No title

Volume 5, Number I, 1986)� JOURNAL OF SEMANTICS AN INTERNATIONAL JOURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANT...

Author: Oxford University Press

11 downloads 422 Views 17MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Volume 5, Number I, 1986)�

JOURNAL OF SEMANTICS AN INTERNATIONAL JOURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANTICS OF NATURAL LANGUAGE MANAGING EDITOR:

Peter Bosch

EDITORIAL BOARD:

Simon C. Garrod Leo G.M. Noordman Pieter A.M. Seuren

REVIEW EDITOR:

Rob A. van der Sandt

ASSISTANT EDITOR:

Bart

Geurts

CONSULTING EDITORS: J. Allwood (Univ. G5teborg), R. Bartsch (Amsterdam Univ.), J. van Benthem (Groningen Univ.), H.E. Brekle (Univ. Regensburg), G. Brown (Univ. of Es�x), H.H. Clark (Stanford Univ.), H.-J. Eikmeyer (Univ. Bielefeld), G. Fauconnier (Univ. de Vincennes), P. Gochet (Univ. de Liege), J. Hintikka (Florida State Univ.), St. Isard (Sussex Univ.), D. Israel (SRI, Stanford) P.N. Johnson-Laird (MRC Appl. Psych. Unit, Cambridge), E. Keenan (UCLA), S. Kuno (Harvard Univ.),

W. Levelt (Max Planck Inst. Nijmegen), J. Lyons (Trinity Hall, Cambridge), W. Marslen-Wilson (Max Planck Inst. Nijmegen), J. McCawley (Univ. Chicago), B. Richards (Edinburgh Univ.), R. Rommetveit (Oslo Univ.), H. Schnelle (Ruhr Univ. Bochum), J. Searle (Univ. Cal. Berkeley), A. von Stechow (Univ. Konstanz), M. Steedman (Edinburgh Univ.), G. Sundholm (Nijmegen Univ.), Ch. Travis (Tilburg Univ.), B. van Fraassen (Princeton Univ.), Z. Vendler (UCSD), Y. Wilks (New Mexico State Univ.), D. Wilson (UCL).

EDITORIAL ADDRESS: Journal of Semantics, P.O. Box 1454, 6501 BL Nijmegen, The Netherlands

Published by Foris Publications, P.O. Box 509, 3300 AM Dordrecht, The Netherlands. e N.I.S. Foundation

ISSN 0167-5133 Printed in the Netherlands by ICG Printing

JOURNAL OF SEMANTICS CONTENTS VOLUME 5 (19H,.67)

Artick.t KEITH ALLEN: Interpre�ing English comparatives

1

RENATE BARTSCH: The construction of properCies under perspectives

293

ROBERT DE BEAUGRANDE: Semantics and text meaning: retrospects and prospects

89

D. S. BREE and R. A. SMIT: Temporal relations

345

DIRK GEERAERTS: On necessary and sufficient conditions

275

ROBERT E. LONGA CRE: The semantics of the storyline in East and West Africa

51

LINDA M. MOXEY and ANTHONY J. SANFORD: Q uantifiers and focus ANTON J. SMOLENAARS and ALEX

189

J. H.

SCHUTZELAARS:

On'cognitive' semantics of emotion words: Solomon quasiecologically tested

207

BffiGIT WESCHE: At ease with "At"

385

WOLFGANG WILDGEN: Processual semantics of the verb

321

Rev1ew article PIETER A. M. SEUREN: Dan Sperber and Deirdre Wilson.

Rdenance commumcatzon and cognition

123

Discussion PETER BOSCH. Pronouns under control

65

FRANCIS CORNISH: Anaphoric pronouns: under linguistic

control or signa11ing particular discourse representations?

233

DEIRDRE Wll.SON and DAN SPERBER: The self-appointment of Seuren

as

censor,

a

reply to Pieter

Seuren

145

Book N!view8 COMMENTS ON MARVIN MINSI\."Y: The Society of Mind, with

contributions by Osten Dahl, John C. Marshall, Keith Oa tley, Ragnar Rommetveit, and A. J. Sanford

163

AD FOOLEN: Paul Werth, Foctl.3, coherence and empha8U

i9

JAMES D. McCAWLEY, Pieter

A.M.

Seuren, Di3couru

8emantic8

261

LEON STASSEN: Ray Cattell,C'ompo.!ite predicate, in Engli8h

185

MAURICE VLIEGEN, Klaus Robering,Die deutschen Verben des Sehen.!

268

II

Dear Subscriber, This issue is the first of Volume 5. It should have appeared already in April 1986. The long delay is due to a number of difficulties of an organizational nature, which we hope we have now resolved. We apologize to our readers and subscribers for the irregularity and inconvenience. Since the editors see no way of catching up with the year's delay that has been incurred without doing serious damage to the Journal's quality, we have decided to skip the issues that would have been due for 1986 and publish only one volume for 1986 and 1 987, viz. Volume 5. Volume 6 which would have appeared in 1 987 is thus moved to 1988. For our subscribers this means that their subscription is extended for one calender year with no extra charge. So if you have already paid for 1987, you will not have to pay again untill989, and if you have paid only for 1986 you will still receive all issues of Volume 5, the last of which will appear in December 1 987. The Publishers and Editors of the Journal of Semantics

Revised publication schedule of the Journal of Semantics volume

I

2

3

4

5

6

7

.

.

.

year

'82

'83

'84

'85

'86/'87

'88

'89

.

.

.

ANNOUNCEMENT

ARTIFICIAL INTELLIGENCE ABSTRACTS

Artificial Intelligence is a field of increasing importance, industrially, technologically and intellectually. However, it is not always easy to locate the descriptions of the research and development being done world wide. Not only is there a host of new journals appearing, but a great part of the work is described in memoranda series from university departments and research laboratories, and one must subscribe to these series individually to receive the abstracts of their papers. An additional difficulty impeding smooth and rapid communication of results is the fact that research work, when published in the open literature, appears not only in journals using phrases like Artificial Intelligence or Machine Intelligence, but also under closely related descriptions such as Expert Systems, Knowledge Based Systems, Logic Programming, Fifth Generation Systems, Cybernetics, Human-Computer Interface, and Pattern Recognition. Much original work in these fields also appears in journals within Linguistics, Psychology, Logic and Philosophy. ARTIFICIAL INTELLIGENCE ABSTRACTS will tackle these pro blems for both researchers in all those fields and the wider community that needs urgent access to this work. It will be published six times a year by Blackwells of Oxford and contain about 250 abstracts in each issue (starting in January 1 987). These will be in a standard format with subject descriptors and an indexing code with a hierarchical classification of the whole field. Research institutions, journal editors and individual researchers are encouraged to send abstracts to the editor of AlA directly: either to Yorick Wilks, Box 3CRL, NMSU, Las Cruces, NM 88003, USA or preferably by electronic mail to yori the amount of the cake which he gave his sister

Where c1 is in an Adj P: Jo was {lesscl {angryx than amused-. the degree to which Jo was angry < the degree to which Jo was amused

(2.5)

Where c l is in an AdvP: (2.6) •

Max waxed the car {morec1{carefullyx than Ed had washed it.the care with which Max waxed the car > the care with which Ed had washed the car

Where the cl is clause predicate: (2.7)

{ Thisx is {morecl than I expected.the quantity of this> the quantity that I expected this was going to be

(2.8)

{ Lizx is {widercl than Bill is tall.the degree to which Liz is wide > the degree to which Bill is tall

(2.9)

{ Maxx is {as oldc1 as my father. Max's age � my father's age

Where cl is an adverb: (2.10) Bill {ranx{morecl than he walked.the degree to which Bill ran > the degree to which Bill walked (2.11) Ed's mother{influencedx him {morecl than his father did. the extent to which his mother influenced Ed > the extent to which his father influenced Ed

Downloaded from jos.oxfordjournals.org by guest on January 1, 2011

(2.3)

6 In (2.12) the c I is clefted: (2.12) It was more Ed's mother than his father who influenced him. The X-functor is not 'Ed's mother' - if it were we would have had of

Ed's

mother because of the determiner 'Ed's', cf. (2.4) and (2.13) There's more of Al's mother than there is of his father, who's just a little runt of a man. In fact the X-functor in (2.12) is the zero anaphor of the verb 'influenced', as is shown in (2.14) (this analysis is further discussed in §14).

And we interpret the comparative in (2.12) the same way as in (2.11), viz. in terms of (2.15): (2.15) It was more Ed's mother than his father who influenced him. the extent to which his mother influenced Ed> the extent to which his father influenced Ed Let me return to the point that the scope of the comparative is defined by the denotata of X and Y within the context of the clauses in which X and Y appear, and not just the X- and Y-functors. Consider what is being compared in (2.16): (2.16) Jo was {lessc1 {angryx {thanc2 {amusedy.the degree to which Jo was angry < the degree to which Jo was amused We see that it is not just the 'angry' and 'amused' that are compared, cf. (2.17) Jo was less angry than amused.- angry< amused Another way to express (2.16) is as follows: (2.18) Jo was less angry than amused.For the domain I Jo was angry! & I Jo was amuse dj , the degree of anger< the degree of amusement 'Angry' is the X-functor, and 'amused' is the Y-functor in (2.16)-(2.18). For another example:


(2.14) It was {morecl Ed's mother [who {influencedx him] than his father who influenced him.

7 (2.1 9) { Morec.{antsx killed termites jthanc2 jtermitesy killed ants. - the number of termite-killing ants > the number of ant-killing termites . (2.20) l Moree! jantsx killed termites jthanc2jtermiteSy killed ants.the number of ants> the number of termites [at least, not without the constraint captured in (2.19)) An

alternative to (2.20) is (2.21), but this needs refming:

(2.21) {Morecl jantsx killed termites {thanc2ltermiteSy killed ants.For the domain I ants killed termite � & I termites killed ant �, the number of ants > the number of termites

3. THE EXCEEDS RELATION

Exceeds: X> Y ["X exceeds Y"). What this means is shown by the relations between the following formulae, any of which may be inferred from either of the others: (X>

Y)- (X-$= Y)- (X= Y+$)

where ·-· is mutual implication or synonymy, and '$' has a positive value equivalent to the difference between X and Y. 3 This difference, which we may call 'the degree of excess', is usually not specified, but may be spelled out in such phrases as $ more Xthan Yor$Xmore than Yor$X-erthan Y, as in (3.1 ). (3. 1 )

Max has six more cherries than Mary. Max has six cherries more than Mary.the number of cherries Max has= the number of cherries Mary has plus six


Obviously such paraphrases will not do in their present form, because there is no mechanism to bind e.g. 'the number of ants' bylants killed termite� rather than I termites killed ant� ; in other words, there is no mechanism (other than our intuitions) to identify 'ants' as the X-functor and 'termites' as the Y -functor. I will suggest how this binding can be 1!-Ccomplished in§ 14, and in the meantime use paraphrases like those in (2.1 6) and (2. 1 9) because they are easily interpretable. And I shall explain in §9 where the glosses 'the degree to which', 'the amount of, 'the number of, etc. come from. I have sought to show two things in this section: how the X-functor is recognized; and the fact that the X-functor and the Y -functor are interpreted as X and as Y respectively by reference to the context of the clauses in which they appear. Before showing how such clauses are identified, we shall take a closer look at the comparative relations, which I have been invoking on intuitive grounds without explanation.

8 In (3.2) the difference is not spelled out: (3.2)

This wine is better than the one we drank earlier.

(3.3)

Max has three times more cherries than me. the number of cherries Max has = three times the number of cherries I have

Intuitively this paraphrase seems preferable to its mathematical equivalent in (3.4): (3.4)

Max has three times more cherries than me. the number of cherries I have = one third the number of cherries Max has

The reason for this is that Y, which in Max has three times more che"ies than me is "the number of cherries that I have", is the standard for comparison and therefore should be the secundum comparationis, i.e. in second place following the comparative relation '=' - which it does in the paraphrase of (3.3) but not (3.4). Nevertheless, the following is semantically legitimate: ((X = MX Y)- (X/ M = Y)) - (X > Y) where 'X/ M' means "X divided by M". In (3.3)-(3.4), X is "the number of cherries Max has", M is "three", and Y is "the number of cherries I have". Here, semantic legitimacy does not necessarily correspond to pragmatic felicity: a standard for comparison is a pragmatic entity not a semantic one. The grammar of comparatives has further examples ofsemantic legitimacy dissonant with pragmatic felicity, as we shall see.


But if asked to describe the degree of excess in (3.2), the speaker might explain the evaluation in terms of nose, body, acidity, etc. Notice that the exceeds relation,>, does not only range over numerical quantities, but in this case over a scale of goodness. In fact, as I will show in §9, comparative relations like > identify a location on a C scale. Consequently '>' should only be interpreted exactly like the mathematical symbol for "greater than" when the scale is a number or quantity scale. Mutatis mutandis, the same holds for the symbols + ' ', '- ', < ' ',·�·. '=', and·�·; i.e. these symbols only have their strict mathematical values when the c1 identifies either the number or the quantity scale.4 The degree of excess symbolized by '$' is not the only one; there are also, M times more X than Y and M times as many, much X as Y, where 'M' is a multiplier, and where X= MX Y, as we see from (3.3).

9 4. THE SUBCEEDS RELATION

Subceeds: X < Y ["X subceeds Y"). There is a degree of subcess, which is the converse of the degree of excess, cf: (X < Y) - (X +$ = Y) - (X = Y-$) This is spelled out in phrases like $ less, fewer Xthan

Yand$Xless, fewer than

Y.

There are subceeds comparatives having the form which are interpreted ((X X M = (4.2)

M times fewer than Y,

Y)- (Y/ M = X)) - (X < Y):

There are three times fewer students this year than last year. "The number of students this year,multiplied by three, is approxima tely equal to the number of students last year" or "The number of students last year, divided by three, is approximately equal to the number of students this year", thus "The number of students this year subceeds the number of students last year."

The degree of subcess, like the degree of excess, can in principle be expressed as a function on either X or Y, but once again the preferred paraphrases range it over secundum comparationis and standard for comparison, Y. There are some further comments on this matter in §14. In subceeds comparatives it is usual to have an F or "fraction" operator in place of the 'M times' operator, i.e.( only) Fas( much, many, few, little) X as Y. Where there is a fraction the following relations occur, exemplified in (4.3), ((X/F = Y) - (X = F X Y)) - (X < Y): (4.3)

Only half as many people came as last time."The number of people who came this time, divided by one half, is roughly equal to the number who came last time" or "The number of people who came this time is roughly equal to one half of the number who came last time", thus "The number of people who came this time subceeds the number who came last time".


I have three fewer cherries than Ed. the number of cherries I have= the number of cherries Ed has minus three

( 4.1)

10 S. THE EQUALS OR EXCEEDS RELATION

Equals or exceeds:� Y ["X either approximately equals or exceeds Y"]; e.g. no less, fewer X than Y and( at least) as ( much, many) X as Y. (5.1)

Max is as tall as Ed.

In §9 I will explain why these are � relations and not just=. Negation over the � relation is equivalent to the subcessive ('"'-'' is the negation symbol): �X � Y) - (X < Y)

6.

Max is not as tall

as

John.- Max is less tall than John.

THE EQUAL S OR SUBCEEDS RELATION

Equals or subceeds : X :s;; Y ["X either approximately equals or subceeds Y]; e.g. no more X than Y or as little, ftw X as Y or at most as X as Y. This relation will also be discussed further; but cf. (6.1) Ed is no more reliable than is Harry. Negation over this comparative is equivalent to the excessive: -(X :s;; Y) - (X > Y) E.g. (6.2)

7.

There are not so few people here as there were last time. the number of people here this time > the number of people here last time

SIMILARITY AND D IFFE RENCE

Under the usual conditions of the co�perative principle (see §1) any two individuals, sets or ensembles of phenomena can be compared in terms of the contraries similarity and difference. Similarity: X = Y ["X is (about) the same as Y"]. The notion of similarity varies between e.g. X is( exactly) the same as Y, or X is identical to Y, and X is about the same as Y or about as ( much, many, little, few) X as Y. Cf. (7 .1)

Jo is about as pretty as her mother used to be.


(5.2)

11

8.

SOME FURTH E R P ROPERTIES OF,

AND

INTER RELATIONS BETWEEN,

COMPA RATIVE RELATIONS

Exceeds and subceeds are converse relations, so that strictly speaking the following interrelations are legitimate:

((X> Y)- (Y<X)) and therefore ((X< Y) - (Y> X)) and -(X> Y) - (X-> Y)-(X� Y) and-(X< Y) -(X-< Y)-(X� Y) However, converses are pragmatically constrained in a number of ways. One is that Y is the standard for the comparison and the secundum comparationis; thus the choice of which denotatum is to beY rather thanX (and vice versa) is determined pragmatically, on the basis of familiarity, topicality, etc. as described in §2. Further constraints on the viability of free variation between converses have to do with committedness and pull, which are examined in §10. Similarity and difference are, in principle, symmetrical relations:

((X:::::.: Y) - (Y=X)) and therefore (-(X= Y) - -(Y=X)) But the semantic symmetry is once again constrained by the fact that Y is the standard for the comparison and the secundum comparationis and this function is pragmatically determined. And consider the following relation, exemplified in (8.1) :

(-(X> Y) & -(X< Y))- (X= Y):


The contrary of similarity is difference: Difference:X =I= Y, e.g. X is different from Yand X is not th� same as Y. The relation of difference is unlike other comparative statements of inequality because it does not relate the differences to a C scale. However, similarity and difference are themselves related by a C scale going through stages such as very similar to, a bit like, little different from , etc.; indeed, all comparatives can be hedged, though I shall not be discussing the hedges (see Atlas 1984). The relations of similarity and difference are interpreted from a particular point of view, presumably determined by the topic: for instance, two things are said to be the same or identical (in normal language useJ5 if any perceivab le differences- such as different spatio-temporal locations- are presupposed insignificant to the matter in hand. The locutions X is identical to Y and X is different from Y, and less certainly X is the same as Yhave a different syntax from other comparatives; in particular theX- and Y-functors are invariably noun phrases, and the c2 constituent is a preposition. Also unlike other comparatives, these locutions do not name a difference relative to a C scale of one kind or another.

12 (8. 1)

Jo is not more intelligent than Ellie, nor is she less so. Jo and Ellie are of about equal intelligence

One might e xpect this relation to be symmetric; logically it should be, but (8.2) shows that it is not pragmatically symmetric: (8.2)

Ma x is about the same height as Ed.Ma x is not obviously taller than Ed, nor is he obviously shorter than

Ed

(-(X> Y) & -(X< Y))-(X= Y) in English, it is pragmatically constrained in the manner just discussed. (8.3}-( 8.6) give the four sets of transitive relations among comparatives; the permissible kinds of transitivity are indicated in the implicata, and should require no further discussion. (8.3)

((A > B) & (B >G) & (G � I) & (I� J) & J = X) & (X= Y) ((A > Y) & (B > X) & (I� Y) & (W = Y))

E.g. (8.3')

J o is taller than Fred and Fred is taller than Ed and Ed is about as tall as Max. - Jo is taller than Ed and Jo is taller than Ma x and Fred is taller than Ma x

Logically,= should not be a transitive relation because it may be affected by the heap of beans effect: remove one bean from the heap and the heap is about the same as it was before; imagine a set of states of the heap which are differentiated by the removal of one bean at a time, until say half the original number of beans has been removed, then any pair of adjacent states of the heap will be about the same size, but the difference be"tween the initial and final states will be significantly different. Logically then, for any string of = relations between all the letters of the alphabet from A to Z:

(A= B)& (B=C)&(C ... etc . . X) . &(X= Y)&( Y=Z)-- D(A=Z) and therefore ¢-(A = Z ) .


It is interesting that 'obviously' (it could have been 'significantly' or 'marked ly') in (8.2) is almost essential to preserve the acceptability of the paraphrase. The e xplanation is that the = relation is vague here, cf. 'about'-, thus by definition it cannot commit the speaker to either excess or subcess, nor to their negations. To maintain the co-operative maxim of quantity, if the speaker says that X= Y, then it must be the case that s/he has no ground for claiming either that X> Y or X< Y; the lack of such ground is e xactly what is indicated by the 'not obviously' in (8.2). Consequently we conclude that although there is a semantic relation

13 means "possibly not". 6 However, lan guage users treat (X = Y) as if it has the properties of()(= Y); the reason they can do so is that the co-operative maxim of manner will be violated before the heap of beans effect appears: only a very small number of chained com parative relations like those in (8.3)-(8.6) is acceptable. Consequently, the transitivity of the = relation is strictly a conversational implicature, but for the reasons given I shall treat it as if it is a semantic implication. Once again we find pragmatics overriding semantics.

