex (1996: 1±2), a book on variety in written English. The choice of the introduction is deliberate, since, as noted above, interactional choices tend to be associated with argumentation: in academic texts this is likely to be salient in the introduction, discussion, and conclusions sections. [4] In this book, then, I am attempting partly to show why this variety [of types of texts] exists, and why it is that readers and writers

GEOFF THOMPSON 63

characteristically group certain texts together as being of the `same kind' and, by extension, di€erent from other `kinds'. At ®rst sight, this may seem a simple task. Most of us acknowledge that there is a di€erence between such writings as shopping lists, diaries, letters to friends, job applications, the presentation of CVs, etc. However, we are often inclined to believe that such di€erences arise naturally: that it is the situation which calls for a particular type of text. I shall be arguing that the interrelationship between texts and contexts is much more complex than this. Of course, it is true that particular situations call for di€erent types of texts, but this is because in any given situation we want our linguistic contribution to have speci®c e€ects. The second paragraph here has a Hypothetical±Real pattern (see Winter 1994; Hoey 1983). The HypotheticalÐi.e. the idea that is presented as not to be taken as trueÐis that the writer's goal is `a simple task'. Its hypothetical status is signalled by `at ®rst sight' and `seem'. The RealÐi.e. the point that is to be taken as true for the writerÐis that the interrelationship is `much more complex'; the main signal of the status of this point is `I shall be arguing'. Within the Real, there is also a concessive relation (`it is true that . . . but'). With both the Hypothetical-Real and the concession, the writer takes the option of performing the argument interactionally. In the Hypothetical member, `may' signals that the Senser to whose `®rst sight' this `seems' simple includes the reader rather than referring only to the impressions of the writer or a third person.2 The reason for the presence of `may' is, I would argue, that the writer needs to avoid making categorical assertions about the reader's thoughts, feelings, and beliefs (Al-Sharief 1996). If the Senser is assumed to be the writer himself or a third person, the writer would not be seen as imposing on the reader and could state baldly that this is how the task `seems'; if on the other hand the Senser includes the readers, it would constitute a face-threatening act, or FTA (Brown and Levinson 1987), to claim to know categorically what they think, especially when the writer is about to say that what they think is wrong. The sense of interaction is reinforced by the use of inclusive `most of us' and `we' in the following sentences (which are part of the Hypothetical), both as Sensers in mental processes (`acknowledge', `believe'): the writer is simultaneously projecting and sharing a position for the reader. Including oneself in the belief is an alternative strategy for mitigating the FTAÐthis comes out clearly if we compare the e€ect of replacing `us/we' with `you' without adding any modalization. In the concessive relation also, the use of `Of course, it is true' signals that the writer assumes that the conceded proposition is one which the reader might want to express (Thompson and Zhou 2000). I said above that the writer has chosen to present the clause relations in interactional terms. It may help to bring this out if we consider a version with the signals commented on above removed:

64

INTERACTION IN ACADEMIC WRITING

[4']

This seems a simple task. Most people acknowledge that there is a di€erence between such writings as shopping lists, [ . . . ] etc. However, they are often inclined to believe that such di€erences arise naturally: that it is the situation which calls for a particular type of text. I shall be arguing that the interrelationship between texts and contexts is much more complex than this. Particular situations call for di€erent types of texts, but this is because in any given situation we want our linguistic contribution to have speci®c e€ects.

It is particularly noticeable that the dialogic overtones in the concession± assertion pair in the last sentence have all but disappeared. In Thompson and Zhou (2000) we argue that a number of kinds of argumentation can be presented either logically or interpersonally. In the case of concession, the logical presentation (as in [4'] ) construes the argument as: `A and B are both true; A is not entirely compatible with B; A does not invalidate B (and B has greater validity)', whereas the interpersonal presentation (as in [4] ) construes it as: `You may think A; I accept that A is true; for me, A does not rule out B (and B is the point I want to make)'. Similarly, although the possibility of the reader identifying with `most people' who hold the Hypothetical belief still remains, it is much weaker than in the original. This suggests that we can express the e€ect of `may' and `Of course, it is true' (and, allowing for some di€erences, of `most of us acknowledge' and `we . . . believe') in slightly di€erent terms from how it has been put above: these signals can be seen as assigning propositions to the reader-in-the-text. Not all Hypothetical members, by any means, are attributable to the readerin-the-text; but it is an inherent part of the pattern that they need a sourceÐ someone who is projected as (potentially or actually) entertaining the belief which the Real member then contradicts, quali®es, or con®rms. This someone may be the writer, the reader, other people, or any combination of the three. The source is usually signalled in some way; and, in academic discourse, a modalized unattributed verbal or mental process introducing a Hypothetical member (the single most frequent exponent in my data is `it might be argued') will normally be taken as signalling that we are hearing the voice of the reader-in-the-textÐin other words, that the source could plausibly be the reader. This is something that the writer has to judge fairly skilfully: individual readers need not accept the Hypothetical idea as representing their own views, but must agree with the writer that some at least of their peers might well accept it. One aspect of the ®eld that novice writers have to become familiar with is the kinds of ideas that are likely to be seen as plausible in this sense, even if they have not been explicitly advanced. In the case of concession, the signals are of a di€erent kind (see below) but the same points about source and plausibility are also true.

GEOFF THOMPSON 65

Contexts and signals of dialogic text From the discussion of example [4] above, and from other analyses of similar data, it is clear that two of the main discourse contexts in which the reader-inthe-text is given a voice in academic text are Hypothetical±Real, and concession.3 These are closely related, in that both have a ®rst member that is normally quali®ed or contradicted in the second member, and a second member that is normally taken to be the one to which the writer attributes greater validity.4 Both raise the question of who is the source of the hypothesis or the conceded proposition. Both also raise the question of why the writer feels it necessary to mention the ®rst member at all. If the source is assumed to be the reader-in-the-text, the reasons for mentioning it relate to the factors discussed earlier, of collaboration, involvement, and persuasion. By stating overtly in the text an idea that is projected onto the reader, the writer is publicly performing awareness of the reader, and, at the same time, inviting the reader to share in the construction of the argument and, ideally, to accept more readily that the ®rst memberÐthe reader's own potential opinionÐneeds to be quali®ed or rejected (or explicitly con®rmed). As noted above, not all cases of Hypothetical±Real and concession involve the voice of the reader-in-the-text. The lexico-grammatical signals which, in certain con®gurations, trigger this reading of the two types of relations appear to be the following: . low-value subjective modalization (modal verbs expressing possibility). These raise the question of why the writer is uncertainÐin particular, whether the uncertainty relates to the proposition itself or to the attribution of the proposition to a particular source. One context in which it is rhetorically politic to display uncertainty about attribution is when the source is the reader-in-the-text: what are in e€ect the writer's claims about the reader's opinions clearly need to be presented with care. This use of modalization is typically associated with unattributed mental and verbal processes such as `the argument may be advanced' or `it might be expected that . . . 'Ðin most such cases, the presence of modalization indicates that the proposition is to be attributed to the reader-in-the-text. It therefore occurs mainly in Hypothetical±Real contexts (though see also example [5] below).5 . high-value objective modalization (expressions of certainty: modal disjuncts such as `of course', `certainly'; modal comment clauses such as `it is true/ certain that'). This raises the question of why the writer feels it necessary to comment on the validity of the propositionÐas Halliday (1994: 89) points out, `you only say you are certain when you are not'. One potential reason is that the proposition is projected as coming from the reader-in-the-text and thus needs to be accepted/conceded before a counter-assertion is made, in order to mitigate the FTA. Modalization of this kind is used when it is the writer's quali®ed acceptance of an idea that is highlighted, rather than the attribution of that idea to the reader-in-the-text; there is therefore no need

66

INTERACTION IN ACADEMIC WRITING

to express cautious uncertainty. This tactic is associated with concessionÐas in the last sentence of [4]Ðrather than Hypothetical±Real. . unattributed mental and verbal processes (normally with modalizationÐ e.g.: `it might be thought/argued' [by whom?], `it may appear' [to whom?] ). These raise the question of who the Sayer or Senser isÐthat is, the source of the ideas or impressions. One reason for leaving the attribution open is again to mitigate the potential FTA of attributing an idea to the reader via the reader-in-the-text. Such processes are mainly associated with Hypothetical±Real (as in `this may seem' in [4] ). Hypothetical and conceded propositions whose source is the reader-in-the text, as exempli®ed in [4], are, of course, not the only interactional resource available to writers of academic text. How do they ®t in with other more familiar patterns such as question and answer? I have discussed those patterns above in terms of exchanges, with the writer and reader-in-the-text each being assigned complementary roles in the exchange. One way of looking at the kinds of Hypothetical±Real and concession relations investigated in this paper is to see them in terms of statements initiated by the reader-in-the-text in order for them to be responded to by the writer. Statements in written text are normally initiated by the writer, and the expected response is acceptance by the reader; but in these cases the roles are reversed and the response is contradiction or quali®cation. This is therefore a case in which the discretionary alternative or dispreferred second is still liveÐindeed, obligatory. Overall, it can be argued that there are three main options conventionally open to academic writers to perform overt dialogic interaction with their readers. One is commands initiated by the writer, which the reader-in-thetext obeys. These might be thought to be of marginal importance, but Swales et al. (1998), in an investigation of imperatives in academic articles, show them to be comparatively frequent and used in complex ways. They argue that the phenomenon deserves more attention both in text analysis and in the training of novice writersÐan argument which chimes in with the views advanced in the present paper.6 Another interactional resource is questions. These are occasionally projected unequivocally onto the reader, as in the following extract from Widdowson (1984: 238)Ðnote the use of `might' in assigning these utterances to the reader: [5] Now what, you might ask, has all this to do with course design? However, in the texts I have analysed this kind of question occurs very infrequently. More often questions are best seen as assignable to the readerin-the-text, `held up' for consideration by the writer on the reader's behalf, as in example [3] above (see S. Thompson 1997: 162±3, and Thompson and Thetela 1995: 124, for further discussion). The other main interactional choice in academic writing appears to be the one focused on in this paper: statements from the reader-in-the-text brought in to be contradicted.7

GEOFF THOMPSON 67

LEARNING TO BRING IN THE READER'S VOICE The preceding section has provided an overview of some of the main resources by which the underlying dialogic nature of argumentation can be brought to the surface in written academic discourse. As mentioned in the opening paragraphs, the impetus for this study came from working with postgraduate students on improving their drafts of written work; and I want at this point to turn to the practical application of the ideas, in particular the interactional use of Hypothetical±Real and concession relations. In the following, I shall present a number of representative examples taken from students' work. I shall discuss the problematic areas, and outline a possible procedure for sensitising students to aspects of interaction in their texts.

Examples of student drafts In helping students to improve their drafts of assignments and theses, much attention is inevitably focused on content and organization. However, I have found that it is often necessary to deal with the way in which interaction with the reader is performed: occasions regularly occur when transitions can be more e€ectively handled and the text can be made smoother to read by a judicious use of interactional resources. The following extracts illustrate this in the essays of student writers, both native speakers and non-native speakers, at various levels of interactional pro®ciency. In a few cases, the drafts show little or no sign that the writers are aware that such resources can be drawn on. Example [6] shows the student still at the stage of trying to construct an argument that could accommodate evidence of her reading. The extract comes from the introduction to an essay on vocabulary teaching in an ESP context: [6] The aim of this work is to examine ways of improving methods for teaching vocabulary to students of ESP on the Tourism course at the University of [NSD]. The reason for my interest in this topic is that as in any ESP ®eld the future professionals are going to need some specialized words of vocabulary of their ®eld which will be relevant to their future work. I want to make reference to what Taylor (1990: 1) says: In order to live in the world, we must name it. Names are essential for the construction of reality . . .

There is clear signalling of where the text is going in the ®rst two sentences, but the introduction of the quotation from Taylor is extremely clumsy: in interactive terms, it does not seem to answer any plausible question that the reader might want to ask after reading the second sentence. Leaving aside the issue of whether the quote is particularly apposite here or not, there were a number of ways in which its introduction into the text could be improved; but in discussing the draft it became clear that the writer saw it as justifying her focus on vocabulary, and we built on that. Justi®cation has a strong inherent

68

INTERACTION IN ACADEMIC WRITING

element of dialogue: you justify your decisions in response to potential criticisms. The revision made this interactional aspect explicit: [6'] The aim of this work is [ . . . ] The reason for my interest in this topic is [ . . . ] It might be argued that there has typically been too great a focus on vocabulary in ESP at the expense of other aspects of language; but, as Taylor (1990: 1) says: In order to live in the world, we must name it. Names are essential for the construction of reality . . .

The basic pattern introduced here is Hypothetical±Real: the objection is construed simultaneously as both a possible opinion and a mistaken one; and the quotation now serves to express the Real. The original version presented one side of an embryonic internal dialogue: the writer was in e€ect working with a speech-based model, in which she was responsible only for producing her own utterances (`I want'); the revision re¯ects a change towards a writing-based model in which she performs both roles in the interaction. However, as well as such one-sided drafts, it is equally common to ®nd that student writers have a sense of the importance of interacting with the reader but are unsure how to do so in a conventionally accepted way.8 Example [7] is a simple one which shows the writer attempting to allow explicitly for his readers' expected reactions: [7] Literally translated `Landeskunde' means `knowledge about the country'. This is misleading in that geographical notions are expected. The reason for the unease I felt on reading this draft was the unattributed mental process `expected'. It was clear that the writer, a native speaker of German, was thinking of his English readership; but he was making a categorical statement about their presumed state of knowledge, which I felt needed to be softened through modalization. [7'] Literally translated `Landeskunde' means `knowledge about the country'. This is misleading in that geographical notions might be expected. Example [8] shows the writers of a joint assignment, on the implications of using the Communicative Approach in di€erent cultures, aiming to construct a sense of dialogic interaction, in this case through the use of questions: [8] The next section of this paper will focus on the main question of this assignment: Why have Hong Kong and Vietnam, with their very distinctive Asian cultures, adopted the CA? Is this approach widely and willingly accepted by the teachers and students? Did they feel culture shock when this approach was ®rst introduced to their educational systems in the 1980s? This does its job fairly well; but the reasons why it was worth asking these questionsÐwhich had ®gured largely in initial discussions at which the topic

GEOFF THOMPSON 69

was agreed onÐwere left unstated. Bringing them to the surface took two steps. The ®rst addressed interactive concerns, spelling out the basic grounds for questioning the adoption of the CA: [8'] Why have Hong Kong and Vietnam, with their very distinctive Asian cultures, adopted a teaching approach, the CA, which is so strongly rooted in BANA (British, Australian, North American) culture? The second step addressed interactional concerns, projecting the ®rst question as a problem potentially voiced by the reader: [8''] The next section of this paper will focus on the main question of this assignment, which relates to the cultural implications of the CA. It might be expected that Hong Kong and Vietnam, with their very distinctive Asian cultures, would ®nd it dicult to adopt a teaching approach, the CA, which is so strongly rooted in BANA (British, Australian, North American) culture. Is this approach in fact widely and willingly accepted by the teachers and students? Did they feel culture shock when this approach was ®rst introduced to their educational systems in the 1980s? The remaining questions are projected as ways of deciding whether the problem raised by the reader-in-the-text is a real one. The essay is now organized in terms of a Hypothetical for which the writers will move towards providing the Real in the Conclusion; and the reader is drawn into the exploration by being invited to accept that the Hypothetical represents a plausible opinion to hold. This second step is certainly less essential than the ®rst in clarifying the context of the discussion; but it makes the text rhetorically more e€ective. The same is true of example [9], the conclusion to the evaluation of an EFL textbook in terms of how far it uses a task-based approach. This was already successful in constructing a fairly complex argument and in deploying interactive resources well. In this case, the discussion of possible revisions was designed to help an already pro®cient writer to move a step further in her command of argumentation. [9] This task is goal-orientated, having a clear purpose and endpoint, and it involves a real outcomeÐthe satisfaction of solving the puzzle. [ . . . ] Given that the learners' attention is wholly on the resolution of the puzzle, the task is meaning-focused and it seems very likely to motivate the learners. However, the balance of each unit is too clearly weighted in favour of structural exercises. Furthermore, not all of the form-focused exercises have any relationship with the task. Thus, in unit 4, only the structures connected to the function of `stating location' have a bearing on the task. Since this was a key point in the summing-upÐthe writer's overall conclusion was that the tasks in the EFL textbooks she examined were mainly windowdressingÐI felt that it would be useful both to prepare the reader more fully

70

INTERACTION IN ACADEMIC WRITING

for the switch at `However' and to recapitulate the main issues. One way of doing this was to use the voice of the reader who, having read the analysis up to that point, might plausibly be assumed to have formed a hypothesis about the extent to which the materials are task-basedÐthe hypothesis that was about to be contradicted: [9'] This task could be said to be goal-orientated, having a clear purpose and endpoint, and it involves a real outcomeÐthe satisfaction of solving the puzzle. [ . . . ] Given that the learners' attention is wholly on the resolution of the puzzle, the task is meaning-focused and it seems very likely to motivate the learners. All in all, the task itself would seem to be very e€ective. It might appear then that [TSD] has successfully incorporated tasks into the course, using the controlling exercises as pre-communicative preparation for the task. Unfortunately, this is not the case. The balance of each unit is too clearly weighted in favour of structural exercises. Furthermore, . . . One reason why this revision is an improvement, I believe, is that it projects onto the reader the process of discovery and reasoning that the writer herself went through. One fairly strong constraint in academic text is that writers are not usually permitted to write in `real time': they are expected to write as if they already had the complete text in front of them at every point, and to remove traces of the process they go through in deciding what to include, what line of reasoning to follow, and so on. However, it is permissible to project the reader-in-the-text as going through the process of discovery, as moving from relative ignorance or error to enlightenment. This allows the writer to create a sense of an unfolding process without losing the overall sense of the text being fully mapped out in advance. So far, I have illustrated the revising of drafts with writers at di€erent levels of pro®ciency, but each case has involved mainly a process of bringing underlying interaction to the surface at a speci®c point in the text. The ®nal example that I wish to discuss presented a rather more complex problem; and it also allows me to exemplify brie¯y a method of training students in performing interaction in writing beyond working with them individually on isolated parts of their drafts. Example [10] is problematic in a number of ways, but the key diculty is that the writer gives unclear signals as to who is to be seen as responsible for which propositions: in interactional terms, the text is more like a hubbub than a dialogue. The sentences are numbered for ease of reference, and I have underlined the signals which are related to the issue of voice. [10]

[S1] It is true that di€erent texts in di€erent genres have a distinctive discourse organization which matches the purpose for which they are read and the requirements of their writing. [S2] I o€er the following examples of di€erent kinds of texts and their respective (supposed) purpose of reading:

GEOFF THOMPSON 71

text purpose catalogues [ . . . ] get information/to buy [ . . . ] [S3] The assumption might seem to be that a certain kind of general organization of the text suits better its direct purpose in being read. [S4] By general organization I mean the linguistic features that commonly characterize that particular genre, and by direct reading purpose I mean what the text was supposed to be written for. [S5] On the one hand, the analysis of the organization of a `written discourse' is rather emphasized in the ways in which the surface of the discourse contains sucient clues for the reader to perceive its organization. [S6] On the other hand, the reading purpose of the `audience' or the `readers' may appear to vary. One major diculty with this text is confusion over which viewpoint the writer is arguing for and which arguments are intended to be seen as coming from the reader-in-the-text or from elsewhere. In discussion, it emerged that the writer mainly had the module tutor in mind as his reader. The ideas about text and reading purpose were those which had been introduced in the module, and the writer felt that he should avoid a possible charge of simply repeating the input by showing critical distanceÐeven though he essentially agreed with the ideas. We went through the text together sentence by sentence, and I asked him to decide whether each proposition (and in some cases an individual term) was meant to be seen as him speaking, and which could be assigned to the reader, or a third person. The results are shown in Table 1, and indicate that most of the signals of another voice in the original version are misleading: it is only in the last sentence that the writer wishes to distance himself as a preparation for re-arming the validity of the general link between text organization and reading purpose. On the basis of this, we agreed on the following revised version: [10'] Di€erent texts in di€erent genres have a distinctive discourse organization which matches the purpose for which they are read and the requirements of their writing. I o€er the following examples of di€erent kinds of texts and the possible purposes for reading them: text purpose catalogues [ . . . ] get information/to buy [ . . . ] The assumption is that a certain kind of general organization of the text suits better its direct purpose in being read. By general organization I mean the linguistic features that commonly characterize that particular genre, and by direct reading purpose I mean what the text was written for. Thus the analysis of the organization of a written discourse can highlight the ways in which the surface of the discourse contains sucient clues for the reader to perceive its organization. It is certainly possible that the reading purpose of the audience or the readers may vary. However, . . .

72

INTERACTION IN ACADEMIC WRITING

Table 1: Sources of propositions in example [10] actual reading S 1 Who believes that di€erent texts have a distinctive discourse organization? 2 Who supposes these purposes of reading? 3 Who assumes that text organization suits the purpose of reading? 4 Who supposed the purpose for writing the text? 5 Who is saying `written discourse'? 5 Who is emphasizing the analysis (or the organization)? 6 Who is saying `audience' and `readers'? 6 Who is under the impression that the reading purpose varies?

writer

reader

intended reading other

3

3

?

reader

3

?

?

writer

3 3

3

3

?

3

?

3

?

3 3

Although problems still remain, some of the other areas of confusion became easier to deal with in the process, and have also been revised in this version. It is worth noting that this draft, and others like it that I have collected, lend themselves well to discussion using the type of grid shown in Table 1, and can be exploited as the basis for awareness training with groups working on academic writing. One useful activity is to ask students to produce texts which correspond to di€erent versions of the grid. Producing these versions highlights the fact that the linguistic signals involved in constructing a reader-in-the-text are actually fairly restricted and easily learnt (this is especially useful for students whose mother tongue is not English). It makes them aware of the subconscious argumentation that may be going on in their minds as they write, of which they are often expressing only their own side. It can also lead on to discussion of when it might be appropriate to draw on these resources, and when it might notÐwhether it is more e€ective at a particular point in the argument to leave the other side submerged or to bring it to the surface.

GEOFF THOMPSON 73

IMPLICATIONS FOR TRAINING In terms of the distinction that I am using, the main emphasis in the training of academic writing skills is normally on interactive signals (see, e.g., Grabe and Kaplan 1996; Waters and Waters 1995). This bias towards interactive aspects can be justi®ed, in that interactional signals are typically less frequent and less overt in academic text. Nevertheless, pedagogically-inspired research into metadiscourse (e.g. Vande Kopple 1985; Crismore et al. 1993; Stainton 1996), has consistently shown that both aspects need to be taken into account. The view of what constitutes metadiscourse varies from study to study, but it generally includes categories which have an `interpersonal function' (Crismore and Farnsworth 1990: 122). The most salient amongst these categories tend to be those involving markers of modality and attitudeÐ that is, of the writer intervening to comment on the content of the text. This overlaps with the extensive literature on hedging and evaluation (e.g. Myers 1989; Hunston 1994; Hyland 1998), and this aspect of interactional signalling has therefore become increasingly well covered in recent academic writing courses (e.g. Swales and Feak 1994). However, Crismore et al. (1993), looking speci®cally at persuasive writing by students, have a category of `commentary', which `allow[s] authors to draw readers into an implicit dialogue' (Crismore and Farnsworth 1990: 124), and which includes the most overtly interactional resources of imperatives and questions.