'- o' means "not necessarily", and

'

0

-'

((A < B) & (B B)& -(B>G)&-=--(G �I)& -(1 � J) (-(A> J)&-(G �J))

( 8 .6)

( -(A< B)& -(B [the degree to which Ed is good 1ooking]1

(9.2)

Ellen is my favourite student, but Jo is as good. the degree to which J o is a good student � [the degree to which Ellen is a good student]

The topic is the speaker's evaluations of students, so we readily infer what 'good' is attributed to. These comparatives are indicated by c l, and the comparison is inferred from context on the basis of syntactic parallelism between the clauses containing X and the actual or imputed Y. So let's examine cl more closely. A cl consists of two parts, a comparative marker and a scale marker,which together identify the comparative relation. c l also ensures that the X-functor is syntagmatically identifiable, and consequently is the key to identifying both X and Y (see §12). It seems appropriate, therefore, to call it the comparative operator. A cl is easily recognized through the form of the comparative marker. There are fewer than a dozen such markers:

more indicating exceeds less indicating subceeds -er indicating either exceeds or subceeds depending on the scale marker as (negative so) indicating either equals or exceeds or equals or subceeds depending on the scale marker

the same, identical. about as indicating similarity different indicating difference There may be something I have omitted, but the number of comparative markers is very small. It will be seen that, despite what anyone may have claimed in the past, both -er markers and as markers indicate one of two oonverse relations, disambiguated by the scale markers. Let's tum our atten tion to that matter next.


There is an implied comparative in (9 .1), and another in (9.2). In these and in all such cases the comparison is recoverable from context, provided the speaker has observed the co-operative principle (in particular the maxims of quantity and manner).

15 I will come at scale markers through discussion of the -er morpheme. -er is usually said to mean "to a greater degree", and there is good evidence for this. For instance there are a number of synonymous expressions like (9. 3)-(9.4), where .ADJk-er- more ADJk: (9.3)

B is cleverer than D- B is more clever than D

(9.4)

B is politer than D- B is more polite than D

There are no examples of ADJk-er superficially equivalent to less ADJ�c There are, however, converses of the following kind: Ed is politer than AI-AI is less polite than Ed

(9.6)

Ellen is less clumsy than Phil- Phil is clumsier than Ellen

Hence it would appear that (B is ADJk-er than D- B is more ( (B is less ADJk than D))

ADJk than D)

-

But appearances are deceptive. My assumption is that -er is the comparative marker and the stem it is bound to is the scale marker. But bound to a stem like few or short, for instance, the excess interpretation of -er seems inappropriate: although one can, by stretching the imagination, interpret fewer as "few to a greater degree" or shorter as "short to a greater degree", one's intuition calls for these comparatives to be understood subcessively. By way of confirmation, the opposite of politer is not normally more impolite (which sounds joky) but less polite, because one is less than happy to speak of a "greater degree of impoliteness". Constraints against such locutions as more impolite are dis cussed at the end of this section, cf. (9.53)-{9.56). The problem stems are all members of a class of what I will call 'sub' forms of relational predicates. Relational predicates minimally consist of two pair ed antonyms each identifying the end of a C scale relating then. Two predi cates F and G are binary relational predicates if each is a fuzzy set of adjacent points on a scale C; and FU GU MID = C. F is a fuzzy set any of whose members can be assigned a value VF between 0 and I such that 0 < VF::::;;; I ; and similarly for a member of G: 0 < VG::::;;; 1. For X is F, VF (1 and VG = 0 . These are values predicated of X in respect of the C scale. MID identifies the middle region of the C scale, such that '((X is C) & -(X is F) & -(X is G))- (X is MID)'. The nature of MID varies from scale to scale; on some scales it intersects with both F and G, including membership of these sets to values necessarily less than 0.5 and typically considerably less. X is not F implies that X has a C scale membership value such that ((VF = 0) & ((0 < VG::::;;; I) v


(9.5)

16 (MID))): in other words, X is not F implies X is MID or X is G to some extent. Conversely for X is not G. I do not intend to discuss antonymy in any

(9.7)

John is tall . John's height > typical height e xpected for such men

Where subs occur as predicates in non-comparative constructions they have the specific "subceeds" sense: (9.8)

Ed is short . Ed's height < typical height expected for such men

In these (so-called) non-comparatives, the subject N P's height is implicitly Table 2.

SUB FORM

C SENSE "of a certain . .

short

"height from base [rei. to human]" tall "length" long "height" high wide "cross dimension" "dimension from top surface down" dup "age " old "weight" Mavy

short low NlTTOW

shallow yowtg light slow few /itt/�

"speed"

"number of' "quantity of [mass]"

."

SUPRA FORM

fa.Jt many much


detail (see Cruse 1976, 1980; Lehrer 1985; Lehrer & Lehrer 1982), but there are relational scales involving more than two terms, e .g. hot, warm, cool, cold , among which interesting relationships arise (see (9.36)-(9 .46). For the time being, I will discuss only binary relational predicates and their associa ted C scale. Each of the two paired antonyms identifies a region towards one end of the C scale relating them: one region is 'supra' , and the other 'sub'; and I shall use these terms for the predicates naming the respective regions. In certain constructions, the supra predicate, and sometimes the sub, is used in a neutral sense, which I will call the C sense because it means "of a certain C' where 'C stands for an attributional category which identifies8 the scale. Supras have the sense "exceeds what might be e xpected as typical for the set denoted in such a context"; while subs have the sense "subceeds what might be e xpected as typical for the set denoted in such a context". Examples are given in Table 2. Let's e xamine the meanings and interrelationships of comparative and non-comparative terms on a given C scale, say, height. Where supras occur as predicates in non-comparative constructions like (9.7) they have the specific "exceeds" sense:

17 compared against a fuzzy norm (cf. Lakoff 1 972, Rusiecki 1985). I n a comparative like (9 .9)-(9. 1 0) X's height is explicitly compared with Y's particular height: (9.9)

Sue is taller than five foot four! Sue's height > a height of 5'4"

(9. 10) Ed is shorter than John. - Ed's height < John's height

(9. 1 I) John is

as

tall as AI! - John's height � AI's height

(9 . 1 2) John is as tall or taller than AI. (John's height � Al's height) & O(John's height > Al's height) Here English or has been interpreted as ..and possibly"; in (9. 1 2) the conjuncts are compatible, but not in (9. 1 3). (9. 1 3) *John is as tall or shorter than AI. *(John's height � Al's height) & O(John's height < Al's height). In order for, say, a philosopher to e xpress the disparate propositions in (9. 1 3 ) e xclusive disjunctions using natural English, they would have t o be expres sed without conjunction reduction, as in (9. 1 4). as

(9. 1 4) John is (either)

as

tall as AI or John is shorter than

AI .

And even this leaves out the third logical possibility or John is taller than A I. From (9. 1 1 )-(9. 1 3) we can deduce (9. 1 5). (9. 1 5) AI is not as tall as John. -(AJ's height � John's height) - (At's height < John's height) There is nothing anomalous in either of (9. 1 6)-(9. 1 7). (9. 1 6) John is about as tall, or only slightly taller/shorter than Ed. Here the first clause means .. John's height = Ed's height" and this can obviously flex in either direction, so that (9. 1 6) means: (9. I 7) John's height

=

Ed's height ±$, for '$' is a small distance


We see, e.g. in (9. 7)-(9. 1 0), that the C sense is invoked in both comparatives and non-comparatives; and further that where the scalar term is a supra we get the exceeds relation, where it is a sub we have the subceeds relation. Now consider the following.

18 Here '+$' indicates a distance up towards the supra end, and '-$' a distance down towards the sub end. Similar explanations to those given for (9. 1 1 )-(9. 16) obtain for (9. 18) (9.22).

(9 . 18) Ed is as short as Harry ! - Ed's height :s;;; Harry's height (9 . 19) Ed is as short or shorter than Harry. (Ed's height :s;;; Harry's height) & O(Ed's height < Harry's height) (9.20) * Ed is as short or taller than Harry. *(Ed's height :s;;; Harry's height) & O(Ed's height > Harry's height)

(9.22) John is about as short, or only slightly taller/shorter than Ed. Other binary C scales have comparable interrelations among their terms. My analysis of comparative -er diverges significantly from that of other scholars in proposing that -er is a comparative marker suffixed to a scale marker; and if the scale marker is a sub then the comparative relation is subceeds; or if the scale marker is a supra then the comparative relation is exceeds. In both cases the scale marker is interpreted in the C sense. We can begin to see the justification for the analysis by contrasting the comparisons in (9.23)-(9.24).

(9.23) Ridgebacks make {bettercl guard dogs than pets. the degree to which ridgebacks make good guard dogs > the degree to which ridgebacks make good pets The exceeds relation is sanctioned by the combination of the comparative marker -er and the scale marker bett-. Bett- is suppleted to the supra good on the goodness (good-bad) scale . By way of contrast, in (9.24) the combination of the stem slow and its suffix -er give rise to a subceeds relation, because slow is the sub on the speed (fast-slow) scale.

(9.24) Max ran {slowerc1 than Joe. the speed at which Max ran < the speed at which Joe ran The proposed analysis of -er comparatives leads to a superficial asymmetry between f ewer and more. Compare (9.25)-(9 .26).

(9.25) There are {morecl books on my desk than on the shelves. the number of books on my desk > the number of books on the shelves


(9.2 1 ) Harry is not as short as Ed. - Harry's height > Ed's height

19 (9.26) There are {fewerc l books on the shelves than on my desk. the number of books on the shelves < the number of books on my desk

(9.27) Jo came earlier than Sid. Tablt 3.

COMPARATIVE

SUB

SUPRA

ftwtr

ftw

many

ltll

little

much

ltn or

mort ltn

ftwtr

COMPARATIVE

mort

"greater degree, quantity, number of, etc." "lesser degree, quantity of, etc." and for many people also "lesser number of' "lesser number of'


Whereas fewer transparently consists of a comparative marker -er and a scale marker in the sub few, more has an opaque morphology. Bresnan ( 1 973) (attributing the idea to Elizabeth Selkirk) has proposed that more is morpho logically equivalent to both • much-er and • many-er, while less is morpholo gically equivalent to • !ittle-er(which, of course, is not intended to be identical with the existing surface form littler). In other words, more and less are portmanteau morphs containing both the comparative marker and a scale marker. This analysis seems well justified on semantic and distributional grounds: fewer is semantically classed with the comparatives more and less, as is illustrated in Table 3 . More is the suppleted scale marker for both of the supras many and much in > comparatives. In standard English less, although the primary antonym of more, is the suppleted comparative only to little. In view of the asymmetry in the system, it is hardly surprising that many speakers use less as the sub marker for both quantity and number scales, suppletive to few and supplanting fewer in < comparatives. I have talked a great deal about scales and scale markers without, so far, demonstrating why scales rather than quantities need to be invoked in analyzing comparatives: after all, the terms in Table 3, which include the canonical comparatives more and less, are all quantifiers, and the mathemati cal symbols I am using to represent comparative relations were originally relative quantifiers. However, quantities are readily scale related; and scalar relations are essential to deal with the semantics of punctual temporal comparatives like the ones in (9.27}-(9.28).

20 (9.28) Jay arrived sooner than e xpected.

(9 .29) Jo came earlier than Sid. - The time at which J o came is closer to the early end of the time scale than the time at which Sid came (9.30) John is taller than Ed. - John is closer to the tall end of the height scale than is Ed; therefore John has greater height (9 .3 1 ) Sid has fewer cherries than Mabel. - The number cherries Sid has is closer to the few end of the number scale; therefore Sid has a smaller number Let us now tum to the question of how comparatives use the C scales. I propose that comparative relation symbols other than =! (difference), viz. the relations , represent the relative locations of arguments on the scale denoted by the c l . Thus the comparative quantifiersfewer, less, and more identify relative locations on a number or quantity scale; taller, as tall and shorter identify relative locations on a height scale. More generally, Table 4 has the c i s identify X's place on a C scale relative to Y. Table 4.

SUB

SUPRA

short

shorter

as tall

taller

tall

colder

as short as cold

about the same height

cold

about the same temp.

as hot

hotter

hot

early

earlier

as early

about the same time

as late

later

late


What we understand by the expressions earlier or sooner is a temporal relation between two points on a time scale. The relationship between loca tions on a scale has come up before: subs and supras were described earlier as paired antonyms each identifying a region towards one end of the C scale relating them. And it is easy to conceive number, quantity, height, goodness, temperature and so forth, as well as time, all to be scales (cf. Table 2) on which a tendency towards one or the other pole can be indicated. Thus, for instance, the comparatives fewer and less locate their arguments towards the sub end (rather than towards the supra end) of the C scales named by these predicates; but, note, not necessarily at the end, nor even over the halfway mark - though this last point depends on the particular terms in a scale, as we shall see in § 10. Conversely, more locates its arguments towards the supra end rather than towards the sub end of one or the other of the same scales (and again not necessarily at the end, nor even, perhaps, over the halfway mark). Similarly for other scales, cf.

21

(9 .32)

{ Moreci ants killed termites than termites killed ants. -

the number of termite-killing ants > the number of ant-killing ter mites

(9.33) Max waxed the car { lessci carefully than Ed had washed it. the degree of care with which Max waxed the car < the degree of care with which Ed had washed the car (9.34) Max ate {no morec1 of the cake than George did. the amount of the cake which Ma x ate � the amount of the cake George ate (9.35) Max arrived {earlierci than Ed . the time at which Ma x arrived < the time at which Ed arrived The emphasized glosses derive from the C scales indicated by the c l . In (9.32) 'more' is a supra on the number or quantity scale, and because it ranges over a countable X-functor the number scale is the relevant one: so the relation is > on the number scale. There is another way of looking at more , though I shall not adopt it: we might assume that it is only a supra marker on the quantity scale, and that wherever it ranges over the head of a countable N P the quantity is understood in terms of number. In (9.33) 'less' is a sub on the quantity scale (or, possibly, the number scale), and because it ranges over an adverb X-functor it is the quantity scale which is invoked. Ranged over an adjective or adverb, the quantity le xicalizes to degree or amount in the metalanguage. The comparative relation marked is relation; in §4 we showed this to be equivalent to �- Here it is partitive and so invokes the quantity not the number scale; when applied to nouns the quantity scale usually le xicalizes to quantity or amount. In (9.35) 'earlier' is a punctual temporal, and an -er marked sub on the time scale; it identifies the < relation. We see from these e xamples that when


Table 4 clearly shows the force of the argument that more is the portmanteau version of • many-er and • much-er , while less is the suppletive for the -er marked sub little on the quantity scale. It is also clear from Table 4 that -er marks the < and > relations, while the distinction between these two is determined by whether the scale marker is a sub or a supra; similarly, as marks the � and � relations, with the scale marker again determining between sub and supra. It can be no surprise that in = relations a neutral scale marker is often used. The implications of not using a neutral scale marker will be discussed in § 10. I postulate that all comparatives other than those named in the locutions X is identical to Y, X is different from Y, and (perhaps) X is the same as Y are scale based in the manner shown in Table 4. We are now in a position to explain where the emphasized glosses come from in (9.32)-(9.35).

22 X-functors are le xicalized in glosses they are also nominalized: this is a quirk of the metalanguage which I will comment upon in § 14. Relational scales involving more than two terms, e.g. hot, war m, cool, cold e xhibit some interesting relationships.

(9.36) X is hot - (""'-(X is cool

v

cold)) & X is more than warm

(9.37) X is warm - ( the degree to which Fred is polite A comparative presupposes gradability in the denotatum of X and Y; thus the hyperbole whiter than white presupposes a metaphorical scale of whiteness beyond literal possibility. Not all comparatives involve antonymous scales with le xical antonyms at opposing poles; for instance, 'greener' has none. (9. 5 1 ) The grass is greener where it's been watered.

Greener presupposes a scale of greenness and a location towards the supra, i.e. focal green, end. There is no le xical antonym (and for most speakers perhaps no firm concept) for the other end of this scale, so the suppleted comparative operator for the sub end of the quantity scale is )lSed; thus (9. 5 1 ) implies the converse (9.52): (9.52) The grass is less green where it has not been watered. The quantity scale is the default scale for comparison, consequently more and less, and the other comparatives on the quantity scale, are the canonical comparative operators and are used with all kinds of scales which, for one reason or another, do not take a comparative marker; for instance as careful ly as normally indicates the � relation on the carefulness scale; but its sub converse is with as little care as. Why should it be that subs are more marked


Although the comparative marker is essentially the same in both sentences, the scale marker is different and so is the X-functor (and consequently the Y-functor will be different too). In (9.47) 'politer' identifies a location further towards the supra end of the politeness scale for X "Jo", than for Y "Fred". In (9.48) 'more' identifies a location further towards the supra end of the quantity scale (the default scale) for X "Jo's politeness", than for Y "Fred's politeness" (see § 1 2). In each case the X- and Y-functors are interpreted within the context of the clause in which they appear, so we end up with alternative ways of saying the same thing, viz. synonymous expressions; cf. (9.49}-(9.50).

24 than supras, such that comparison with a sub requires using the default scale whereas comparison with the supra does not? Put another way, why do we interpret greener as a supra rather than a sub? The answer lies in the pollyanna principle discussed in § 1 0. Before going to § 10 I will briefly discuss such expressions as more impolite, more badly-behaved, more unnatural etc. in which the default scale supra c 1 more is ranged over a n adjective X-functor whose negative o r sub prefix marks it as a sub. (Moreover, all such sub X-functors are committed, see § 10.) Consider the following: (9.53) Ted is less polite than Mo.

(9.55) Ted is even more impolite than I recall. The normal < relation for these scales uses the default less with the scale marker in the supra form: less polite, less well-behaved, less natural, compare (9.53) with (9.54). H owever, the sub form can apparently function as some thing like a scale in its own right when the speaker wants to make a particular ly strong denunciation - compare (9.53) with (9.55), where 'even' seems almost obligatory; nevertheless, impoliteness, bad behaviour, unnaturalness and the like are not scales but X-functors; and the scale marked in c 1 is the default quantity scale. Consequently we find the contrary of, for instance, (9.55) in (9. 56). (9.56) Ted is less impolite than I recall. In this section I have argued that the comparative operator, c l , has two constituents: a comparative marker from a very small set that readily enables recognition of the comparative construction; and a scale marker which is the head of the c 1 , and which identifies not only the scalar basis for the particular comparison, but also defines the particular relation involved, depending whether the term used is a sub or a supra. I have shown that the comparative relations symbolized should not be interpreted as quanti fiers (except when the scale is the number or quantity scale) because they identify relative locations on the scale named by the c 1 . That is, X< Y and X � Y locate X further towards the sub end of the scale named in c 1 than Y, whereas X� Y and X > Y locate X further towards the supra end of the scale. X � Y, of course, locates both X and Y at about the same place on the scale.


(9.54) ??Ted is more impolite than Mo.

25 10. COMMITIEDNESS AND PULL

Cruse ( 1 976) and Lehrer ( 1985) talk about worse being 'committed', whereas better is uncommitted. Lehrer ( 1 985:403) explains why: We can say ( 1 4) Y is better than X, but both are bad. but not ( 1 5) •x is worse than Y, but both are good.

there's about the same number of X as Y - ((there's about as many X as Y) & D(there's about as few X as Y)) -

but X happened at about the same time as Y early) as Y)

-

(X happened about as (late v

The second, which also introduces Rusiecki's 1985 notion of the 'pull effect', has to do with the motivation for choosing between semantically equivalent expressions like not so many people versus fewer people or do worse versus not do so well. And the third is to show that committed terms, whether subs or supras, do not venture beyond that notional halfway mark on their C scale; whereas the uncommitted ones may do so. It seems that whenever a supra is committed, the corresponding sub is always committed; but there are many instances of subs being committed while the corresponding supra remains neutral. An explanation for this is the pollyanna principle: The Pollyanna Hypothesis asserts that there is a universal human tendency to use evaluatively positive words more frequently and diversely than evaluatively negative words in communicating. [ ... ) most people most of the time in most places around the world talk about the good things in life [. . . ] they tend to see and report the good qualities of things. (Boucher and Osgood 1 969) The principle results in a preference for the supra over the sub. I will leave it to psychologists to offer explanations for what we find, but there is certainly evidence for such preferences and for the precedence given to the supra (cf.


I shall not attempt to do more than touch on the notion of committedness here; for a fuller treatment, see Lehrer ( 1 985). However, I do interpret committedness somewhat more widely than Cruse or Lehrer, and I concen trate on its relevance to comparatives and interrelations on the C scale, which they do not. I have three interests in committedness: the first is to account for the following asymmetry, and others like it:

26 Allan 1987). Where relevant, the supras tend to be perceptually more promi nent - bigger, taller, louder, brighter, and otherwise ear and eye-catching. Given the pollyanna principle, supras should less marked than subs; and, of course, in general they are - we saw this in the contrast between the supra greener and the sub less green in (9.5 1 )-{9.52) above. And we also get subs marked by negative prefixes, cf. the supra polite and its sub impolite. Then, the following supras are all neutral in how-questions (questions of degree) but the corresponding subs are not: SUPRAS: many, much, tall, good, beautiful SUBS: few, little, short, bad, ugly

( 1 0. 1 ) How tall is John? >> John is of a certain height ( 10.2) How short is Ed? >> Ed is in the short region of the height scale We could paraphrase the presupposition in ( 1 0. 1 ) as 'the speaker does not know whether John is tall or short or of average size." By contrast, the presupposition in ( 10.2) means "the speaker either knows, or at least believes it probable, that Ed is short." Despite the different presuppositions9 in these questions of ( l O. l )-{ 1 0.2), both seek an answer to the C scale related question "What height is . . ?"; the different presuppositions indicate whereabouts on that scale the questioner believes the answer to lie. However, a few supras, e .g. hot, beautiful, are also committed in how questions (as are their corresponding subs, of course): .

( 10.3) How hot is it outside? >> it is in the hot region of the temperature scale ( 10.4) How cold is it outside? >> it is in the cold region of the temperature scale So much for questions of degree; now to comparatives. In � and � relations, i.e. positive as . . . as comparatives, a sub implies 10 that the sub predicate holds true for X, in other words a sub is committed; but so are many supras, e.g. much, many, hot, beautiful, happy. We begin with some uncommitted supras and their corresponding subs. ( 1 0.5) John is as tall as AI! (John & AI are tall) v (John & AI are short)


In fact all subs are committed in how-questions, viz. a sub form presupposes that the sub predicate holds true for the argument; most supra forms have no such presupposition. In ( 10. 1 )-{ 1 0.4) '>> P' means "implies the speaker presupposes P".

27 The speaker of ( 1 0.5) presumably knows whether or not John and AI are tall, but he is not indicating it here; thus the hearer infers that John and AI may be either tall or short (unless he happens to know from some other source which of these obtains). The 'may be' in this inference could be represented by a - o operator as in ( 1 0 . 6), which of course can be calculated directly from ( 10.5). ( 1 0.6) John is as tall as AI! -o(John & A1 are tall) & -o(John & Al are short) Against (10.5)-( 1 0.6) compare ( 10.7) with the committed sub:

As a consequence of the implicata of ( I 0.5)-( 1 0. 7) there are the interrelations between terms on the height scale shown in ( 10.8)-( 10. 1 0). ( 1 0.8) John is as tall as AI, but both are quite short. The insertion of 'quite' seems to be necessary to maintain the co-operative maxims: Why should the speaker use the predicate tall in one clause if it is to be completely, and blatantly, contradicted in the next clause? Cf. ( 10.9) • John is as tall

as

Al, but both are short.

( 10. 10) • Ed is as short as Harry, but both are quite tall (really). Compare the anomalous ( 1 0. 10) with ( 10. 7). The hedges in ( 1 0. 1 0) may seem to render it less blatantly contradictory than ( 1 0.9) by indicating the speaker's unwillingness to make an affirmation of Ed's and Harry's heights, but it is still completely unacceptable. For some other examples: ( 10. 1 1 ) Max is as good at math as Bill . (Max & Bill are good at math) v (Max & Bill are bad at math) I.e. both Max and Bill are at about the same location on the goodness scale with respect to math , wherever that location may be. Thus ( 1 0 . 1 2) is accep table: ( 10. 1 2) Max is as good at math as Bill, but neither is any great shakes. With the committed sub: ( 10. 1 3) Max is as bad at math as Bill . Max & Bill are both bad at math [i.e. both at the bad end of the goodness scale]


(1 0. 7) Ed is as short as Harry! both Harry & Ed are in the short region of the height scale

28 Thus ( 1 0. 14) is unconditionally acceptable whereas ( 1 0 . 1 5) is only acceptable if the first clause is taken to be ironic. (10. 14) Max is as bad at math as Bill , i.e. neither is any great shakes. (10. 1 5) Max is as bad at math as Bill , who is really quite good. As we have said, some supras are also committed in � and � comparatives. E.g.

(10. 1 7) It is as hot inside as it is outside. it is hot both inside & outside Consequently ( 1 0 . 1 8) is anomalous: ( 1 0 . 1 8) *It is as hot inside as it is outside, yet it is cool in both. The committedness of the supra can be neutralized by ranging only or no more over the c l , e.g. (10. 19) It is only as hot inside as it is outside. it is not very hot either inside or outside (10.20) It is only as hot inside as it is outside; in fact it is downright cold everywhere. Subs are, of course, committed, e.g. ( 1 0 . 2 1 ) It is as cold inside as it is outside. it is cold both inside & outside Let us now turn to committedness in the < and > comparatives. Many subs which are committed in other environments are not committed in < or > comparatives, and nor are the corresponding supras. E.g. 1 0 . 22) John is taller than Ed.

-

(1 0.23) Ed is shorter than John.

John is tall -

Ed is short

Consequently ( 10.24)-( 1 0.25) are fine.


(10. 1 6) Max has as many as five Cadillacs. Max has many Cadillacs [& -(Max has few Cadillacs); i.e. the num ber of Cadillacs Max has is towards the many end of the number scale for Cadillacs possessed by individuals]

29 ( 1 0.24) John is tall. Ed is shorter, but still tall. ( 1 0 . 25) Ed is short. John is taller, but still short. Number scale sub few and width scale sub na"ow are also committed in other environments, but not in this one, e.g. ( 1 0.26) There were more people here yesterday than there were the day before, but still not enough to cover our expenses. ( 1 0. 27) There are fewer people here than there were yesterday, but still quite a lot.

( 10.28) Max is better at math than Bill. - (-o(Max is good at math)} Contrast ( 1 0.28) with ( 1 0.29): ( 1 0.29) Ed is worse at math than AI. - Ed is bad at math Here it is a necessary implication that Ed is bad at math , but not that A1 is. Contrast ( 1 0. 30) with ( 1 0 . 3 1 ). ( 1 0 . 30) a. Ed is worse at math than AI; in fact AI is good. b. * Ed is worse at math than AI; in fact Ed is good. ( 10.3 1 ) a. * Ed is as bad at math as AI; in fact AI is good. b. * Ed is as bad at math as AI; in fact Ed is good. ( 1 0 . 32) X is worse than Y - (X is bad) & -o(Y is bad) The converse of ( 10.32) is ( 10.33): ( 1 0.33) X is not as bad as Y - (Y is bad) & -o(X is bad) ( 1 0.34) X is as bad as Y - (X & Y are bad) The converse of ( 10.34) is ( 10.35): ( 1 0. 35) X is no worse than Y - -o(X & Y are bad) Some supras are committed in < and > comparatives. E.g.


Subs expressing negative evaluations such as bad, unacceptable, inaccurate, revolting, etc. are all committed, but the corresponding supras are not.

30 (10.36) It is hotter inside than it is outside. it is hot both inside and outside (& it is not not-hot either place) The corresponding subs are, of course, committed too; because whenever a supra is committed in any of the constructions we have considered, its corresponding sub is also committed. It is the committedness of plain and beautiful that warrants such sentences as ( 1 0.36)-{ 1 0 . 37): (10.36) Fanny isn't plainer than Lisa, she's just not as beautiful.

I don't know where the line between being plain and being beautiful lies, but ( 10. 36) attempts to move Fanny over it, while ( 1 0.37) moves Sheila in the opposite direction. We are now in a position to account for the following asymmetries, mentioned earlier:

there's about the same number of X as Y - ((there's about as many X as Y) & -o(there's about as few X as Y)) X is about the same height as Y - ((X is about as tall as Y) & -o(X is about as short as Y)) but X happened at about the same time as Y - (X happened about as (late v early) as Y) X is about the same temperature as Y - (X is about as (hot v warm v cool v cold) as Y) All these exemplify the = relation which says that X and Y are located in about the same place on a scale. The scale names number, height, time and temperature are neutral terms and by definition uncommitted, so their use does not indicate in which region of the scale X and Y are to be found. Similarly with the uncommitted supras many and tall (but see below). The subs are all committed, as are the supras late and hot, so by choosing them the speaker does identify whereabouts on the scale X and Y are located. We have seen that subs are normally committed in how-questions, =, � and � comparatives, and that 'negative evaluation' subs are committed in < and > comparatives (further analysis will probably reveal other classes of subs to be committed with < and > comparatives, too). Supras are much less likely to be committed than subs, though hot seems always to be committed. Interestingly, warm is not committed, cf.


(10 .37) Sheila is no more beautiful than Fanny; indeed, both are distressingly plain girls.

31 ( 1 0. 38) It's a s warm a s toast, i n fact i t is downright hot! ( 1 0. 39) It's warmer outside than in, in fact it's downright hot! ( 1 0.40) The pool's warmer than I expected, although it's still pretty cold. Contrast ( 1 0.40) with ( 1 0 .4 1 ) ?The coffee's hotter than I expected, although it's fairly cool none theless.

( 10.42) X is less good than Y ( 10.43) X is less bad than Y We do not get comparable comparatives on the height scale: ( 10.44) •x is less tall than Y ( 10.45) •x is less short than Y Instead of ( 1 0.44)-( 1 0.45) we have ( 1 0.46)-( 1 0.47) respectively: ( 10.46) X is shorter than Y ( 10.47) X is taller than Y neither of which is committed. ( 1 0.42)-( 1 0.43) are both committed, and they are alternatives to ( 1 0.48)-( 10.49) respectively:


( 1 0 .4 1 ) would be less acceptable with cold instead of'cool', and fully accepta ble with warmer in place of 'hotter'. We have seen how committedness affects the interrelations of terms on a C scale. It should be obvious that the choice of terms from a C scale may depend on whether or not a term is committed. These choices reflect the presupposi tions of the speaker, and his or her concern for the implications of the locution to be used. To help in choosing, where there is a committed term on a scale, there are formulae for avoiding unwanted committedness. This can be deduced from the set of possibilities on the temperature scale, described in (9 .36 )-(9 .46), or from the choice between saying X is the same temperature as Y rather than X is about as hot as Y, or vice versa; however I will exemplify it from an asymmetry between terms on the height and goodness scales. We find on the goodness scale:

32 ( 1 0.48) X is worse than Y ( 10.49) X is better than Y

( 1 0. 50) There are not so many people here today as there were yesterday. ( 10.51) There are fewer people here today than there were yesterday. Although these sentences should, strictly speaking, be synonymous, the choice of not so many in ( 10.50) suggests that the number of people is towards the supra end of the number scale - presumably as a function of 'many', whereas the choice offewer in ( 10.5 1 ) suggests a number down at the sub end of the scale (the number scale being constrained by the expected norm for the occasion). The pull effect is a conversational implicature, and is open to cancellation without contradiction, e.g. ( 10.52) There are not so many people here today as there were yesterday, indeed it's been a very disappointing turnout on both days. ( 1 0. 53) There are fewer people here today than there were yesterday, al though still a lot came. Speakers pay regard to the implicatures resulting from the pull effect when choosing between competing possibilities. For instance, the choice between ( 10.54) and ( 10.55) will often be motivated by the fact that the former conversationally implicates that a lot of people came, which the latter does not. ( 10. 54) There's about as many people came as yesterday. many people came


of which ( 10.48) is committed in the opposite direction from ( 1 0 .42), and ( 10.49) is uncommitted. It is clear that ( 1 0 .42)-{ 1 0 .43) exist because of the committedness of the sub on the goodness scale; if the speaker wishes to indicate that "X is not as good as Y" (which is uncommitted) and at the same time indicate which end of the scale X is on, s/he may choose to say either ( 1 0.42), locating X over the halfway mark towards the good end; or they may choose to say ( 1 0 .48), locating X firmly at the sub end of the scale. It may appear that another asymmetry arises because ( 1 0 .49) is uncommitted! Al though this is true, the default interpretation of ( 1 0.49) locates X at the good end of the scale unless anything to the contrary is said or mutually known. This characteristic is the 'pull effect', and we shall be discussing it immediate ly. The pull effect was first mentioned by Rusiecki ( 1 985:98). Compare ( 1 0.50)-{ 10.5 1 ):

33 ( 1 0.55) There's about the same number of people came as yesterday. [(1 relevant] I use · - ' to mean conversationally implicates. More significantly, if the pull effect of a certain locution could affront the hearer's face, the speaker will generally avoid using that locution and choose a face saving alternative, unless s/he wishes to inflict the insult. E.g. ( 1 0.56) is polite, it conversationally implicates that the hearer's roses not only did well last year, but also are towards the supra end of the goodness scale this year: compare it with ( 10.57).

( 1 0.57) Your roses did worse this year than last year (but they still did better than mine) - your roses did badly To negate a supra instead of asserting a sub is a pragmatic device for keeping to the supra half of the scale and avoiding the invocation of the wrong half of the scale. We see from ( 1 0.52)-( 1 0.57) that the pull effect is independent of committedness, yet it is particularly significant with a positive-negative scale like goodness where the sub is committed in all environments. Compare ( 10.58)-( 1 0.60), all of which express the same propositions: ( 1 0. 58) Your son does better than mine at school; Max is cleverer than Ed Max is clever ( 1 0.59) My son doesn't do so well as yours at school; Ed's not so clever as Max. - Max is clever ( 1 0.60) My son does worse at school than yours; Ed's more stupid than Max - Max is stupid ( 10.58)-( 1 0 . 5-9) differ only in result of the different choice of grammatical subject - presumably because of a difference in contextual topic; both sen tences use supras. By contrast ( 1 0.60), by invoking the committed subs, implies that Ed is a real dummy and implicates that Max is stupid, which is a face affront to the hearer. Now compare the uncommitted terms in ( 1 0.6 1 )-( 1 0 .62), either of which could be used to describe the same state of affairs. ( 10.61) Max is not much shorter than his wife. ( 10.62) Max is almost as tall as his wife.


( 1 0.56) Your roses didn't do so well this year as last year (but they still did better than mine) - your roses did well

34

1 1 . THE c2 IN ENGLISH

In § 1 we identified two comparative construction types: (i) c1 X c2 Y; (ii) X c l c 2 Y . We also said that for the locutions X is identical to Y, X is differentfrom Y, and X is the same as Y, c2 is a preposition and Y is invariably a NP, so that the comparison is essentially one of NP denotata. There are three reasons for thinking that c2 is a preposition in these comparative constructions: firstly c2 bas the form of a preposition; secondly it ranges over a NP; and thirdly, it can be supposed to identify the participant role of that NP's denotatum within the denotation of the comparative clause. E.g.

( 1 1 . 1 ) My toothbrush is different { fromc2 yours. My toothbrush * your toothbrush ( 1 1 .2) My toothbrush is identical {toc:2 yours. My toothbrush = your toothbrush (More precisely, the relation in (1 1 .2) is = . ) ( 1 1 . 3) This is the same {asc2 that. This ""' that We also find these locutions in type (i) constructions in which, once again, c2 is a preposition ranging over a NP which has the Y -functor as head. ( 1 1 .4) Max is the same height {asc2 his father. Max's height ""' his father's height ( 1 1 .5) Max is of identical build {toc:2 his father. Max's build ""' his father's build


The motivation for choosing to say ( 10.6 1 ) rather than (10.62) would be where the contextual topic was something like "men shorter than their wives"; whereas an appropriate contextual topic for (1 0.62) would be some thing like "tall women". If there were no such contextual contraints, then the pull effect of 'not much shorter than' and 'almost as tall as' would be to establish such topics. What other reason could there be for choosing one locution rather than the other? In this section I have sought to show how committedness affects the interrelations of terms on a C scale, and how the pull effect motivates the choice between semantically equivalent alternatives. It is clear that the choice of terms from a C scale may depend on whether or not a term is committed. The choice reflects the presuppositions of the speaker, and his or her concern for the implications and implicatures of the form of words to be used.

35 ( 1 1 .6) Max has a different build {fromc2 his brother. Max's build =F his brother's build However, with all other comparative relations, viz. , c2 marks a boundary between the end of the clause containing X and the onset of the clause containing Y. Despite Pinkham (1982: 1 3 1 ) (and many others), c2 in these other comparatives, which is either than or as, is not a preposition because its scope is a clause containing the Y-functor, instead of invariably being a NP with the Y-functor for its head. Thus, althou'gh we find compara tives like ( 1 1 . 7), in which the 'me' is simply the default form of the pronoun, we also have comparatives like ( 1 1 .8), in which c2's score is undoubtedly clausal.

( 1 1 .8) Max gets more pocket money than I do. I shall call c2 'the comparative conjunct' (cf. Ryan 1983). Although as may be a preposition in the same as, there is no convincing evidence that than is prepositional at all . 1 1 The etymology of than (it is cognate with the additive conjunct then, its meaning and function, shows it to be more of a clause linker than a preposition: the nature of the linkage is co-ordination; the purpose of the linkage is comparison; hence the name 'comparative conjunct'. It is a co-ordinator not a subordinator, because although the two clauses of the comparative are mutually dependent semantically and syntactically, there is no sign of subordination in the than-clause, cf. ( 1 1 .9) Max visits his mother more often than Mary does. ( 1 1 . 10) •Max visits his mother more often than Mary to do. Thus in (2.5) and (2.8), clauses linked by than were replaced in the interpreta tive gloss by two parallel (what Stassen 1984 calls 'balanced') clauses linked by '&' (and see also § 1 4). Hankamer ( 1 973) and Brame ( 1983) suggested that clauses like ( 1 1 . 1 1 ) treat than as a stranded preposition by analogy with ( 1 1 . 1 2). ( 1 1 . 1 1 ) Who is he bigger than? ( 1 1 . 12) Who is he similar to? But whereas To whom is he similar? is acceptable, •Than whom is he bigger? is certainly not. Furthermore, there are only a few occasi ons where such wh-questions about the Y clause are permissible. For instance, Q's question in ( 1 1 . 1 3) is a whole lot more acceptable than Q's question in ( 1 1 . 14).


( 1 1 . 7) Max gets more pocket money than me.

36 ( 1 1 . 1 3) P: Max washed the car more carefully than Ed waxed it. Q: Max washed the car more carefully than what? ( 1 1 . 14) P: Max washed the car more carefully than Ed waxed it. Q: ?*What did Max wash the car more carefully than?

( 1 1 . 1 5) Max is less stupid than his wife tried to persuade him to believe that people think he is. the degree of Max's stupidity [in fact] < [the degree of stipidity that] Max's wife tried to persuade him to believe that people attribute to him Let's now consider how we identify Y.

1 2. IDENTIFYING THE Y-FUNCTOR

The basis for identifying the Y-functor in a comparative is a syntactic parallelism between the c l clause and the c2 clause or, in the case of the specified relations of similarity and difference , the cl NP and the c2 NP. We can see from ( 1 2 . 1 ) below that both X- and Y-functors are subjects of their respective clauses, and from ( 1 2.2) that both are objects of their respective clauses. ( 1 2. 1 )

{ Morecl {antsx killed termites { thanc2 {termiteSy killed ants.

(1 2.2) Ants killed {morecdtermiteSx {thanc2 termites killed {antSy. Observe that the location of the Y-functor varies in relation to c2. The recognition that 'termites' is the Y-functor in ( 1 2 . 1 ) and 'ants' is the Y-functor in (1 2.2) is cued by the semantic-syntactic role of the X-functor. In (1 2.3) ( 1 2.5), X- and Y-functors are clause subjects:


And note that * Than what did Max wash the car more carefully? is utterly impossible. We must presume that the basis for such wh-questions as ( 1 1 . 1 1 ) is merely formal analogy and not a systematic recategorization of than. None of the other arguments that than is a preposition is any more convincing than this one. In fact, Hankamer ( 1 973), Brame ( 1 983) and Hoeksema ( 1984) all claim that there is both a prepositional than and a clause co-ordinating than; but, even if they were correct, there is no change in its semantic and syntactic function, so than is always a comparative conjunct. I shall assume that, with the excepted locutions already allowed for, c2 is a co-ordinator located immediately after the clause containing c I and X, at the onset of the clause containing Y. However, c2 can be a long way from Y, as shown in (1 1 . 1 5) where Y is elided from sentence fmal position:

37 (1 2.3)

(Morecdantsx killed termites (thandantsy were killed by termites.