Awareness-raising of interactional choices It would therefore seem useful in training novice writers to complement attention to interactive aspects with exploration of interactional aspects, and in particular of the issue of dialogic interaction in writing. Admittedly, the problems discussed above imply a fairly high level of sophistication on the part of the writers, and it could even be argued that the suggested revisions represent relatively optional icing on the cake (though this would certainly not be true of example [10] ). However, there are a number of reasons why I believe that it is valuable to focus on sensitizing students to the dialogic perspective on academic written text. As noted earlier, the skill of judging the most appropriate way of conducting interaction with the audience is one that can be seen very clearly in operation in pro®cient writers: the choice between performing argumentation interpersonally or logically at each stage is an important decision which can have a marked e€ect on the overall tone of the text. However, the skill does not come automatically, and conscious attention to the skill does seem to help novice writers acquire it more rapidly. In addition, at advanced levels, particularly with students, such as those on postgraduate courses, who are being expected to act at least temporarily as members of the academic discourse community, it can be dicult to give concrete advice on how to improve their writing that does not appear to largely reiterate what they have been told before.

74

INTERACTION IN ACADEMIC WRITING

Discussing text in terms of whose voice is being heard (and how we know) o€ers a fresh slant on writing and gives students a sense of a new area that they can aim to incorporate into their developing mastery of academic conventions. The general notion of explicitly allowing for the reader's views to be incorporated, and deciding whether a proposition is intended to be your own or the reader's (or someone else's) seems to appeal to students and to be readily understood. In practical terms, the procedures mentioned aboveÐ picking up on individual points in drafts, or using the grid of sources for propositions as a way of examining more extensive stretches of textÐare simple but e€ective in prompting investigation of the voices in the text. It should also be borne in mind that there is no reason why these issues cannot be raised with students whose command of writing and/or English is not as advanced: they would ®t in very well, for example, with the genre-based approach to the teaching of writing in schools pioneered by Martin (e.g. 1985) and his colleagues. At the same time, there are of course notes of caution that need to be sounded. The writing of assignments and dissertations is in some respects a special case because the main reader is actually known: the student writer is likely to have a fairly clear idea of the views of the tutor who will read the work to assess it. The writer may feel understandably uncomfortable about challenging those views openly (as noted above, this was an important source of the problems in example [10] ). However, this can also be turned to advantage if the tutor is explicitly seen as representing the wider academic community: the personal interaction of supervisions and comments on drafts, in which possible reactions are expressed or alternatives raised, can become the stimulus for constructing interaction in the text. This does mean, though, that students may need to be made aware that the dialogue in formal written text is not natural but stage-managed, and that there are relatively ®rmly established conventions for constructing the dialogue. These conventions are di€erent from those in face-to-face interaction; and for non-native speakers the conventions might be di€erent from those that they may have internalized in writing in their mother tongue. For instance, informal evidenceÐsuch as example [8] aboveÐsuggests that Chinese writers writing in English make more frequent use of questions than British writers, and tend not to exploit other interactional options (cf. Connor 1996). In addition, even within argumentation, performing interaction is only one possible way of handling the communicative problem of taking your reader convincingly through your text: it aims to draw the reader into the process of constructing the argument, but this may not always be desirableÐfor example, repeated use within an essay will almost certainly become counterproductive. It seems likely that e€ective writing strikes a balance between more monologic `logical' argumentation and the more dialogic collaborative kind (see also Thompson and Zhou 2000); and students respond well to exploration of how this balance can be achieved.

GEOFF THOMPSON 75

CONCLUSION In this paper, I have set out to describe a relatively unexplored area in the way writers of academic texts interact with their readers. In Thompson and Thetela (1995) we drew on the concept of the reader-in-the-text and the distinction between interactive and interactional resources in exploring how interaction works in advertisementsÐopenly persuasive texts which typically exploit many of the features of spoken registers. What I have set out to demonstrate here is that the same approach is equally illuminating when applied to written text types which are much less overtly dialogic. It provides a robust framework which brings together a number of features, all related to a perspective on written text as a series of exchanges in an interaction conducted half by proxy, which have tended to be looked at separately, if at all. I have focused on the way in which clause relations of Hypothetical± Real and concession are deployed in this interaction because, in comparison with questions and commands, they represent a less obvious form of exchange which has so far had little attention and yet which my research suggests is one of the more commonly used interactional resources in academic writing. I have restricted the study to writing in Applied Linguistics; one obvious way of extending the research, therefore, is to look at the frequency, distribution, and realization of this resource across di€erent disciplinesÐas, for example, Swales et al. (1998) have done for imperatives. On a broader scale, Widdowson (1984: 62) argues that genres could be characterized `by reference to the degree to which the discourse is textualized' (or, in the terms employed here, the extent to which writers draw on interactional resources). It would therefore also be worth exploring how statements from the reader-in-the-text are deployed in di€erent genres, both academic and non-academic. The paper also re¯ects an explicit attempt to combine discourse analysis with consideration of the training of writing skills, and to show that the second is inextricably linked with the ®rst. The original impetus for the study was working on drafts of written work produced by my students; and the ®ndings have fed directly back into helping them with the revising process and from there into techniques for more general training. There is no space here to report on the implementation of these techniques; but the approach rearms the basic principle underlying much recent work in English for Academic Purposes (and language teaching in general), that novice writers bene®t not just from process-oriented practice in producing texts but from guided scrutiny of how texts work (see, e.g., Swales 1990, Martin 1985). Sustained exposure to texts and discovery through trial and error clearly have a crucial role to play in building up a sense of what sounds appropriate and what works; but rather than simply leaving this to emerge by a kind of osmosis, it is widely accepted that intervention in the form of language awareness activitiesÐconscious examination of what may be called the `discourse value' of particular lexico-grammatical choicesÐcan strengthen

76

INTERACTION IN ACADEMIC WRITING

and speed up the acquisition of pro®ciency. This involves encouraging students to take on the role of discourse analysts; but in order to do this eciently, teachers themselves need a clearer picture of the areas which repay investigation. The perspective on written text as partly enacted dialogue illuminates speci®c features that would otherwise risk being overlooked. In addition, it provides a novel `handle' on text which can be easily understood and applied, and which contributes in a stimulating way to the deconstruction of the mystique of e€ective writing that is an essential step in progressing from novice to initiate. (Revised version received May 2000)

ACKNOWLEDGEMENT I would like to thank three anonymous Applied Linguistics reviewers for their helpful comments on an earlier draft of this paper.

NOTES 1 On academic discourse as essentially persuasive, see e.g. Hunston (1994). 2 The terminology used here to discuss processes and the participants involved, such as `Senser', is that developed by Halliday in his account of transitivity (1994, ch. 5). 3 A di€erent phenomenon which can bring in the voice of the reader-in-the-text is negation. It is generally accepted that negatives typically deny a positive proposition that is somehow `on the table'Ðoften because it is assumed that someone else, potentially the reader, believes it (Jordan 1998). Unfortunately, there is no space here to explore this area of interaction. 4 The main di€erence is that in concession the validity of the conceded proposition is not in question (as Winter (1994), points out, it is the consequences that might be expected which are implicitly denied by the asserted proposition). In Hypothetical±Real, on the other hand, it is the validity of the Hypothetical which is presented as open to questionÐeven if the Real actually con®rms its validity (`It might be thought . . . And indeed this is so'). In the particular cases I am focusing on, this is typically re¯ected in the type of signal: concessions are tagged as accepted by the writer, whereas hypotheses are often attributed to the reader-in-the-text through reporting structuresÐcompare:

Of course, it is true that particular situations call for di€erent types of texts, but . . . (concession) It might be argued that particular situations call for di€erent types of texts, but . . . (Hypothetical±Real) 5 There are cases where `may' in the ®rst member of a concession±assertion pair functions by itself as a signal of attribution to the reader-in-the-text. This typically occurs in more informal texts, though I have come across a few examples in academic writing such as the following from a student essay: The TOEFL test may claim eciency and scoring reliability; however, in terms of validity, according to my investigation, it is doubtful whether it re¯ects one's real linguistic competence. 6 Swales et al. (1998) focus only on imperatives. For a full picture of commands in academic text it would be necessary to take account also of modulated statements functioning as writer commands (`It should be borne in mind that . . . '). 7 To give an approximate sense of the relative frequency of these interactional options, a survey of 3,752 sentences from the introduction, discussion, and conclusions sections of a range of academic articles and books in Applied Linguistics showed the following distribution patterns: statements attributable

GEOFF THOMPSON 77

to the reader-in-the text occurred on average every 29 sentences; questions `held up' by the writer every 38 sentences; and commands from the writer every 94 sentences. Questions clearly assigned to the reader-in-the-text occurred only every 375 sentences. This survey was merely exploratory, and a number of problematic cases were left out of the calculations to simplify matters; but, rough as they are, the results suggest, in line with the ®ndings of Swales et al. (1998), that interactional performance is frequent enough to be worth including as a topic in the training of novice writers. (Interestingly, the ®gures reported in Chang and Swales (1999: 72), for three academic disciplines show a much higher frequency of imperatives than of questions;

this is true even if the results for the Applied Linguistics texts are looked at in isolation. However, they also support the point that these two interactional features are fairly widely used.) 8 It has been convincingly argued that we need to be wary of simply demanding that novice writers conform to the dominant conventions (see e.g. Pennycook 1995; Zamel and Spack 1998). At the same time, we have to take into account the fact that written work of the type that I am focusing on is typically judged according to these conventions. For a balanced discussion of the issues that comes down on the side of awareness raising rather than prescription or rejection of the conventions, see Johns (1997: 51±70).

REFERENCES Al-Sharief, S. 1996. `Interaction in written discourse: The choices in Mood, reference and modality in medical leaflets'. Unpublished MA dissertation, University of Liverpool. Bakhtin, M. M. 1986. Speech Genres and Other Late Essays (tr. V. W. McGee, ed. C. Emerson and M. Holquist). Austin, TX: University of Texas Press. Bex, T. 1996. Variety in Written English. London: Routledge. Brown, P. and S. Levinson. 1987. Politeness: Some Universals in Language Usage. Cambridge: Cambridge University Press. Chang, Y.-Y. and J. Swales. 1999. `Informal elements in English academic writing' in C. N. Candlin and K. Hyland (eds.): Writing: Texts, Processes and Practices. London and New York: Longman. Connor, U. 1996. Contrastive Rhetoric: Crosscultural Aspects of Second Language Writing. Cambridge: Cambridge University Press. Coulthard, M. (ed.) 1994. Advances in Written Text Analysis. London: Routledge. Crismore, A. and R. Farnsworth. 1990. `Metadiscourse in popular and professional science discourse' in W. Nash (ed.): The Writing Scholar. London: Sage. Crismore, A., R. Markkanen, and M. S. Steffensen. 1993. `Metadiscourse in persuasive writing: A study of texts written by

American and Finnish university students.' Written Communication 10/1: 39±71. Frank, J. 1989. `On conversational involvement by mail: The use of questions in direct sales letters.' Text 9: 231±59. Grabe, W. and R. B. Kaplan. 1996. Theory and Practice of Writing. London and New York: Longman. Halliday, M. A. K. 1994. An Introduction to Functional Grammar 2nd edn. London: Edward Arnold. Hoey, M. P. 1983. On the Surface of Discourse. London: George Allen and Unwin. Reprinted as Reprints in Systemic Linguistics No. 3, University of Nottingham 1991. Hoey, M. P. 1988. `Writing to meet the reader's needs: Text patterning and reading strategies.' Trondheim Papers in Applied Linguistics IV: 51±73. Hunston, S. 1994. `Evaluation and organization in a sample of written academic discourse' in M. Coulthard (ed.): Advances in Written Text Analysis. London: Routledge. Hunston, S. and G. Thompson. (eds.) 2000. Evaluation in Text: Authorial Stance and the Construction of Discourse. Oxford: Oxford University Press. Hyland, K. 1998. Hedging in Scientific Research Articles. Amsterdam and New York: Benjamins. Johns, A. M. 1997. Text, Role and Context:

78

INTERACTION IN ACADEMIC WRITING

Developing Academic Literacies. Cambridge: Cambridge University Press. Jordan, M. 1998. `The power of negation in English: Text, context and relevance.' Journal of Pragmatics 29: 705±52. Kirsch, G. and D. Roen. (eds.). 1990. A Sense of Audience in Written Communication. London and Newbury Park, CA: Sage. Martin, J. R. 1985. Factual Writing: Exploring and Challenging Social Reality. Geelong, Vic.: Deakin University Press (republished by Oxford University Press 1989). Myers, G. 1989. `The pragmatics of politeness in scientific articles.' Applied Linguistics 10: 1± 35. Nystrand, M. 1986. The Structure of Written Communication: Studies in Reciprocity between Writers and Readers. Orlando: Academic Press. Pennycook, A. 1995. `English in the world/The world in English' in J. W. Tollefsen (ed.): Power and Inequality in Language Education. Cambridge: Cambridge University Press. Pomerantz, A. 1984. `Agreeing and disagreeing with assessments: Some features of preferred/dispreferred turn shapes' in J. Atkinson and J. Heritage (eds.): Structure of Social Action. Cambridge: Cambridge University Press. Stainton, C. 1996. Metadiscourse: The Rhetorical Plane of Text. Nottingham Working Papers No. 2, University of Nottingham. Swales, J. 1990. Genre Analysis: English in Academic and Research Settings. Cambridge: Cambridge University Press. Swales, J. and C. Feak. 1994. Academic Writing for Graduate Students: Essential Tasks and Skills. Ann Arbor: University of Michigan Press. Swales, J., U. K. Ahmad, Y.-Y. Chang, D. Chavez, D. F. Dressen, and R. Seymour.

1998. `Consider this: the role of imperatives in scholarly writing.' Applied Linguistics 19: 97±121. Tadros, A. 1985. Prediction in Text. ELR Monograph No. 10, Birmingham University. Thompson, G. and P. Thetela. 1995. `The sound of one hand clapping: The management of interaction in written discourse.' Text 15/1: 103±27. Thompson, G. and J. L. Zhou. 2000. `Evaluation and organization in text: The structuring role of evaluative disjuncts' in S. Hunston and G. Thompson (eds.): Evaluation in Text: Authorial Stance and the Construction of Discourse. Oxford: Oxford University Press. Thompson, S. 1997 `Presenting research: A study of interaction in academic monologue.' Unpublished PhD thesis, University of Liverpool. Vande Kopple, W. J. 1985. `Some exploratory discourse on metadiscourse.' College Composition and Communication 36: 82±93. Waters, M. and A. Waters. 1995. Study Tasks in English. Cambridge: Cambridge University Press. Webber, P. 1994. `The function of questions in different medical journal genres.' English for Specific Purposes 13: 257±68. Widdowson, H. G. 1984. Explorations in Applied Linguistics 2. Oxford: Oxford University Press. Winter, E. 1994 `Clause relations as information structure: Two basic text structures in English' in M. Coulthard (ed.): Advances in Written Text Analysis. London: Routledge. Zamel, V. and R. Spack (eds.). 1998. Negotiating Academic Literacies: Teaching and Learning across Languages and Cultures. Mahwah, NJ and London: Lawrence Erlbaum Associates.

Applied Linguistics 22/1: 79±103

# Oxford University Press 2001

Preposition Entries in UK Monolingual Learners' Dictionaries: Problems and Possible Solutions SETH LINDSTROMBERG Hilderstone College, Kent, UK This paper presents a selective examination of the entries on the preposition on in ®ve advanced learners' dictionaries published in the UK. The principal concern is the preposition on as a signi®er of contact with a surfaceÐespecially an upper surface. The expressions Located Object (LO) and Landmark are glossed in accordance with their use in Cognitive LinguisticsÐe.g. the catLO on the matLANDMARK. With respect to `contact' on, key functional roles of LOs and Landmarks in metaphorical expressions are informally characterized in terms of Cognitive Linguistics metaphor theory. There follows an analysis of the entries in light of these viewpoints. The general conclusion is that all the entries show potential to obscure or conceal important information about the meaning of the headword. Most signi®cantly, (1) sense information is sometimes vague or misleading; (2) sense information is often mismatched to examples; (3) coverage of similar uses is often dispersed throughout an entry; and (4) paradigmatic semantic contrasts are ignored. The suggestions made for the improvement of entries for on may be relevant to entries on spatial prepositions generally.

1. INTRODUCTION 1.1 The scope of interest Among the best known UK-published intermediate-to-advanced monolingual learners' dictionaries1 are Cambridge International Dictionary of English/CIDE (1995), Collins Cobuild English Dictionary/Cobuild (1995), Essential Chambers English Dictionary/Chambers (1995), Longman Dictionary of Contemporary English/ LDOCE (1995) and Oxford Advanced Learner's Dictionary of Current English/OALD (1995). There have been occasional criticisms of each of these dictionaries but few reviewers have expressed such a stark reservation as Amritavalli (1999: 267) who judges Cobuild not to be a learner's dictionary at all (principally on account of the diculty of the language in its de®nitions and examples). Within ELT generally the consensus seems to be that all ®ve dictionaries are fundamentally sound and useful. (Shalom 1996 is a teacher's review; see also Herbst and Herbst 1988 and Landau 1989: 29.) There is one respect though in which the quality of these dictionaries has apparently been largely unexaminedÐtheir portrayal of prepositional

80

PREPOSITION ENTRIES IN UK MONOLINGUAL LEARNERS' DICTIONARIES

semantics. Low (1988: 141±2) observes that a number of dictionaries ignore metaphorical patterning in entries on up, a criticism he supports with speci®c reference to the 1977 edition of OALD. Boers and Demecheleer (1998: 202) note that OALD (presumably 1995) fails to give information crucial to understanding a key metaphorical use2 of beyond and suggest as well that the entry for behind is similarly ¯awed. Given these indications, further scrutiny seemed in order.3 As a report on the treatment of prepositions in these ®ve dictionaries could easily expand impracticably even though only a few score words are at issue, the central concernÐaddressed in Section 5Ðis how these dictionaries deal with selected aspects of the semantics of on, those discussed in Sections 2±4. The discussion in Section 5 is summarised in Section 6.1 and recommendations are o€ered in Section 6.2. Section 7 states the general conclusions.

1.2 Why learners need to know the meanings of spatial prepositions There are at least three reasons why it matters whether this or that spatial preposition is adequately treated in a learner's dictionary. 1 The commoner an expression is, the more important it is to clarify its meaning(s) well. And most spatial prepositions occur with high frequency in discourse of all kinds. 2 Accurate understanding of spatial prepositions can be crucial as it is principally through them that we characterize locations, paths and orientations in physical space. This is true even for encounters at some remove. Morrow (1985), for example, ®nds that a reader's understanding of prepositions plays a signi®cant role in construal of narrative texts in which they occur. Categorical statements that prepositions are `lexically empty' (CIDE: viii) are therefore dicult to make sense of, especially given cases in which (1) context is constant, (2) prepositions vary and (3) meaning varies accordingly. A number of researchers have observed instead that prepositions actually have considerable potential to force particular types of construal on their objects, or `Landmarks' (e.g. Gibbs 1994: 48). For instance, frog in the grass tends to suggest one image and frog on the grass another. It will be argued below (especially in Sections 3 and 4) that key aspects of two spatial senses of on remain constant across quite di€erent uses, even across non-literal uses. 3 It cannot be assumed that advanced learners already understand prepositions adequately. For the last ®ve years I have checked in all my upper-level EFL classes whether anyone can assign a preposition to an illustration showing the side view of an object placed against (i.e. in contact with it) a wall. Of over 200 advanced learners during this time, fewer than 20 have o€ered against as the appropriate preposition even though I always ®rst talk the class through a similar depiction (including a human ®gure to clarify

SETH LINDSTROMBERG

81

viewpoint) of an object placed near a wall and elicit an appropriate prepositionÐe.g. near or by. Similar exercises have suggested other frequent gaps in knowledge of prepositions. My impression is that the percentage of advanced learners who are confused even about central spatial senses of at least some prepositions (e.g. inside/within vs in) is considerable. The results of my informal surveys are very much in line with Coppieters' (1987) ®nding for learners of French that even near-native (i.e. post-advanced) speakers misunderstand prepositions.4 Related to these three points are the three below. 4 Learners who understand prepositions well are better equipped to spot clues about the meaning of unfamiliar vocabulary than learners who do not. For instance, a learner who knows that behind tends to construe a reference object, or `Landmark', as a thing of signi®cant height (viz., hidden/ lying behind the bush/*lake) should, on encountering the unfamiliar word screen in She stood behind a screen, have a better chance at guessing what a screen might be than a learner who thinks that behind merely means `on the other side of '. Certainly, the more words (including prepositions) that one understands in any string of words, the quicker and better the string can be interpreted. With respect to reading, it has been claimed (Laufer 1997: 24, 29) that if learners are to (1) understand a text reasonably well and (2) be able to guess a signi®cant number of unknown words from context, they must know about 95 per cent of all the word tokens the text contains. Poor understanding of only a few high frequency prepositions thus has signi®cant potential to hinder a learner's reading comprehension (by continually dragging down the percentage of word tokens correctly understood) and so reduce his or her chances of guessing unknown lexis of other classes.5 Much the same must be true of spoken texts, it being the case that most prepositions (though not on) are phonologically fairly prominent (e.g. under[neath], across, through, towards, over, above, up, down, beside, along, near[by], between). 5 Many prepositions have at least one salient spatial sense which is systematically employed in shaping the expression of one or more important but essentially non-spatial concepts. Boers (1996), Dirven (1994), Johnson (1987: 30±40), and Lindner (1981) are among those who have argued this claim for English prepositions; it is, as well, powerfully implicit in Lako€ and Johnson (1980, e.g. pp. 29±32) and Reddy (1993: 166±71). A far larger number of studies in linguistics, psychology, and philosophy o€er evidence for the view that ®gurative thought of various kinds is central to human cognitionÐGibbs (1994), Johnson (1987), KoÈvececs and Szabo (1996), Lako€ (1987, 1993), Lako€ and Johnson (1980) and Lako€ and Turner (1989), to mention just a few. The recent interest in tropes, including prepositional metaphor, has been partly responsible for the vastly increased interest in the underlying spatial semantics of prepositions in the last two decades. Dozens of studies of

82

PREPOSITION ENTRIES IN UK MONOLINGUAL LEARNERS' DICTIONARIES

English prepositions have come out during this time. Among those with special relevance in a discussion of on areÐCoventry et al. (1994), Garrod et al. (1999), Herskovits (1985), Jackendo€ and Landau (1991), Lako€ (1987: 416±44), Langacker (several, e.g. 1991), Rice (1996), Sandra and Rice (1995), Svorou (1993), Talmy (several, e.g. 1983), and various papers in Zelinsky-Wibbelt (1994.) These researchers are not in full agreement about the mental structures that constitute the meanings of spatial prepositions. However, most seem to see prepositional meaning as including schematic prototypes which are (1) to some extent gestalt-kinetic in character (e.g. Lako€ 1987; Langacker 1991; Mandler 1996) and (2) suciently endowed with geometrical and functional structure to enable metaphorical extension. The most important category of evidence considered in these studies of prepositional metaphor or of ®gurative language generally is the nonrandom patterning of ®gurative language. It is assumed or argued by all these researchers that some kind of underlying propensity for ®gurative thought is the only plausible explanation for the fact that many ®gurative expressions fall into structured setsÐe.g. the set of expressions portraying life as a journey (e.g. Lako€ 1993; Lako€ and Turner 1989). But the role in cognition of tropes of all kinds has interested a number of experimental psychologists as well. The evidence reported for widespread ®gurative interpretation of ®gurative language, if not yet utterly compelling, is impressive (see, for example, Gibbs (1994) and contributions in Katz et al. (1998)). To summarize, there is a great deal of evidence that prepositional metaphor is hugely important. Learners who are unclear about the basic spatial meanings of prepositions and about the connections between these meanings and metaphorical uses are likely to be disadvantaged. 6 Frank Boers has observed (personal communication): `Languages that are closely related to English tend to have equivalent spatial prepositions as far as the central senses are concerned. Once you move into the metaphoric extensions, however, deviations are likely to occur. A quick look at a French±English and English±French dictionary, for example, reveals that most central spatial senses of on can be translated as sur. Very few of the ®gurative senses of on . . . can be.' Thus, a dictionary which clearly set out plausible links between spatial uses and derived metaphorical uses could be a very useful resource. It might be objected that the total number of preposition look-ups made by the average learner is far smaller than the total number of look-ups of socalled lexical vocabulary. But there are at least three reasons why this undoubted fact should not lead one to suppose it matters little whether prepositional entries are of good or of indi€erent quality. 1 It may be that individual prepositions come fairly high in a ranking of the number of times that individual words are looked up by a large sample of

SETH LINDSTROMBERG

83

learners. I know of no study which casts clear light on this matter, but my own experience as a language learner suggests that this possibility is very strong. 2 In the case of high frequency vocabulary such as many prepositions, it is particularly possible that entries might be consulted for study purposes rather than for text-speci®c decoding. Though he does not mention prepositions in this connection, BeÂjoint (1981) reports that 54 per cent of the learners he surveyed claimed to browse when using a monolingual dictionary. 3 Learners could soon tire of looking up prepositions if they ®nd that the preposition entries in their dictionaries are unhelpful. That is, the frequency with which individual prepositions are looked up could be low, not primarily because learners have no interest in ®nding out what prepositions mean, but rather because they have become pessimistic about ®nding out what they want to ®nd out. It will be argued in Section 5 that learners might be justi®ed in concluding that attempts to ®nd clear information about prepositional meaning in any of these entries are all too likely to be fruitless and take too long.

1.3 Why focus on on? On seemed a particularly good candidate for this study because it falls in the middle range of semantic complexity between, for instance, toward(s) and by. Thus, a reasonably comprehensive dictionary entry for on is unlikely either to put too many semantic complexities in our way or be so short and simple that choice of lexicographical approach hardly matters.

2. ON: AN OVERVIEW OF ITS MEANINGS Spatial on has two distinct meanings. In expressions such as the bottle on the table, on indicates that one thing (the `located object', or LO) is in contact with a surface of another thing (the `Landmark'). This on may be called `contact' on.6 On is also used adverbially, or `intransitively' (Jackendo€ and Landau 1991: 152), to indicate: (1) continued motion after a (potential) interruption (e.g. We hiked on despite/after the rain) and/or (2) motion in an understood direction, especially that in which someone/something is already facing (e.g. Don't just sit there. Come on). This on may be called `progressive' on. The chief concern in the following sections, however, is literal and metaphorical uses of `contact' on.

3. CONTACT METAPHORS Although on can indicate contact from any direction (e.g. lower surface: a lamp on the ceiling; edge: a cafe on the road to London), the `cat on the mat' arrangement (where the lower surface of the LO is supported by the upper

84

PREPOSITION ENTRIES IN UK MONOLINGUAL LEARNERS' DICTIONARIES

surface of the Landmark) is probably salient in cognition. There are three categories of evidence for this. 1 The Oxford English Dictionary (OED) notes that the `cat on the mat' sense has come increasingly to the fore over recent centuries. 2 Experimental studies carried out by Rice (1996) and Sandra and Rice (1995) indicate that this sense has some kind of psychological prominence for adults. 3 First language developmental studies have shown that the cat-on-the-mat sense of on is acquired very early (e.g. Bowerman 1996: 412 and examples in van Geert 1985/86 and Tomasello 1987) and that `situational' learning (i.e. learning with reference to here and now physical movements and con®gurations) is very important (e.g. Johnston 1984 and Tomasello 1987; but see Bowerman 1996 for evidence that attending to and learning patterns in adult speech is also important). If this sense of on is acquired very early and in considerable part situationally, it seems especially likely that it has enough psychological pre-eminence and semantic structure to make its metaphorical application not just possible but informationally worthwhile. All this evidence ®ts with the assertion in the OED that this sense of on underlies a large number of metaphorical expressions. These seem groupable into a small number of overlapping, fuzzy sets according to the roles played by the LOs and Landmarks (Lindstromberg 1997: 60±67). Before considering these, however, it is necessary to consider what it might mean to say that a literal sense might `underlie' a metaphorical use or, in other words, what it means to say that a given expression is cognitively metaphorical. Gibbs (1994: 445±49 and 1998: 93±107) o€ers four hypotheses. In order of increasing strength, they are as follows. 1 Metaphoric thought `plays some role in changing the meanings of words and expressions over time but does not motivate ordinary speakers' contemporary use and understanding of language' (1994: 446). 2 `[It] motivates the linguistic meanings that have currency within linguistic communities, or may have some role in an idealized speaker's/hearer's understanding of language [but] does not actually play any part in individual speakers' cognitive systems to facilitate their ordinary understanding of language' (1994: 446; italics in the original). 3 `[It] motivates an individual speaker's use and understanding of why various words and expressions mean what they do but does not play any role in people's ordinary on-line production or comprehension of everyday language' (1994: 446). For example, by this hypothesis some people think in terms of conceptual metaphors to the extent that these may play a role in conscious interpretation of metaphorical expressions. 4 `[It] functions in people's immediate on-line use and understanding of linguistic meaning' (1998: 103).

SETH LINDSTROMBERG

85

Hypothesis 1 is assumed in such standard works as the OED (e.g. on, `General Sense'; see Sweetser 1990 for an account of the role played in language change by ®gurative thought processes). However, if hypothesis 1 is the case, then hypothesis 2 must be true to a degree since it is unlikely that human cognition today is so fundamentally di€erent from what it used to be when the scope of this or that word was being ®guratively extended in the past. Gibbs (1994: 255±8 and 1998: 104±5) notes the existence of some experimental evidence for hypothesis 4, but also states that methodologies presently available to the laboratory psychologist seem unlikely to have direct bearing on this issue (1998: 107). He concludes that hypothesis 3 is probable in light of what is known from both linguistics and experimental psychology (1994: 103). In sum, there are a number of reasons why compilers of learners' dictionaries should take metaphor into account. Let us now turn to a consideration of `contact' on in metaphorical expressions.

3.1 The LO as burden There are many expressions which suggest that the Landmark is metaphorically and disadvantageously burdened by the LO. Frequently, the linguistic expression of the burden is a clauseÐe.g. His car died on himÐa type of expression which portrays not an object (`his car') but an event (`his car died') as the burden. The metaphor here is doubtless very remote, or abstract; indeed, here on is nearly `grammaticalized' although still far from meaningless. At the more explicit end of what may be a continuum of metaphoricity are: It weighed on his mind, the onus is on s'one, crack down on crime, a tax on beer, impose on s'one. More abstract burden expressions areÐe.g. It's not fair on her, shame on you, a block on aid, hold out on s'one, don't try that on me, tell/inform/spy/cheat on s'one, a drain on our resources, a joke/prank on s'one, turn one's back on s'one, take a toll on s'one.7 Explicit burden expressions such as the following constitute particularly good evidence for the cognitive reality of this metaphorÐbear a heavy/light responsibility (cf. *bear a privilege), a load o€ my mind, get o€ my back, bear a burden of guilt/debt/shame, we all have our cross to bear and the straw that broke the camel's back. (See Boers 1996: 47±8, 82 for use of under and down in expressions of the burden metaphor.) Further, the hypothesis of a burden metaphor is useful. It explains, for instance, why have s'thing on your mind is apt where the LO is a worry while have something in mind is apt where the LO is something like a plan.

3.2 The LO as burdensome impact Probably very closely related to the above are expressions containing verbs or LO nouns which entail burdensome impactÐe.g. an attack/assault on, ®re on,

86

PREPOSITION ENTRIES IN UK MONOLINGUAL LEARNERS' DICTIONARIES

wage war on, march on Paris, they turned on their sponsors (cf. jump/stamp/step on s'thing, pounce on s'thing).

3.3 The Landmark as platform/basis or as a way Also numerous are expressions in which the Landmark plays the role of basis for an explicit or implied action, event or state of a€airsÐe.g. on (an) equal/(a) ®rm footing, on a level playing ®eld, on the grounds that, rely/depend/count on s'one, stand on principle, act on the condition/assumption/premise that, act on purpose/a whim/an impulse, decide on the strength of, build/capitalize on, a landslide victory on a nationalist platform (Cobuild). The expressions X hinges/turns on Y portray the Landmark as a fairly speci®c kind of basis (i.e. a pivot). A few expressions such as agree on (cf. a basis for agreement) and insist on seem to represent an overlap between two Landmark rolesÐthat of basis and that where the Landmark is metaphorically touched but does not support, e.g. a report on the economy. Such `on + topic' expressions are not included in the analysis reported in Section 5. (Results reported in Rice (1996: 156) suggest that basis and topic expressions with on are seen by native speakers as being quite similar to each other but di€erent from burden expressions.) `Way' expressions (e.g. on the road to destruction) are also likely to be related to basis expressions since a way is a basis along which one can move. However, the metaphor involved here has unique entailmentsÐe.g. movement/change and destination/result.

3.4 Other kinds of metaphorical contact: objects of action as contacted substrates, mental contact, visual contact The high frequency of `basis' expressions must derive from the perceptual salience of our early and repeated experience of physical bodies, including our own, resting, standing, or landing on physical bases such as ¯oors. But we make other kinds of perceptually salient physical contact from aboveÐmost notably, perhaps, when working. Most of the commonest traditional forms of work involve physical contact with something located below eye level, be it a ¯intstone or a sheet of paper. In any case, instances of especially prolonged attention are frequently spoken of as if they involved physical contactÐe.g. work/focus/dwell/re¯ect/meditate on a problem. The di€erence in meaning between wait on s'one and wait for s'one is relevant here. Because on entails `contact', wait on is especially apt in cases where the Landmark is present (as in wait on six tables). In wait for, however, for signals that the Landmark is the reason for waiting, hence contact is only potential (Lindstromberg 1997: 221± 5; cf. call s'one vs. call on s'one). Much the same is true of looking, especially if one looks attentively or with emotionÐe.g. focus on the letter `E', keep an eye on, lay/clap one's eyes on, keep a close watch on, gaze on, look on s'one with contempt. Note also the expression visual contact itself. (Sweetser 1990: 28±48 includes a cross-lingual account of how words for seeing have derived from verbs for physical manipulation.)

SETH LINDSTROMBERG

87

Of course, verbs and nouns of looking collocate with other prepositions as wellÐe.g. to, towards and at. But each preposition contributes to such expressions at least something of one of its well-attested spatial senses (Lindstromberg 1997: 172±3). For example, one can throw a stone `at' a bottle without actually hitting it. Here, at indicates only that the Landmark is a target, not that contact is necessarily made with it. But to throw something `on' something clearly entails contact. In the case of looking, it seems less possible in practice to look at something yet somehow `miss' it. (Missing tends to be conveyed by such expressions asÐe.g. overlook, fail to notice and not see.) So, in the case of `looking' verbs, the main di€erence in meaning between at and on seems to be that on suggests visual contact which is spatially or temporally extended. At, since it does not construe a Landmark as having dimension (e.g. at a point), often ®gures in `sharp focus' expressions (e.g. peer/ aim at) and `target' expressions (e.g. throw/talk at as opposed to throw/talk to). Thus, gaze on s'thing, conveying prolonged and probably emotionally tinged visual contact, is a familiar expression (cf. also dote on, fawn on) whereas *glance on s'thing is odd. `Contact' on seems to resist direct collocation with verbs of impolite looking such as stare, peep, gawk and gape which very much construe the Landmark as a target.

3.5 Control, in¯uence, and e€ect as contact A prototypical instance of in¯uence, control or e€ect is almost certainly one in which a physical LO (oneself, for example) is in perceptible `hands on' contact with a physical Landmark (see KoÈvecses and Szabo 1996: 337±43 for lists of relevant expressions). Being the only English preposition of pure contact (Section 4), on occurs in a number of common control, in¯uence, and e€ect expressions in which Landmarks may refer to both concrete and abstract entitiesÐfor example, keep a tight rein on, get a handle/grip on, have an impact/ e€ect/in¯uence on.

3.6 Discussion A recent strand in TESOL methodology is the Lexical Approach (see especially Nattinger and DeCarrico 1992) which is predicated on (1) the hypothesis that ®rst language acquisition involves a great deal of learning of `lexical phrases' (e.g. depend on . . .) more or less as if they were single words and (2) that adult second language learning can proceed in a substantially similar fashion. Advocates of this approach seem to take little notice of prepositions except when describing them as lexically empty.8 There is in this `collocational' approach the strong potential, even the intent, to direct the attention of teachers and learners towards the task of item-by-item learning of lexical phrases and away from generalizable insightsÐsuch as those discussed aboveÐ about meaning of the words which make up phrases. As it happens, at least

88

PREPOSITION ENTRIES IN UK MONOLINGUAL LEARNERS' DICTIONARIES

one prominent lexicographer, John Sinclair, editor-in-chief of Cobuild, has been in the theoretical forefront of this approach (Sinclair 1991). With respect to prepositions, certain de®ciencies in a strict collocational presentation of prepositional usage can be identi®ed already. Such a presentation is ill-suited to giving learners an accurate idea of why certain words collocate. In particular, it seems to o€er learners no means of dealing with low frequency combinations such as talk at and look on. Further, proponents of the Lexical Approach seem wholly uninterested in such matters as the semantic coherence of on in the myriad expressions of contact metaphors. Ignoring word meanings is not, in fact, a sine qua non of a collocational approach merely a strong tendency in practice. In Section 5 it is argued that this tendency is all too manifest in some of the entries examined.

4. THE ISSUE OF PARADIGMATIC RELATIONS: THE CASE OF TWO `CONTACT' PREPOSITIONS Knowing a word adequately includes knowing when to use some di€erent word instead. Let us consider the case of on vs. against. On collocates with verbs that entail force (e.g. push) even though on itself entails mere contact rather than forceful contact. AgainstÐthe other unambiguous `contact' prepositionÐhas, additionally, considerable inherent potential to suggest `force'. In one informal survey (brie¯y mentioned in Lindstromberg 1997: 175±6) 20 adult native-speakers were asked to sketch `a diagram or very simple picture' indicating the meaning of against. Ten produced schematic depictions of a book-like LO leaning against a wall-like Landmark. Four drew a horizontal force vector meeting a vertical line. There was one each showing: static contact with a wall, two opposing force vectors (?/), an arm horizontally outstretched to resist a force, something resisting force vectors coming from four directions, and a `No bomb!' sign. Horizontal force was thus a major element in virtually all the depictions. The participants were then given a set of 20 sentences exemplifying the full range of uses of against as indicated in Cobuild. They were asked to (1) choose sentences which matched the meaning shown by their sketch; (2) group the remaining sentences by meaning(s) of against; and (3) diagram the meaning that against contributed to each group of sentences. Each participant drew one or more horizontal force vectors for at least one group of examples. The most unexpected result was that the sentences Sit against that and You can ®nd it against the X (which I had thought would be interpreted as referring to situations with no horizontal force) were matched by several participants with `leaning book' sketches. This result in particular suggests that force, especially horizontal force, is fairly central to the meaning of against. It is likely, on the other hand, that best example uses of contact on involve top down contact. Push on X, push against X, and pull on X are all acceptable expressions in which the prepositions are used in the ways just discussed. However,

SETH LINDSTROMBERG

89

collocation of pull with against is likely to be odd (?He pulled against the door) except when the derived sense against = `in opposition' can be seen to apply (e.g. In tug o' war, two teams pull against each other, which speaks of two opposing forces). In short, for against to aptly describe a concrete situation in which force is being exerted, at least one force must be directed away from an LO, either like this, LO)Landmark, or like this, LO,LO. Thus, the use of on seen hereÐHer dress caught on a nail (OALD)Ðmay arise as follows: against and on are both candidates for use given the explicitness about contact and the lack of detail suggesting, for instance, containment (cf. My kite got caught in some branches). But against is not apt because the direction of the force is wrong. So on is used even though its central sense of `upper side' contact does not necessarily apply. For the bene®t of learner users, all this should be put succinctlyÐfor example, `Contact which holds or stops s'thingÐHer dress caught on a nail'.

5. HOW THE DICTIONARIES DEAL WITH PREPOSITIONAL MEANING The intention here is to focus on features which in some way re¯ect the compilers' methodologies and underlying beliefs. On the assumption that the overall high quality of each of these dictionaries will be evident to anyone who takes the time to examine them, I concentrate mainly on features that might be improved. In the following discussion I use the term division to indicate a numbered or headed subsection which seems intended to cover one type of use. Note that almost all the corpus examples quoted below have been abbreviatedÐthose from CIDE and Cobuild particularly severely.

5.1 De®ciencies shared by two or more dictionaries Jumbled or fragmented positioning of information There are places in all of the entries where like is unnecessarily separated from like. A number of examples follow. . In CIDE, metaphorical basis expressions (e.g. rely on, depend on, dependent on and based on) are treated under the capitalized rubric NECESSARY. Other `basis' expressions, and `way' expressions, are covered under the rubrics TRAVEL (e.g. on the buses), FOOD/FUEL/DRUG (e.g. on drugs), and FINANCIAL SUPPORT (e.g., on welfare). Although the expressions covered under NECESSARY and TRAVEL are closely related, between these two divisions come three others, NEXT TO, TIME, and AFTER. These contain expressions which are neither closely related to those in NECESSARY and TRAVEL nor closely related themselves. For example, the expressions in NEXT TO are very distantly related to those in TIME. The clumping of TRAVEL, FOOD/FUEL/DRUGS, and FINANCIAL SUPPORT is an island of semantic coherence. However, there is a good deal of fragmentation elsewhere.

90

PREPOSITION ENTRIES IN UK MONOLINGUAL LEARNERS' DICTIONARIES

. Division 11 of Cobuild includes on an area of land. Then come two divisions on, respectively, uses such as on Fifth Avenue (i.e. contact at a side) and on the plane (= airplane). A close link with division 11 is not re-established until division 14 (on a piece of paper). On a list is not covered in division 14 (with on a piece of paper) but independently in division 15. The rationale for creating a separate division on the basis of such a ®ne distinction is not obvious. Further, the following uses seem quite similar, yet they are widely scatteredÐdivision 13 deals with literal vehicle expressions such as on a bus; division 17 with on principle, on credit, 18 includes on the computer; 23 on a cruise/course, 32 on medication, 33 run on batteries, and 34 on low wages. . Division 3 of the `Preposition' section of the Chambers entry includes (a) an expression on your face and (g) on a journey. Division 7 includes (a) have an e€ect/in¯uence on and (c) on ®re. These juxtapositions seem totally random. Division 9 includes (a) working on a task, (d) a tax on something, (e) a pro®t on something and (f) the drinks are on me. These examples are not as unsystematic as those in divisions 3 and 7; still, (d) and (f) are burden expressions and the others are notÐjust to pick out one respect in which this grouping is mixed. . In LDOCE, syntactic rather than semantic criteria have determined whether certain information is given under `on1 preposition' or `on2 adj, adv'. The ®rst three divisions under `on2' concern progressive on, as does division 11 (division 3 concerns temporal extensions of this sense, e.g. from that day on). In between are, for instance, two divisions concerning physical contactÐ e.g. Put your coat on (division 4) and The bus stopped and everybody rushed to get on (division 5). Division 4 in LDOCE covers `in a particular direction: On my right sat the Chancellor. | As the troops marched on the city'. These are two quite di€erent uses. The former use is related to on's functionally neutral contact-at-anedge sense whereas the latter is an expression of the burden metaphor. Division 5 reads, possibly somewhat obscurely, `used to show the person or thing a€ected by an action or someone's behaviour: a tax on cigarettes'. This is where marched on the city belongs. Division 6 reads, `on one of the sides of something such as a river or road: a cafe on the river'. This use is the same as that covered ®rst in Division 4. Division 21 reads, `If a machine stops, breaks etc. on you, it stops or breaks while you are using it: Suddenly the telephone went dead on me.' This should go with or just after division 5. Also, the de®nition conceals the breadth of the applicability of this construction while giving the false impression that, in such cases, on indicates a purely temporal connection. . Division 9 of OALD reads, in part, `at or near a place or time: a town on the coast; a house on the main road; a house on a council estate . . ., drive on the M25' The ®rst two examples involve contact at a side; the last two do not.

SETH LINDSTROMBERG

91

Potentially unhelpful positioning of divisions on temporal uses Temporal uses are poorly, or not at all, segregated in all ®ve entries, probably owing to `by frequency' ordering. Examples include the following. . OALD places I moved to Leeds just on a year ago under `on1 prep' division 9 where the gloss is `at or near a place or time'. This division also includes on the coast, on a council estate, on both sides of the river. Other temporal uses are covered in division 3 as well as in the second half of the entry under `on2 adv part'. . Chambers, division 5, which includes (a) on a certain day and (b) on hearing the news is in among divisions on non-temporal uses. (This division also includes do something on certain grounds, which is related to work on [i.e. on the basis of] a certain system in division 10.)

The longer an entry is the more such mixed-in positioning will hamper `temporal' look-ups. Such mixing seems all the less desirable given evidence that native speakers maintain a cognitive separation of temporal senses from other senses (Rice 1996; Sandra and Rice 1995).

Lack of attention to metaphor None of the entries has adequately taken metaphor into account. This has resulted in (1) a massive loss of potentially helpful information about meaning and (2) mislocation of many divisions and examples.

Lack of information about paradigmatic semantic contrasts None of the entries gives explicit information about how or why using one preposition rather than another might result in a change in meaning. Examples of this are as follows. . Division 9 of the LDOCE entry begins, `in a bus, train, aircraft, etc.' This does nothing to explain why one sometimes says, for example, on a bus and sometimes in a bus. A more helpful explanation (extending an observation in Jackendo€ and Landau 1991: 166) might run as follows. `We use on with ¯at vehicles (like magic carpets) and with ones which have ¯oors big enough to walk on. Also, if you are on a bus/plane/train, the idea is that you are travelling, or that you are going to travel fairly soon. But if you are in a bus/plane/train, the idea is just that you are not outside of it. So, we are not talking about travelling here ? He lives in an old broken down bus in his back yard.| We sat in the plane for hours waiting to take o€. | They were trapped in the sinking ship. But we are talking about traveling here ? She travels so much she practically lives on airplanes.' . The OALD entry, unusually, alerts users to a few potential sources of confusion such as that there is a di€erence between by bus, in the bus, and

92

PREPOSITION ENTRIES IN UK MONOLINGUAL LEARNERS' DICTIONARIES

on the bus but in fact gets no closer to actual clari®cation than this notice in division 2, `Compare: travel by bus, etc; sitting in a bus, etc'. Much the same kinds of weakness can be pointed out in other divisions in each of the ®ve entries.

Lack of pictorial illustration A good deal of information about the meaning(s) of most spatial prepositions can be conveyed pictorially, a fact which none of these dictionaries exploits. Cobuild and Chambers are not illustrated at all. OALD has pictorial illustrations of, for instance, [wood] shavings and contortionist; LDOCE ones for daisy wheel and rosette; CIDE ones for canal and types of trunk. None of the three has an illustration for any preposition. My experience suggests that the compilers were wrong to assume that advanced learners already know so much about prepositions that pictorial illustration of basic senses can be dispensed with. Further, if reference could be made to illustrations, compilers would probably ®nd it easier to compose clear and concise explanations of the functional roles of LO and Landmark in particular metaphors.