( 1 2.4) The Smiths and the Robinsons went strawberry picking and pigeon shooting respectively. As you might guess, (morecl (strawberriesx_ (thanc2 (pigeonsy got eaten during the day. ( 1 2.5) is a contradiction: (1 2.5) • ( Morecdantsx killed termites (thandantsy killed termites. • the number of termite-killing ants > the number of termite-killing ants

( 1 2.6) The Smiths and the Robinsons went strawberry picking and pigeon shooting respectively. As you might guess, the Smiths picked ( mo rec l {strawberriesx(thanc2 the Robinsons shot (pigeonSy; but all were delighted with the fruits of their labours. ( 12.7) Ed has taken (morec l (bribeSx (thanc2 I've had (hot dinnersy In ( 1 2.8) the X- en Y-functors are verbs: ( 1 2.8) Liz (talkedx (morecdthanc2 Jo (Jistenedy. The parallelism even works in sequences like (1 2.9)-( 1 2. 1 0), which I find unacceptable because, to my mind, there is no reasonable conte xt in which to use them. ( 1 2.9) (•) (Morecl {ant5x killed termites {thanc2 {childreny ate ice-creams. ( 1 2 . 10) (•) Ants killed {morecdtermitesx{thanc2 Russia has { missileSy_

I have put the asterisks in brackets because some people can imagine contexts where such comparisons are acceptable; I shall be commenting on this below. Where there is a syntactic misfit, as in ( 1 2. 1 1 ), the strings are likely to be Jess than wholly acceptable. ( 1 2. 1 1 ) ??Ed and Max went strawberry picking and pigeon shooting respecti vely. As you might guess, {morecllstrawberriesx got eaten {thanc2 Ma x shot {pigeonsy; and Ed felt very sick during the evening. (12. 1 1 ) is disfavoured because it violates the co-operative principle (at least, my version of it in Allan 1986, § 1 .2) by making it tougher than necessary for the hearer to readily understand; yet in order for ( 1 2. 1 1 ) be understood at all,


The X- and Y-functors are clause objects in (1 2.6)-( 1 2.7).

38 the speaker relies o n the condition o f the co-operative principle which holds that every normal utterance is intended to make sense: the interpretation I have put on the comparative in ( 1 2. 1 1 ) is the only interpretation which does make sense. Compare ( 1 2 . 12}-( 1 2. 13):

( 1 2. 1 2) a. Muslims eat {morecdporkx{thanc2 hindus eat {beefy. b. { Morecl {porkx is eaten by muslims {thanc2{beefy is eaten by hin dus. ( 1 2 . 1 3) a. ?Muslims eat {morecl {porkx{thanc2{beefy is eaten by hindus. b. ? { Moree! {porkx is eaten by muslims {thanc2 hindus eat {beefy.

( 1 2 . 14) Jack picked {morecl {fruitx{thanc2 there were {basketsy to put it in. ( 1 2 . 1 5 ) { Morec dfruitx was picked { thanc2 there were {basketsy to put it in. Although it is undoubtedly smart-alecky and abnormal, ( 1 2. 1 6) would just about be acceptable if addressed to someone immediately able to count or reasonably estimate the number of chairs in the hall referred to:

(12. 1 6) ?Max has written {morecd bookSx { thanc2 there are {chaiTSy in this hall Once again it relies heavily on the clause of the co-operative principle which requires that all utterances normally be intended to make sense: a clause which causes the hearer to make an effort to interpret even very degraded material. We see in the discussion of e xx . ( 1 2.9}-( 1 2 . 16) that comparison depends on some likeness perceived between the comparands; in other words there must be relevance in the comparison. Because I cannot see any likeness whatsoever between the comparands in ( 1 2.9}-( 1 2. 10), nor any reasonable basis for the comparison, I find them unacceptable. Even people more imaginative than me, who will accept them, admit the comparisons are forced. Perhaps they


Once again , although the structural dissimilarity in 12. 1 3 jars and the parallelism in ( 1 2. 1 2) is undoubtedly preferable to it, ( 1 2. 13) are almost acceptable, given my interpretations - which are the only ones to make sense of the comparative. The reason for this is that like is compared with like: the NPs compared in these active/passive pairs have the same participant roles and denote similar edible phenomena, viz. meat. That is what is relevant to their understanding; but we cannot ignore the stylistic difference between (12. 1 2) and ( 1 2. 1 3) which results from the superficial mismatch. Structural parallels are the norm for compared clauses, but there are indisputably-grammatical syntactically mismatched comparatives where there is only one contender for Y-functor; e.g. ( 1 2 . 14}-( 1 2. 1 5).

39 are swayed by the fact that c l identifies the quantity scale in both these examples; compare them with the radically anomalous ( 1 2 . 1 7). ( 1 2 . 1 7) *My driveway is narrower than ants killed termites.

1 3. < OR > THAN A.DJ COMPARATIVES

There is a set of purposive comparatives which evaluate the sufficiency or adequacy of something for a given purpose. The string always concatenates< or >, c2 and the adjective of sufficiency, e.g. ( 1 3 . 1 ) His more than adequate supply kept us stoned all night. ( 1 3 .2) His less than sufficient funds led to the demise of the project. cl and c2 are identified in the usual way, and the adjective functions as both X and Y, Y always being interpreted as a contingent norm: i.e. what one would expect to be sufficient enough for the particular purpose. X is the degree of sufficiency for this occasion, which c l compares with Y. It is not only purposive comparatives that have such characteristics, but other than ADJ constructions too. E.g. ( 1 3 . 3) Hal is m ore than wealthy, he's a millionaire. ( 1 3 .4) Eve's worse than stupid brother came in and interrupted us at the crucial moment. Here too the adjective functions as both X and Y , and Y is interpreted as a contingent norm. These constructions are a rare exception to the requirement that the X-functor precede c2.


There is not a lot in common between the comparands in ( 1 2. 1 1 ), either, although the conjunction 'strawberry picking and pigeon shooting' in the first sentence suggests some degree of likeness, and the comparands are both being collected as foodstuffs; so ( 1 2. 1 1 ) is better than ( 1 2.9)-( 1 2 . 1 0). ( 1 2. 1 3) is much better still, because of the similarity between the eating of pork and the eating of beef (cf. the perfectly good comparisons in ( 1 2. 1 2)). In (12. 1 4} ( 1 2 . 1 5) there is little likeness between the comparands, but there is an obvious connexion: the baskets are to put the fruit in and there need to be sufficient of them for that purpose; so the comparison of the relative quantities of fruit and baskets is well motivated. ( 1 2. 1 6) only works because the point of the first clause is to focus on the quantity of books that Max has written, and it is this focused-upon-quantity, invoked by the number scale marker 'more', which forms the basis for the forced, and hence smart-alecky and abnormal, compa rison.

40 14. INTERPRETING COMPARATIVES

In the course of this paper I have shown how it is we recognize and interpret c l , X, c2 and Y. It remains to draw this material together. The first step from the original surface comparative is to identify c l , X, c2 and Y, as in ( 14. 1). (14. 1)

{ Moree1 {antsx killed termites {thane2 {termiteSy killed ants.

( 14.2)

{ Moree1 {antsx killed termites {thane2 {termitesy killed ants. the number of X > the number of Y for the domain I { antsx killed termite � & I { termitesy killed ant�

Note that the interpretative metalanguage statement includes all of the following. ( 1 ) a parsing of the object-language (i.e. English) sentence to determine c l , and where feasible the X-functor, c2, and Y-functor. (2) a statement of the particular comparison made in the form of two NPs, 'the number of X' and 'the number of Y' predicated by the comparative relation '>'. In this statement I have retained something like an SVO order, but it could just as well be written ">(the number of X, the number of Y)", depending how far from normal English one wants the metalanguage to diverge. The comparative relation, or predicate in the statement of compari son, is determined by the meaning of c 1 , as was shown in §§9- 10. The NPs which form the scope of the predicate consist of a head element determined from the scale marker in c 1 , which in ( 14.2) is 'the number or ranged over the variables for the primum and secundum comparationis. These are nominali zed in the metalanguage just because they are arguments of the comparative relation, and the statement of comparison in the metalanguage is modelled on English fmite clause structure. Part (3) of the metalanguage statement picks out the relevant clauses which bind the variables X and Y; notice that these clauses are conjoined, and the location of the conjunction correlates with the location of c2.


c1 is recognized via the comparative marker and the semantics of the head constituent which identify the appropriate scale. More is a portmanteau morph meaning "the quantity of X > the quantity of Y" where '>' symbo lizes "further towards the supra end of the scale named".12 In ( 1 4 . 1 ) 'more' is located within a countable NP, so the quantity is understood in terms of "number". More precisely, ' more' is ranged over the head noun 'ants', which therefore constitutes the X-functor. The use of 'more' gives rise to the expectation of the c2 than introducing the Y clause. The X-functor is located within the subject NP of the clause, so we anticipate the Y-functor to hold the comparable position in the Y clause. This seems confirmed when we find that the Y clause is a syntactic parallel to the X clause. It seems perfectly coherent to assume that 'termites' is the Y-functor in (14. 1), and so we conclude that the following gloss is the relevant one:

41 Another example: ( 14.3) Max waxed the car { lessc J { carefullyx{thanc2 Ed had washed it.

( 1 4.4) I Max waxed the car {carefully,d & j Ed had washed the car {carefully� Where no Y-functor is supplied it is assumed to be semantically identical with the X constituent, but of course contextually distinct; this is so whether or not c2 is present. Putting the various strands of this interpretation together, we get: ( 14.5) Max waxed the car { lesscl {carefullyx {thanc2 Ed had washed it the degree of X < the degree of Y for the domain I Max waxed the car {carefully,d & I Ed had washed the car {carefully� And another: (14.6) Max ate {no morecdof the cakex { thanc2 George did. Here the c l , 'no more', is the negation of more. More means "the quantity of X > the quantity of Y", so no more means "the quantity of X -> the quantity of Y" or "the quantity of X � the quantity of Y" - see §4 and §6. ·�· symbolizes "either further towards the sub end or at about the same location on the scale named." In (14.6) 'no more' ranges over the NP 'the cake' (the 'of is inserted following 'more' because of the determiner 'the'), so the X-functor is the partitive 'of the cake'. Once again the Y clause is understood on the basis of semantic and syntactic parallelism to the X clause, yielding the interpretation in ( 1 4.7). ( 14.7) Max ate {no morecl { of the cakex {thanc2 George did. the quantity of X � the quantity of Y for the domain I Max ate {of the cake,d & I George ate {of the cakey.


Here 'less' is recognized to be a portmanteau morph meaning "the quantity of X < the quantity of Y" where ' Y's height for the domain I { Lizx is of a certain width! & I { Billy is of a certain height Here the shift of scale from width to height is what makes such sentences peculiar. This particular example is joky, but serious comparisons of this nature do occur provided there is available a common metric for measure ment; cf. Thtir bed is widtr than it is long. Next, consider the interpreting of a committed comparative: (14. 1 4) Max {isx {worsecdat math {thanc2 Ed {isy in English. Notice that this is a statement about the relative capabilities of Max and Ed in these two different disciplines; a statement about their relative performance would have used performs or does in place of 'is'. The capability predicate is


Here we have the fairly uncommon situation of an X-functor being determin ed by semantic-syntactic parallel with the Y clause. Once the structure of (14.8) is understood, the translation into our interpretative metalanguage is straightforward:

43 the discontinuous 'is ... at math' in the first clause, and really the whole predicate is the X-functor; in context it will nominalize to " Max's capability at math" in the gloss, to be matched with "Ed's capability in English" for the Y clause. The c I is, of course, an adverb by function, though never in form (cf. the question How is he at math?). Worse is a committed portmanteau compa rative meaning "X is in the bad region of the bad-good (goodness) scale and further toward the bad end than is Y"; we might symbolize this as follows: X is bad & X < Y, where 'is' reflects the tense of the original, 'bad' names the appropriate end of the [goodness] scale, and ' Y's cleverness for the domain I {Maxx is of a certain cleverness! & I {Edy is of a certain clevemes�

Compare ( 14.2 1 ) with (14.22)

it fit in? Consider

( 14.22).

Max is { morecllstupidxdthanc2 Ed . ..... Max is stupid & the degree to which X > the degree to which Y for the domain I Max is { of a certain stupidityJ & I Ed is { of a certain stupidity�

(14.23)

{ Maxx is { the same heightcl{ asc2{his fathery

Here the c l means X's height = Y's height and the interpretation is straight forward: (14.24)

{Maxx is {the same heightcl { asc2 {his fathery . ..... X's height = Y's height for the domain I {Maxx is of a certain heigh� & I { Max's fathery is of a certain heigh�

Comparatives with the locutions X is the same as Y, Xis identical to Yand Xis Y are not scalar. Consider ( 14.25)-( 14.27).

different from (14.25)

{ My toothbrushx is {differentc1{fromc2 yours. ..... X :f:: Y for the domain I { My toothbrush,d & I { Your toothbrush�

(14.26)

{Thisx is {the samecdasdthaty . ..... X = Y for the domain I t ThisJ & I t thatv

Where same, identical or different is ranged over the head noun of the NP in which it is located, that noun is the scale marker for the comparison; i.e. it is a constituent of c l . E.g. ( 14.27)

{Maxx has {a different buildcl{ fromc2{his brothery . ..... X's build =F Y's build for the domain I { Maxx is of a certain build! & I { Max's brothery is of a certain buildl


Here there is an implication which overwhelms a corresponding conversatio nal implicature. And now:

45

( 1 4.28) Today we had { many fewerc t { visitorsx{thanc2 yesterday. Here 'fewer' means "the number of X identifies the degree of subcess, thus:

< the

number of Y"; while 'many'

(14.29) Today we had { many fewerc t{visitorsx{thanc2 yesterday. the number of X = the number of Y $, for $ = many [i.e. a large number] for the domain I We had { visitorsx toda� & I We had {visitorsy yesterda� -

In expressions like M times more X than Y and M times as many, much X as Y we have a degree of excess such that X = M X Y . The arithmetic is obvious, but how does it correspond with the surface sentence? Consider the follow ing: (14.30) There were { three times as manycdsupportersx{asc2 { opponentSy. Here the head of the c 1 , 'as many' means "the number of X � the number of Y", where ' �· symbolizes "either further toward the supra end or at about the same location on the named scale". Ranging over the head is the expres sion 'three times', which disambiguates 'as many', selecting the ">" sense of it. Moreover, 'three times' states the degree of excess of X over Y, namely "three times further toward the supra end of the [number] scale" which we readily interpret in context as "the number of X is three times the number of Y" or symbolically "the number of X = 3 X the number ofY". Thus ( 1 4.30) is interpreted as shown in ( 14.3 1 ).


It is notable that in ( 1 4.27) we once again have the scale named in the statement of comparison, and conjoined clauses in the metalanguage state ment of the domain for X and Y. And note that in none of ( l4.25)-{ 14.27) is there a relevant implication or conversational implicature. One question I have not properly addressed in this paper so far is the leftward extent of a c l . By this I mean the question whether the negatives in e.g. no less, not as many and the emphasized quantifiers in one more, some more, three times as many, !!E!fas much, and the like, are strictly speaking part of the c l . Certainly they ( 1 ) range over (the rest oO the c l and therefore affect its interpretation; and (2) fall within the same sentence constituent as (the rest oO the c l ; which surely shows that they are functionally part of the c l . So I shall assume from now on that a c 1 extends leftwards to the leftmost negative or quantifier within the clause constituent containing the comparative mar ker, provided that this negative or quantifier is itself concatenated with the constituent bearing the comparative marker, or that it is concatenated with a quantifier that does so and which forms part of the c l . In ( 14.28), there is a supra ranged over a sub head of a c l :

46

(14. 3 1 ) There were {three times as manyc. { supporters {asdopponentsy. (there were many supporters) & the number of X = 3 X the number of Y for the domain I There were { supporters,d & I There were {opponents� -

( 1 4. 32) We have {three times fewercdstudentsx this year {thanc2 last year (there were few students this year) & the number of X X 3 = the number of Y for the domain I There were {studentsx this yea� & I There were {studentsy last yea� -

(14.33) There were {three times3x{as manycdsupportersx {asc2{opponents;y. - the number of X I 3 = the number of Y for the domain I There were { supporters,d & I There were {opponents� There is an interesting discrepancy between language and mathematics here. In language ( 1 4. 34) and ( 1 4.35) are converses: (14.34) There were three times as many supporters as opponents. (14.35) The number of supporters was three times the number of opponents There is a direct mathematical translation of ( 14.35) into X = 3 X Y, as shown in ( 1 4.36): (14. 36)

{ The number of supportersx {was,. {three times the number of oppo nents3xv·

But there is no direct mathematical translation of ( 1 4. 34). The mathematical converse of X ""' 3 X Y is X/3 ""' Y, which corresponds not to ( 1 4.34), but to the very unnatural ( 1 4. 37)-( 14. 38). { 14.37) ?The number of supporters divided by three was (equal to) the number of opponents. (14. 38) ?One third of the number of supporters was (equal to) the number of opponents.


Here 'fewer' means "the number of X < the number of Y", with 'three times' stating the degree of subcess of X below Y on the number scale. Thus we interpret 'three times fewer' as "3 X X = Y". As I pointed out in §3, the metalanguage statement of the comparative {1 4.3 1 }{ 14.32) is the one preferred to the semantically equivalent ipterpreta tion emphasized in ( 14.33).

47 Another discrepancy between language and mathematics is the following. English does not have ( 14.39) •x · IS

{

twice fifty-nine times

}

l

longer shorter better worse

than Y

But we do get idiomatic hyperbole like ( 14.40). ( 14.40) X is ten times

��;r

better worse

than Y

'Ten times' is an idiomatic hyperbole that usually means "very much" or "a lot, even as much as ten times". For instance, X is ten times worse than Y would be interpreted: "X is bad, & X is very much further towards the sub end of the goodness scale than is Y." I take it that because MX expressions (like twice) are implicitly supra they do not sit well with a numerical sub; hence twice asfew contains a uncomfor table juxtapositioning of supra and sub. As I remarked in §4, fractions like half as many are preferred to twice as few. In half as many, the 'many' is uncommitted, whereas the sub few is committed.

IS. SUMMING UP

It has been my intention in this paper to relate the forms of English compara tives with their meanings, and to create an English-based metalanguage to capture those meanings. Interpretation begins with a parsing of the object language sentence to determine c 1 and the X-functor, and if they are present, the c2 and the Y-functor. Just occasionally, the X-functor has to be recogniz ed as a zero anaphor of the Y-functor, cf. ( 14.8)-( 1 4. 1 1 ). Although both c 1 and c2 are readily recognizable on a formal basis, I have sought to show that interpreting a comparative depends primarily on the recognition of the comparative operator c l ; indeed, since the Y clause can be omitted altoge ther, c1 is necessarily the key to comparison. I claim that c1 has two constituents: a comparative marker and a scale marker. In the c 1 s listed in § 1 , Q-er, ADl-er, AD V-er, as/so Q, as/so ADJ, as/so AD V, the same (N), identical (N), different (N), more, less Q, ADJ, AD V and N are scale markers; -er, as, so, the same, identical and


1� l

48


different are comparative markers. A very few c l s bundle the two consti tuents together in a portmanteau morph, e.g. more, less, worse, but these are readily recognized and all other c 1s are easily identified from the comparative marker. We saw in § 14 that a c 1 may extend to the leftmost negative or quantifier within the same clause constituent as the comparative marker that is also ranged over and chained to the comparative marker. Although there is a very small number of comparative markers, there is a large number of scale markers. Indeed, they form an open-ended set, even though much comparison relies on the default scale of quantity. The fact that quantity is the default scale seems to have misled some scholars into assuming that only quantities are compared; but even though quantity is the most prevalent scale, I have shown beyond doubt that comparison is scale based. Consequently, the symbols , which I use to represent the comparative relations indicated by the c 1 , actually identify relative locations on a scale and not relative quantities - except, of course, when the scale denoted is a number or quantity scale. Scales have a supra end and a sub end, and we found that some supras and many subs are 'committed' in compara tive constructions. The uncommitted term allows the relative locations of X and Y to be anywhere on the scale named in the comparison (within the definitions of the comparative relation); by contrast, a committed term locates either X or Y or both in one region of the scale, i.e. on just one side of the halfway mark in a binary scale. However we found that the use of even an uncommitted term normally conversationally implicates the end of the scale which that term identifies; this is the 'pull effect'. The more marked a term is, the stronger the pull; i.e. the more likely is the implicature to be true. Not surprisingly, a speaker's choice between semantically equivalent comparative expressions is typically determined by their respective implications and impli catures, where it is not already contextually determined by the choice of what is to be X and what is to be Y (cf. §2). c i s identify relative locations on a scale for the comparands X and Y. X is (the denotatum of) the scope of c 1 , but is necessarily interpreted within the context of the clause in which it occurs. c2 marks the end of the X clause (except in cases of heavy-NP shift, and other rightward placings of clause constituents to the end of a sentence), it also introduces the Y clause - indeed, this is its very function. c2 is either than or as except in the locutions X is identical to Y and X is differtntfrom Y, where it is 'to' and 'from' respectively. These two locutions along with X is the same as Y are monoclausal, conse quently X and Y are NPs instead of constituents within separate clauses, as they are in all other comparatives (cf. § 1 1 ). Y is recognized via the fact that both it and the Y clause are typically semantically and syntactically parallel to X and the X clause. I proposed an English-based metalanguage for interpreting English com paratives. Statements in the metalanguage have at least three and possibly four parts: ( 1 ) A parsing of the object language sentence to determine c l , and where feasible the X-functor, c2, and Y-functor.

49

NOTES • I should like to thank Tony Back:house, Adrienne Lehrer, Jan Rusiecki, Aubrey Townshend and the anonymous reviewers for JS for their enlightening comments on earlier versions of this paper. None of them is in any way to blame for the faults in this one. I. See below and § 1 2. 2. Cf. Allan ( 1986, § 1 .2) for post-Gricean aspects of this. 3. Others might use 'entailment' where I have used 'implication', but I prefer not to for reasons explained in Allan ( 1986:11, 385). 4. I shall show in §9 that the C scale divides up as follows: SUB MIDDLE SUPRA > < � � For the reinterpretation of ±$ see (9. 1 7). S. True identity is expresse d by X is Y, which is equative, not comparative. 6. I shall use •-op• where it might be appropriate to read "possibly P'', given my intention to stick as closely as possible to natural English. In natural English '(Q - 0 P) - (Q - 0- P)' see ms intuitively legitimate, although it is logically false. The reason for the intuition is that ·o.,.p ..oP' and 'OP - -op• where ·- · symbolizes conversational implicature. Incidentally, it may be assumed for this essay that 'A - P' is equivalent to 'A - DP'. 7. I ass ume that the relationship between good and betttr is not controversial. The implication in (9. I ) that AI is not obnoxious is no part of the kind of comparative construction I am discussing.

8. The C scale does not necessarily have a simple name, i.e. it is not necessarily lexicalized. 9. Although supras like tal� good, m011y etc. are supposedly uncommitted in how-questions, it ain't necessarily so. There is the following prosodic difference between the committed and the uncommitted how question (prosodic conventions as in Allan 1986, ch. 6.): Uncommitted: I How glkld was the dinner party? I i:Xxl w4 s Committed: I How g the dinner party? I


(2) A statement of the comparison in the form of two NPs predicated by the comparative relation from the set , and =1=. The comparative relation, or predicate in the statement of comparison, is determined by the meaning of c l , as was shown in § §9- 10. The NPs which form the scope of the predicate consist of a head element determined from the scale marker in c l ranged over the variables for the primum and secundum comparationis, X and Y. (3) A conjunction of the clauses which bind the variables X and Y; or, in the case of the locutions X is the same as Y. X is identical to Y and X is different from Y a conjunction of NPs that bind X and Y. These are determined by interpreting the X-functor and the Y-functor within the context of the X and Y clauses respectively. (4) Where relevant, a statement of the implication and/or implicature arising from the committedness or pull of the term used in the comparison, or of the X-functor in comparisons where the default scale is used. In my view these metalanguage statements offer a more comprehensive interpretation of the meanings of English comparatives than any other account so far presented. But it remains necessary to widen the range of data discussed and to improve on the very coarse and informal metalanguage I have employed.

50 10

II.

See note 3 .

A possible exception is in the substandard different than. 'Than' has apparently been

attracted to the c l different which compares unequals because than is so frequently the c2 in inequality comparatives. Yet one hears constructions like Max i.J different than I am in which 'than' is definitely not a preposition, as well as Max i.J different than what I am where 'what' =

"that which", so 'than' could be a preposition. Incidentally, even if I am wrong about than and it

is sometimes a preposition, there would be no serious consequences for the thesis of this paper.

the object of the purported preposition than would be Y, and Y would necessa rily be interpreted

in a context denved from the X clause. 12.

The supra end of the quantity scale is, of course, "greater quantity"; the sub end is "lesser

quantity" - which leads us into an inescapable tautology in exposition.

REFERENCES

Allan, Keith 1 987: Hierarchies and the choice of left conjuncts (with particular attentiOn to Eng lish). Jowrnal of Lingwistics 23. 1 . Atlas, Jay 1984: Comparative adjectives and adverbials of degree : a n introduction t o radically radical pragmatics. lingwi.Jtics and Philosophy 7:347-77. Boucher, Jerry & Osgood, Charles E. 1 969: The pollyanna hypothesis. Jowrnal of Verbal Lear

ning and Verbal Behavior 8: 1-8.

Brame, Michael 1983: Ungrammatical notes 4: smaner than me . Lingwi.Jtic Analysi.J 1 2:323-38.

Bresnan, Joan 1973: Syntax of the comparative clause construction in English. Lingwistic ln

qwiry 4:275-343.

Bresnan, Joan 1975: Comparative deletion and constraints on transformations. Lingwistic

Analysi.J 1 :25-74. Cruse, D. Alan 1 976: Three classes of antonyms in English. Lingwa 38:28 1 -92. Cruse, D. Alan 1980: Antonyms and gradable complementaries. In: Dieter Kastovsky (ed.)

Perspektiven der lexikalischen Semantik, Bouvier Verlag Herbert Grundmann, Bonn, 1 4-25. Grice, H. Paul 1 975: Logic and conversation. Peter Cole & Jerry Morgan (eds.). Syntax and

Semantics 3: Speech A cts. Academic Press, New York, 4 1 -58.

Hankamer, Jorge 1973: Why there are two than:r in English. Papers from the Ninth Regional

Muting of the Chicago Lingwi.Jtic Society, 1 79-88. Hellan, Lars 1 98 1 : Toward.J an Integrated Analysi.J of Comparatiw:s. Tilbingen, Narr. Hoeksema, J. 1 984: To be continued: the story ofthecomparative.Jowrna/ ofSemantics 3: 93- 107. Klein, Ewan 1980: A semantics for positive and comparative adjectives. Ungllistics and Philoso-

phy 4:3 2 1 -58. Klein, Ewan 1982: The interpretation ofadjectival comparatives. Jowrnal ofLingwistics 1 8 : 1 1 3-36. LakoiT, George 1 972: Hedges: a study of meaning criteria and the logic of fuzzy concepts. Papers

from the Eighth Regional Muting of the Chicago lingwi.Jtics Society, 1 83-288. Lehrer, Adrienne 1985: Markedness and antonymy. Jowrnal of Ungwi.Jtics 2 1 :397-429. Lehrer, Adrienne & Lehrer, Keith 1 982: Antonymy. Linguistics and Philosophy 5:483-501 . Rusiecki, Jan 1985: Adjectives and Comparison in English: a Semantic Stwdy. Longman, London. Ryan, Karen L. 1 983: Than as coordinator. Papers from the Nineteenth Regional Muting of the

Chicago Lingllistics Society, 353-6 1 . Pinkham, Jessi e E. 1982: The Formation of Comparative Cla��Ses in French and Engli.Jh. Indiana University Linguistics Club, Bloomington.

Sperber, Dan & Wilson, Deirdre 1982: Mutual knowledge and relevance in theories of compre hension. In: Neil Smith (ed.). Mwt��al Knowledge, Academic Press , New York, 6 1 -85.

Stassen , Leon 1984: The comparative compared. Jowrnal of Semantics 3: 1 43-82.

Von Stechow, Arnim 1984: Comparing semantic theories of comparison. Jowrnal of Seman

tics 3 : 1 -77.


Allan, Keith 1986: Linguistic Meaning. Routledge & Kegan Paul, London.

Journal of &mantics 5: 5 1 -64

THE SEMANTICS O F THE STORYLINE IN EAST AND WEST AFRICA

ROBERT E. LONGACRE

ABSTRACT

0. INTRODUCTORY CONSIDERATIONS

The sine qua non of a narrative is that it presents actions and events which are (a) sequential, (b) punctiliar, (c) at least in part causally connected. Since a discourse is not a story unless it is so characterized, it follows then that the structurally most important part of the story is its storyline, i.e. the sequence of sequential, punctiliar and (at least partially) causally connected actions and events which are represented in the narrative. A storyline reports not simply events but actions, i.e. voluntary doings of animate (usually human) agents. Events, in the sense of contingencies, need not be occasioned by agents nor be caused by preceding actions. Thus a story may contain the actions: ( I ) Maurice eloped with the banker's daughter; (2) They commandeered a company car, (3) They set out for a resort in the Adirondacks. But it may also record events which are sequential and punctiliar (as are l -3 above) but which are not the doings of agents nor causually consequent on previous actions: (4) They had aflat tire 30 miles down the road; and (5) They got caught in a violent electric storm . Certainly such events as well as voluntary actions of agents are


llus paper starts out with the definition and exemplification of storyline and non-storyline elements in narrative with emphasis on their structural and semantic functions as narrative universals. The schema thus developed is then confronted with languages of diverse word order typologies: SOY (Ethiopia), YSO (certain Nilotic languages of Sudan), and SYO (a number of West African languages). The major storyline consideration in the Ethiopian languages which are represented is the matter of the storyline scheme status of various kinds of "gerunds" (non-fmal verbs) as opposed to final verbs in chaining structures; this is seen to entail some basic semantic distinctions. Some Nilotic languages, of which Luwo of Sudan is representative, have strict YSO structures on the storyline but various kinds of NV structures ofT the story line; here again various semantic distinctions are entailed. The function of consecutive tenses in various YSO and SYO languages is then considered along with the consideration of storyline schemes in SYO languages which do not have special consecutive tenses. A general paraUelism of the medial and final clauses in SOY languages to imtial and consecutive clauses in YSO and SYO languages is noted. Several further parallelisms and differences among SOY, YSO, and SYO languages are noted.

52


part of the storyline. Events often clearly cause subsequent actions and events without being themselves necessarily caused by what precedes them. In most languages, clauses that advance the storyline of a narrative are formally distinguished in some more or less consistent way from clauses that do not advance the storyline. Often storyline clauses require a distinct type of verb. In any event, however, they are in some way recognizable as represen ting happenings which are puncti/iar and sequential. The identification of a storyline within a narrative constitutes, however, only a port of entry into the semantics of narrative structure. Storyline clauses, except in the most rudimentary and stripped-down kind of narrative, are accompanied by further clauses which report other sorts of information and which may be ranked in order of progressive degrees of structural departure from the storyline (cf. Longacre 1 98 1 and 1 982; Grimes 1 975; Jones and Jones 1979; Hopper 1979; Hopper and Thompson 1980). Thus, if storyline clauses are ranked ( 1 ), we may also find: (2) clauses which report backgrounded actions/events which are considered to be preparatory to or resulting from those which are reported on the storyline; (3) clauses which report backgrounded activities which are overlapping rather than sequential, and durative rather than punctiliar; (4) setting clauses (the when and where of the story; the circumstances; introduction of participants; background expla nation) which have 'be' verbs, whether existentialllocational or copulative/ descriptive, or other stative verbs; (5) Irrealis clauses (modals, conditionals, and most negatives) that present an alternative world to the world given in the storyline; (6) author intrusions (performatives, evaluation, and morals); and (7) cohesive clauses (adverbial) which refer back to previous parts of the storyline, either by reporting a new but script-predictable event or by anapho ric reference to the former event (cf. Diagram I). In regard to this scheme: ( A) Since it is assumed that verbs are central to their clauses, the ranking of the latter in regard to the storyline of narrative in a language is also a ranking of verb forms (aspects, tenses, and moods) with reference to narrative. (B) Clauses without verbs, i.e. nominal clauses, can occur in setting. (C) The scheme especially classifies independent verbs in main clauses. Adverbial clauses, relative clauses, and most verbals are demo ted. Thus a verb which, if independent, would have been on the storyline may be demoted to (2) when it occurs in an adverbial clause, relative clause, or as a verbal - but not simply cohesive as in (7). (D) Some languages have a special pluperfect form which determines a variant of (2) above; more often, langua ges do not have such a form. (E) The ranking is structural in regard to how a story adds further material to its storyline; semantically, however, an item of fairly low rank may nevertheless be of crucial importance to the story. I summarize this ranking in Diagram I. The accompanying Tabulation I illustrate such forms in English. Where only a part of a sentence is relative to a given ranking that part of the sentence is italicized. The example sentences which are given here are not given in sequence but are assumed to all be taken from the same narrative.

53 1

Storyline

•

2

Backgrounded actions/events

•

3

Backgrounded activities

•

4

Setting

•

5

Irrealis

•

6

Author intrusions

•

•

Cohesive Diagram I

I . We climbed over the wall . . . The guard yelled out a warning.

2. Jumping up and grabbing the edge, I managed to pull myself up. Then I tossed down the rope which Harry had brought. 3. All the time the two 'decoys' were making their way around to the other side of the property. 4. It was a small, somewhat bombed-out little town in Normandy, three weeks after D-day . . . . The wall was about seven-foot high and made of crumbly old masonry. 5. If it had only been two hours earlier, all this wouldn't have happened. 6. I remember that wall and I defy anyone who thinks we could have got over it noiselessly. 7. About

as

soon as we got over the wall, the hue and cry of the chase began. Tabulation I

It is my claim that the scheme here presented is an integral part of story-telling anywhere in any language. Obviously, however, the surface structure textures of languages, including the verb systems which they em ploy, differ greatly around the world. Thus, in crossing central Africa from East to West we find typologically distinct languages} On the extreme East on the hom of Africa we find the Cushitic and Omotic languages, and (by assimilation) Ethiopian Semitic, all of which are SOV languages; in typologi cal structure they resemble Japanese, Korean, most languages of Papua New


7

54

I. ETHIOPIAN LANGUAGES I have already briefly characterized Ethiopian languages (Cushitic, Omotic, and Semitic) as SOV chaining structures with or without explicit morphologi cal switch-reference. The medial verbs in such languages have been called by various names (converb, conjunctive, consecutive, participle, gerund, and gerundive) by students of languages in that part of the world (cf. various contributors in Bender 1976, and Hetzron 1976). I here retain the term 'gerund' with, of course, the caution that 'gerund' here is not so much like the IE structures with that name as like the medial verb structures found in, e.g., Papua New Guinea and northern S.A. The 'gerund' is not so much subordi nated as simply positionally prior. In fact, three successive clauses which contain gerund, gerund, and independent verb respectively may be a close counterpart of three coordinated clauses in English: "/ went over, picked up the spear, and thrust it into the palm tree. " Taking a hypothetical string of clauses with gerund+gerund+independent verb there are three theoretical possibilities regarding the mutual storyline ranking of the clauses within the string: A. Only the final verb is storyline; preceding gerunds indicate rank (2) backgrounded actions/events, (3) backgrounded activities, or some still lower rank. B. Gerunds are of the same storyline rank as the verb on which they depend. C. The final verb is itself off the line (as a somewhat formal, semantically empty device) and the preceding gerunds consistently outrank it. While B above seems to be by far the commonest situation, some evidence


Guinea, many languages of northern South America, and a few languages of the southwest comer of the U.S. These languages are chaining structures, where medial clauses of some sort precede a final clause which is distinctive in form by virtue of having a distinctive final verb. Some of these languages have developed switch reference systems in their morphological structure; others have not. The Nilotic languages of East Africa present, however, a very different picture. Some of these (e.g. Luwo of Sudan) are basically VSO in structure and are typologically resemblant to Biblical Hebrew and certain languages of Mesoamerica. Finally, West African languages present us with SVO structures. In both VSO and SVO languages of Africa we may find a consecutive tense as discussed below. SOV, VSO , and SVO, three of the main word-order typologies found around the world, are all represented in moving from East to West across central Africa. The question arises: How do languages of these diverse types tell a story, or more specifically, how do they mark the storyline and the degrees of devia tion from that line?2 It is this question which I want to explore in preliminary fashion in this paper. I will confine myself to the upper levels of the scheme given in Diagram I.

55

·

. . . hamagus n when-he-went norgnis butter (0) eti taking

mat'nasi . . . . the-other (S)

kotabek'us a-little

yi its

ton top

yici putting

gabm hank'u market (L) went (3m)

"When he went, the other one put a little bit of butter on its top, took it and went to market." Here hank'u 'he went' is the final storyline verb. The clause which ends with yici 'putting' and the clause eti 'taking' have unmarked gerunds which are on the storyline by virtue of preceding a finite storyline verb. Other forms of the gerund in Gimira, e.g. perfective and imperfective gerunds, rank as less salient than the storyline verbs on which they depend. In Kambatta (data from Mirya Saksa), unmarked gerunds may similarly precede a storyline verb and group with it on the storyline:


also exists for (A) and (C). Thus in Koorete (data from Carolyn Ford) possibly only the final verb reports the primary action/event; preceding verbs can best be considered to report secondary actions/events that are preparato ry. In other examples the gerund gives the stimulus or grounds of the action/event reported in the main verb, or gives a paraphrase of the main verb that follows. But in every case the main verb seems semantically to dominate the storyline provided that it has the structure of a simple past. At the other extreme may be Afar (some texts studied by Loren Bliese and Yvonne Genat), which possibly exemplifies hypothetical possibility (C) above. Here the storyline is carried by gerunds ('perfect participles' in Bliese 198 1 ) and the distinctive final verb may b e semantically somewhat vacuous; i n the text which I saw the latter was simply the verb say and was, in effect, a disclaimer of responsibility on the part of the story teller. This is reminiscent somewhat of certain chaining languages of the SW United States (e.g. Tolka paya, Hardy 198 1 ) where final verbs are semantically auxiliary to the verbs which precede them. For most Ethiopian languages it seems reasonable to believe that alterna tive (B) above is to be posited, i.e. final verbs ofthe requisite tense-aspect are on the sto ryline along with certain gerunds that depend on them. It is necessary to specify certain gerunds, since e.g. a language may have several different types of gerunds some of which are clearly off the storyline. Thus in Gimira and Kambatta it is the unmarked gerund which is on the storyline when preceding a storyline verb. For Gimira, note the following example (from Mary Breeze):

56 hikkanniic zakkiin maaliic matu that after of meat k'at'a icci portion eating illisi eyes agurri leaving

(okk'a?i lifting

duubbi jamma"i c'o?u getting-full to skin starting wosicu (uujji the-dog seeing

"wo§aanwo§e" yee?i "Doggie" he-said

Here the unmarked gerunds, which are italicized, precede the fmal verb yee?i 'he said' and, together with that verb have storyline status. On the other hand, Kambatta gerunds that are specifically marked (?ani for simultaneous actions by the same subject, -aniyaan for simultaneous action by a different subject or -aan for successive action by a different subject) are a kind of secondary storyline (cf. (2) in Diagram 1). In Gimara, Kambatta and other languages (Haddiyah and Amharic) a sequence of relatively short sentences can occur in which the storyline is carried mainly by independent verbs with few accompanying gerunds. Thus, for Haddiyah, Elsa Korhonen reports that the storyline gerunds, i.e. unmar ked gerunds, are of fairly low incidence before the main verbs. In fact, the main verb occurring unaccompanied by a gerund or at the most by one gerund is the usual situation in routine narration. At a great moment of a story, however, a long string of gerunds can attach to one main verb. She cites such an example in which ten storyline gerunds are found preceding the main verb. In Gimira, Kambatta, and Haddiyah subordinated verbs (with a formal subordinator) must be distinguished from gerunds and are considered to be still lower rank in reference to the storyline. Amharic (Biiese 1 984) presents a somewhat simpler situation in which there is only one type of gerund which, along with the independent verb, carries the storyline. Amharic has, however, a number ofways to subordinate verbs. To summarize these observations regarding Ethiopian languages (based on data from Gimira, Haddiyah, Kambatta, and Amharic): I . The storyline is carried by independent verbs which are simple past indicative forms and by unmarked gerunds which are dependent on such forms. 2. Other independent verbs, e.g. perfectives, imperfectives, pluperfects and 'be' verbs in various tenses, distribute out along bands (2)-(5) in the main


"After that he ate part of the meat, got full, started to skin it, lifted up his eyes, saw the dog, and said 'Doggie'."

57 diagram above. This enables us to calibrate deviations from the storyline across sentences. 3. The richness of structure in regard to distinctions among gerunds and among subordinated verbs indicates the need to posit degrees ofsubordina tion which can be calibrated according to deviation from the storyline within the sentence.

2. CERTAIN EASTERN SUDANIC LANGUAGES

55.

RuOhw E morrow and

56.