Flawed sense information Four of the entries contain sense information which is either vague, misleading, or inaccurate. Examples of such problems include the following. . In CIDE, the division on INVOLVEMENT reads, `used to show when someone is involved or taking part in something: I'm working on a new book, She's out on a job . . ., We were on page 42.' There are a number of problems here. First of all, the de®nition is potentially misleading since involve/take part typically collocate with in rather than on. In construes the Landmark not as a work surface or basis but as a bounded space, the latter being a typical rei®cation of activities (e.g. in action and in doing that she . . .; see, e.g. Lako€ and Johnson 1980; Lako€ 1993). Beyond that, it is unclear what the rubric excludesÐespecially in light of the motley examples; the last of which (on page 42) is a straightforward and not particularly metaphorical instance of on for surface location (cf. on a table) or, depending on context, an instance of on as a topic marker (as in We were on [= talking about] page 42, right?). Again in CIDE, the division headed TO (which is a sense rather than a topic rubric), includes the followingÐon the left, the attack on the village, her dog turned on me, creep up on me. `To' is a distinctly inaccurate gloss of, especially, the second and third uses as the Landmarks are clearly emburdened, or disadvantaged. (Here, as in a number of other examples, there is more than one ¯awÐin this case, not just poor sense information but also poor classi®cation of examples.) . Division 4 of the Chambers entry reads, `(a) A building is on something such

SETH LINDSTROMBERG

93

as a road if it is situated beside it and faces towards it: A hotel on the seafront . . . (b) On is commonly used in referring to other locations: He works on a farm; Do you enjoy living on an island?' One problem here is that part (a) concerns contact at an edge while part (b) belongs in division 1 with on the ¯oor. Further, (a) is too speci®c; it is not the case that a property absolutely must face a road to be on it; it must merely abut onto and, perhaps, be accessible from it. (As in the case of a house with its main entrance on a side away from the street.) A more general problem with (b) is that the wording `commonly used in referring to other locations' includes almost any sort of locative expression whatsoever. . In LDOCE, division 3 part (a) begins, `in a particular place, building or area of land: Several bombs landed on the runway. | He grew up on a ranch'. This explains nothing about why one uses in with some place Landmarks and on with others. For instance, I cannot think of when on a building might mean `in it'. Division 3 part (b) reads `AmE in a particular road: We live on Mulberry Drive'. This conceals the fact that when talking about roads, as opposed to streets, British English also uses on in its contact-with-an-edge senseÐe.g. a service station on the M25/road to Exeter. Note also the acceptability in BrE of on the High Street. . The de®nition in OALD division 8 reads, in its entirety, `indicating direction'. This surely misses the gist of on in march on the capital, turn one's back on, and draw a knife on. Division 13, `indicating an increase, esp of cost', includes a tax on, charge interest on, a strain on. Bringing in the notion of increase without further explanation is potentially misleading since increase collocates quite commonly with in and of as well as with on. The examples in this division should be with those in the following division (14)Ðe.g. a ban on imports, He's hard on his kids. But the gloss for this latter division, `with regard to sb/sth; so as to a€ect sb/sth', hardly captures the gist of the semantics here.

Dicult language . In Chambers, sub-division 5a of the section entitled `phrases' reads: `One thing moves, or is moved on to or onto another thing when it moves so that it the other [sic] is underneath it or supporting it. ``He got on to his motorbike''.' Even without the typo this might be less than clear to a learner user. This is a perfect place for a simple picture. . Division 9 of the OALD entry reads, in part, `at or near a place or time: a town on the coast; a house on the main road; a house on a council estate . . ., drive on the M25'. A learner somewhat in the dark about on is likely to be more in the dark about at. (For example, at church means `in or near the church' with the added implication of attendance for a church related purpose; see Lindkvist 1978 and Lindstromberg 1997: index).

94

PREPOSITION ENTRIES IN UK MONOLINGUAL LEARNERS' DICTIONARIES

Omitted usages Neither Cobuild nor Chambers covers expressions which characterize an entire event as a burden (as in The phone went dead on her).

5.2 Summaries Cambridge International Dictionary of English (CIDE) A fundamental weakness of this entry is that the high number of divisions, 33, is well beyond any plausible number of actual senses (and probably also cognitively distinct use types). This may make it harder for learners to discern consistent threads of meaning, whether literal or metaphorical. Frequent muddling of examples only increases this potential as does the fact that rubrics and de®nitions (to use the latter term loosely) are sometimes obscure or inaccurate. Rubrics which give reasonably clear sense information (e.g. NEXT TO) are in a minority. What the user encounters instead are characterizations of striking disparity (e.g. PAYMENT, MEMBER, FAULTY, TOOL, HAPPENING, PAIN, and two divisions entitled simply `adv, adj'.) Two additional features of this entry are likely to make it less helpful still. Although the example sentences tend to be rather long, the headword is almost never typographically highlighted (e.g. When you go out, leave the bedroom light on to scare o€ the burglars).

Collins Cobuild English Dictionary (Cobuild) This entry has 41 divisions with ones on `progressive' on being scattered throughout ones about `contact' on. Example sentences tend to be quite long. In general, uses are covered in order of frequency (Rosamund Moon, Cobuild lexicographer, personal communication). While it is recognized that prepositions can have `very clear spatial meaning' they are considered, on the whole, to be grammatical rather than lexical words; apparently, `the dictionary [n]ever sought to give a detailed account of their deep meanings' (ibid.). That being said, the Cobuild entry exhibits relatively little jumbling of uses and examples. However, there is no explicit information about semantic relations among di€erent uses. As with CIDE, the high number of divisions has the potential to conceal the semantic gist. A further criticism that could be made of this entry is that ®nding any particular bit of information is much like chancing upon a needle in a haystack. The entry is long. It lacks conspicuous section or division headings and the headword is not highlighted in examples.

Essential Chambers English Dictionary (Chambers) The Chambers entry has the following macro-divisions: `preposition' (10 divisions), `adverb or preposition' (5 divisions), `adverb or adjective' (4), `adverb' (3) and `phrases' (6). A di€erence from the other four dictionaries is that the divisions are themselves divided to create three levels of

SETH LINDSTROMBERG

95

categorization. For example, division 10 in the macro-division `preposition' includes ®ve lettered sub-divisions of which the ®rst and the last are: (a) on a drug and (e) on a certain system/principle. This creation of syntactic macrodivisions has led to considerable duplication and scattering of semantic information. For instance, much the same information about progressive on is given in `adverb' division 1 and `phrases' division 4. The information given under `preposition' division 1 (`One thing is on another (a) if it is supported by it') is dealt with again in detail under `adverb or preposition'. Such duplication and fragmentation of information is certain to decrease scannability when users are not sure what part of speech on is in a particular context. Further, even when users are certain they want information about an adverbial use, for instance, there are three di€erent macro-sections in which it might be found (not counting the `phrases' section).

Longman Dictionary of Contemporary English (LDOCE) There are two separate entriesÐ`on1 ``preposition'' ' and `on2 ``adjective, adverb'' '. In the former there are 21 divisions. As in Chambers, such syntactic macro-division of the entry leads to a disorganized presentation of sense information. Like the other four entries, that in LDOCE is not as clear and accurate as it might be. It is, however, relatively concise and easy to scan, mainly because the examples are comparatively brief and the divisions relatively broad in coverage.

Oxford Advanced Learner's Dictionary of Current English (OALD) The OALD has 18 divisions for contact on and covers progressive on in a separate entry. The introduction explicitly states that uses are presented in order of frequency (p. ix). Sense information and examples are concise, making for a relatively scannable entry. Otherwise, this entry is problematic in more or less the same ways as the entries in the other dictionaries. As with the other four entries, my account of weaknesses focuses on conspicuous lapses and is not exhaustive.

6. SUMMARY AND RECOMMENDATIONS 6.1 Summary These dictionaries are in some respects quite individual. However, with respect to the treatment of on, it is their similarities that come most notably to the fore.

With respect to on itself There is a good deal of evidence that the compilers of these dictionaries have not taken the signi®cance of metaphor into account. If metaphor plays

96

PREPOSITION ENTRIES IN UK MONOLINGUAL LEARNERS' DICTIONARIES

anything like the role suggested in Section 3, then (1) a worrying number of examples are in the wrong divisions (some, in fact, are in the wrong division no matter how one thinks about metaphor); (2) divisions are too often in an order which disguises semantic relations among uses; (3) there is virtually no indication of how uses in di€erent divisions are semantically related; (4) there is a general dearth of explicit sense information and some of it is misleading or wrongÐeven material on literal uses is occasionally inaccurate or incomplete and/or disorganized; (5) especially in CIDE and Cobuild, there are many divisions which are so speci®c as to topic that general sense information is concealed. These are truly remarkable weaknesses given the availability of useful information about sense relations, including metaphor, in such works as the OED (albeit couched in terms of diachronic semantics).

On vs. other prepositions None of the entries has clear potential to help users make sense of on in expressions in which more than one preposition can also occurÐe.g. look on s'one vs. look at s'one. With respect to the paradigmatic semantics of prepositions generally, it is as if Saussure had never drawn a breath. Even the Longman Language Activator is extremely weak here despite the fact that its architecture provides for a considerable degree of semantic (rather than alphabetic) ordering of sense information.

6.2 Recommendations If, as has been argued, on is robustly meaningful in a wide range of uses, then the following innovations seem worth considering: 1. Clear information should be given about the spatial con®gurations that on is used to describe. One way of being more informative here would be to include some (schematic) pictorial representation of meaningÐfor example: On can mean contact with an edge, like this: Toronto is on Lake Ontario.? Toronto **********

Given the small number of headwords involved, addition of pictorial illustration need not result in a ®nancially signi®cant increase in the size of a dictionary given also that many of the illustrations could be very small. 2. The semantics of individual prepositions should be presented in a more coherent fashion. For instance, (1) literal spatial uses and (2) relations between literal and metaphorical uses should be explicitly characterized whenever possible with metaphorical uses being marked as such.9 For instance, the probable close semantic relation between the contact-at-anedge use and that seen in You'll see it on your left/right should be brought out, if only by means of a cross referenceÐfor example:

SETH LINDSTROMBERG

97

About visual contact?If something is on your left or right, this usually means you can see it if you look in that direction. E.g. Go along this street. You will see a bank on your left. Or, Drive along this highway. On your left, you will see some mountains in the distance. See division x, `CONTACT AT AN EDGE'.

When appropriate, an account of how a spatial use has been extended should include a characterization of the functional roles of LO and Landmark. This would include, for instance, whether the Landmark is emburdened by the LO or is an essential support for it. I see no reason not to use the terms literal and metaphorical; they are, in fact, worth learning on their own account. Terms so academic as LO and Landmark should not be used. Instead, the account should concern the arrangement of particular entities referred to in an example. For instance: `To jump, you need something to stand on, a ¯oor for example. Without the ¯oor you cannot jump, so the ¯oor is your ``basis'' for jumping. It is the reason you can jump.?I chose my car on the basis of price.= ``The price was my basis, or reason, for choosing''. | The police arrested him on suspicion. = ``Suspicion was their basis, or reason, for arresting him'' . . .' Boers (2000), Boers and Demecheleer (1998) and KoÈvecses and Szabo (1996) have all found that explanations of this general kind have improved subjects' performance, relative to control groups, in tasks relevant to the argument being made here. 3. Di€erences in meaning between semantically related prepositions should be made clearÐas in lean on vs. lean against and on a bus vs. in a bus. I am unaware of any telling rationale for omitting information of this kind about high frequency closed-set lexis. 4. Because the practice of treating uses in order of frequency strongly tends to reduce semantic coherence, it ought to be abandoned (pace Schol®eld 1997: 292). It is not obvious that learner users have dependable instincts about the frequency of unfamiliar uses. But since frequency ranking is precisely what they must know in advance if they wish to jump straight to the correct division in a long by-order-of-frequency entry, the single advantage of such ordering is that it may shorten look-up times when and only when a user is looking up a high frequency use. On the other hand, users may well begin a look-up with a rough idea of the meaning of the target expression, an idea gleaned from a context, say. This may be enough to enable them fairly quickly to negotiate their way through any semantic headings near or to the correct division. Once they have found the right division, if they desire further semantic information, they can retrace their steps to a (higher) level where they may ®nd useful, generalized sense information. Further, it seems possible that many users of a dictionary in which entries were semantically well structured might come to understand structural regularities from entry to entry within a word class and learn to exploit this knowledge to save time in later look-ups. One such regularity

98

PREPOSITION ENTRIES IN UK MONOLINGUAL LEARNERS' DICTIONARIES

would be that semantically similar uses are in the same part of the entry. I see no signi®cant possibility of strategy transfer in the case of by-order-offrequency entries. If spatial prepositions generally have the degree of consistent meaning that is argued for above, it seems odd that compilers of dictionaries should so signi®cantly exempt them from the precept that `getting the word meaning' is an essential step in learning lexis (Hatch and Brown 1995: 372±4, 382±5). True, there is a trade-o€ between providing best scannability and best coverage of meaning (Cowie 1981: 205). Still, few can doubt that a common (perhaps the commonest) use of a monolingual learner's dictionary is to look up the meaning of a word encountered in a text (e.g. BeÂjoint 1981). In such look-ups, once a user has located the correct entry, the next step is to ®nd as quickly as possible that part of the entry which is most relevant to the context at hand. A particularly straightforward way of doing this is to take note of the Landmark noun (which is often just after the preposition) and then scan the entry for that noun or for one of similar meaning. CIDE sensibly caters for this strategy by beginning some divisions with a conspicuously boxed `Landmark heading', e.g. ` FOOD . . . live on . . . ` However, regardless of highlighting, catering too slavishly for this look-up strategy results in very long, hard to scan entries that are composed of divisions which give a minimum of generalizable sense information. (Of the ®ve dictionaries examined, CIDE and Cobuild have these two traits to the greatest extent.) Plainly, compilers ought to do their best to reconcile users' need for explanation with their need to ®nd relevant information as quickly as possible. To shorten look-ups without sacri®ce of explanation, two speci®c practices seem desirable. 5. Although a degree of syntactic categorization of uses should not be ruled out, the organization of entries should be standardized primarily according to semantic criteria: (1) temporal uses should be clearly identi®ed and separatedÐpositioned, say, at the end of each preposition entry; (2) divisions should consistently be grouped so that ones which cover similar senses/uses occur consecutively. Where such contiguity is impractical, minimal cross-referencing should be considered. One possibility is to group divisions under a hierarchy such as the following: PREPOSITION (a) `Touching the top of something', (b) `Touching an edge or side' [with a cross reference to the relevant entry under `Visible contact'], (c) `Touching the bottom of something', (d) `Visible contact' [to include a division `Visible if you look left/right'], (e) `Contact which holds or stops', (f) `Moving, Changing', (g) `Time'. ADVERB (a) `Continuation', (b) `Encouragement to move', `Time'.

Section (a) of `PREPOSITION' might group divisions in subsections with such headings as `Literal', `Basis (metaphorical)', `Burden (met.)', `Topic (met.)' and `Way (met.).

SETH LINDSTROMBERG

99

6. Prominent highlighting of succinct rubrics has such obvious potential to facilitate scanning of long entries (such as preposition entries) that it is hard to imagine why it has been not been more widely adopted. For example, in CIDE the expression This meal is on me is given under the rubric PAYMENT . This use of capitals and boxing is obviously eye-catching. The disadvantage in this case is that neither the rubric nor the de®nition given under it say anything about the meaning of on, with the result that users are not helped to see what this use has in common with other burden expressions. The solution here might lie in highlighted `lowest level' rubrics; for example: On: TOUCHING THE TOP OF SOMETHING? BURDEN METAPHOR? PAYMENT . A brief account of the key metaphor would be given at the middle level. (Additionally, the headword should be highlighted in examples.) It might be objected that some of the above measures would be ®nancially impractical on account of space constraints. However, individual entries are unlikely to be greatly lengthened, if lengthened at all. In any case, the relatively small number of semantically complex spatial prepositions means that any overall lengthening of a dictionary is likely to be negligible.

7. CONCLUSION It seems a reasonable assumption that ambitious learners should in general be helped to progress in the direction of some variety of standard (e.g. nativespeaker) competence. Evidence has been presented, mostly with respect to on, that understanding the meanings of spatial prepositions is an element of such competence. It has also been argued that ®ve popular learner's dictionaries, for all their many strengths, currently o€er signi®cantly de®cient portrayals of prepositional meaning and usage.10 It is common knowledge that most language learners prefer to use bilingual dictionaries. What is it that these dictionaries most conspicuously provide?Ðentries which are easy to scan and which present meaning explicitly. I think there is a moral here for those who wish to widen the appeal of monolingual learner's dictionaries. As to presentation of meaning, there is fairly strong evidence that one's ability to learn additional languages in mother tongue fashion diminishes with age (Skehan 1998: 222±32). If there is indeed a `critical period', the majority of monolingual dictionary users are almost certainly beyond it. Writing of the special requirements of such (adult) learners, Ellis (1997: 84±5) observes that explicit instruction, or `consciousness raising', can be useful if accompanied by examples. I believe that the manner of portraying prepositional meaning advocated above has the potential to raise learners' awareness of important features of English semantics and usage in just the sense intended by Ellis. The evidence presented in Boers (2000), Boers and Demecheleer (1998) and KoÈveces and Szabo (1996) is encouraging. I believe additionally that there are

100 PREPOSITION ENTRIES IN UK MONOLINGUAL LEARNERS' DICTIONARIES

bound to be ways of explaining prepositional usage while giving adequate information about high frequency collocations.

ACKNOWLEDGEMENTS I am indebted to Frank Boers for his suggestions concerning a late draft as well as for his kindness in providing me with copies of Boers (1996) and of work in progress. I am grateful also to James Banner and Tessa Woodward for their comments and to R. R. K. Hartmann for valuable bibliographical leads. The following were especially generous in supplying me with various papers: Holger Becker, Hubert Cuyckens, Ray Gibbs, Graham Low, Sally Rice, and Jordan Zlatev. Three anonymous readers o€ered great help in making this paper better than it once was.

(Revised version received May 2000)

NOTES 1 These are all `decoding' dictionaries; that is, they are most useful in word-to-meaning look-ups. For `encoding', or meaningto-word look-ups, learners typically resort to bilingual dictionaries. Longman Language Activator (1993) is a monolingual encoding dictionary; but its entry for on deals only with the literal `cat on the mat' sense of on. 2 I use the term use (= type of use) rather than sense where it seems prudent not to presuppose a particular view on the polysemy/monosemy problem. 3 See Becker (1998: Section 4) for a list of standard works on lexicography which say nothing signi®cant about the representation of prepositional meaning. 4 Naturally, some errors and slips made by very advanced non-native speakers must derive from imperfect knowledge or recall of collocational normsÐe.g. *everybody used to sneer on them, uttered by an exceedingly ¯uent and accurate German (BBC Radio 4, 6 November 1999, `A World in your Ear'). 5 Such has been the lack of interest in prepositions in the ®eld of second language vocabulary acquisition, that it is sometimes dicult to know whether authors actually do or do not mean to include prepositions when they speak of vocabulary and lexis. It does seem though that the term word tokens (Laufer 1997, 24) must be intended to include them. 6 (a) Garrod et al. (1999) posit, for contact on, a `hybrid' meaning which includes both a geometrical element (i.e. `contact') and a

functional (i.e. `locational control'/`support') element. Their hypothesis is particularly good at accounting for the use of on with respect to certain non-prototypical con®gurationsÐe.g. a kite on a string. Here, the string (Landmark) is in a position of locational control/support with respect to the kite (LO) and there is contact, so use of on is appropriate despite the non-prototypical con®guration. This is another aspect of the semantics of on which could easily be dealt with in a learner's dictionary. I do not go into this in detail lest the discussion sprawl. (b) LO and Landmark are generally used here to refer to linguistic forms but occasionally they are used in a short-hand way to refer to real world entities these terms might themselves refer to. 7 The results of a sorting task described in Rice (1996: 156) o€er some support for the view that on does in fact have a stable sense across a range of such putative burden expressions. For instance, Rice's subjects grouped an overt burden expression (She's carrying too much responsibility on her shoulders) with one that is much less overt (He turned the light out on me). Unfortunately, owing to the small number of relevant expressions among the total of 70 sentences that were sorted, it is dicult to reach a ®rm conclusion about subjects' intuitions concerning the conceptual similarity of visual contact expressions (such as gaze on) and burden expressions. It would be interesting to know, for instance, how spy on would be sorted.

SETH LINDSTROMBERG

8 Van Oosten (1977) shows that the views (a) that prepositions are lexically empty and (b) that they are not both have long pedigrees. Of the former she notes (p. 454) that it is usually held `by people for whom the notion of a class of meaningless or empty words is a pre-empirical requirement or preference'. 9 Amritavalli (1999: 266), who does not

101

mention prepositions speci®cally, is a recent example of the suggestion that compilers give information about metaphorical uses. 10 Improving preposition entries in learner's dictionaries seems especially advisable given evidence that the preposition system tends to be poorly covered in popular EFL course books (Lindstromberg 1999).

REFERENCES Amritavalli, R. 1999. `Dictionaries are unpredictable.' ELT Journal 53/4: 262±9. Becker, H. 1998. `Stories of OVER: The Semantics of Prepositions in General Monolingual Dictionaries of English'. MA dissertation. Exeter: Exeter University. BeÂjoint, H. 1981. `The foreign student's use of monolingual English dictionaries.' Applied Linguistics 2/3: 207±22. Boers, F. 1996. Spatial Prepositions and Metaphor: A Cognitive Journey along the UP-DOWN and the FRONT-BACK Dimensions. TuÈbingen: Gunter Narr. Boers, F. (2000) `Metaphor awareness and vocabulary retention.' Applied Linguistics 21/4: 553±71. Boers, F. and M. Demecheleer. 1998. `A cognitive semantic approach to teaching prepositions.' ELT Journal 52/3: 197±204. Bowerman, M. 1996. `Learning how to structure space for language: A cross-linguistic perspective' in P. Bloom, M. Peterson, L. Nadel, and M. Garret (eds.): Language and Space. Cambridge, MA: MIT Press. 385± 436. Cambridge International Dictionary of English. 1995. Paul Procter, editor-in-chief. Cambridge: Cambridge University Press. Collins Cobuild English Dictionary 1995. John Sinclair, editor-in-chief. London: HarperCollins Publishers. Coppieters, R. 1987. `Competence differences between native and near-native speakers.' Language 63/3: 544±73. Coventry, K., R. Carmichael, and S. Garrod. 1994. `Spatial prepositions: object-specific function and task requirements.' Journal of Semantics 11/4: 289±309. Cowie, A. P. 1981. `Lexicography and its pedagogic applications: an introduction.' Applied Linguistics 2/3: 203±6.

Dirven, R. 1994. `Dividing up physical and mental space into conceptual categories by means of English prepositions' in C. ZelinskyWibbelt (ed.): The Semantics of Prepositions. Berlin: Mouton de Gruyter. 73±97. Ellis, R. 1997. SLA Research and Language Teaching. Oxford: Oxford University Press. Essential Chambers English Dictionary. 1995. Elaine Higgleton, managing editor. Edinburgh: Chambers. Garrod, S., G. Ferrier, and S. Campbell. 1999. `In and On: investigation of the functional geometry of spatial prepositions.' Cognition 72/12: 167±89. Gibbs, R. W., Jr. 1994. The Poetics of Mind: Figurative Thought, Language and Understanding. Cambridge: Cambridge University Press. Gibbs, R. W., Jr. 1998. `The fight over metaphor in thought and language' in A. N. Katz, C. Cacciari, R. W. Gibbs, Jr., and M. Turner (eds.): Figurative Language and Thought. Oxford: Oxford University Press. 88±111. Hatch, E. and C. Brown. 1995. Vocabulary, Semantics and Language Education. Cambridge: Cambridge University Press. Herskovits, A. 1985. `Semantics and pragmatics of locative expressions.' Cognitive Science 9/2: 341±78. Herbst, D. and T. Herbst. 1988. `Review of Longman Dictionary of Contemporary English (1987 edition).' ELT Journal 42/4: 315±17. Jackendoff, R. and B. Landau. 1991. `Spatial language and spatial cognition' in D. J. Napoli and J. A. Kegl (eds.): Bridges between Psychology and Linguistics. A Festschrift for Lila Gleitman. Hillsdale, NJ: Lawrence Erlbaum Associates. 145±70. Johnson, M. 1987. The Body in the Mind: The Bodily Basis of Meaning, Imagination and Reason. Chicago: Chicago University Press. Johnston, J. 1984. `Acquisition of locative

102 PREPOSITION ENTRIES IN UK MONOLINGUAL LEARNERS' DICTIONARIES

meanings of behind and in front of.' Journal of Child Language 11/2: 407±22. Katz, A., C. Cacciari, R. Gibbs, and M. Turner. (eds.). 1998. Figurative Language and Thought. Oxford: Oxford University Press. KoÈvecses, Z. and P. SzaboÂ. 1996. `Idioms: a view from cognitive science'. Applied Linguistics 17/3: 326±55. Lakoff, G. 1987. Women, Fire and Dangerous Things: What Categories Reveal about the Life of the Mind. Chicago: University of Chicago Press. Lakoff, G. 1993. `Contemporary theory of metaphor' in A. Ortony (ed.): Metaphor and Thought 2nd edn. Cambridge: Cambridge University Press. 202±51. Lakoff, G. and M. Johnson. 1980. Metaphors We Live By. Chicago: Chicago University Press. Lakoff, G. and M. Turner. 1989. More than Cool Reason: A Field Guide to Poetic Metaphor. Chicago: Chicago University Press. Landau, S. 1989. Dictionaries: The Art and Craft of Lexicography. Cambridge: Cambridge University Press. Langacker, R. 1991. Concept, Image and Symbol. Berlin: Mouton de Gruyter. Laufer, B. 1997. `The lexical plight in second language reading' in J. C. and T. Hucking (eds.): Second Language Vocabulary Acquisition: A Rationale for Pedagogy. Cambridge: Cambridge University Press. 20±34. Lindkvist, K.-G. 1978. At vs On, In, By: On the early history of spatial AT and certain primary ideas distinguishing At from On, In, By. Acta Universitatis Stockholmiensis. Stockholm Studies in English, XLVIII. Stockholm. Almqvist and Wiksell International. Lindner, S. 1981. `A Lexico-Semantic Analysis of Verb-Particle Constructions with UP and OUT'. Ph.D. dissertation, University of California at San Diego. Available from the Indiana University Linguistics Club. Lindstromberg, S. 1997. English Prepositions Explained. Amsterdam: John Benjamins. Lindstromberg, S. 1999. `Spatial prepositions in three U.K. Published general English courses.' Folio 5/2. Longman Dictionary of Contemporary English. 1987. 1995. Della Summers, director. Harlow: Addison Wesley Longman. Longman Language Activator. 1993. Della Summers, editorial director. Harlow: Longman. Low, G. 1988. `On teaching metaphor.' Applied Linguistics 9/2: 125±47.

Mandler, J. M. 1996. `Preverbal representation and language in P. Bloom, M. Peterson, L. Nadel, and M. Garret (eds.): Language and Space. Cambridge, MA: MIT Press. 365±84. Morrow, D. 1985. `Prepositions and verb aspect in narrative understanding.' Journal of Memory and Language 24/4: 390±404. Nattinger, J. and J. DeCarrico. 1992. Lexical Phrases and Language Teaching. Oxford: Oxford University Press. Oxford Advanced Learner's Dictionary of Current English. 1995. Jonathan Crowther, editor. Oxford: Oxford University Press. Oxford English Dictionary. 1984±1989, second edition. J. A. Simpson and E. S. C. Weiner. Oxford: Clarendon Press. Reddy, M. 1993 (1979) `The conduit metaphor: a case of frame conflict in our language about language' in A. Ortony (ed.): Metaphor and Thought 2nd edn. Cambridge: Cambridge University Press. 164±201. Rice, S. 1996. `Prepositional prototypes' in R. Dirven and M. PuÈtz (eds.): The Construal of Space in Language and Thought. Berlin: Mouton de Gruyter. 135±65. Sandra, D. and S. Rice. 1995. `Network analyses of prepositions: Mirroring whose mind, the linguist's or the language user's?' Cognitive Linguistics 6/1: 89±130. Scholfield, P. 1997. `Vocabulary reference works in foreign language learning' in N. Schmitt and M. McCarthy (eds.): Vocabulary: Description, Acquisition and Pedagogy. Cambridge: Cambridge University Press. 279±302. Shalom, C. 1996. `Four Major New ELT Dictionaries.' MET 5/3: 72±78. Sinclair, J. 1991. Corpus, Concordance, Collocation. Oxford: Oxford University Press. Skehan, P. 1998. A Cognitive Approach to Language Learning. Oxford: Oxford University Press. Svorou, S. 1993. The Grammar of Space. Amsterdam: John Benjamins. Sweetser, E. 1990. From Etymology to Pragmatics: Metaphorical and Cultural Aspects of Semantic Structure. Cambridge: Cambridge University Press. Talmy, L. 1983. `How language structures space' in H. L. Pick and L. P. Acredolo (eds.): Spatial Orientation: Theory, Research and Application. New York: Plenum Press. 225±82. Tomasello, M. 1987. `Learning to use preposi-

SETH LINDSTROMBERG

tions: a case study.' Journal of Child Language 14/1: 79±98. Van Geert, P. 1985/86. `In, On, Under: an essay on the modularity of infant spatial competence.' First Language 6: 7±28.

103

Van Oosten, J. 1977. `On defining prepositions.' Berkley Linguistics Society 3: 454±64 Zelinsky-Wibbelt, C. (ed.). 1994. The Semantics of Prepositions. Berlin: Mouton de Gruyter.

Applied Linguistics 22/1: 104±121

# Oxford University Press 2001

Interpreting the Discourse of H. G. Widdowson: A Corpus-Based Critical Discourse Analysis ROBERT DE BEAUGRANDE United Arab Emirates University The discourse of a recent position paper by H.G. Widdowson is subjected to analysis by three methods criticised in that very paper. The paper was converted into a miniature data corpus and analysed with the concepts of systemic functional linguistics, corpus linguistics, and critical discourse analysis. These methods may well harbour a potential for more rigorous and disciplined intellectual debate thematising the rhetorical strategies that serve to construct or deconstruct academic ideas and positions.

ANALYSING A DISCOURSE AS A MINI-CORPUS H. G. Widdowson's (2000) paper starts on a note of `celebration' for the 20th anniversary of the journal Applied Linguistics; and for the `past achievements' and `future prospects' of the ®eld of `applied linguistics' `moving con®dently into the new millennium' (p. 3).1 But the bulk of the paper projects no bright `con®dence'.2 Instead, the ®eld is diagnosed to be fraught with `persistent and pervasive uncertainty'Ðthis last key-word attested four times in two pagesÐ and menaced by ancillary symptoms like `conceptual elusiveness' and `niggling dissatisfaction', and by the `risks' of following `unprincipled expediency', or `becoming an appendage to another discipline' (3±4). To allay such symptoms, Widdowson vows he will `try yet again to get clear' the `distinctive nature of applied linguistics as a ®eld of enquiry' (4). Intriguingly, the lines of argument he expounds turn out to point in just the opposite direction from the editorial in the same anniversary issue, which salutes a `broad range of research paradigms and the possibility of their combination': `applied linguistics, in becoming more interdisciplinary, is better prepared for the principled handling of a range of real-world issues' (Bygate and Kramsch 2000: 2). For such a discourse on such an occasion, I would propose to `apply the approved procedures of close textual analysis' (15); and, more ironically, the procedures of those linguistic approaches the paper criticizes. To this endÐ and to obviate any suspicion of misquoting himÐI shall approach Widdowson's discourse as a mini-corpus, which I have scanned and entered as a `text library' into John Milton's WordPilot 2000, a useful and compact

ROBERT DE BEAUGRANDE

105

corpus programme I am currently developing for pedagogical applications here at the United Arab Emirates University (cf. Milton 1999; Beaugrande in preparation). Its main use is for writers in Word or similar programs to call up and display attested usages of key words from a library of texts; as far as I know, this report is its ®rst use to analyse a single text.3 In one plausible interpretation, the paper prescribes in general terms what the `discipline of applied linguistics' should do, and proscribes in speci®c terms what it should not do. To start us o€, here are the occurrences returned by my `mini-corpus' when I queried the key-word collocation `applied linguistics' and identi®ed Widdowson's prescriptions:4 1 being concerned with language problems as experienced in the real world (3); 2 taking cognizance [of] the contextual conditions that have to be met in the classroom for language to be a reality for the learners (7); 3 pedagogically treating descriptive ®ndings to make them appropriate as prescription (9); 4 mediation [and] interpretation [to make] relevant the ideas and ®ndings from linguistics (5); 5 mediating between linguistics and other discourses and identifying where they might relevantly interrelate (23); 6 relating and reconciling di€erent representations of reality (5); 7 making people more socio-politically aware of the way language is used to manipulate them (9). Such theoretical and abstract formulations might prove dicult to con®rm or contest, especially for language teachers who want concrete and practical advice. Looking back `25 years' to when `linguistics was de®ned along traditional and formal lines', Widdowson suggests that his own discipline had a `relatively straightforward task' at that time (4): 8 [to] recontextualize language [after] linguistics decontextualized language from reality, [and to] reconstruct reality in the process (4); 9 to refer such abstract analyses of idealized internalized I-language . . . back to the real world to ®nd ways in which experienced, externalized E-language could be reformulated so as to make it amenable to benevolent intervention (4). In comparison to the beleaguered dichotomies of `langue and parole' or `competence and performance', Chomsky's newer dichotomy between `I-language' and `E-language' (invoked more elaborately in Widdowson 1991) might seem less disputatious. But `I-language' in Chomsky's sense is a universal code for all languages, genetically and biologically programmed to provide `a highly determinate, very de®nite structure': `as we acquire language', `these things just kind of grow in our minds, the same way we grow arms and legs' (Chomsky 1991: 66); such a growing process precludes

106 INTERPRETING THE DISCOURSE OF H. G. WIDDOWSON

interventions, benevolent or not. In exchange, Chomsky (1986: 25) regards E-languages as `epiphenomena at best'. So our task might sound more straightforward if stated in older terms, for example: 9a to infer, from abstract analyses of ideal competence, the means for benevolent enhancements of real performance. Even so, the task seems highly likely to foster `uncertainty'. In Widdowson's paper, the newer dichotomy can support his particular delimitation of `linguistics', elsewhere called `theoretical linguistics' for greater emphasis: 10 the discipline of linguistics . . . is bound to be an abstraction in some degree, its theories and descriptions at some remove from the reality of experienced language (23); 11 theoretical linguistics . . . in its quest for understanding, cannot do otherwise than idealize reality and produce abstract models which bear no direct resemblance to the actual experience of language (Widdowson 1997: 146). Here, the entire discipline gets subsumed under one `dominant orthodoxy', whose `formalist limits' (4) place linguistics at the safest distance from applied linguistics. Widdowson merely remarks in passing that `some linguists, of course, never accepted them anyway' (4). The ®eld is thus being radically reduced and impoverished by ignoring its venerable traditions of research in close contact with the `reality' and `actual experience of language', namely the documentation and ®eldwork carried out on previously undescribed languages or dialects. Wenker, Wrede, Winteler, GuilleÂron, Jaberg, Kloeke, Schuchart, Schmeller, Gamillscheg, Ellis, Kurath, Sapir, Whorf, Firth, Pike, McDavid, Longacre, Chafe, Grimes . . . some linguists indeed! Their work laid the foundations for linguistics to be accredited as a science long before the retreat into `idealizations' and `abstractions' far `removed from the reality of language' (cf. Beaugrande 1991, 1997a, 1997b, 1998a). Widdowson looks back only `over the past 25 years or so',5 and ®nds that the bounds of linguistics as he de®nes them are being overstepped, whence arises an imminent danger: 12 There is now a fairly widespread conviction that linguistics should concern itself not with idealized constructs but with the reality of language as people actually experience it: as communication, as the expression of identity, as the means for the exercise of social control. . . . So long as linguistics dealt with idealized remote abstractions, there was a role for applied linguists to play in referring them back to the reality of lived experience. But if linguistics now accounts for that reality, what, one wonders, is there left for applied linguists to do? (4f) `Once a more comprehensive view of linguistics is in place', he goes on, `one might argue that applied linguistics' `quite naturally becomes redundant' and

ROBERT DE BEAUGRANDE

107

`has no purpose'; `linguistics' would simply `incorporate the concerns of applied linguistics' (5). He thus arrives at his leit-motif `distinction between applied linguistics' and `linguistics applied' (Widdowson 1980: 5). I must confess I have never seen any compelling motive for this distinction. Surely we ought judge any project of applying on its own merits; and some such projects have proven undeniably useful. The linguistic investigations of phonetics and prosody conducted out at the School for Oriental and African Studies were expressly intended for application to military and diplomatic concerns during and following World War II. But for Widdowson, the second term deserves `in e€ect' to be christened with the sinister antonym `misapplied linguistics' (6; compare sample 31 below). Misapplication is enacted whenever `some language problem' gets `reformulated by the direct and unilateral application of concepts and terms deriving from linguistic enquiry' (5). Widdowson now exhorts `applied linguistics' to defend its own `occupation'6 by `avoiding and indeed resisting the deterministic practices of linguistics' getting `applied' (5, 23). Through this line of reasoning, Widdowson's paper moves to criticize just those trends inside linguistics which seem acutely likely to `extend its range'. With no evident sense of irony, he crosses the boundaries of his discipline to castigate other disciplines for crossing theirsÐjust when, as I noted, the journal's editors are welcoming a `broad range of research paradigms'. Moreover, he briefs several leading ®gures in linguistics to the e€ect that their research would be `potentially highly serviceable' and `of considerable signi®cance' only if they acknowledge their `limitations' and disavow their `mistakes' (8, 23f). I have studied and applied the trends he discusses, and I ®nd myself unable to recognize them in Widdowson's portrayals, least of all in the `assumptions' and `claims' he consigns to them (shown below in data samples 37±41 and 42±45). Not surprisingly, they are the same three trends I have noticed him criticizing in his other papers and speeches (e.g. Widdowson 1991, 1995, 1997; see Beaugrande 1998b, 2000a for discussion): the systemic functional linguistics of M. A. K. Halliday; the corpus linguistics of J. McH. Sinclair; and the critical discourse analysis of Norman Fairclough. As if to single out a contact point among all three, Widdowson cites the work of Michael Stubbs (e.g. 1994), who, I would object, is not a central strong representative of any of them. A key-word in the case against all three trends, and also against linguistics as a whole, turns out to be `reality'. This potentially abstruse term becomes vital once `applied linguistics' is de®ned to be intimately concerned with `reality' and the `real world' (samples 1, 2, 6, 8±9). The major steps in the `reality' argument might run like this: 13 although the scope of linguistic enquiry has been extended, its mode of enquiry is such as to give only a restricted account of experienced language (5);

108 INTERPRETING THE DISCOURSE OF H. G. WIDDOWSON

14 no matter how extended the scope of linguistics into the real world, its unmediated application can never `become' applied linguistics because it will always represent that reality linguistically, on its own terms and in its own terms (5); 15 [in] linguistics applied, problems are reduced and resolved by the imposition of necessarily partial linguistic account [sic] on the reality of language experience. (3) By this argument, linguistics can grasp reality only in restricted and partial ways just because it is linguistics, full stop. The prospect of surmounting such limits by joining an interdisciplinary research programme (e.g. as expounded in Beaugrande 1997a) is not raised for a moment. As evidence for the key role of `reality', I would cite its 31 occurrences in Widdowson's paper, plus those of `real' (20), `really' (6), and `realize/ realization' (always in the sense of `make real, put to use, perform') (13). The same stem is thus attested a remarkable 70 times, like a curious ostinato. If `the term ``real'' indeed is often used freely as a general stamp of commendation' (5), then Widdowson plainly follows suit. In return, `examples of linguistics applied' can be criticized for inappropriately trying to `extend their scope into the real world' (6), as when `corpus linguistics' and `critical linguistics' are said to: 16 make claims for the relevance of their analyses to the formulation of problems as experienced in the real world which I believe to be questionable (6). Scepticism is implied too when these approaches emphasize `real language' or `examples': 17 Now that we know what real language looks like, the argument runs, we expose learners to it and rid our classrooms of contrivance (7). 18 It is sometimes assumed to be self-evident that real language is bound to be motivating (7). 19 [Sinclair] has recently o€ered a number of precepts for language teachers, the ®rst of which is: present real examples only (9). Yet we might wonder whether such diverse approaches can all be arraigned with the same case on grounds of `reality'. So I shall examine three `cases' in turn.

The case against Halliday and functional linguistics Halliday has been a prominent advocate for linguistics to analyse `real-life discourse' instead of `idealized isolated sentences'; and to recognize the `text' as `an active partner in the reality-making and reality-changing process' (Halliday 1985: 159, 318). Such an approach contravenes Widdowson's avowal that `linguistics' must `idealize reality and produce abstract models which bear no direct resemblance to the actual experience

ROBERT DE BEAUGRANDE

109

of language' (sample 11). In parallel, he argues that Halliday's `grammar' must be more theoretical and further from real texts than Halliday himself maintains. Signi®cantly, not Halliday but Stubbs (1994: 203) is called to witness the `Hallidayan assumption that all linguistic usage encodes representations of the world'. This `assumption' is a gross exaggeration; and my own studies of Halliday's works (e.g. Beaugrande 1991) convince me that he assumes no such thing. He has in fact said, for example: 20 the `slices of meaning' which `the categories of our language represent' may not `correspond to our conscious structuring of the world on creatures and things' (Halliday 1985: xxv); 21 `categories of grammar' are on `a purely abstract level of coding with no direct input±output link with the outside world' (Halliday 1985: xxxv). Halliday cites the `Gender system in English' `not corresponding to any of our conscious categorizations of experience' (1994: xxv), but the more complex Gender system of Arabic makes the point far better. In Modern Standard Arabic (MSA), the Cardinal Numerals 1 and 2 as well as 11 and 12 agree in Gender with the Noun, whereas for 3 through 10 and for 13 through 19 the Feminine Numeral goes before a Masculine Noun, and vice versa.7 Surely no one would claim that the world is being represented here, for example on the grounds that when three's a crowd, gender identities get fouled up!8 Besides, such is the `extraordinary complexity' of the `rules' that `only in the prepared reading aloud of MSA do many educated Arabs themselves feel con®dent of getting them correct' (Holes 1995: 173). I would also reject Widdowson's vision of `Halliday sharing the assumption' that `textual meaning is a sum of the parts rather than a set of mutually modifying relations' (16). Halliday says quite the reverse: 22 In the `grammar', `meanings are accepted from di€erent metafunctional inputs and spliced together to form integrated outputs, or wordings' . . . `The wording ``realizes'', or encodes, the meaning. . . . There is no sense in asking which determines which [or] what each symbol as an isolate means; the meaning is encoded in the wording as an integrated whole' (1985: xxxvf, xx). Further, I would contest the Widdowsonian repudiation of Halliday's (1985: xv, xxii) `aim' to `construct a grammar for the purposes of text analysis' and to provide `at once both a grammar of the system and a grammar of the text' (17): 23 this would seem to suggest that if you can identify part of a text as manifesting a grammatical feature of whatever kind, all you need to do is to read o€ the meaning it encodes. . . . The text is thus taken to be a static patchwork. The dynamic interrelationships which grammatical features contract with each other are disregarded as irrelevant (17).

110 INTERPRETING THE DISCOURSE OF H. G. WIDDOWSON

I would recall here Widdowson's (1997: 150, 165) divisive contention that Halliday's is `a grammar of the system and not a grammar of the text at all', `it being misleading to claim' otherwise. As with `linguistics', Widdowson delimits the concept of `grammar' from his own particular viewpoint: 24 a grammar, no matter how functionally informed, is an analytic device. Thus its systems are abstract constructs which separate out the underlying elements of language and assign them separate semantic signi®cation (16). Of course, such an `abstract construct' can hardly apply to real texts or discourses; and Widdowson (1997: 155) raised the stakes by stipulating that the `account' would have to be `exhaustive': 25 the grammar . . . can never be an account of what people can mean, [because] the very account of the potential . . . can never tell you exhaustively what language users will make of the language resources at their disposal. Against the grain of my own earlier work, I myself have had to accept Halliday's (1985: xvi) precept that a `discourse analysis' `not based on grammar is not an analysis at all, but simply a running commentary on a text'; a chapter in my New Foundations expounds a Halliday-style grammar and illustrates its application with 498 authentic text-samples (Beaugrande 1997a: 182±288). Widdowson's case begins to unravel when we drop the improperly attributed `Hallidayan assumption that all linguistic usage encodes representations of the world'. The `functional grammar' in Halliday's sense merely assumes that linguistic usage and grammar can serve to represent the world where the co-text and context are appropriate. This obviously could not be achieved by any `static' operation of `reading o€ ' what has been permanently `encoded' (sample 23)Ða conception far more congenial to formalist than to functionalist linguistics. Allow me to illustrate with my Widdowsonian mini-corpus. In his discourse, the stem `argue' attested in Verbs or Nouns (25 occurrences) is found to carry two opposed views of reality. Used with the First Person Singular in the Present Perfect, it carries the force: `and so we can really accept it', viz.: 26 I have argued that corpus linguistics provides us with the description of text, not discourse (9). 27 As I have argued elsewhere . . . a grammar . . . cannot account for the way language functions synthetically in text (16). 28 As I have argued earlier, the computer can only reveal what is actually attested as overt behaviour (18). 29 But, as I have argued elsewhere,this is a crucial principle (21).

ROBERT DE BEAUGRANDE

111

The force is subtly similar for `I would argue' and `I want to argue' despite the change in modality: 30 casting doubt on the relevance of linguistic description for pedagogic prescription is, I would argue, precisely what we should be about (9). 31 I would argue, then, that linguistics applied is, in e€ect, misapplied linguistics (6). 32 In both corpus and critical linguistics what we ®nd, I would argue, is an analysis of text which is then given unwarranted discourse signi®cance in disregard of crucial contextual factors (22). Other grammatical options for the same Verb suggest a failure to grasp reality. Such is the force if the `arguing' is either agentless, as in the Medial 33 or the Passive 34; or if the agent is the impersonal `one' 35, or a linguist being criticized 36. 33 Now that we know what real language looks like, the argument runs, we expose learners to it and rid our classrooms of contrivance (7). 34 It might be argued, however, that this is not the relevant evidence (18). 35 One might argue that . . . the ergative . . . always carries implications of process and result (14). 36 And this, as Stubbs argues, is where corpus descriptions come in. The Noun and Verb stem `assume', at 22 attestations, never collocates with Widdowson as First Person agent having a privileged access to reality. Instead, we ®nd contexts constructed to cast doubt, as in: 37 There is a widespread assumption that the classroom is of its nature an unreal place (8). 38 People concerned with foreign language pedagogy have expressed reservations . . . about the assumption that the ®ndings of corpus linguistics should determine the content of the language subject (8). 39 He [Stubbs, later said of Fairclough too] makes the assumption that the semantic meaning that the grammarian identi®es as having become inscribed in a particular grammatical category is carried intact into the text (16, 19). 40 And this assumption . . . he shares with Halliday, [who] apparently makes no distinction between grammatical and textual units of meaning (16f). 41 corpus linguistics assumes that the appropriate can be derived from the attested, and critical linguistics assumes that the appropriate can be derived from the possible (23). Wherever a speci®c `assumer' is identi®ed, I could adduce documentation that he does not make these assumptions in the ®rst place. For sample 41, however, I think they are true, although Widdowson wants to suggest the contrary. I cannot imagine a linguistic usage being appropriate if it had never been attested, though it might well not be attested in one particular corpus; and still less can I imagine the appropriate being derived from the impossible!