E and

57.

uyEdhE wihy yaadh and-climbed-3s on tree

nguuh lion

adOOa dOm returned forest

kardahd place-of-hunt

hyin mOOgO acahg bEEhh kEEhd. boy another repeated come again


At least some of these languages have a basic VSO structure. Teposa (North eastern Nilotic, data from Martin and Helga Schroeder) has a consistently VSO structure. Murle (data from Cindy Miller) has a VSO structure which shifts, e.g. to SV(O), when the subject is put into focus. I assume that the situation in Murle is somewhat like what we fmd in Luwo as described in the balance of this section. Luwo (Northern Nilotic, Randall Buth 1984) has a strict VSO order for storyline clauses. Other clauses with word orders SVO, OVS and even SOV are considered to reflect degrees of deviation from the main storyline. The situation is complicated, however, by the presence of VSO consecutive clauses (with an u- preftx) which have a two-fold function: ( 1 ) As main storyline of narrative discourse; and (2) as continuing a construction in any other tense, aspect or mood. In the ftrst function u- verbs are punctiliar and sequential. They begin sentences (and even paragraphs) and are not simply a continuation of some other verb form. In the second function they take the semantic value· of the tense in the preceding clause, whether habitual, impera tive, present, or future. In the ftrst function they are autonomous while in the second they take their cue from what precedes. The initial problem in describing storyline clauses for Luwo is one which we ftnd for several languages of this area: In place of one storyline verb form we apparently have several. In accordance with a basic assumption of dis course analysis we do not simply assume that the forms are equivalent and used interchangeably. What then are the semantic/functional distinctions which are involved? One of the basic oppositions in Buth's analysis is between u- verbs and a- verbs. Take, for example, the following sentences of a folktale, 'Uriid and the Lions', which Buth has analyzed:3

58 58.

uchagE bOOgE kE thOOhr yi repeated-3s leaves to throw in

pany mortar

"The next day the lion went off again to hunt in the forest and again, another boy came (to the house). He climbed up on a tree and again threw leaves into the mortar."

cahq nyaahE repeated the-girl

59.

Maa and

60 .

ucOOnE ngO piny and called-3s 3s(O) ground

ngO 3s(O)

kEE niid to see

"Once again the girl saw him and called him down." In 59 a further and quite important verb form in Luwo narration is found: viz. verbs introduced with maa. Here a further opposition u- verbs versus maa verbs is illustrated. Notice that in the paragraph given, i.e. lines 55 ff., while 55 and 56 present prelimina ry events and 57 and 58 initiate the main storyline of the paragraph, it is 59 and 60 that bring us to the main concern of the story at this point, i.e. (another) boy meets girl. Here maa marks the onset of the more thematic material. The paragraph continues with u- on a verb of saying and gives the girl's warning to the boy (the lion might eat him the way a boy was eaten the day before). Another maa verb introduces the boy's refusal to go away, while u-verbs continue the story line: (the boy) went inside the house, and hid in the hole of the grindstone; the girl came and covered him with a sieve. Thus while the storyline of this paragraph is mostly carried by clauses which contain u-verbs, maa verbs mark the more significant actions: the girl discovered the boy and warned him, but he refused to listen (and embarked on the same


Two typical uses of a- verbs as opposed to u- verbs are seen here: ( 1) a- verbs typically occur early in a paragraph and indeed an initial spate of them is typical of the onset of a story. (2) They permit a noun (here subject) to come in front of the verb, while u-.verbs must stand clause-initial. From a study of many such instances - and from further data - Buth has generalized as follows: 'While a- verbs are usually punctiliar and perfective they are back grounded in reference to u- verbs. Clauses with a- verbs are preparatory or ancillary to clauses with u- verbs. This seems plausible in that rotation of nouns to the fore of an a- verb apparently throws a participant (e.g. the lion and the boy) into focus rather than their actions.' Buth goes on to note that a- verbs, unlike u- verbs can be used for 'failed' events (things that don't happen), as well as in relative and adverbial clauses. Continuing with the text, a portion of which was given above, I now give the next sentences:

59

------- + thematic maa ( I ) +

.,

------- � s.q;;;,;;; �

+perfective

-thematic u- (2) + thematic akE (3)

+ ancillary

�

-thematic a- (4)

Diagram /1

To complete Buth's storyline scheme, the imperfective (= the imperfect) is opposed to the perfective, while all the independent forms are opposed to dependent forms in clauses which are subordinated with naa 'when' (+ a verb) and gE 'whenever' (+ imperfective). At the bottom of his scheme (which does not go into matters of collateral or author intrusion) are clauses with 'be' verbs. Interpreting his scheme in terms of my diagram I, we can clearly see the ·


course of action which had Jed to the death of the first boy the day before). All this illustrates Buth's contention that both maa verbs and u- verbs (which are similar in their exclusive use of VSO order and rejection of noun fronting) are storyline but that maa marks a sentence (or in a few instances the last clause of a sentence) as thematically prominent, or more salient, as opposed to u-verbs, which mark the more routine parts of the storyline. One further verb type occurs, i.e. verbs preceded by a particle akE. Clauses which contain a verb so marked are SV rather than VS and may not contain a definite direct object but only an oblique direct object. These features contrast akE verbs with both maa and u- verbs which are VS(O) and unrestric ted as to definite versus oblique object. On the other hand, akEverbs occur in clauses which indicate some portentous, although backgrounded event. Thus, in the staging of the story which is referred to above, a pair of clauses with a-verbs record that a woman gave birth to a girl who grew and matured while a following clause has an akE verb "and she became quite beautiful". The latter piece of background information is of considerable importance because her beauty provoked the lion to steal her and carry her off to his home. Likewise, after she has hidden the first boy in the house and is trying to assure the lion that no one is there, 'akE the rooster crowed' and told the lion where the boy was hidden. All such clauses marked with akE indicate pivotal points where a "change in fortune" occurs. Since however akE clauses resemble more the background clauses with a-verbs, Buth considers that akE indicates a thematic although backgrounded element. All this suggests the arrangement for Luwo as in Diagram II (taken from Buth).

60

3. WEST AFRICA

Here we briefly consider storyline and related matters in languages of basical ly SVO structure. Languages represented in this study are spoken in Nigeria, Benin, Togo, Ghana, and Liberia. Some of the stocks that these represent are however, not restricted to West Africa as such but range over into East Africa (e.g. Niger-Congo is represented by MUndO, Ndogo and Biri, which figured in the Sudan part of our research program). Our interest here is in the narrative structures of these SVO languages as opposed to such structures in the typologically SOV languages of Ethopia and in the VSO Nilotic languages. As we have mentioned, however, a point of comparison arises between the latter and certain SVO languages which have a consecutive tense. Many of the languages represented in our sampling of West African structures have a rather straightforward structuring of the storyline, i.e. there is a special narrative form (whether marked or unmarked) which carries the storyline and there is no distinct consecutive verb form - even though serialization may occur and require some feature such as repetition of the pronoun before each clause. Thus for Izi, Kaje, Mumuye, Nawdm, Gangam, Frafra (all of which are Niger-Congo) as well as Bass a (Kru), Ewe (Gbe), and Mwaghavul (Western Chadic).


storyline band (with a thematic versus non-thematic split); backgrounded activities (imperfectives); and setting (use of 'be' verbs in Luwo). Subordina ted clauses are very rare in Luwo narrative because most of their function is taken up by ancillary a-verb clauses. Nevertheless, it is important to remem ber that, unlike English, which distinguishes only main clauses from subordi nated clauses, Luwo has a three-way distinction: main clause, ancillary clause, subordinated clause. The above distinction, exemplified for Luwo, where foregrounded and backgrounded events create, in effect, primary and secondary storylines, is also seen in some other area languages, e.g. Sabaot (Southern Nilotic), Jiir-Modo and Avocaya (both Central Sudanic), and in MOndO (Niger-Con go). In spite of many mutual differences all these languages have, like Luwo, a sequential tense which is storyline while backgrounded events are specifically marked as perfectives. These primary and secondary storylines (if one prefers to call them that) are both opposed to imperfectives which encode back ground activities. All these languages also have subordinated clauses which are distinct from all the above. In MUndO and Sabaot it is clear that the consecutive (or sequential) tense, while found on the storyline of narrative, can also be consecutive on verbs of other types when they precede it and take on the function of what precedes. This feature also proves important in many West African languages - and significantly enough - MOndO is Niger-Congo in its affinities although spoken in Sudan.

61


Several of these languages have, however, vanatwns of the storyline clauses which determine various further semantic distinctions. Thus, in Mwaghavul (Western Chadic data from Stephen Niyang) be 'then' regularly precedes the neutral (i.e. storyline) form, dang be,4 a somewhat stronger 'then' occurs at the onset of a major paragraph division, and (be) yaksi occurs introducing one or more clauses at the peak of a story. In Mumuye (Niger Congo data from Dan Juma Gambo) the connective tUU 'then' 'now' " marks the beginning of the next event or longer unit" (paragraph), while the connective wEE dUU naa bee 'in a short time' , has a somewhat opposite implication: "This connective keeps the sequence of actions together" (Gambo). In Bassa (Kru) (data from Donald Slager) all the storyline forms (neutral, i.e. unmarked) must be preceded by a connective; in the absence of a connective they do not advance the storyline but rather paraphrase, amplify, or explain an already reported event. But the connectives do not all have the same semantic import. Thus pu marks the onset of a pivotal event, while hwe ke marks an event which is represented as resultant on a previous event, and ke is a less specific and more neutral connective. Connection before a storyline verb can also take the form of a time phrase or an adverbial clause; these forms of connection probably mark divisions of the text as well. Other languages have a special consecutive verb form which must be preceded by a regular storyline form if it is to be on the storyline, otherwise a consecutive form is some non-storyline category according to whatever precedes it. Thus in Yom (Gur, Niger-Congo, data from D. Forsberg) there is a consecutive tense (called 'aorist' by Beacham 1968). The storyline forms are perfect (barring Ia man + perfect to indicate a flashback, and perfects in subordinated clauses). When the consecutive tense follows a storyline perfect then it itself is also storyline. It can, however, follow other non-storyline forms which it can be considered to continue. Likewise in Tern (also Gur, Niger-Congo, data from Robert de Craene) there is a special storyline form that is aspectually completive while background activities are incompletive or stative. But any of these forms and others can be followed by a consecutive form which continues the mode and tense of what precedes it; again, the on-or-off status of a consecutive form in respect to the storyline depends on what precedes it. I have a note from E. Osam to the effect that Fante (Kwa, Niger-Congo) makes a similar use of consecutive verbs. Obolo (also Niger Congo, data from Uche Aaron) is similar but offers a somewhat more complicated situation. Notice both the parallelism and the contrast of structures of this sort with those seen in Ethiopian (Cushitic, Omotic, Semitic) languages (Diagram III). In summary: (I) the SVO languages here cited are of Head +Accompa nying structure, while the Ethiopian languages are Accompanying +Head structure. The storyline status of the whole chain depends on the Head. Notice also the difference between the SVO structures with consecutive tenses and Luwo (VSO), which is superficially similar. In Luwo the model is not +Head +Accompanying clauses all of which are (or are not) storyline.

62

Ethiopian SOV Language (Cushitic, Omotic, and Semitic)

special initial storyline form

special final storyline form

consecutive clauses follow initial clauses

medial clauses precede final clauses

consecutive clauses have the same meaning and narrative function as initials

medial (unmarked) gerunds have the same meaning and narrative function as finals

D�agram III: Comparison of SVO Chaining and SOV Chaimng

Rather in Luwo the form which is storyline in narrative has the same form as the consecutive tense off the line. However, the occurrence of the storyline forms in narrative is not dependent on any obligatory initial clause. Any clauses which precede are secondary in regard to the storyline. One of the overt clues here is that the a- forms (ancillary) which precede the u- forms (storyline) may rotate a noun to the front of the verb while u-clauses are strictly VSO. Also, as was seen, a-clauses sometimes represent activities rather than events and occur in both subordinated and negative clauses while u-clauses can occur in neither. We encountered three types of storyline conventions in SVO languages vis-a-vis with Luwo: ( l ) Languages with only a marked/neutral storyline form and no consecutive form. (2) Languages with an initial storyline form along with a consecutive form which is storyline only when it follows an initial storyline form and which elsewhere is off the line. (3) Languages with a storyline form which is not dependent on the occurrence of a special initial storyline form, but which does "double duty" as an off the line consecutive form following various off the line elements. I summa rize all this in Diagram IV.


Certain SVO Languages (e.g. Yom, and Tern)

63 YOM, TEM (SVO)

LUWO (VSO=NV)

I. Storyline forms have no initial consecutive split

Storyline forms split: A. Initial form B. Consecutive form

Storyline forms have no initial consecutive split

II. Various off-the line forms

Various off-the line forms split: A. Initial form B. Consecutive form (can be same or different from above)

Various off-the line forms split: A. Initial form B. Consecutive form (which is the same as the storyline form)

Diagram IV: The storyline and consecutive forms

4. CONCLUSIONS

Within a narrative, storyline elements and various categories of non-storyline elements encode fundamentally different sorts of semantic functions. In this article a universal storyline scheme with semantically graded functions is posited, and this scheme is calibrated with languages of strikingly different word-order typologies: ( 1 ) SOY structures of the sort described here and found elsewhere in the world are chaining structures with a medial-final distinction in their verbs and in the clause types which are determined by their verbs. Both medial and final verbs must be semantically evaluated language by language as to their functions in storyline schemes. (2) VSO and SVO chaining structures represent the reverse of SOY chaining in languages of Africa which develop a special initial verb in the initial clause which consecu tive verbs in following clauses. Here again the semantic evaluation of the various types of initials and consecutives is vital to storyline analysis. (3) While conjunctions and other · sequence signals can never be ignored in · analyzing the storyline scheme in any language they are especially vital to such schemes in languages which have neither a special consecutive (or narrative tense) nor use word-order variation to mark on the line versus off the line status. I commend again to the reader Diagrams III and IV for their summary of some of the parallelisms and distinctions which are summarized in this paper. University of Texas at Arlington,

7500 W. Camp Wisdom Road.

Dallas. TX 75236. USA.


IZI, MUMUYE (SVO)

64 NOTES

I. 2.

For a presentation of word-order typologies of the sort here indicated see Comrie ( 1981 ). Thi! was essentially the research question which guided my field word in Africa during 1984.

This project, a fuller report of which is currently being written, embraces field workshops in various countries with investigation ranging over some 30 languages. I want to acknowledge here

the help and cooperation of the following organizations and agencies: The Bible Society of

Ethiopia, The Nigerian Bible Translation Trust, The Sudan Interior Mission and the Summer Institute of Linguistics.

3.

In the transcription of Luwo I have used E and 0 to represent low front and low back vowels

in the phonological structure of that language. cf. Buth ( 1980). A similar convention is followed

in transcribing similar vowels in languages cited in Section 3. The velar nasal is symbolized as ng

in the transcription of Luwo. 4.

In the transcription of Mwaghavul and Mumuye, d and b transcribe i mplosive voiced stops.

Bender, M. Lionel, (ed.), 1976: The Non-semitic Languag�s of Ethiopia. East Lansing: African Studies Center. Michigan State University.

Bliese, Loren F., 198 1 : A generative grammar of Afar. SIL Publicatiofl3 in Linguistics 65. University of Texas at Arlington. Arlington. Bliese, Loren F., 1984: A discourse analysis of Amharic narrative. (Paper presented at the Eighth International Congress of Ethiopian Studies). Buth, Randall, 1 980: The twenty vowels of Dheluwo (Jilr-Luwo, Sudan). In M. Lionel Bender and Thilo Schadeberg,

(eds.):

Nilo Saharan. Procudings from the First Nilo-Saharan

Colloquium, Leiden Sept. 1-3. Buth, Randall, 1984: Luwo clause types and their saliency. Unpublished paper, Juba: Summer Institute of Linguistics. Comrie, Bernard, 1 9 8 1 : Languag� Univ�rsals and Linguistic Typology. University of Chicago

Press . Chicago.

Grimes, Joseph, 1975: Tit� Tltr�ad of Discours�. Mouton, The Hague. H ardy, Heather, 198 1 : The Tolkapaya auxiliary system: subject copying, predicate raising and auxiliarization in Pai. Paper presented to the Conference on the Syntax of Native American Languages, Univenity of Calgary. Hetzron, Robert, 1976: Ethiopian Semitic. Journal of S�mitic Studi�s 2. Hopper, Paul, 1 979: Aspect and foregrounding in d iscourse. in: T. Givon, (ed.), Discourse and Syntax. (Syntax and Semantics 1 2), 2 1 3-242

Hopper, Paul and Sandra Thompson, 1980: Transitivity in grammar and discourse. Languag�

56: 25 1-299 . Jones, Larry and Linda K. Jones, 1979: Multiple levels of information in discourse. Discouru Stud�s in M�soamtrican Languagts, Vol. I , pp. 3-33. S I L Publications m Linguistics 58. University of Texas at Arlington. Arlington. Longacre, Robert E., 198 1 : A spectrum and profile approach to discourse analysis. Ttxt 1 :

337-359. Longacre, Robert E., 1982: Verb ranking and the constituent structure of discourse. Journal of tht! Linguistic Association of tht Southwest 5: 177-202.


REFERENCES

Journal of Semantics 5: 65-78

D IS C U SS I ON PRONOUNS UNDER CONTROL? A Reply to Liliane Tasmowski and Paul Verluyten•

PETER BOSCH

ABSTRACT Liliane Tasmowski and Paul Verluyten have recently expressed their misgivings about a propo was put forth in Bosch ( 1 980, 1 983) and have re-emphasized their ideas towards a uniform treatment of anaphoric pronouns, as originally published in Tasmowski and Verluyten ( 1 982). In the following pages I shall point out some limitations of the uniform pronoun treatment Tasmowski and Verluyten have in mind and I shall propose some amendments and extelll ions of my earlier proposals in order to take phenomena of gender and number agreement of pronouns into account which were ignored in earlier versions.

0. INTRODUCTION

Despite almost universal agreement among language scientists that within the class of anaphoric personal pronouns different ways of pronoun functioning must be distinguished, Tasmowski and Verluyten (1 982, 1 985) are proposing a uniform treatment of anaphoric pronouns. 1 The price they pay for the uniformity is the limitation of their proposals to phenomena of gender and number agreement. Semantic as well as syntactic problems of anaphora, which have motivated the distinctions among anaphoric pronouns that others have found necessary, are outside the scope of Tasmowski and Verluyten's theory. Their criticism of theories of anaphora that assume distinct types of anaphoric pronouns was first (cf. Tasmowski and Verluyten 1 982) directed against Evans, and in their recent paper ( 1 985) against Reinhart ( 1 983) and Bosch ( 1 983). Since I have discussed Evans' distinctions at length in my ( 1983) and Tanya Reinhart may speak for herself, I shall here only defend my own proposals. Tasmowski and Verluyten's criticism concentrates on some unclarities in the distinction between referentially and syntactically functioning pronouns which I have been advocating and on questions of number and gender agreement between pronoun and antecedent. Particularly with regard to the latter they are pointing out problems that have so far remained unaccounted •

Cf. Liliane Tasmowski and Paul Verluyten: "Control Mechanisms of Anaphora", this Jour

nal, vol. 4.4: 341 -370 ( 1 985).


sal for a di5tinction between syntactically and referentially functioning anaphoric pronouns that

66

for. Before I attend to these problems however (in Section 3), I shall try to , clarify my distinction between syntactic and referential pronouns (Section 1) and I shall try to show that Tasmowski and Verluyten's uniform pronoun treatment has nothing that could substitute for the required distinction (Section 2).

I. SYNTACfiC vs REFERENTIAL PRONOUNS

(1)

a. b.

Fred said he was sick. Nobody said he was sick.


My hypothesis (in 1980, 1 983) was that the crucial distinction that is needed in order to account for a variety of problems in the interpretation of anaphoric personal pronouns is a distinction between pronouns that function purely syntactically and pronouns that function referentially. I have called the two classes syntactic and referential pronouns respectively (for short, SPs and RPs). It follows from this notion of basing the SP-RP distinction on the difference in referential status of pronoun occurrences that all that can possibly matter for the interpretation of SPs is their structural relation to their controller. In this respect they behave like agreement morphemes that mark subject agree ment in the finite verb. There is no issue as to whether the third person -s in a sentence like x leaps refers to anybody or anything and if so to whom or what. Its function is to mark the fact that leap is interpreted as a function that takes x as its argument, no matter what you fill in for "x": an expression of definite reference, indefinite reference, generic reference, or of no reference, such as a quantifier expression. The referential status of an SP-controller is immaterial to the control relation . Hence such relations will remain undisturbed by any syntactically appropriate substitutions for the antecedent. This is different for RPs. They need a referent in order to be interpreted, just like any other referentially occurring NP. If, in a particular syntactic structure, this referent is introduced by means of another definite NP and the RP occurs co-referentially with that NP, then the interpretation of the RP depends on the reference of that NP. Thus if we substitute a non-referential NP for the original NP to which a pronoun is anaphorically related via co-reference, the anaphoric relation must break down. This difference between SPs and RPs is illustrated by the following cases. If we take ( 1 a) in the reading on which he is anaphorically related to Fred, we can see, after the substitution of the non-referentially functioning nobody for Fred in ( l b) that the "anaphoric" relation between the pronoun and the subject is maintained. Hence we may conclude that also in ( I a) the anaphoric relation cannot depend on co-reference and does not require a referential occurrence of he.

67 For (2a), on the reading that links he to Fred, we get the opposite result for the substitution of a non-referential NP for the antecedent. So we may conclude that, other things being equal, the reference of the antecedent plays a crucial role for the anaphoric relation in (2a), i.e. that the pronoun occurs referential ly as an RP and co-refers with the antecedent. (2)

a. b.