112 INTERPRETING THE DISCOURSE OF H. G. WIDDOWSON

Compared to `assume', the contextual range of the Noun and Verb stem `claim', at 19 attestations, is even more drastic. Not only it is never used with Widdowson as agent, but the `claims' are made to sound dubious, especially in: 42 Each claims to have something quite radical to reveal about language . . . each also makes claims for the relevance of their analyses (6). 43 Both claim that they reveal the reality of experienced language, hitherto inaccessible (23). 44 claims are made that this provides the only language worth teaching (7). 45 the claim for relevance is not informed by pedagogic considerations (9). Such collocations might signal what Widdowson implies by invoking `the kind of enquiry that this journal, nominally at least, claims to promote' (4)Ðnamely that it may be promoting `misapplied linguistics'. I would note how the editorial in the same journal issue expressly approved the `emergence' of `corpus linguistics' and `critical linguistics' (Bygate and Kramsch 2000: 2). My own point would be that we surely do not `read o€ formally encoded functions' for these Nouns and Verbs attested in Widdowsonian discourse. The grammar of English does not encode whether an `assumption', or `claim' is or is not plausible, justi®ed, and so on. All the grammar does is allow a discourse producer to create a non-committal frame for a concept or idea somebody else holds or expresses. Stated in Halliday's (1994: 107) terms, these expressions are posed on the border between Mental Processes and Verbal Processes as `symbolic relations constructed in human consciousness and enacted in the form of language'. But the attestations of this mini-corpus show these expressions being exploited to formulate what Widdowson presents as accurate positions of Halliday, Sinclair, and so on, even if they never said so. Perhaps this tactic illustrates `mediating between linguistics and other discourses' (sample 5), but I would strongly prefer the exegesis of the discourse itself, as I am attempting it here.

The case against Sinclair and corpus linguistics For corpus linguistics, Widdowson's `arguments' take a di€erent tack. For Halliday's linguistics, a key objection was essentially that Halliday's linguistics is (or should be) con®ned to a theoretical mode of `grammar' that cannot be applied to texts. For Sinclair's linguistics, the application to texts is impossible to deny; so Widdowson argues that these `texts' are not `real' after all, contrary to the ®rm conviction among corpus linguists (myself included). Arraigning the approach on grounds of `reality' thus grows more intricate. On the one hand, corpus data are implied to be only super®cially real by terming them `actual language behaviour', `overt behaviour', `material products', `textual traces', or just `print-outs' (6f, 18f). These, Widdowson vows, are all `the computer can cope with', since it `readily recognizes' `formally marked criteria' rather than `semantic subtleties' (6f, 14); but

ROBERT DE BEAUGRANDE

113

software is now available for highly complex automatic parsing and lemmatizing (Aston and Burnard 1998). Also, he has accused Sinclair of claiming that `real English' consists only of corpus data: 46 You do not represent language beyond the corpus: the language is represented by the corpus. What is not attested in the data is . . . not real English (Widdowson 1991: 14). Like Halliday, Sinclair says just the opposite of what Widdowson represents him saying, for example: 47 language users treat the regular patterns as jumping o€ points, and create endless variations to suit particular purposes (Sinclair 1991: 492). The valid formulation, in my view, is that the data in a very large corpus do not embody or set the boundaries of `real English', but rather manifest the major regularities underlying a large portion of `real English'Ðthe English of real users of the language, in real experience. On the other hand, Widdowson characterizes corpus data as `contextual abstractions' deprived of `reality' until they have been `recontextualized', viz. samples 47±50; and we might recall `recontextualizing language' and `reconstructing reality' being picked out as concerns for applied linguistics (e.g. back in sample 8). 48 The texts which are collected in a corpus have a re¯ected reality: they are only real because of the presupposed reality of the discourses of which they are a trace (7). 49 This is decontexualized language, which is why it is only partially real. If the language is to be realized as use, it has to be recontextualized (7). 50 language units [that have been] contextually abstracted . . . have to be recontextualized in the classroom so as to make them real for learners (8). 51 The textual ®ndings of frequencies and co-occurrences have to be contextually reconstituted in the classroom for their reality to be realized (7). These arguments blur the distinction between Chomsky's linguistics of isolated sentences and Sinclair's linguistics of corpus dataÐtwo approaches that just could not be more disparateÐby ascribing to both, without convincing justi®cation, the study of `decontextualized language'. To advance his case, Widdowson reverts to another of his favourite dichotomies, this one between `text' and `discourse'. Such a dichotomy originated in the 1960s and 1970s from the unproductive but attractive idea of the text being a set or sequence of sentences,9 such that `sentence linguistics' and `sentence grammar' could be extended to accommodate `text linguistics' and `text grammar' with minimal revisions (for a detailed discussion see Beaugrande 2000b, in preparation). Widdowson's motivation now is just the reverse: de®ning the `text' so it will not be accommodated, at least not until applied linguistics has turned it into `discourse' (samples 52±53). Yet the whole dichotomy remains empty for a simple reason: I have never

114 INTERPRETING THE DISCOURSE OF H. G. WIDDOWSON

seen a convincing instance of the theoretical text isolated from all discourse. The text linguistics and text grammar predicated upon this text have largely disappeared, whilst text linguistics and discourse analysis have briskly converged, much to their mutual bene®t (Beaugrande 1997a); the linguists who still `idealize reality' and shun `the actual experience of language' prefer their invented John-and-Mary sentences anyway. I would submit that a real text cannot be decontexualized, that is, removed from any context; we can only shift it into a di€erent context, which is an ordinary transaction not just in language classrooms, but in most reports or discussions of what somebody said. With a real text, you cannot help getting implicated in interpreting it. Widdowson seems to envision two separable stages, one of them reserved for applied linguistics: 52 text recognition is not the same as text realization, for this latter involves interpretation, that is to say the deriving of a discourse from it (10); 53 the textual constraints on interpretation that people should be aware of, [and] whereby people derive diverse discourses from the same text . . . are questions for applied linguistics (23). Yet recognition without realization or interpretation is not an empirically real operation, despite what some theoretical models of the reading process may have suggested (survey in Beaugrande 1980±81). You can of course recognize isolated letters or words without understanding them as discourse, especially if you are not ¯uent in the language, but then they do not constitute a text as a communicative event. And this is certainly not the situation of corpus linguists anyhow. What corpus queries do is extract samples of speci®ed length from their original texts and display, at a glance, multiple occurrences of an expression or pattern that were not originally placed in such close proximity. But when corpus linguists read just one sample, we are not proceeding too di€erently from a reader in the intended audience who also reads just a line or two at a time; and we need merely touch a few buttons to expand that sample up to as many lines as we desire. As a further reservation against corpus linguistics, Widdowson devises a trichotomy among ®rst, second, and third person data: 54 frequencies of occurrence of words, and regular patterns of collocation cooccurrence . . . are third person observed data (`When do they use the word X?') which are di€erent from the ®rst person data of introspection (`When do I use the word X?'), and the second person data of elicitation (`When do you use the word X?') (6). 55 Corpus analysis reveals . . . third person facts of what people do, but not the facts of what people know, nor what they think they do: they come from the perspective of the observer looking on, not the introspective of the insider. . . . In ethnomethodogical [sic] terms, we do not get member categories of description (6).

ROBERT DE BEAUGRANDE

115

56 The description of internalized language requires a ®rst person perspective . . . to prise knowledge out from the recesses of the mind: knowledge which is not realized as behavioural evidence available to the observer. . . . Corpus linguistics . . . adopts the third person perspective (Widdowson 1991: 15). But everyone has a three-fold vision, and it is rarely a debilitating obstacle. Since as discourse participants we cannot observe meanings, we observe what we can see or hear and make sense of it with the help of introspection, using a ®rst person point of reference (e.g. `what would I mean if I said that?'), but also a second person (e.g. `what do you mean?') and a third person (e.g. `what do they mean?'). None of this would seem problematic, had not theoretical linguistics forced a wedge in between observation and introspection in order to declare that the `observed use of language' `surely cannot constitute the subject-matter of linguistics' as a `serious discipline' (Chomsky 1965: 4). At the same stage, elicitation was ousted from the primary role it has occupied in the ®eldwork linguistics I cited above. All these divisions were artefacts of a restrictive academic politics, not empirically real demarcations within human processing. When linguistics was enjoined to `construct a description' or `explanation for the enormous mass of unquestionable data concerning the linguistic intuition of the native speaker, often himself ' (Chomsky 1965: 20), we see who was really behind the `ideal speaker-hearer'. Authority from one's statements about the languageÐparticularly when the `data' proved to be `questionable' after allÐwas claimed by virtue of holding an advanced degree in theoretical linguistics, which guaranteed a privileged access to intuitions. Today, corpus linguistics is ®nally reinstating observation, and on a scale previously not feasible. The linguist is rejoining the community of real speakers and hearers rather than hiding behind an ideal one. Authority is now claimed from carefully examining large sets of real texts with a reasonably open mind, investing ®rst, second, and third person perspectives, along with our introspection and intuitions, in the full awareness that we may need to go and look for more data. To demand complete or exhaustive coverage from any corpus of texts is just as unreasonable as demanding a display of all the grammatical sentences of a language. What corpus research seeks is some critical mass where a set of data converge upon a reasonable certainty about their structure and meaning. We do not claim that our interpretations are `correct' or `privileged', but rather that they are more methodical and systematic than if we were `prising knowledge out from the recesses of the mind', which sounds like shelling a stubborn walnut. A systematic look just at the tiny corpus I generated from Widdowson's anniversary paper (11,418 words by my count, excluding notes and references) can be a substantial aid for `interpretation' and not just for `®nding frequencies' within a merely `re¯ected reality' (samples 48 and 51). I can observe clusters of Noun±Verb occurrences that help to verify

116 INTERPRETING THE DISCOURSE OF H. G. WIDDOWSON

Widdowson's `arguments' and falsify the `claims' of those he criticizes; or occurrences that assign di€erential values to what counts as `real' or `reality'. And I can follow in some detail how he carves out niches for `applied linguistics' and warns various linguistic approaches that they do not belong there. Mine remains a `partial interpretation' for which I allege no `privileged status' but which I propose as an instance of what Widdowson calls a `stimulus for enquiry' (22), for instance, when I ask to see an instance of a text that is not discourse.

The case against critical discourse analysis The `argument' is once again di€erent for `critical discourse analysis' (hereafter CDA) although Widdowson suggests it is in cahoots with the other two, viz.: 57 It is not surprising then that critical discourse analysts should ®nd Halliday's grammar so well-suited to their purposes. It is tailor made for them (17). 58 And this, as Stubbs argues, is where corpus descriptions come in. If Fairclough's claim of `usual association' is supported by evidence of collocational frequency (18) Yet Sinclair's work is always called `corpus analysis' and never `discourse analysis', despite his massive contributions to the latter ®eld based squarely upon real language experience in the classroom (e.g. Sinclair and Coulthard 1975; Sinclair and Brazil 1982). Nor is the latter term applied to Halliday's work even though the monumental Introduction to Functional Grammar of 1985 and 1994, adduced by Widdowson, partly `grew out of class notes prepared for a course' `in discourse analysis' (Halliday 1994: ix). The term is always collocated with `critical', as if there were no other approach.10 In addition, CDA does not properly belong to the discipline of linguistics, whose `applied-ness' is after all Widdowson's declared target. Fourteen times he slips into calling it `critical linguistics' (10±11, 15, 19, 22±23), the old name that was abandoned precisely because the linguistic framework had proven too restrictive. Also the examples he picks ®t the old term better than the new. Taking a di€erent tack, he detaches CDA from `discourse' and consigns it to `text': 59 In spite of its name, critical discourse analysis is . . . an exercise in text description. And it, too, has a way of assigning discourse signi®cance to textual facts. The supposed area of relevant application is here, however, much broader and of much greater moment (9). If `applications' are judged a danger to applied linguistics, as seems clear from sample 12, then this `broader area' is all the more threatening. Interestingly, a standard Widdowsonian defence from previous papers does

ROBERT DE BEAUGRANDE

117

not reappear, namely to deny the third term in the name of CDA: what Fairclough and others do is not `analysis' but `interpretation' (Widdowson 1995, 1997). This move was supported by requiring that `analysis' be `exhaustive', `arbitrary', and non-'selective' (1997: 153): 60 Analysis, in principle, takes everything into account that is encoded (1997: 153) This support doesn't apply now that Widdowson is arguing against the idea of `grammatical features' having `meanings encoded' in advance of the `text' (sample 23). Yet he reverts to the dubiously `Hallidayan assumption' (quoted from Stubbs) in order to reproach CDA for not taking everything into account: 61 in practice [you] select certain features and disregard the rest, and this is what Stubbs does. [But] in selecting a particular feature for special attention, you in e€ect accept that the disregarded parts of the text are redundant in that they are irrelevant to the identi®cation of ideological stance. But how does this square with the idea that `all linguistic usage encodes representations of the world?' (11f). Yet even if we accepted the extreme idea about `all linguistic usage', this argument is untenable. Selecting one feature need not imply the `irrelevance' of others, but merely the intention to keep one's analysis from getting too long or repetitive and incurring the wrath of editors and publishers. Whereas the cases against the linguistics of Halliday and Sinclair were argued with no data at all, the case against CDA adduces a few examples, albeit brief ones chosen mainly from Stubbs and all vulnerable to charges of over-interpreting. In one instance, he adopts the (for him) most unusual recourse of looking up data in a very large corpus (the BNC), anticipating that the data will go against Fairclough. Otherwise, he seems content in `making appeal to plausible pragmatic uptake' (20), which I suppose would be a `®rst-person perspective' in the sense of samples 54±56. This handful of data is duly adduced as grounds to repeat the charges of `treating the text' as a kind of `static semantic patchwork' (17, 22), directed before against Halliday in sample 23; and of claiming a `privileged status' for one's `own partial interpretations' (22). Without further evidence, a severe judgement is passed: 62 This example of what can only be called tendentious interpretation is not, regrettably, just a momentary lapse, for it seems to be endemic in a good deal of critical discourse analysis (18). Perhaps I might be allowed one bit of counter-evidence of the data CDA really wants to address: 63 An inspection carried out by the Council for Nuclear Safety . . . showed more than 1000 workers in Harmony Gold Mines . . . have received an annual radiation dose ®ve times higher than it should be. `Essentially,

118 INTERPRETING THE DISCOURSE OF H. G. WIDDOWSON

these workers were being fried,' said a council source. `They are not provided with protective clothing or even instruments that would allow them to measure radiation levels.' `Mining has a social bene®t and we can't make it so costly that workers' jobs are at risk. So perhaps you say radiation will kill you, but no jobs will also kill you,' said Anglogold representative Johan Botha (Mail & Guardian, Johannesburg, March 1, 1999). My intuitions expected the mining company to express surprise and regret, and promise some investigation or remediation such as protective clothing. But not so. The company's spokesman cynically o€ered its workers the choice between two ways to be `killed': radiation or joblessness. Disturbingly, the `empirical ethnographic considerations' which `locate the texts in its sociocultural settings' (Widdowson, 22) include the facts that the workers are black Africans and the spokesman is a white AfrikanerÐwhich, in the historical context of gold mining in South Africa, makes the statement leap into sharp perspective. I hope I am not being tendentious if I call attention to the expression `social bene®t'. I would adduce corpus data11 to back up my claim that a `manipulation' of the usual meaning has indeed been performed, such as (my emphases): 64 A number of important social bene®ts, such as human health, environmental quality, improvements in family life and community development, and public decision making are typically omitted from the social rate of return on a research investment. 65 Another aspect of this element of performance is what might be described as social bene®ts such as human rights, employment standards, employment equity. Equating `social bene®ts' with company pro®ts leads to the perverse logic of protecting jobs by frying job-holders. To me this presents an exquisite emblem for the new style of business known as `killer capitalism', portrayed with conclusive documentation by Martin and Schumann (1996). Much valuable work remains to be done here by CDA, as is now coming together at the internet website `Language in Neo-Capitalism'.12

2. THE CASES REST The three `cases' brought by Widdowson against three approaches have now been reviewed in light of textual evidence and may `rest' on their merits. By treating his discourse, and the discourse of the linguists he discusses, as smallcorpus data, I hope to have uncovered some interpretative possibilities that might otherwise not be realized. Any privileged status that might be claimed for my interpretation would have to emerge from my exegesis of the data, as contrasted with Widdowson's own proceedings of `arguing' from the

ROBERT DE BEAUGRANDE

119

`assumptions' and `claims' other linguists supposedly make, whilst disregarding what they have actually said in print. With consummate (though perhaps unintentional) irony, he announces six times in his paper, like a mantra, that the real value of other approaches lies in `provocation'Ðthey get people like him to straighten matters out through `critical responses' and `closer enquiries', for example: 66 I would argue that the value of [Sinclair's] proposed precepts is precisely because they provoke a critical response (9); 67 work in critical discourse analysis . . . provokes us to enquire more closely (10); 68 from the applied linguistic point of view, their importance [i.e., of corpus linguistics and critical linguistics] lies not in the answers they provide but in the questions they provoke (20). The sombre, cautionary tone of his paper could suggest that applied linguistics is in real danger of getting `absorbed' or deprived of its `occupation' (see footnote 5) by armies of `misappliers' `converging' in from the schools of Halliday, Sinclair, and Fairclough. Such a parlous vision seems a bit alarmist. All I can think of remotely moving in this direction is Tim Johns' (1991: 30) advocacy of using a corpus to `cut out the middleman' and `give the learner direct access to the data'Ðand only if the applied linguist were indeed a `middleman', which is hardly an apt ascription. I would see three other dangers. First, the very linguistics of `idealization' and `abstraction', which Widdowson evidently accepts, may atrophy within Chomsky's (1991: 88) precept that `your professional training as a linguist' `just doesn't help you to be socially useful'. Second, attempts to `apply' such a non-useful linguistics anyway may foster truly unwarranted `impositions', such as Krashen's (1985: 55) precept that `materials' and `lesson plans, etc. are not necessary', because he literally believes in a `language acquisition device' (Beaugrande 1997b). Such precepts could lead to a drastic retrenchment of linguistics as a whole, as already indicated by the closing of entire programmes or departments. At that stage, educators and administrators may mistrust all disciplines with `linguistics' in their names. Unless `applied linguistics' changes its nameÐas in South Africa, where `applied language studies' has taken holdÐa sound strategy for survival would be to seek active alliances with those countertrends inside linguistics that point to real language and social relevance, instead of warning about improper `extensions'. I see a third danger in attempting to judge in advance whether or not the research produced by such trends can or should be applied to classroom practices. That question should be answered in the classroom, and not preempted by theoretical deliberations, inaccessible to practising language teachers, about recondite dichotomies like `I-language' vs. `E-language', `text' vs. `discourse', or `analysis' vs. `interpretation'. As applied linguists, we should wait and see, respecting the competence of these teachers to tell us

120 INTERPRETING THE DISCOURSE OF H. G. WIDDOWSON

which applications work better or worse, and to join in discussions on a level they prefer. Revised version received August 2000

ACKNOWLEDGEMENTS I would like to thank John Sinclair and the two anonymous reviewers for helpful and careful work. I am also grateful to Henry Widdowson for many fresh and lively discussions, even though he remained adamant in all his positions. Other papers on corpus work in applied linguistics can also be downloaded at www.beaugrande.com.

NOTES 1 To conserve much journal space, I cite page numbers in Widdowson (2000) without the year. All emphases are mine unless otherwise noted as `his emphasis'. 2 The term recurs only once, and for unfounded `con®dence' in one's `privileged access to covert signi®cance' (11). 3 WordPilot has many other capabilities, including a TESOL and SAT word-test coach; a web browser for dictionaries, writing workshops, and tutorials; and a speech synthesizer for any text or web page. More information is found at the Website address http://www.compulang.com/. 4 Here and elsewhere I make minor changes in the grammar to ®t the context, such as turning participles into in®nitives. These should be allowed among the `dynamic interrelationships which grammatical features contract', as Widdowson acknowledges (17). Some minor inaccuracies caused by the scanning process may have survived my multiple proof-readings, for which I apologize. 5 In 1975, Sinclair and Coulthard published the ®rst British introduction to the `analysis of discourse'. A coincidence? 6 The quote, `Othello's occupation's gone', sounds piquant, since the occupation was `big wars that make ambition virtue' (Othello III iii, 357, 349). 7 More precisely, 1 is an Adjective agreeing with its Noun; 2 is a Dual Noun in

8

9

10

11 12

Apposition with the Noun; 3±10 are Status Constructus (called `iDaafa' [annexation] by Arab grammarians) with an Inde®nite Noun in the Genitive Plural; and 11±19 govern the Noun in the Accusative Singular (Wickens 1980: 98f; Holes 1995: 173€). Actually, Wickens (1980: 99) sees this `feature' `doubtless holding a number of implications in the areas of both psychology and anthropology', though he doesn't elaborate. Widdowson notes how `linguists still routinely talk about text as being stretches of language above the sentence' (10), but I suspect this talk has grown less routine and less complacent over the years. Widdowson makes no mention of the fact that his own unpublished thesis was on an Applied Linguistic Approach to Discourse Analysis (1973). His `approach' was presumably quite di€erent from those of Halliday, Sinclair, or Fairclough. These attestations were taken from a 500,000-word corpus of public speeches compiled by John Milton (see note 3). The URL of the Language in Neocapitalism website is: http://www.uoc.es/hum®l/nlc/LNC-ENG/ lnc-eng.html#Pinboard In the event of problems contacting it, a helpful person is Joan Pujolar at the Universitat Oberta de Catalunya, e-mail [email protected].

REFERENCES Aston, G. and L. Burnard. 1998. The BNC Handbook. Edinburgh: Edinburgh University Press.

Beaugrande, R. de. 1980±81. `Design criteria for process models of reading.' Reading Research Quarterly 16: 261±315.

ROBERT DE BEAUGRANDE

Beaugrande, R. de. 1991. Linguistic Theory: The Discourse of Fundamental Works. London: Longman. Beaugrande, R. de. 1997a. New Foundations for a Science of Text and Discourse. Greenwich, CT: Ablex. Beaugrande, R. de. 1997b. `Theory and practice in applied linguistics: Disconnection, conflict, or dialectic?' Applied Linguistics 18/ 3: 279±313. Beaugrande, R. de. 1998a. `Performative speech acts in linguistic theory: The rationality of Noam Chomsky.' Journal of Pragmatics 29: 1±39. Beaugrande, R. de. 1998b. `On ``usefulness'' and ``validity'' in the theory and practice of linguistics: A riposte to H.G. Widdowson.' Functions of Language 5/1: 85±96. Beaugrande, R. de. 2000a. `Large corpora and applied linguistics in language teaching: H.G. Widdowson versus J.McH. Sinclair' in P. Battaner et al. (eds.): Conferencias sobre la aplicacioÂn de los corpora linguÈõÂsticos en la ensenÄanza de las lenguas: VI Jornadas de Corpus LinguÈõÂsticos. Barcelona: Universitat Pompeu Fabri. Beaugrande, R. de. 2000b. `Text linguistics at the millennium: Corpus data and missing links.' Text 21. Beaugrande, R. de. In preparation. A New Introduction to the Study of Text and Discourse. Bygate, M. and C. Kramsch. 2000. `Editorial.' Applied Linguistics 21/1: 1±2. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge MA: MIT Press. Chomsky, N. 1986. Knowledge of Language. New York: Praeger. Chomsky, N. 1991. `Language, politics, and composition' (interview with G. Olsen and L. Faigley) in G. Olsen, Gary and I. Gales (eds.): Interviews: Cross-Disciplinary Perspectives on Rhetoric and Literacy. Carbondale: Southern Illinois University Press. 61±95. Halliday, M. A. K. 1985. An Introduction to Functional Grammar. London: Arnold. Halliday, M. A. K. 1994. An Introduction to Functional Grammar: 2nd Edn. London: Arnold. Holes, C. 1995. Modern Arabic: Structures, Functions, and Varieties. London: Longman.