When Fred arrived, he was tired. When nobody arrived, he was tired.

(1)

c.

Fred said the old malingerer was sick.

(2)

c.

When Fred arrived, the old malingerer was tired.

Quite apart from the question of how sharp the boundary between these two ways of pronoun functioning can be drawn (a point to which I attend in a moment), one may reasonably expect of any theory of pronouns that it should be able to account for these substitution phenomena and their seman tic effects. It seems pretty clear that a pronoun theory that claims that all pronouns function alike will be in big trouble when it tries to account for these observations. Tasmowski and Verluyten seem to think - and this is why I am labouring the point - that the function of these observations is only in providing criteria for a taxonomy of pronoun uses. What they ignore is the status of these observations as problems that need to be accounted for by pronoun theories. From the theoretical considerations above it is easy to derive the following two tests to decide between SPs and RPs, which I proposed in Bosch ( 1980, 1 983): if a 3rd person personal pronoun that stands in an anaphoric relation with some antecedent can be replaced by another referential expression with out thereby affecting the anaphoric relationship, then the pronoun func tions as an RP. For short: pronoun replaceable - RP


Furthermore, we may consider the effects of substituting a referential NP for the pronoun. If the anaphoric relation remains, then this is clear evidence that the relation between pronoun and antecedent, also in the original version of the sentence, may rest on co-reference and that the pronoun functions referentially as an RP (cf. (2c)). lfthe outcome of this test is negative however, i.e. if the NP that we have substituted for the pronoun cannot link up to the antecedent, we must be more careful with our conclusions: perhaps we have only not found the right kind of NP and another NP may do the job. Bearing this caveat in mind, a negative outcome may count as evidence that the anaphoric relation in the original sentence was not based on a referentially functioning pronoun but on a pronoun functioning as an SP (cf. ( lc)).

68 II if the antecedent, A, of a 3rd person personal pronoun that stands in an anaphoric relation with A can be replaced by another NP that clearly functions non-referentially and the anaphoric relationship remains unaf fected, then the pronoun functions as an SP. For short: antecedent replaceable - SP

(3)

Nobody will get into this building, unless he has a valid identification.

In view of such cases, Tasmowski and Verluyten believe that the "nobody test" (viz. test II above) becomes worthless, because they think that test II would mark the pronoun as an SP. - Well, properly used, the test yields correct results also here. The test requires nota bene replaceability of the antecedent: my claim is that an SP links up to a syntactic position, indepen dently of the referential status of an NP that fills that position. Thus the mere occurrence of nobody (or any other non-referential NP) as an antecedent does not satisfy the condition of the test. When we substitute a referential NP for nobody in (3), the anaphoric relation cannot be preserved except by twisting the meaning of the sentence. This shows that the pronoun does not link up to the syntactic position occupied by nobody, but to a (generic) discourse referent that is established in the course of the processing of earlier parts of the utterance (cf. for more detail Bosch 1 983 : 1 33[). A more essential point is the following. In my ( 1980, 1 983) I assumed that the class of anaphoric pronoun occurrences is exhausted by the two disjoint classes of SP and anaphoric RP occurrences. And although I was perhaps not very explicit about this, I took for granted that the classes of pronoun occurrences that are marked as SP- and anaphoric RP-occurrences by the tests should share this characteristic. But if this were so, we would have a number of potentially troublesome cases: pronoun occurrences that are marked as RP-occurrences by test I and as SF-occurrences by test II. Tas mowski and Verluyten are quite right to point out this difficulty. In Bosch ( 1983) two types of such cases are mentioned, which I tried to treat as cases of transition in historical language development. I am not yet


When Tasmowski and Verluyten ( 1 985) believe that "such criteria do not allow one to recognize SPs", they are, as is evident from their attempts to substantiate their criticism, getting the direction of these consequence rela tions wrong, not reading them as crittria that determine the type of pronoun occurrence (right side of the i mplication sign) on the basis of the test results (left side of the implication sign), but reading them as predictions: if some thing is an SP (or RP), then the tests must yield such and such results. Another point which Tasmowski and Verluyten bring up in order to show specifically that test II is not reliable, concerns anaphoric relations between non-referential antecedents and subsequent pronouns which are referential in nature and depend on an inferred referent, such as in the following case:

69

2. A REVIVAL OF SUBSTITUTIONALISM

The uniform theory for "all true pronouns" (i.e. roughly all pronouns one would loosely refer to as "anaphoric pronouns"2) that Tasmowski and Verluyten propose postulates something they call "linguistic control of the pronoun by its antecedent". But what does "linguistic control" mean in this connection, and what exactly is meant by an "antecedent"? I first thought (cf. Bosch 1 984) that linguistic control might be syntactic control, which made me wonder how syntactic control relations could cross sentence boundaries in cases where the antecedent is in another sentence than the pronoun. But Tasmowski and Verluyten have corrected me: linguistic control is not syntac tic control. It would appear that by "linguistic control of a pronoun by its antecedent", at least in the most straightforward cases, they do not mean anything else than that the pronoun agrees with its antecedent with respect to gender and number. There need not be a syntactic relationship between the two, and Tasmowski and Verluyten have given no hints at any semantic relation connected with what they call "linguistic control" . The second question concerns the notion o f the pronoun's antecedent. According to Tasmowski and Verluyten all anaphoric pronouns are linguisti cally controlled by an antecedent. If "antecedent" here means what it nor mally means, viz "overt antecedent", then we may expect quite a variety of different relations under the name of "linguistic control", as might appear from some selected examples in (4) (where the star indicates only that no


certain that they are not. But I am certain that the postulate of disjointsness of the classes of SP- and RP-occurrences rests on a confusion, and having seen this, it becomes irrelevant for the theory whether or not any cases in the intersection are cases of historical transition. Let me explain where I think I went wrong. I still believe that the theoreti cal considerations that lead to the SP-RP distinction are correct. And as the tests follow from these considerations, they ought to be equally correct. The tests however show nothing about occu"ences of anaphoric pronouns (as I used to believe), but are strictly concerned with the functioning of pronoun occu"ences in processing. More precisely: the distinction is one between the strategies a linguistic processor employs rather than between properties of occurrences. Thus if a pronoun occurrence is tested with positive result for SP-hood we know that it may successfully be processed as an SP, namely independent of any assumptions with respect to reference. And if the tests show the pronoun to be an RP we know that it can be processed as an RP, i.e. on the basis of the assumption that it has a referent. In the absence of a full specification of these processing strategies, however, there is no reason to think that there should not be pronoun occurrences that may successfully be processed by either strategy and possibly even with superficially indistinguishable results.

70 anaphoric reading of the pronoun with respect to the italicized antecedent is possible). (4)

a. b. c. d. e. f.

1.

k. I.

The observations I want to make with respect to these cases are very straight forward and there is no need to comment on the examples one by one. My main point is that either the claim that the antecedent determines the number and gender of the anaphoric pronoun must be false or most of the italicized antecedents in (4) are not what Tasmowski and Verluyten mean by "antece dent": we find the same overt antecedent wiih different pronouns and we find pronoun-antecedent pairs that once allow for an anaphoric relation and once don't. And additionally there are cases, like (4m), where the overt antecedent is of a syntactic category that could not possibly determine the gender and number of an anaphoric pronoun: (4)

m. Shakespearean imitators don't usually capture his style.

Now Tasmowski and Verluyten ( 1 985) do indeed suggest that the overt antecedent of an anaphoric pronoun is not always the antecedent that linguis tically controls the pronoun. Not only in the case of antecedentless anaphoric pronouns the controlling antecedent must somehow be reconstructed, but also in cases where there is an overt antecedent, they occasionally consider another noun that does not occur in the discourse to be the "antecedent" that determines the pronoun's gender and number. But since they neither give a general rule for when such reConstruction is required nor how it would be accomplished, their notion of a controlling antecedent (as differing from regular overt antecedents) remains somewhat elusive. It is no art (nor science) to look at the pronouns in (4) and in each case think up some NP or noun that agrees with the pronoun in gender and number and would thus willingly yield to the role of the pronoun's controlling "antecedent". The question is how we get from identical overt antecedents to wildly different controlling antece dents.


g. h.

I met a colleague in the theatre. He was wearing a black tie. I met a colleague in the theatre. She was wearing a black tie. I met a colleague in the theatre. They all wear black ties. I met every colleague in the theatre. *He was wearing a black tie. I met every colleague in the theatre. They were all wearing black ties. I didn't meet a colleague in the theatre. *He was wearing a black tie. Nobody took his hat off. Nobody took their hat off. • Nobody took her hat off. Watch out with that chair. It's not very stable. Watch out with that chair. They're not very stable.

71

(5)

The pilot who shot at it hit the plane that chased him.

I fail to see the substantial difference between the difficulties this type of sentence raises for pronominalization and the problems it would raise for Tasmowski and Verluyten's controlling "antecedents" . Of course one could postulate the noun man as the covert "antecedent" for him and the noun plane as the covert "antecedent" for it. But such a move would only exemplify the arbitraryness of "antecedent recovery". From what Tasmowski and Verluy ten have written on the subject so far, I cannot see how their proposals have risen above this level.

3. AGREEMENT

The important insight in Tasmowski and Verluyten's work to my mind is a matter quite separate from, and of more general impact than, their substitu tionalist anaphora theory. They have pointed out that practically all ap proaches to pronouns have overlooked the fact that the shape of anaphoric


Tasmowski and Verluyten are not the first in this predicament. A n ap proach to pronouns of fundamentally the same substitutional nature was pursued by Bloomfield ( 1 935) and Hockett ( 1958). They, too, never succeed ed in getting rid of the air of arbitrariness in what for Tasmowski and Verluyten is the recovery of the controlling antecedent. In their formulation pronouns are "substitutes" for full linguistic forms. The relation between the pronoun and the full linguistic form it substitutes for is not a relation of stylistic variation; there is no claim that the form substituted could actually stand in place of the pronoun. The idea is rather that the form substituted for must, in some sense, be in the speaker's mind when he utters the sentence in question and in the listener's mind when he understands it (that Bloomfield or Hockett would not have spoken of things in anybody's mind is not the issue here). It is roughly this intuition of pronouns being substitutes that early trans formationalist accounts of pronouns attempted to make explicit in terms of Deep Structure (which would explicitly contain the full forms) and a pronom inalization transformation (as an explication of the substitution process). But even before the pronominalization approach was abandoned, transfor mationalists came to accept the fact that the pronominalization transforma tion could not be all there is to an account for anaphoric pronouns. This insight was brought home by the Bach-Peters Paradox (cf. Bach et. al. 1 974, Karttunen 197 1 ), as exemplified by sentences of the kind of (5). Whatever the Deep Structure representation of one of the two pronouns, the Deep Struc ture representation of the other pronoun must be part of it, and the latter Deep Structure representation in tum must contain the former Deep Struc ture representation as its part, and so on and nauseam.

72

(6)

a.

b.

(Watching a Frenchman trying to put a large table - Ia table (f.) into his car): "Tu n'arriveras j amais a {Ia/ •te} faire entrer dans Ia voiture." [You11 never manage to get it into the car]. (same situation but with a desk - le bureau (m.)): "Tu n'arriveras jamais a {•Ia/ le} faire entrer dans Ia voiture." [You 11 never manage to get it into the car].

This phenomenon is by no means peculiar to French, but is commonly found in languages that employ formal noun gender. Cf. the following cases from German. (6)

c.

d.

(watching a German trying to move a wardrobe - der Schrank (m.)): "Wenn du die Kleider nicht rausnimmst, kriegst du {ihn/ •siej •es} nie von der Stelle." [If you won't take the clothes out, you'll never make it move]. (a German looking at his typewriter in dismay - die Schreibma schine (f.)): "{Sie/ •Er I • Es} tut's nicht me hr. Ich muB {sief •ihn / •es} reparie ren lassen." [It doesn't work any more. I must have it repaired].


pronouns cannot regularly be derived from properties of their intended referents, and that the agreement features of particular nouns or NPs a language would employ to describe the relevant referent plays a crucial role in the determination of the shape of the pronoun. This oversight may indeed be due to the fact that most of the discussion of pronouns has been carried on in English and with English as an object language. Since formal noun gender plays a marginal role in English, the determination of the agreement features of pronouns could easily pass for a matter of denotation or reference, which it certainly is not. But I am afraid that Tasmowski and Verluyten are overesti mating the role of their insight for a comprehensive account of anaphoric pronouns when they offer a generalization of this notion as a uniform pronoun theory. Still, the core of their proposal deserves close attention, because it can, I believe, lead to a better understanding of the form of mental representations that are relevant for the processing of anaphoric expressions. The clearest case that could be made for a formal determination of pro noun features can be made with antecedentless anaphoric pronouns. And Tasmowski and Verluyten have shown convincingly, already in their ( 1982), that the gender and number of antecedentless pronouns is determined by the gender and number of a particular NP that is not overtly present in discourse. They refer to such NPs as "absentee antecedents". The following are two of many examples they provide:

73


The story that Tasmowski and Verluyten tell about these cases is one of a two-step control: there must be a salient object in the situation (e.g. the table in (6a)) which "pragmatically controls" an absentee antecedent (in the cases at hand the French noun Ia table) which in turn "linguistically controls" the pronoun. The open question is, as in the general case already discussed, how we can recover these absentee antecedents. Tasmowski and Verluyten speak of "severe recoverability conditions", but in practice this boils down to the bare requirement that the absentee antecedent must be recoverable. How it is recovered was, at least in their (1 982) paper, left to the reader's own fantasy, except for the constraint that the absentee antecedent must agree with the pronoun (and, presumably, denote the pronoun's referent). In order to account for these cases I suggested in my ( 1 984) that the agreement features of the pronoun should be taken as inherited from the default description of the referent in question. In the unmarked case this is the basic level common noun (cf. e.g. Rosch 1 977) that classifies the intended referent. Cornish ( 1 986) has, independently I believe, arrived at the same proposal, and also Tasmowski and Verluyten seem to be adopting the idea in their most recent paper ( 1 985). They add, quite correctly, that it will not always be the basic level common noun that determines the pronoun's agreement features, but that there may be contextual varition: t he default value for an object's description may be re-set for specific contexts. This is a point already reasonably well established in studies of categorization and naming (cf. e.g. Tversky, 1 977; Osgood, 197 1 ). Basic level categorizations are default categorizations. The notion that there is, for each object and context, one unique common noun that would form the least marked, and hence the most naturally expectable, description of the object in that context gives us at least a sketch of the kind of relation that Tasmovski and Verluyten call "pragmatic control"; and for unmarked contexts we not only have a sketch but can make the empirical prediction that the appropriate basic level common noun is the least marked description and that it is this noun that determines the gender of the anaphoric pronoun. There are many details still to be worked out until these mechanisms of the determination of the shape of anaphoric pronouns are entirely clear. This is not the place to attempt to do this (but cf. Bosch in preparation). But if there is a way of making the idea of constructio ad sententiam, or "notional agreement", as it is sometimes called, explicit as far as feature agreement of anaphoric pronouns is concerned, I believe that is along these lines. The only point in this connection I want take up here is the question of how these agreement mechanisms for referential anaphoric pronouns (APs, in order to distinguish them from the wider class of RPs of which they form a subset) differ from the syntactic control mechanisms that determine the shape of syntactic anaphoric pronouns (S Ps).

74 3. 1 Feature agreement for

S Ps and A Ps

(7)

a. b.

Nobody was tired. •But he left. Nobody was tired when he left.

The quantifier nobody does not occur referentially (cf. Bosch 1983:43ff), certainly not in the sense that it would introduce any referents into the discourse. It does not result in a referential representation to which an AP could link up at a later stage (cf. (7a)). But the quantifier NP can control SPs for the simple reason that it can control VP-agreement (cf. (7b)). (8)

a. b.

Nobody was tired. But they left. Nobody was tired when they left.

Somewhat puzzling in this connection may be (8) where the pronoun does not agree with the singular number of nobody. Here, however, they is not an SP but functions referentially. It is not anaphorically linked to the quantifier (which would be impossible because of the disagreement) but to the contex tually understood set of people over which it quantifies. And if such a set of people is contextually assumed, there must be a representation of this set in the context model which is accessed by the pronoun.


Since the forms of third person personal pronouns are used both in the function as SPs and APs, we may assume that the agreement features in SP and AP-agreement are also the same. Thus if we assume that the processing of an NP results in a representation that contains, among other things, also the NP's agreement features, then it is these feature specifications with which a pronoun must agree in order to link up anaphorically with the NP. And on this general level there is no difference between SPs and APs. The difference is that in the case of referentially used NPs the feature representation is linked to (or perhaps forms part of) a representation of the NPs referent, which is not the case for non-referentially occurring NPs. Since we have seen that SPs do not care about the referential status of their antecedents it is plausible to assume that they can be processed with respect to the feature representation alone. And if the relation between SPs and their antecedents is indeed a relation of syntactic control, it is also plausible to assume that these feature representations, as long as they are not linked to referential representations, are rather short-lived representations that are available only during the actual syntactic processing and are deleted when the processor leaves the relevant syntactic domain (the Processing Unit, in terms of Bosch (1983)). In the referential case, however, these representations are integrated into dis course or context models that contain representations of discourse entities. Accordingly they are accessible via referential devices also after the processor has left the syntactic domain in which the antecedent NP occurs. Cf. the following examples:

75 That they in (8b) can indeed not be an SP is demonstrated by the fact t hat the finite verb that is controlled by nobody must take the singular form. Hence if SP-agreement functions like verb-agreement, as I have been claim ing, also SPs must take the singular (cf. 9).

(9)

a.

Nobody {has/ *have} gone out.

In this respect nobody contrasts with the quantifier none, which is not specified with respect to the number feature. Accordingly we may have t he verb in either singular or plural ( l Oa), and also the SP ( l Ob) . When we remove the underdetermination in t he NP only one of the alternative verb forms or pronoun forms remains (cf. (11) where the stars, as before, apply only to t he intended anaphoric reading). a. b.

None of the la9ies {has/have} gone out. None of the ladies took {her I their} luggage.

( I I)

a'. b'. a". b".

Not one of the ladies {has/ *have} gone out. Not one of the ladies took {her I *their} luggage. All of the ladies {*has/ have} gone out. All of the ladies took {*her I their} luggage.

The lack of nu mber determination in t he subject of (10) thus would seem to lead to a feature specification t hat is matched by feature specifications with arbitrary values for t he number feature (and is thus in agreement with singular as well as plural VP features). Nobody not only has a definite feature value for number, as we have seen, but it is also specified with regard to its gender value, viz as generic personal. The pronoun his in (12) does not permit the conclusion that we are concerned with males, and it is only in virtue of this use of the masculine pronoun form as an i mplicit generic pronoun that his is possible in (12) . Her is excluded for precisely the reason that it does not have a generic use. The (composite) explicit generic pronoun he or she is of course possible. (12)

Nobody took {his / *her / his or her} luggage.

Despite the superficial similarity between French personne plus negation and nobody, the two do not behave quite the same way. Cf.

( 13)

a. b.

Personne n'a admis {qu'il/ *qu'elle} connaissait le coupable. [Nobody admitted that {he/ *she} knew the culprit] J'ai interroge toutes ces dames. Personne n'a admis {qu 'ellef •qu'il} connaissait le coupable. [I have questioned all the ladies. Nobody admitted that {she/ *he} knew the culprit.


(10)

76 c.

Personne n'est satisfait. [Nobody is satisfied (m.)]

d.

Personne de ces demoiselles n'est satisfaite.

[None of these young ladies is satisfied (f.)] Although French grammars usually insist that the indefinite pronoun per sonne is of masculine gender, we see that this is only a default value and that it can be over-written by specific contextual information. That we are concer

ned with a value in the feature specification and with agreement, rather than with pronouns functioning with reference to the set of people over which

mation is available feminine agreement is excluded (cf. ( 1 3a)). Since however the masculine forms of pronoun or participle do not allow for the inference that perso nne quantifies over males, the correct analysis should be analogous

to nobody and would postulate a specification as generic personal rather than as masculine. The difference between personne and nobody is that for personne we are concerned with a default value for the gender feature, whereas nobody has its gender value fixed. ( 1 21 is not a good English expressions for the reading on which nobody quantifies over the set of ladies introduced in the first clause. cr.

( 1 21

The ladies have left, but nobody took: her luggage.

The theoretically interesting point about

personne is that we find clear

evidence for the intrusion of context specific information into feature specifi cations, which cannot be removed by postulating a referential functioning of the anaphoric pronouns, as it could in the case of the plural pronoun following nobody

(cf. (8) above). If we were to say that the pronouns in ( 1 3a,

b) function referentially, we would also have to say that the participle in (1 3c, d) functions referentially. The model I am proposing maintains agreement relations as syntactic and context independent relations and relegates the referential relation into the interpretation of personne. This does not mean that the quantifierpersonne is ascribed a referential occurrence. Similarly, the NP

none of these ladies

cannot occur referentially and none does not occur referentially either in this NP. The referential sub-constituent that establishes the link to the context and enables the use of contextual information is the NP makes the domain of quantification of

these ladies, which

none explicit.