121

Johns, T. 1991. `From print out to handout: Grammar and vocabulary in the context of data-driven learning.' ERL Journal 4: 27± 37. Krashen, S. 1985. The Input Hypothesis: Issues and Implications. London: Longman. Martin, H.-P. and H. Schumann. 1996. Die Globaliserungsfalle: Der Angriff auf Demokratie und Wohlstand. Reinbeck bei Hamburg: Rowohlt. Milton, J. 1999. `Lexical thickets and electronic gateways: making text accessible by novice writers' in C. Candlin and K. Hyland (eds.): Writing: Texts, Processes and Practices. London: Longman. 221±243. Sinclair, J. McH. 1991. `Shared knowledge' in J. Alatis (ed.): Georgetown University Round Table on Languages and Linguistics 1991. Washington, DC: Georgetown University Press. 489±500. Sinclair, J. McH. and D. Brazil. 1982. Teacher Talk. Oxford: Oxford University Press. Sinclair, J. McH. and M. Coulthard. 1975. Towards an Analysis of Discourse. London: Oxford University Press. Stubbs, M. 1994. `Grammar, text, and ideology.' Applied Linguistics 15/2: 201±23. Wickens, G. M. 1980. Arabic Grammar. Cambridge: Cambridge University Press. Widdowson, H. G. 1973. An Applied Linguistic Approach to Discourse Analysis. Unpublished Ph.D. Thesis, Dept. of Linguistics, University of Edinburgh. Widdowson, H. G. 1980. `Models and fictions.' Applied Linguistics 1/2: 165±70. Widdowson, H. G. 1991. `The description and prescription of language' in J. Alatis (ed.): Georgetown University Round Table on Languages and Linguistics 1991. Washington, DC: Georgetown University Press. 11±24. Widdowson, H. G. 1995. `Discourse analysisÐ a critical view.' Language and Literature 4/3: 157±72. Widdowson, H. G. 1997. `The use of grammar, the grammar of use.' Functions of Language 4/ 2: 145±68. Widdowson, H. G. 2000. `On the limitations of linguistics applied.' Applied Linguistics 21/1: 3±25.

# Oxford University Press 2001

REVIEWS Douglas Biber, Susan Conrad, and Randi Reppen: CORPUS LINGUISTICS. INVESTIGATING LANGUAGE STRUCTURE AND USE. Cambridge University Press 1998.

Until comparatively recently there was a shortage of books describing the general methodology and applications of corpus linguistics, but this has been remedied to some extent by publications on the subject over the last few years. This book, unlike some of its predecessors, does not examine corpus linguistics in detail as a topic for discussion in its own right: instead it explores it through the areas of language that it can be used to investigate. This approach is immediately evident in the structure of the book. After a brief introductory survey of the background to the corpus approach and its tools and resourcesÐ`The goals and methods of the corpus-based approach'Ð it considers areas of investigation of language subdivided into `the use of language features' and `the characteristics of varieties'. The ®rst of these covers lexicography, grammar (with a separate chapter on lexico-grammar), and discourse characteristics, the second register variation (including English for Speci®c Purposes), language acquisition and development, and historical and stylistic investigations. These two main sections are brie¯y summarized and followed by a consideration of corpus linguistic methodology. This takes the form of a set of `methodology boxes' in which corpus design, concordancing, tagging, frequency analysis, and statistical analysis techniques are dealt with in very brief summaries showing the main points to be taken into account. The main emphasis, then (215 pages out of 280), is given to language-based study of the potential of corpus linguistics, bearing out the authors' statement in the preface that they `chose to write a book that re¯ects our enthusiasm for corpus-based research . . . a book about language as well as corpus linguistics'. This is an interesting approach to the subject, and one which sets corpus linguistics ®rmly in its practical context. The subjects chosen to illustrate its uses, as already mentioned, are suciently closely linked for the book to work as a complete unit, rather than as a set of papers on unrelated topics. Lexicography, grammar, and lexico-grammar, in particular, form a usefully connected set of topics. The chapters dealing with them provide examples of possible approaches to the investigation of these areas, and highlight the di€erences in practical techniques needed for each of them together with the inescapable linkage between studies at word and at clause level. The chapter on discourse characteristics which completes Part I extends the study to investigations beyond clause boundaries. Part II uses a di€erent dimension as its basis as it introduces the

REVIEWS 123

investigation of general characteristics of language varieties, again using several illustrative types of study. While the areas chosenÐregister, language acquisition, and stylisticsÐare not so tightly homogeneous as in Part I, they provide a useful sample of linguistic research topics and show the universal relevance of the corpus-based approach. Part III summarizes the wide-ranging issues raised by Parts I and II before Part IV launches into a series of brief but very interesting methodological appendicesÐthe `methodology boxes'. These provide basic summaries, sometimes tantalizingly compact, of the main methods of corpus linguistics. The ten areas covered range from corpus design through software selection and tagging into statistical evaluation. At ®rst sight this may seem a rather bizarre organizational scheme for any book, resembling as it does a hybrid of the general text book and the collection of loosely related research papers, but for this particular subject it works rather well. Many publications that deal with corpus linguistics get bogged down in justi®cation of the approach or in the details of methodology. This often limits their contributions to the wider realms of linguistic research to such an extent that corpus linguistics begins to seem like a cult in which the methodology is revered for its own sake. This book avoids this problem precisely through its hybrid approach, which allows it to deal with any remaining need for justi®cation and the methodological basics alongside reports of real research. There are, inevitably, shortcomings. It is a compact book which deals with a wide range of issues, and they cannot all be dealt with adequately. The methodological boxes sometimes intrigue more than they inform, especially when they deal with statistical methods, which need rather more detailed treatment if they are to make practical sense. Similarly, the research areas highlighted cannot form a comprehensive survey of current applications of corpus linguistics. Neither of these criticisms, however, detracts in any serious sense from the value of the book, which makes excellent use of its restricted space and provides a very useful summary of the most important of current approaches to language analysis. (Received June 2000) Reviewed by Geo€ Barnbrook The University of Birmingham

124 REVIEWS

Lynne Hansen (ed.): SECOND LANGUAGE ATTRITION IN JAPANESE CONTEXTS. Oxford University Press 1999. Sandra G. Kouritzin: FACE[T]S OF FIRST LANGUAGE LOSS. Lawrence Erlbaum 1999.

The year 1999 saw the publication of two signi®cant books on language attrition: Lynne Hansen's Second Language Attrition in Japanese Contexts and Sandra G. Kouritzin's Face[t]s of First Language Loss. These two books show a wide variety in their research agendas and methodologies, which I take as an indication of the maturity of the sub®eld. Of the two, the former employs a more conventional methodology; that is, the 7 studies in it try to capture the deterioration of linguistic knowledge of the attriting language per se, mainly in the production of lexicon and morphosyntax. Tomiyama (Chapter 3), as well as two other chapters, attempt to capture a long-term change in so-called Japanese `returnee' children's L2 English. Tomiyama examined the English of one such subject, Ken, who came back from the US after almost 7 years there. The study aims to describe an overall change in the subject's English rather than to test a particular hypothesis. Ken's attrition was revealed in less ¯uency and greater diculty in lexical retrieval, whereas loss of morphology and syntax was only marginally observed, and little loss of productive phonology and receptive skills was found. As Tomiyama argues in her discussion section, these results are very similar to attrition patterns reported elsewhere. Though one sometimes wishes that one could look at the whole data, rather than the selected bits that are relevant and most convenient to endorse the author's argument, this manner of presentation, for its very simplicity and selectivity, helps readers understand the study. The next study, Yoshitomi's (Chapter 4), takes a very similar approach to Tomiyama's, examining the attrition of the highly pro®cient L2 English of four returnee girls. Although the results resemble Tomiyama's and those of previous studies, one of the highlights of Yoshitomi's study is the ®nding that it is in `the speakers' ability to utilize and to coordinate their linguistic knowledge fully' (p. 93: ll.23±24) that their attrition is revealed. Yoshitomi further presents a very insightful discussion on some commonly held notions about language attrition phenomena, such as `the earlier attrition of vocabulary than grammar,' `the initial plateau,' `residual learning,' etc. She argues that we need to reinterpret our data because attriters do cover up loss by using various compensation strategies which obscure the data, and we always have only a limited picture of the attriter's total changeÐlimited by the instruments we use in our attrition studies. The paper is stimulating and thought provoking. Readers are recommended to read her Ph.D. dissertation, of which this chapter is a condensed version, to fully appreciate the discussion presented in this paper.

REVIEWS 125

Reetz-Kurashige's study (Chapter 2) is the most ambitious of the three studies, in terms of the subjects included and the variables considered in a single study. The core part of the study is presented in the analysis of verb usage (target-like use or TLU) in story-telling/picture description data. One of the important claims that the author makes is that `children over eight years old and with at least two years of school abroad retained over 80% TLU' (p. 31: ll.18±19), and the ones who do not meet these conditions tend to lose their L2 English. Although the complex data do show a tendency for a high level of L2 English attained by older children to be retained, it is unfortunate that the author does not discuss details of controversial portions of the data within the limitation of space. I will point out here three of the puzzling areas in ReetzKurashige's analysis. First, when we observe TLU scores in her tables, we notice that at least ®ve subjects (Nos. 6, 7, 9, 15, and 18) violate her general rule. Secondly, we also notice mysterious gains in some subjects' TLU at Time 3 (the ®nal testing). If these gains are the results of English lessons upon returning to Japan, some measure should have been used to disentangle this variable. I believe these points deserve discussion by the author in this chapter. The third area that we need to be aware of is Reetz-Kurashige's rather unique use of the words `retention' and `attrition'. Literally, and in most attrition studies in the past, attrition refers to deterioration of certain language skills examined over time compared to those at a certain reference point, whereas the author uses the word `high retention' to mean 80 per cent of TLU in verb morphology at Time 3. Thus some subjects who lost very little of their TLU scores between the two test times but whose scores were low at both times are labelled as members of the `moderate' or `low retention' groups (e.g. Nos. 9 and 10). The chapter presents much more interesting information that this review could touch upon, especially the qualitative analysis of error types in verb morphology. If only more elaborate discussion of details and better proofreading in the tables were available, the chapter would have done full justice to the richness of the study. The four papers in Part II deal with L2 Japanese attrition. Russell's study (Chapter 5) examines 20 university students' L2 Japanese after returning from a two-year long missionary experience in Japan. Their production data were collected, and their lexicons in the data were analysed in terms of total word counts, lexical variability, lexical density, lexical accuracy, lexical frequency orders, frequency of ®llers, loan words, and English words. Russell dedicates three pages (pp. 116±18) to the discussion of the validity of his analytical instruments and the results of previously reported studies. It seems that other than face validity no measurement enjoys a well-established relationship with language pro®ciency levels. Possibly for this reason, or for some other plausible reasons such as the attriters' avoidance strategies (p. 127) and the subjects' being at the `initial plateau in the L2s of highly pro®cient attriters' (Hansen, in the introduction to the volume; p. 12), the results reveal

126 REVIEWS

very few areas of signi®cant di€erences between test times (i.e. attrition). Russell's study o€ers a lot of information on the 20 JSL speakers' lexicons, and it also presents a huge amount of statistical data, which will be a signi®cant reference for researchers who conduct similar studies in the future. Due to this very statistical treatment, however, I at times could not help feeling that these numbers calculated looked so remote from how the actual JSL speakers' monologues may sound. With few actual sample monologues, it was hard to imagine how the speech would come to sound when, for example, the errorfree T-units decrease by 0.8 in a group of 10 speakers from Time 1 to Time 3. The next two papers (Hansen's and Hayashi's) are the only hypothesis testing studies in this book. They both address the regression hypothesis, a relatively frequently addressed issue in the ®eld of language attrition. In both studies the negation formations elicited by picture cards were examined to see if the regression hypothesis is upheld, i.e, whether the order of attrition is, as the hypothesis predicts, the reverse of that of acquisition. Hansen's data from returned missionaries from Japan shows the predicted order of attrition. Knowing that some studies have found areas where the regression hypothesis was not upheld, Hansen very reasonably says that the questions that have to be addressed are `when and under what conditions its predictions hold true, and what the causal mechanisms are' (italics in original, p. 150: ll.8±9). In the case of Japanese negation, the frequency in natural input was the explanation of the acquisition and attrition order. Hansen's study is well designed, carefully carried out, and very clearly presented. It is a good example of how to study the regression hypothesis in naturalistic language attrition. Hayashi (Chapter 7) replicated Hansen's study, using elderly Micronesians, who were forced to learn Japanese at elementary schools (3±5 years) during the early twentieth century when the Micronesian islands were colonized by Japan. Hayashi's replication of Hansen's study is motivated by the di€erences of her subjects from Hansen's in terms of their L1 and their ages of learning Japanese (at elementary schools vs. as adults). The results show that the regression hypothesis was upheld. The important di€erence from Hansen's results was, however, the lack of di€erence in performance in Hayashi's study between the subjects who learned Japanese for a relatively short period and the subjects who were exposed to Japanese for a longer period. Hayashi's analysis of errors, juxtaposed with data obtained from interviews, found that formulaic expressions, especially emotion laden ones (mostly used by the teachers to scold the pupils) were retained well. Thus, Hayashi points out the importance of a€ect in language retention for future studies. This replication is also well designed, adding a new ®nding to the original. In the last paper in the volume, Nagasawa (Chapter 8) deals with seven graduate students in an American university who spent the summer prior to the research period in Japan. These subjects, unlike most attrition study subjects who have little exposure to the attriting language, were taking Japanese classes during the 7 month research period, and they had access to Japanese to various degrees outside of these classes depending on their

REVIEWS 127

eagerness to improve their Japanese. The author tries to address the e€ect of factors such as the initial pro®ciency level, motivation and attitudes, and language learning backgrounds in attrition. Studying language attrition while subjects still have some access to the language is obviously more complicated but is a necessary line of study. This should be attempted more often in the future, because this is a very frequent language learning situation, and thus such studies have a high possibility of being able to o€er pedagogical implications. Although Nagasawa's intention is well taken and respected, unfortunately her study design is not the best one to disentangle the complicated phenomena, because none of the many variables was controlled beforehand and with only seven cases it is impossible to sort them out using statistical treatment. With the present number of studies compiled in the volume, it is dicult to cover all the issues and theories important in the ®eld of attrition. The studies vary in their degree of focus as well as in the depth of analysis and discussion. Nevertheless, the studies compiled here are very informative and interesting for educators and researchers. Particularly those who work with the Japanese language should, by all means, read this book. Sandra G. Kouritzin's Face[t]s of First Language Loss was written for a considerably di€erent goal, using a method far removed from what we ®nd in Hansen's book. At the outset Kouritzin clearly states that her study was motivated by the desire `to understand the meaning of the experience [of ®rst language loss] from an insider perspective'(p. ix: ll.7±8). Thus the book does not seek to analyse linguistic change, nor does it pursue the cause(s) of loss, which has been the primary targets of, respectively, psycholinguistically oriented and sociolinguistically oriented attrition studies. Furthermore, because the author wanted to privilege `individuals' understandings of a phenomenon over the phenomenon itself ' (p. 20: ll.17±18), the method of `life history' was chosen, which avails us a private and retrospective perspective. By approaching the issue in this way, Kouritzin claims, the study can describe the intersection between language, identity, culture, and marginalization in former minority-language speakers. This introduction is followed by ®ve selected narratives. It is impossible to summarize the ®ve life histories in the way I did with the seven studies in Hansen's book, because every story refers to many issues, and each one does so using powerful episodes; only with such full contextual information can the interviewees communicate their perspectives. These ®ve stories, if read in their entirety, give readers a good clue as to how serious an e€ect the loss of ®rst languages had on the speakers' lives. However, if a reader cannot a€ord to take the time to tune in to listen to the interviewees' stories in their complete form, what he or she can do is to read the excerpts of the 21 stories around the ®ve themes (family relationships, self-image and cultural identity, school relationships, school performance, and meaning of loss) in Section II. Since, in this section, the stories are further edited and reorganized according to the speci®c themes, readers

128 REVIEWS

would have an easier time digesting the messages, by taking them a bit at a time. Loss of the ®rst language resulted in a number of negative consequences in family relationships. It sometimes led the young to separation from extended family members. In the cases when the parents did not develop English, however, it even caused the lack of means of communication between parents and children. Some life stories reveal occasions when children are the family link to the world outside of home, and children often edit information (deceiving the parents) to get bene®t or avoid harm. In such cases, parents rarely keep up with children's growth in the dominant culture, which makes children feel ashamed of their parents. Later in life, some bear a grudge against their parents for not having forced them to maintain their ®rst languages. Negativism lurks in their self-image and cultural identity as well. Kouritzin says that their negative image of their own cultural heritage, when turned inward, manifests itself in being ashamed of their heritage and aspirations for total assimilation into the dominant culture, while, turned outward, it takes the form of racism against their own race. In both cases, many still feel that they belong to neither culture. As for school relationships, one important course of ®rst language loss during dominant language acquisition was revealed in this study. Contrary to the commonly held notion that immigrant children fail in school because their parents cannot a€ord to support their children's education, Kouritzin found that the very enthusiasm of the parents, coupled with erroneous guidance from school, also drives children into monolingualism. Memories related to school performance show how some children have no choice other than cheating in doing their homework when teachers do not o€er sucient help and still force them to do work they are in no way able to accomplish. The meaning of loss, Kouritzin found, varied among the 21 interviewees. For some, it meant the total or partial loss of linguistic abilities. Loss of ease in using the minority language to some meant leaving the community which speaks the languageÐthe feeling that one can never `go home' again. What ®rst language loss implies, Kouritzin says, seemed to di€er, at least to some extent, depending on the interviewee's age, in other words, what stage of life they were at. It meant a lost opportunity for a particular career, an inability to understand many facets of their parents and their cultures, and the loss of something too deep for words. The book is absolutely unique in terms of its goal and its methodology. In fact I cannot think of any other work that shares even one of these two elements with this book. The book is easy to read, yet highly academic. Like many other qualitative research methods, the one employed in this book might look simple, but it is obviously not. This research project seems to be possible only with this author's ¯exibility in human interaction, clear analytical insight in observing her own reactions and thoughts in

REVIEWS 129

encountering a unique life story each time, and artistic, articulate, and scienti®c ways to present the complex data. (Revised version received June 2000) Reviewed by Emiko Yukawa Notre Dame Women's College of Kyoto, Japan

Guy Cook: LANGUAGE PLAY, LANGUAGE LEARNING. Oxford University Press 2000.

Recently within the ®elds of Applied Linguistics and Second Language Acquisition, a smattering of articles (cf. e.g. Cook 1997; Lantolf 1997; Ohta 1995; Sullivan 2000; Tarone 2000) and conference presentations (cf. e.g. Belz 1997, 2000; Broner and Tarone 1999; Ohta 1998) on the importance of language play to second language acquisition has begun to appear. This wellwritten and insightful text, in which Cook rightly argues for an intensi®ed exploration of this ubiquitous, but often marginalized phenomenon, represents the ®rst book-length contribution to the theoretical relation between language play and language learning from a strictly academic perspective.1 Language play, Cook (p. 5) assumes, `involves the patterning of linguistic form, the creation of alternative realities, and the social use of both of these for intimacy and con¯ict.' Play in general and language play in particular have been avoided in many contexts because they are considered to be `immature, trivial, and super¯uous' (p. 186), impediments to the work-oriented focus of contemporary, industrialized Western society. This aversion to the perceived non-serious quality of play has carried over into current utilitarian language teaching orthodoxies which emphasize grammatical and semantic norm approximation, frequently in the form of transactional classroom tasks. To counter this laic characterization, Cook draws on an impressively broad body of research to discuss the potentially innate basis of language play and its role in cognitive adaptation and ¯exibility, ultimately arguing, in the case of language teaching methodologies, `to make the periphery the centre and the centre the periphery, so that language play is no longer seen as a trivial and optional extra' (p. 204), but rather as both a means and an end in second language learning. The text is divided into three parts, which re¯ect the description (Chapters 1±3), theory (Chapters 4 and 5), and pedagogic applications (Chapters 6 and 7) of language play, respectively. Part I is subtitled The Interlocking Levels because Cook posits a causal link between the three levels of language play: the patterning of linguistic form creates alternative realities which are used to e€ect both social collaboration and competition. In Part II, Theories and

130 REVIEWS

Explanations, language play is characterized from both evolutionary and sociocultural perspectives. Cook's use of the term `sociocultural' should not be confused with current Vygotskian approaches to second language research and learning; his sociocultural account of play draws on Huizinga (1949) and Caillois (1969), not Vygotsky. The basic claim, which is surely reasonable in relation to social language play, is that `a theory of play is stronger if it acknowledges the complementarity of the two positions, rather than seeing them in opposition' (p. 125). In Part III, Language Learning, Cook (p. 6) does not argue for language learning through popular notions of fun and games, but rather for the incorporation of a play element into language teaching and learning that would `exploit the universal human liking for competition, ®ction and arti®ce, repetition, and the manipulation and analysis of form.' It would be interesting to consider the universal human liking for competition in Cook's account of language play against the backdrop of language and gender. In her well-known analysis of conversation in primarily Western society, Tannen (1990) suggests that men orient toward competition and oneup-manship, while women tend to engage in more supportive and collaborative conversational moves. Advantages of the incorporation of a play element into language teaching methodology are outlined in admittedly programmatic fashion (p. 194) in Chapter 7. At the outset of Part I, Cook (p. 11) clari®es that his aim is to theorize the child's (and later the adult's) reception of playful language; thus, the reader should not expect learner-produced exempli®cation of this phenomenon, but rather an examination of the `language in which the child is silently absorbed.' Further, the choice of reception narrows the object of investigation to primarily social language play. Characteristic of this language is linguistic patterning in the form of rhyme and rhythm (a manifestation of repetition) and its co-occurrence with social interaction as seen, for example, in the 4/4 stress verse of tickling rhymes, e.g. This Little Pig. It is suggested that children have innate predispositions to both rhythm (grammar) and social interaction (pragmatics). An understanding of conventionalized sign±meaning pairings (semantics) may come about as a result of these two innate responses. As Cook (p. 23) writes: `Rhythm and interaction provide a pathway to language through existing gateways.' Thus, a central claim of this book, that meaning may arise from random form (pp. 192±3), is rooted in biological inheritance. Fiction is one type of meaning which arises from the patterning of linguistic form.2 As Cook (p. 36) describes it, ®ctions, or alternative realities, appear to be social entertainment created through language, e.g. dramas, novels, and soap operas. Although ®ction functions in social education and the establishment of group solidarity, its primary function is to allow for greater cognitive ¯exibility and adaptability in unforeseen circumstances. In this way, language play is again tied to biological inheritance and evolutionary advantage. Fantasy, which is `created in the mind but not shared with others' (p. 39), is posited to be a precursor to ®ction. Cook's description of fantasy is strikingly similar to Vygotsky's (1986) uncited notions of private and inner speech. For

REVIEWS 131

example, it is `less likely to be structured or articulated in a way which is clear to a second party'; it is `not social'; it can occur sub-vocally when the speaker thinks he is alone; if overheard by others, it is a `source of embarrassment.' It is interesting to compare the status of this phenomenon in Cook's theory of language play and Vygotsky's theory of the development of higher psychological processes. For Cook, it is a use of language which operates in the externalization of an innate response to the patterning of linguistic form, the production of ®ction for social consumption. For Vygotsky, it is a use of language which operates in the internalization of social constructs for the development of the human mind. The uses of language play are the focus of Chapter 3. In line with Wolfson's (1988) `bulge theory' of discourse, Cook writes that language play typi®es non-transactional, emotionally charged interactions between either intimates or rivals. Therefore, uses of language play can take the form of verbal duels, jokes, riddles, and puns. In analogy to ball games, language play is paradoxical; it can serve collaboration, such as the gentle passing of a ball from father to son, or it can be aggressive and disruptive, such as the fastball of the opposing pitcher. The disruptive function of language play can often be seen in humorous uses of language. Raskin (1985) suggests that script incongruity, roughly, the ¯outing of co-occurrence expectations, is one source of humor. In my opinion, the conceptualization of language play as an inversion of the established order (p. 84), whether grammatical, semantic, or pragmatic, may be an appropriate interpretive frame for investigations of learner identity and agency in post-Native Speaker approaches to SLA where learners are not necessarily conceptualized as defective communicators in pursuit of an idealized target language native speaker norm (cf. Firth and Wagner 1997; Roebuck 2000). In such an analysis, rule subversion may not be indicative of de®ciency, but rather an indication of the learner's privileged stance on L2, his informed right to ¯out. The major contribution of Chapter 4 is the presentation and examination of Caillois' (1969) fourfold division of play into agoÃn (competition), alea (chance), mimicry (simulation), and ilinx (vertigo) with a qualifying continuum which oscillates between ludus, or rules, and paedia, or `improvisation and gaiety.' (p. 115). For example, spontaneous races or games of tag may register on the paedia end of the agoÃn division, while chess may belong on the ludus end. According to Cook (p. 115), the origin of agoÃn is relatively easily traceable to animal behavior, while mimicry may serve a learning function.3 The role of alea is less readily explained and consequently becomes the focus of Cook's own examination. At ®rst glance, Cook's de®nition of language play, given in Chapter 5 on p. 123, is dangerously, but only super®cially and, in actuality, visually reminiscent of necessary-and-sucient-feature approaches to lexical semantics (i.e. lists of de®ning characteristics). Cook (p. 113) wisely notes the ease with which the explanatory power of such an approach can degenerate into `quibbles over one criterion or another' and ultimately adopts, a la Huizinga,

132 REVIEWS

`the kind of ¯exible and adaptable de®nition which an activity such as play demands' (pp. 113±14). In my opinion, this de®nitional stance is in line with previous play researchers (e.g. Fromberg 1999: 27; Garvey 1977: 5; Peck 1980: 154) who stress that play is not a list of particular activities but rather a manner of performing any activity. Although Cook does not technically say so, he appears to interpret language play and play in general as a pluricentric or `radial' prototype category (cf. Lako€ 1987: 91±114) with the four types of play suggested by Caillois serving as independent, but systematically related foci (the nature of these concept-internal systematic inter-relations is, perhaps, the topic of another book). The recalcitrant concept of play might pro®t from an explicit recasting of Caillois' de®nitional schema in terms of a radial category. In Cook's theory of language play, it is alea which takes centre stage and the innate quality of this phenomenon is further underscored in the remainder of Chapter 5 by drawing an, at times admittedly limited (p. 143), analogy between alea in language play and genetic mutation. In short, Cook (p. 140) argues that language play is a means by which R-random elements enter the language system and re¯ect and rebound upon it `underwriting change, dynamism, and adaptability.' In two well-written and judicious ®nal chapters, Cook discusses the distinctions between the metaphors of work and play (although he does not explicitly refer to them as conceptual metaphors) in second language acquisition research as well as their implications for and re¯ections in current language teaching orthodoxies, in particular, task-based instruction. Task, Cook argues, has been associated with work, not play, and, therefore, elements of language play, in particular, its focus on form-driven meaning and its manifestation in emotional, intimate, humorous, and aggressive discourse, have been under-represented in those types of task-based instruction which take transactional, needs-based interaction as their core. In his last chapter, Cook provides a very helpful schematization (p. 191) of various functional taxonomies of language and pro€ers that Jakobson's (1960) taxonomy be augmented with a seventh, ludic function which allows (requires) a back¯ow between meaning and form, and, thus, a justi®cation for the pursuit of form-driven creativity in language learning and teaching. The adoption of this suggestion has ®ve advantages for language learning: (1) the validation of explicit rule instruction; (2) the authentication of a focus on form; (3) the authentication of arti®ce; (4) the expansion of permitted instructional patterns within the classroom; and (5) the co-existence of change and tradition. (Received June 2000) Reviewed by Julie A. Belz The Pennsylvania State University

REVIEWS 133

NOTES 1 David Crystal's (1998) Language Play takes a more popularized approach to the phenomenon, although it does brie¯y consider the relation of language play to (second) language learning. Cook (p. 7, fn. 4) remarks that Crystal was published as he was ®nishing the current text. 2 A detailed, if speculative and ultimately analyst-imposed, account of how this may occur is given on pp. 138±9 for the case of a playful insult. It is somewhat more dicult to imagine how the process might work in the case of a novel or a soap opera. 3 Language play as mimicry may pro®t from an exploration of the Vygotskian construct of internalization, whereby imitation forms one of the primary mechanisms of learning. Internalization, however, is not mere mimicry, but rather transformative in nature.

REFERENCES Belz, J. 1997. Discourse analysis at the cultural faultline: Capitalizing on the multilingual reality of language study. A paper presented at the American Association for Applied Linguistics, Orlando, Florida. Belz, J. 2000. Language play. A Colloquium presented at the American Association for Applied Linguistics, Vancouver, Canada. Broner, M. and E. Tarone. 1999. Is it fun? Language play in a ®fth grade Spanish immersion classroom. A paper presented at the American Association for Applied Linguistics, Stamford, Connecticut. Caillois, R. 1969. `The structure and classi®cation of games' in J. W. Loy and S. Kenyon (eds.): Sport, Culture, and Society: A Reader on the Sociology of Sport. London: Macmillan. (Original work published in 1955.) Cook, G. 1997. `Language play, language learning.' ELT Journal 51/3: 224±31. Crystal, D. 1998. Language Play. London: Penguin. Firth, A. and J. Wagner. 1997. `On discourse, communication, and (some) fundamental concepts in FLA research.' The Modern Language Journal 81: 285±300. Fromberg, D. 1999. `A review of research on play' in C. Seefeldt (ed.): The Early Childhood Curriculum: Current Findings in Theory and Practice. New York: Teachers College Press. 27±53. Garvey, C. 1977. Play. Cambridge, MA: Harvard University Press. Huizinga, J. 1949. Homo ludens. London: Routledge and Kegan Paul. (Original work published 1944.) Jakobson, R. 1960. `Closing statement: Linguistics and poetics' in T. A. Sebeok (ed.): Style in Language. Cambridge, MA: The MIT Press. 350±77. Lako€, G. 1987. Women, Fire and Dangerous Things. What Categories Reveal about the Mind. Chicago: University of Chicago Press. Lantolf, J. P. 1997. `The function of language play in the acquisition of L2 Spanish' in A. PeÂrezLeroux and W. R. Glass (eds.): Contemporary Perspectives on the Acquisition of Spanish. Somerville, MA: Cascadilla Press. 3±24. Ohta, A. S. 1995. `Applying sociocultural theory to an analysis of learner discourse: Learnerlearner collaborative interaction in the zone of proximal development.' Issues in Applied Linguistics 6/2: 93±121. Ohta, A. S. 1998. The role of language play in the acquisition of foreign languages by adult learners: Evidence from the classroom. A paper presented at the American Association for Applied Linguistics, Seattle, Washington. Peck, S. 1980. `Language play in child second language acquisition' in D. Larsen-Freeman (ed.): Discourse Analysis in Second Language Research. Rowley, MA: Newbury House Publishers. 154±64. Raskin, V. 1985. Semantics Mechanisms of Humour. Dordrecht: Reidel. Roebuck, R. 2000. `Subjects speak out: How learners position themselves in a psycholinguistics task' in J. P. Lantolf (ed.): Sociocultural Theory and Second Language Learning. Oxford: Oxford University Press. 79±95. Sullivan, P. 2000. `Playfulness as mediation in communicative language teaching in a Vietnamese

134 REVIEWS

classroom' in J. P. Lantolf (ed.): Sociocultural Theory and Second Language Learning. Oxford: Oxford University Press. 115±31. Tannen, D. 1990. You Just Don't Understand: Women and Men in Conversation. New York: Morrow. Tarone, E. 2000. `Getting serious about language play: Language play, interlanguage variation and second-language acquisition' in E. Tarone, M. Anderson, C. Klee, F. Morris, and B. Swierzbin (eds.): The Interaction of Social and Cognitive Factors in SLA: Proceedings of the 1999 Second Language Research Forum. Somerville, MA: Cascadilla Press. Vygotsky, L. S. 1978. M. Cole, V. John-Steiner, S. Scribner, and E. Souberman (eds.): Mind in Society. The Development of Higher Psychological Processes. Cambridge, MA: Harvard University Press. Vygotsky, L. S. 1986. A. Kozulin (ed.): Thought and Language. Cambridge, MA: The MIT Press. Wolfson, N. 1988. `The bulge: A theory of speech behaviour and social distance' in J. Fine (ed.): Second Language Discourse: A Textbook of Current Research. Norwood, NJ: Ablex.

A. Suresh Canagarajah: RESISTING LINGUISTIC IMPERIALISM IN ENGLISH TEACHING. Oxford University Press 1999.

Robert Phillipson's book Linguistic Imperialism (1992) began an important debate by postulating that there was a vast hegemonic intent (conscious or not) by the ELT community. A. Suresh Canagarajah's book Resisting Linguistic Imperialism in English Teaching (1999) continues this debate but from a very di€erent perspective. Canagarajah centres his eight-chapter book on the following premises. First, since language is not value neutral, ESL mainstream pedagogies (MP) (those developed in `centre' countries such as the USA and Great Britain) cannot be applied in `periphery' countries (such as India and Sri Lanka) because they `fail to accommodate the type of questions arising in the periphery educational context' (p. 13). Second, students and teachers must resist MP thus enabling them `to develop critical consciousness and resist domination' (p. 31). And third, students and teachers must not only resist MP, they must appropriate English to ®t their own local needs (p. 77). Canagarajah claims that Phillipson's macroscopic theoretical perspective failed to capture the reality of periphery classrooms. Canagarajah seeks to correct this by using a micro-social approach. He uses an ethnographic methodology to document and analyse how MP is ine€ective or at least suspect in his native Sri Lanka. He provides classroom data, interviews, analysis of `marginalia' (notes students have written in their textbooks), materials analysis, and case studies of three student writers. All of these highlight how the implementation of MP in this periphery country is inappropriate. Chapters 1 to 3 give a theoretical overview of critical pedagogy (CP) and demonstrate how Sri Lanka is ripe for its implementation. CP is an educational approach that recognizes that learning is not `a detached cognitive activity' but rather a personal, political and cultural experience (p. 15). According to Canagarajah, English language professionals (particularly in the periphery) should approach their class from a CP perspective which allows

REVIEWS 135

students, as members of the community, to construct knowledge instead of being passive recipients of a pre-packaged pedagogy. Chapter 1 opens this theoretical section with a vignette about an English class in Sri Lanka where the teacher bases her lesson on a text that describes an ideal student life in Great Britain. This is juxtaposed with the violent reality of Sri Lanka where students may or may not be able to attend class due to civil war. The teacher in this vignette is painfully unaware of how her choice of materials and teaching methods are inappropriate and deny any chance for students to resist Western cultural imposition. To remedy this situation, Canagarajah advises teachers to avoid a centre-based pedagogical approach. He concludes these three introductory chapters by giving a detailed history of English in Sri Lanka. Canagarajah notes that despite its colonial heritage, there is a growing L1 monolingual elite in Sri Lanka that rejects English. Chapters 4 to 7 are perhaps the most enlightening. In these chapters, Canagarajah describes how students and teachers resist cultural domination in English classes. Students do so by scribbling in their text, focusing on discrete grammar points and codeswitching; all of which serve to avoid and resist the cultural values inherent in an MP approach. Sri Lankan English teachers also resist Western colonial heritage, even though it might not always be on a conscious level. Their resistance most often surfaces as inconsistencies between their professed teaching philosophies and their classroom implementation. These include: professing that English is the sole medium of instruction yet codeswitching with the vernacular; and claiming to teach standard British/American English while using Sri Lankan English in the classroom (p. 110). The book concludes with Canagarajah o€ering his own solutions to ELT linguistic imperialism in periphery settings. He maintains that while students should be taught the contextual appropriateness of certain varieties (including a standard variety), `they must also be taught that any dialect has to be personally and communally appropriate in order to be meaningful and relevant for its users . . . (thus) leading to the pluralization of standards and democratization of access to English' (p. 181). The strength of this book lies speci®cally in the picture it paints using the data listed above. It provides an invaluable glimpse into student reactions to English in this one periphery setting. Indeed, we see that students do want to resist the imposition of cultural values that ELT often brings with it while still wanting to learn English. The book also provides insight into the challenges for English teachers as more students and governments begin to question the validity of traditional ESL methods. Canagarajah provides some suggestions for how students and teachers may begin to resist these traditional methods. These suggestions go beyond simply developing communicative competence; students and teachers must also re¯ect on the ways in which English is used in their contexts and develop pedagogies accordingly. Canagarajah also provides a response from an author of one of the so-called mainstream

136 REVIEWS

texts used in Sri Lanka (p. 100) that he criticizes for promoting Western values. The inclusion of this debate is fascinating. The book does have several shortcomings. More examples of the ethnographic data would have added to the largely theoretical perspective. In addition, Canagarajah says that he is doing an ethnography of one speci®c ELT context, but he writes as if the results are applicable to any periphery classroom (p. 79). The book's biggest weakness lies in the author's stance on how students should overcome the cultural domination of the West. After discussing at some length how students must appropriate English for themselves, he makes the following statement: `Although students display their strategies of linguistic appropriation in largely non-re¯ective ways, the wider rami®cations of this option need to be theorized with the help of periphery scholars who have given this matter some thought' (p. 182). Why must scholars be the leaders in this linguistic appropriation? If Canagarajah wishes to be faithful to his theoretical perspective, then he must allow students to complete the work they have started with teachers acting as facilitators rather than scholars theorizing. Despite these shortcomings, the book provides a much needed glimpse into a periphery ELT setting and can be an invaluable tool for ELT professionals interested in periphery contexts. It will make educators aware of the complexity of issues surrounding ELT as well as how the centre often hinders what it claims to promote. (Revised version received May 2000) Reviewed by David Johnson Kennesaw State University Georgia, USA

Deborah Cameron: GOOD TO TALK? LIVING AND WORKING IN A COMMUNICATION CULTURE. Sage 2000.

What a nice start! Why would anyone complain? The smiling female ¯ight attendant looks me in the eye, wishes me good morning, addresses me as `sir', and directs me deferentially to my seat. Her friendliness persists through the long ¯ight: always a smile, never a cross word. Then there are lengthy announcements: the steward's cheery o€ers, the pilot's reassuring drawl. The words are not the speakers' own, though. These same phrases can be heard on all the ¯ights run by this airline. We have all grown used to this kind of behaviour from service providers. We ®nd it pleasant, annoying, or ridiculous, depending on mood or personality. On this occasion I found it mildly annoying. The announcements

REVIEWS 137

seemed prolix and uninformative, and they were stopping me reading my book. As I read on, however, I began to pay them more attention. My book was Deborah Cameron's Good to Talk?. It concerns a phenomenon of which cabin crew behaviour is an example: the `styling' of human beings to speak and act in designated and pre-scripted ways, no matter how badly the customer treats them, and no matter what their own personalities, or feelings at the time. In Cameron's view it is a phenomenon of major political and linguistic signi®cance. In the new economic order of expanding service industries, such `language work' and `emotional labour' are so commonly encountered, and constitute the occupations of so large a proportion of the workforce, that they can no longer be treated as marginal phenomena. The classic example is the worker in the telephone call centre, the subject of the book's fourth chapter `Communication Factories'. She (or much less frequently he) endlessly repeats the same script, and is in danger of losing her job if she departs from itÐby answering back, for example, or engaging in time-wasting small talk. Through long shifts, subjected to random or undercover monitoring, she always talks other people's words, sometimes dictated down to the ®nest detail (`How may I help you today' and not just `May I help you today'). She always has to be both ecient and politeÐ imperatives which, as we know from Pragmatics, often pull in opposite directions. Cameron documents e€ectively the exhaustion and frustration of people trapped in this kind of work. Many of her informants were simply scared to talk openly. They are temporary low-paid workers, often in economically disadvantaged areas, who have little support from trade unions or politicians. Their situation does not conform to a classic model of oppressive management. It is not a question of people being made to do something (and thus perhaps having the freedom to be surly or grumpy as a form of protest while doing it) but being made to speak and communicate something: a constant friendliness they simply may not feel. This is surely an issue for applied and sociolinguistics. Yet it is one that has elsewhere elicited little comment. Potentially relevant approaches have been preoccupied with higher status arenas. Critical Discourse Analysis, for example, has largely (though not entirely) concentrated on manipulation of the population at large in public discourse such as advertising, the press, and political propaganda. It has, in addition, been over preoccupied with text as object, treating ideology as something wholly encoded, rather than as a matter of interpretation emerging in the interaction of texts with actual people in variable contexts. Another approach from which one might have expected interest is the growing study of workplace communication. Here there is attention to user interpretation, revealed in protocols and interviews. Yet, as Cameron points out, most studies to date are of the talk of professionals, who are allowed much greater freedom than the lowly call centre worker. Cameron escapes both of these limitations. As she explains in an appendix

138 REVIEWS

on research methods and ethics, her data are predominantly from two sources, and her conclusions derive from relating the two. The ®rst source of data is textual: written materials seeking to regulate spoken discourse (`workplace training materials, employee manuals and appraisal checklists, educational policy documents, examination syllabi and assessment criteria, and self-help, and advice literature'). The details of such advice sources and the behaviour they recommend are often bizarre. In one McDonalds training session for example, a smartly dressed young man walks up to a slightly younger woman. He asks: `how are you feeling today?' `Fine,' she replies. `That's good,' he says, `but when I ask you how you're feeling, I want you to say `outstanding'. As he utters the last words, his voice rises in pitch and intonation; he punches the air with both hands. Cameron's second source of data is observation and interviews of `people engaged in the practices under investigation'. By uniting the two, she develops an argument for the major importance of her topic, not only in itself (because of the extensive distribution of the texts in question and the numbers of people involved in the practices) but as a key to attitudes to language use in the new global capitalist society in general, and thus to the nature of changes in that society itself. For the `styled' communication in the service industries, though it forms a striking and memorable centre-piece for her argument, is by no means its whole. Cameron approaches it as part of a larger phenomenon: the obsession of modern society with `good communication' (as re¯ected for example in the new emphasis on spoken communication skills in school syllabuses) and the assumption that all ills can be healed if only we talked to each other `better'. All we need to do, as one platitudinous advice manual puts it, is to `reach down for the smile deep inside us'. This faith in talk, Cameron argues, leads to underlying injustices and disagreements being ignored. And it is an international phenomenon. These attitudes and practices are invading languages very di€erent from English, and cultures which were previously distinct. It is in this expansion of the argument that the major achievement lies. This is one of those booksÐin my view the best kind of academic workÐwhich takes a very speci®c phenomenon, and shows it to be the key to issues of much larger signi®cance. It is clear to us all that human life has recently undergone immense changes, and it would seem likely, given this fact, that language use has undergone changes too. This, more than any other book I have read, seems to have put its ®nger adroitly upon what those linguistic changes are. Its achievement is to have hit upon a major and ominous development: the notion that communication can be styled, improved, and controlled. Relentlessly, it reveals the contradictions within that stance. `Be yourself ' urges one particularly platitudinous self-improvement course (BT's `Talk Works: How to get More out of Life through better conversation') while simultaneously urging people toÐin e€ectÐbe someone else, by changing the way they speak. `Good' communication, Cameron argues, can not be legislated. It must be determined by speci®c people in speci®c circumstances.

REVIEWS 139

There can be no general all-purpose training manual, however much employers would like one. In amusing illustration of this point, Cameron tells of an employee of WalMart supermarkets slavishly following his `oath' that `every time a customer comes within ten feet of me, I will smile, look him in the eye and greet him. So help me Sam.' He became the subject of a complaint from a woman customer who happened to cross his path repeatedly while he was shelf stackingÐand got eye-contact and a smile every time. Cameron is known for her writings on feminismÐand most of the workers whose plight she considers in this book are women. Women have remained the majority in service work, despite the numbers of unskilled males left unemployed by the decline in heavy manufacturing. The reasons for this bias are complex. Not only has `language work' traditionally been lower-paid than manual labour, it also draws upon qualities of verbal behaviour traditionally (and in some feminist linguistic literature) associated with womenÐmore `face work' and attention to others. Given this, it would have been easy to slip into a simplistic feminist discourse, describing another instance of male oppression. One strength of the book is that it does not do so, but rather views `the communication society' as oppressive of both sexes. As a male reader, I found the understanding of contemporary boys and men impressive for the empathy it demanded from a female writer. Citing the way educational assessment classi®es more and more boys as under-achievers and poor communicators, Cameron argues convincingly that such a trend says less about the shortcomings of the male population than about the changes in the behaviour which society demands from its lower status workers. Cameron seems wary of speculation, and everything she says is meticulously documented. The argument might have bene®ted, however, from more speculation on the underlying causes for this polarization of the traditionally male and female communication skills, and the favouringÐat least among the less advantaged sections of the populationÐof the latter. I could not help wondering (though Cameron might not like such evolutionary psychological conjectures) whether this growing emphasis on avoiding confrontation might be one side of our species' reaction to overcrowding. The other side is the straightforward aggression evident, on the macro scale, in the growth of intranational wars, and, on the micro scale, in such phenomena as road rage, and playground shootings. I found this book entertaining, moving, inspiring, informative, and intellectually stimulating. It has altered my ideas about applied linguistics, and about what happens in my daily life. I was still immersed in it as the plane commenced its descent, amid the usual surfeit of unwanted announcements. What were they telling me? The captain and his crew had enjoyed having me on board! It had been a pleasure looking after me! They hoped I would ¯y with them again, and have a safe onward journey! They even looked forward to seeing me again! As I left the plane the ¯ight attendants smiled and caught my eye, and wished me (and a few hundred others) a personal goodbye. What

140 REVIEWS

am I to make of this? Applied linguistics should surely be able to inform opinion about common current language practices. One of Cameron's major achievements, here and in earlier work (especially 1995), is to have reintroduced to us the notion that, instead of merely retreating behind the old adage that linguistics is descriptive not prescriptive, we should have something to say about the prescriptions which dominate popular discourse about language use. Her writing demonstrates that evaluation can be incorporated into scholarly description and rational argument in ways which are neither crassly prescriptive nor merely assertive. Implicit in Cameron's criticisms of the communication society, is a strongly felt notion of what communication ought to be, of what really does make it good to talk. If one of the most valuable functions of language is to establish honest relationships by keeping each other informed of our feelings and thoughts, then Cameron has drawn our attention to the fact that a good deal of the behaviour currently demanded by employers and educators is a threat to this, and insulting and demeaning to all concerned. In this book she has made thisÐas it should beÐa major topic for future linguistics. (Received July 2000) Reviewed by Guy Cook University of Reading, England

REFERENCE Cameron, D. 1995. Verbal Hygiene. London: Routledge.