It see ms that the feature specifications that result from the processing of

the antecedents of anaphoric pronouns are always specified with respect to gender. In the first instance this specification comes from the head of the NP and

if the head is not specified for gender the value is supplemented from its


personne quantifies is clear from the gender in the participle in (c) and (d). The default specification of the gender value of personne must not be mistaken to mean that personne is unspecified with respect to the gender feature (as English none is for the number feature). When no specific infor

77

4. CONCLUSION

Let me try to sum up. Tasmowski and Verluyten and I seem to agree that the shape of anaphoric pronouns is not explicable on the basis of their objects of reference alone. Tasmowski and Verluyten say that anaphoric pronouns agree with their antecedents (in their special sense of "antecedent" noted earlier). I want to say that anaphoric pronouns agree with feature representations that result from the processing of their antecedents. In the case of SPs these feature representations are identical to the feature specification of the controller, unless this specification is incomplete with respect to any features or specifies only a default value for them. In such cases the feature representa tion is completed with respect to the missing feature values and default specifications may be overwritten by information that is drawn in from the context. In the case of referential APs there is still the same agreement relation between the pronoun and a feature specification, but first it is not a relation of


explicitly given or implicitly understood domain description. If the gender value of the head, as in the case of personne, is only a default value, it can be over-written by domain descriptions. The value of the number feature must either be supplied by the determiner of an explicit antecedent, in which case it may be singular or plural or undefined, or alternatively, it may be supplied by the domain of denotation or quantification of the antecedent, in which case it is plural. If, as in the case of none, the head is unspecified with respect to number then the feature specification of the whole NP will also lack a value for the number feature. So far I have argued that anaphoric pronouns agree with the feature specification that results from the processing of their antecedent NPs. The processing of the NP may, but need not, take context specific referential information into account. The relation between the pronoun and the feature specification is one of strict VP agreement: the pronoun behaves like the fmite verb. To this general rule we have seen one type of exception, where the anapho ric pronoun does not behave like the finite verb and does not agree with the feature specification of this antecedent but functions referentially with re spect to a contextually available referent (cf. the case of nobody . . . they). In order to maintain the generality of the rule that anaphoric pronouns agree with the feature specification of their controller or antecedent, we should say that referentiallyfunctioning anaphoric pronouns agree with the feature speci fication of a contextually available description of their referent. This descrip tion, as we can see from the nobody case, must have the shape of an NP, i.e. it must not only contain a noun but also a determiner, for otherwise it could not prescribe a value for the pronoun's number feature.

\

78

syntactic control, but mere feature agreement, and second, it is not necessari ly the overt antecedent that is the origin of the feature representation but the currently most relevant description of the pronoun's referent, which may or may not be identical to an overt antecedent. I should be glad to offer this proposal as either a further specification, with some amendments, of what Tasmowski and Verluyten are proposing or alternatively as an extension, with some amendments, of what I advocated earlier. If they want to accept the first of these ways of looking at our discussion, I am a happy man and we are in the same boat. If they don't I shall opt for the second alternative. Dept. of Language and UteratiO'e

.5000 LE Tllburg, The Netherlmui.r

NOTES I. I am grateful for the comments on this paper by an anonymous referee of the Journal. 2. There are some problemB with respect to the delimitation of Tasmowski and Verluyten's class of "true pronouns", which however I cannot take up here without going far beyond reasonable limitations of space. For instance a no�mphatic English It in many occurrences would not count as a "true pronoun" for them while it would certainly qualify as an anaphoric pronoun on just about any notion of anaphora.

REFERENCES Bach, Emmon, J. Bresnan, T. Wasow (1974): Sloppy Identity. Ling. Ingu. 5. Bloomfield, Leonhart ( 1 935): Language. Allen and Unwin. London. Bosch, Peter ( 1 980): The Modes of Pronominal Reference and Their Constraints, in: Kreiman, J. and A. E. Ojeda (eds.): ProfWwu and Anaphora. Paper$from the Paraunlon on PronOIDI$ and Anaphora. Chicago Linguistic Society. Chicago, Ill. Bosch, Peter ( 1 983): Agreement and .Anaphora. Academic Press . London. Bosch, Peter ( 1 984): Coherence and Cohesion. Paper presented at the Intern. Conference on Text Coherence at ZiF, Bielefeld, 1 5th-19th October, 1984. To appear in the Proceedings (ed. J.S. PetOfi). Bosch, Peter (in preparation) Anaphora and Representation. Cornish, Francis ( 1 986): .Anaphorlc Re/atiOfl$ In Engilih and FreTJCh. Croom Helm, London. Hockett, Charles ( 1 958): A COllTse In Modern Lingui.stlc.r. MacMillan. New York. Kantunen, Lauri ( 1971): Defmite Descriptions on Crossing Coreference. Fowtd. of Lang. 1. Osgood, Charles ( 197 1 ): Where do Sentences Come From ? in: Danny D. Steinberg and Leon A. Jakobovits (eels.): Semantlc.r. Cambridge University Press , Cambridge. Reinhart, Tanya ( 1 983): .Anaphora and Semantic Interpretation. Croom Helm. London. Rosch, Eleanor ( 1 977): Human Categorization, in: N. Warren (ed.): Studies in Cross-Cultural Psychology, vol. I , Academic Press, New York, pp. 1-49. Tasmowski-De Rijck, Liliane and Paul Verluyten ( 1 982): Linguistic Control of Pronouns, JS 1 :323-346. TasmoWJki, Liliane and Paul Verluyten ( 1985): Control Mechanisms of Anaphora, JS 4: 341-370. Tversky, Amos ( 1977): Features of Similarity, Psych.. Rev. 84:327-352.


Ulliver$lty of Brabant at Tilburg P. 0. Box 90 153

Journal of Stmantics 5: 79-86

BOOK REVIEW

Paul Werth: Focus, Coherence and Emphasis. Croom Helm, London, 1 984. Pp. viii+293, £ 1 6,95 (cloth). AD FOOLEN

I. RESEARCH BACKGROUND


Focus, Coherence and Emphasis (FCE) is not written within the framework of one of the current models of grammar or discourse theory. Werth has set himself the goal of developing a model of his own. As he says in the preface, he has worked on it for several years, and in the bibliography we find publications of his on the subject going back as far as 1968. The fact that Werth envisages a model of his own, does not mean that he has developed it in strict isolation from existing models. In fact, the book contains many references to, and discussions of the work of others. Unfortu nately this causes many interruptions of Werth's line of argument and makes the book difficult to read. The 1 3 pages of the Bibliography bear witness to the quantity and diversity of the literature that are taken into consideration. Werth's approach is characterized by a strong text-linguistic or discourse theoretical background and perspective. To use the words of the author: "In very general terms, our approach falls into the category of the 'functionalist view of language'. We hold, this is to say, the view that language is a purposive instrument, whose properties reflect the expressive needs of speak ers within given situations" (p. 1 3). But "functionalist view of language" does not refer to any specific model like for example Halliday's or Dik's; it is rather meant as a general characteri zation of approaches that are opposed to formal approaches like "REST/GB, GPSG, LFG or any other acronymic schools of thought" (p. 262). Werth addresses much of his discussion to Generative Grammar, and he shows throughout the book that he is well acquainted with its history and different models. Werth is generally opposed to what he calls "S-grammar" (sentence gram mar) and in favour of "0-grammar" (discourse grammar); in several places he makes this explicit in a rather polemical way. It has not become clear to me, however, whether Werth rejects S-grammars in principle, or if he only wants to expand them. The latter position could be inferred from the follow ing quote: "We propose that a 0-grammar is a S-grammar, with additional constraints making it sensitive to sequential properties of sentences in texts. As for the S-grammar in question, we assume that any of the current models, insofar as it can express the basic syntactic properties of sentences, will be adequate" (p. 259). But four pages later, Werth formulates a more reductio nist Giv6n type of view of the status of formal grammar: "All such

80 constraints [rule constraints as proposed in GB, A. F.] are at base fossilized versions of semantic requirements on textuality" (p. 262). To sum up: I am under the impression that Werth has not achieved full clarity on the relation between functional and formal approaches in the analysis of sentence structure. But in this respect he is not alone.

2. TITLE AND STRUCTURE OF THE BOOK

3. THE DISCOURSE THEORY

Having discussed the general outlook and structure of the book, I now turn to its actual contents. As I said already, Chapters 2 to 5 (pp. 1 6-94) are on discourse theory. The contents of these chapters is comparable to what one can find in books like Halliday & Hasan ( 1 976), Van Dij k ( 1 977), De Beaugrande & Dressler (198 1 ) or Brown & Yule ( 1 983). Terms that the reader will expect and find in these chapters are coherence, cohesion, connectivity, context, common ground, chains of reference, etc. I find it difficult to give a clear and concise summary of the discourse model that Werth develops in these chapters, nor would it be easy to say exactly in which respects the model differs from the models that are presented in the


FCE has ten chapters, the first and last of which being the 'Introduction' and ' Conclusions'. The remaining eight are divided into two sets of four chapters. In Chapters 2-5 Werth provides an outline of his discourse theory, and in Chapters 6-9 he analyzes some aspects of sentence structure (accent structure, anaphora, and word order) against the background of the discourse theory developed in the first part. Two of the terms that appear in the title of the book can be linked to the parts just mentioned, namely 'Coherence' and 'Emphasis'. 'Coherence' is seen as the main property of a text or discourse, and the concept thus plays a central role in the first part, whereas 'Emphasis' is Werth's general term for those aspects of sentence structure that can be related to the functioning of sentences in a discourse context. That ' Focus' turns up in the title of the book, is somewhat misleading: this notion does not play any important role in the book. Section 6. 7 is a review section devoted to 'focus', and in particular to the question of what Halliday and Chomsky meant by the term, but the review does not lead to an integra tion of this notion into Werth's own model. So 'focus' plays at most a small, negative role in the book as a whole. The appearance of the word 'focus' in two of the chapter titles, namely Ch. 1 : 'Introduction: bringing things into focus' and Ch . 9: 'Syntactic variation: getting movement into focus' is equally m isleading. The word is used here in its everyday sense, and not as a technical linguistic notion.

81

4. LINGUISTIC STRUCTURE

In the second part of the book, some aspects of English sentence structure are analyzed from the point of view of the discourse theory developed in the first part. Ch. 6 and 7 deal with the accent structure of English sentences, Ch. 8 covers anaphora and Ch. 9 word order. I will discuss each of these topics in turn.

4. 1 Accent structure

Chs. 6 and 7, 70 pages in total, are devoted to an extensive analysis of accent structure in English spoken discourse. Ch. 7 discusses the question which items (lexical and grammatical) can bear a contrastive accent and which cannot. I will not expand on this topic, and concentrate on Ch. 6.


books mentioned. The author i s at least partly responsible for this difficulty: the argument of the book constantly refers to the work of other authors, which makes it difficult to get a clear idea of Werth's own model. A certain compensation may be found in the fact that Werth illustrates his text with many analyses of pieces of real discourse, which reduces at least some of the conceptual unclarity and helps to see what the author is aiming at. It does become clear however that the concept of 'coherence' is central to Werth's discourse theory. Every utterance in a text must cohere in some way or other with the information that is built up in the preceding text (the "common ground", cf. p. 53); otherwise, the text would not be well-formed. This semantic-pragmatic condition on texts is called the 'Coherence constraint'. Coherence is not a property of the form of utterances or texts, but of their content: "The notion of coherence refers to a complex set of semantic relationships between concepts, including (partial) synonymy, hy ponymy (inclusion), and implication" (p. 74). "Normally, coherence ope rates through inference, i.e. partial or implied semantic-pragmatic relationships" (p. 75). In order to get a first impression of how coherence works, the reader may concentrate on the analysis of the " Fred, the hungry alligator" example (p. 1 6), a newspaper passage of nine lines, that is analyzed in Ch. 2 and in Section 4. 5 1 (p. 73-75). A distinction of some importance for the remainder of the book is the one between positive and negative coherence (see Sections 5.2 and 5.3, p. 83-89). There is positive coherence, if there is a semantic-pragmatic relationship of synonymy, which may be of varying degrees, between concepts or proposi tions in the text, whereas antonymic relations, or oppositeness, lead to negative coherence. The relation of antonymy is based on synonymy, but requires an extra negative operator that is explicitly or implicitly present (p. 87).

82

"(a) Mark as accented each terminal predicate in a tree Pi: < Pred >-< Pred A >. (b) If a predicate is predictable within the common ground, i.e. if it positively coheres: A- R. (c) Under the semantic condition for contrast,. i.e. negative coherence: R - c·. .

.

.

.

.

.

.

.

.

.

.

.

What I miss in Werth's account, however, is a grammatical-phonological rule system that takes these markings from the underlying semantic structure to real surface structure accents. Let me give an example. On pp. 106- 1 07, an utterance is analyzed, which was taken from spoken discourse: "You pass through the country". In the transcription , pass has an A-accent. In the underlying structure, pass is analyzed into two semantic predicates: MOVE and CONTINUOUS. The accent rule system takes the foregoing discourse into consideration and results in an R for the first predicate and an A for the second. But the link between these underlying R +A and the surface A is missing. I want to round off the discussion of the proposed accent-emphasis system with two methodological remarks, First, the examples analyzed in Chs. 6 and 7 show "that the set of rules accords with the 'facts' of emphasis" (p. 105). But I cannot quite get rid of a certain suspicion of circularity here. The accent transcriptions of the discourse fragments and their semantic analysis were carried out by the same researcher, and there is a real danger in this situation that, for example, when a reduced accent is heard, semantic coherence will be found, and vice versa. Werth's rule-system could be tested much more


The reader will note that the term 'emphasis' was used earlier in a wider sense, not only in relation to accent structure, but also in relation to other aspects of sentence structure that are sensitive to coherence phenomena. In fact, Werth uses the term both' in this wider sense (e.g. p. 260) and in the narrower sense in which it only applies to accent structure. Such terminologi cal looseness unfortunately also applies to other terms in the book. Emphasis in the narrower sense thus refers to the accent system. Werth distinguishes three types of accent: A (for normal accent), R (for reduced accent), and C (for contrast accent). Chs. 6 and 7 contain several transcriptions of spoken English discourse, transcriptions in which the words in each utterance are marked for A, R, or C. Werth then shows that the distribution of these different accents can be predicted from the discourse. I will try to sketch briefly how the discourse induced 'generation' of accent works. The accent rules first apply to the semantic structure of each sentence. A semantic structure is a series of semantic primitives ('terminal predicates') that are related to each other in a tree diagram, much like in Generative Semantics. Now the following set of rules is proposed for the marking of the terminal predicates as A, R, or C (p. 105):

83 reliably if the transcription and the semantic analysis of the discourse were carried out independently by different researchers. From the semantic analy sis predictions about the accent structure should be made, and these predic tions should then be compared with the accents that actually occur in the data. Secondly, the whole approach hinges, of course, on the assumption that it is possible for the transcriber to distinguish phonetically between A, R, and C accents. From several formulations of Werth, however, I infer that he has his doubts here himself: " . . . the presence of stress does not necessarily distin guish between Accent and Contrast . . . " (p. 1 80); . . . that A-items may be no different suprasegmentally from C-items, and R-iterns may not differ suprasegmentally from unmarked items" (p. 1 00). If this is so, then the empirical basis for the undertaking becomes rather weak. "

Ch. 8 is on anaphoric connectivity, i.e. on antecedent-anaphor relationships in texts. 'Anaphora' should not be understood here as a strictly defined category like, for example, in GB, where it is opposed to the category of pronouns, or in discussions that contrast anaphora and deixis (Bosch 1983). For Werth, 'anaphora' covers reflexives and reciprocals, pronouns, definite NPs, epithets, repetitions, and gaps. Thus e.g. in 'Basil was in a cupboard, examining the hinges' (p. 1 67), the hinges is considered as anaphoric to cupboard. According to Werth anaphoric relationships in texts must obey semantic pragmatic discourse conditions. The condition of coherence (the Coherence Constraint, see 3, above) must hold for the anaphor and its antecedent and, additionally, it is required that "the two items are ( . . . ) related to the �arne conceptual entity" (p. 1 86). If I understand the author well, this means that coherence can in principle be found with all parts of speech (which, then, in principle, are all sensitive to the accent rules as presented in Cbs. 6 and 7), whereas anaphoric relationships are found primarily between N Ps. The main purpose of the chapter is a refutation of a strictly formal account of anapho ric relations as has been.developed in Generative Grammar. Werth wants to demonstrate that "sententional anaphora is merely the thin end of discourse anaphora, and any apparent syntactic restrictions are to be explained either as obeying the coherence constraints themselves, or as in fact emanating from a different and coincident cause, or in some cases from a full or partial syntacticisation of the cohesion requirements (cf. Giv6n, 1 979, Ch. 5)" (p . 1 87). To test his claim, Werth discusses "a type of anaphora which is usually assumed to be among the most sententially-restricted of all, namely recipro cal or each other anaphora" (p. 1 87). In particular, Werth confronts his view with the GB analysis of each other. This discussion takes up (the second) half


4. 2 Anaphora

84 of the chapter (p.

1 87-2 1 0). In Werth's alternative, each other is considered as each and the disjunctive or negative anaphor other" (p. 193). 'Negative anaphor' here means that other takes its value to be the converse of the value of its antecedent (p. 196). Each quantifies over, and is distributive-anaphoric to, other. being made up of "more than one item, namely the quantifier

Although Werth's discussion of the matter is rather original the reader doesn't end up with the feeling that the GB view has been falsified. As Werth concedes, reciprocals and reflexives have "for a constellation of reasons which we do not yet fully understand, ( . . . ) a strong tendency to confine themselves to the domain of the sentence" (p. Werth and GB then seems to

206). The difference between

be a difference in the evaluation of the facts.

ives as "a further idiosyncrasy of an already idiosyncratic category" (p.

206).

His conclusion is, therefore, that "it is a serious tactical error to regard this behaviour as diagnostic for anaphora, and thus to justify Procrustean at tempts to force all anaphora into the same inappropriate bed" (p.

206).

4.3. Syntactic variation In the last chapter of Part Two, syntactic variation is taken into considera

tion, passives and cleft constructions in particular. A review section on

Topic-Comment Articulation (TCA) leads to the formulation of Werth's 'TCA-constraint' on word order: "Semantic material is deployed in a dis course so as to respect the order: Anaphoric - Non-anaphoric. This corres ponds to the emphatic structure

{�} . . . A. Corrolary: to maintain this order

in surface structure, syntactic elements may be moved, or a variant allowing the order may be preferred to one which loses it" (p.

220). (NB: the

clustering of R and C emphasis should not surprise us, if we realize that both cohere with the foregoing discourse, positively and negatively respectively, whereas items with A emphasis do not cohere). In Section

9.5 Werth analyzes several examples of passive constructions in

written discourse. He shows that passives obey the TCA-constraint and concludes that their use is motivated by this constraint. According to Werth "passives typically occur when the logical object is anaphoric and the logical subject is not" (p.

236).

But what is meant by 'typically'? How often must a construction occur to

be called 'typical7 Werth does not use any quantitative results from a corpus study. Werth also discusses (pp. 237-239) some other functions of the passive that have been proposed in the literature and argues that these proposals would follow from his own analysis. I doubt though that this result could be generalized, in particular to functional analyses of the passive, as e.g. Pinks ter

( 1 985: 1 1 5). And I also wonder if the function(s) of the passive in (certain

varieties of) English would necessarily coincide with the function(s) of the passive in natural language in general.


Werth tends to consider the syntactic constraints on reciprocals and reflex

85

5. CONCLUSION

I hope this review has made clear what Werth was aiming at: "an account which relates relevant features of 'settings' to features of grammatical surface structure in a direct and natural way" (p. 262). As I have pointed out discussing the chapters on accent structure, the way Werth relates setting and surface structure is not as direct as be suggests: a semantic deep structure is involved. Although such a semantic level must probably play a crucial role in a model such as Werth's it also makes the undertaking complicated and vulnerable. In particular, questions remain as to the exact relation between deep 'underlying' accents and surface accents. The analysis of anaphoric connectivity, passives, and clefts is stimulating in parts, but on the whole not precise enough to be ultimately convincing. Werth does not claim that he is presenting a fmal model (who would?): "Further analysis of a great many real discourses along the lines suggested here will reveal how viable the system is, and how much modification it requires" (p. 262). He does claim, however, psychological reality for his


Section 9.6 finally is devoted to cleft and pseudo-

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

Recommend Documents