Quantifier Polarity and Referential Focus During Reading

JOURNAL OF MEMORY AND LANGUAGE 39, 290±306 (1998) ARTICLE NO. ML982561

Quantiﬁer Polarity and Referential Focus during Reading

Kevin B. Paterson University of Nottingham, United Kingdom and Anthony J. Sanford, Linda M. Moxey and Eugene Dawydiak University of Glasgow, United Kingdom

We report the results of two eye-tracking experiments that examine how readers process sentences containing anaphoric pronouns when the referent is provided by a preceding quanti®ed statement. Previous studies (Moxey & Sanford, 1987; Sanford, Moxey, & Paterson, 1996) have shown that positive and negative quanti®ers (e.g., a few and few, respectively) cause subjects to focus on different aspects of the described situation and have direct consequences for the interpretation of subsequent anaphoric pronouns. In the present studies, we consider whether positive and negative quanti®ers make different sets available as the referents of subsequent anaphora or if readers must infer the nature of these sets on encountering the anaphor. The results suggest that positives do make sets available as referents, whereas in the case of negatives, readers must infer the referent set. The ®ndings are consistent with linguistic arguments concerning the differences between positive and negative quanti®ers and add to our understanding of complex plural anaphora. ᭧ 1998 Academic Press

A major task in reading is to determine if expressions refer to the same person or object, expressions that appear in different parts of and much effort has been expended on under- the text refer to the same persons or objects standing the processes that enable a reader in the discourse. Such reference is frequently to identify an expression as an anaphor and signaled by the use of anaphors, such as de®- recover a coreferential antecedent from the nite noun phrases (e.g., the man) and pronouns preceding text. In particular, studies have (e.g., he, she, they), which usually take their shown that the time spent reading a sentence meaning from expressions that appeared ear- containing an anaphor depends in part on the lier in the text. Quite often there is a simple ease with which a reader can identify a co- coreferential relationship between an anaphor referential antecedent (e.g., Haviland & Clark, and its textual antecedent, such that the two 1974; Garrod & Sanford, 1977). Moreover, by monitoring subjects' eye movements during Experiment 1 was carried out while KBP held a Carne- reading, it has been possible to establish which gie Postgraduate Scholoarship at the University of Glas- factors have an immediate or early in¯uence gow. The remainder of the work was supported by a on the antecedent search process (Ehrlich & British Academy Grant to AJS. Thanks are due especially Rayner, 1983; O'Brien, Shank, Myers, & to Martin Pickering and Simon Garrod for helpful com- Rayner, 1988; Duffy & Rayner, 1990; Garrod, ments. We also thank Phil Johnson-Laird and two anony- mous reviewers for comments on an early version of this O'Brien, Morris, & Rayner, 1990; Garrod, paper. The full set of materials and ®ller items used in Freudenthal & Boyle, 1994). Eperiments 1 and 2 are available from Kevin Paterson, However, there are very many cases in E-mail [email protected]. which a pronoun does not bear a simple core- Address reprint requests to Kevin Paterson, Department ferring relation with a textual antecedent, in- of Psychology, University of Derby, Mickelover Campus, Derby DE3 5GX, United Kingdom, or Tony Sanford, cluding those in which either the anaphor or Department of Psychology, University of Glasgow, Hill- the antecedent expression is within the scope head, Glasgow G12 8QQ, Scotland. of quanti®cation or negation. Examples of such

0749-596X/98 $25.00 290 Copyright ᭧ 1998 by Academic Press All rights of reproduction in any form reserved.

AID JML 2561 / a015$$$$21 07-06-98 15:26:20 jmla AP: JML POLARITY AND FOCUS DURING READING 291 sentences are prevalent in the formal semantics quanti®ers, such as not many X, few X, or less literature (cf. Kamp & Reyle, 1993), yet there than 30% of X, reference to the set of fans is little experimental research on how such sen- who did not go to the match does seem to tences are interpreted. Many of the more com- be possible and even preferable. Thus (2Љ)is -plex cases are interesting because they pose acceptable, and (2؅) seems to be less obvi serious problems about which aspects of a dis- ously acceptable than when it appeared as course become the focus of attention as a result (1؅). It was argued that such a pattern of focus of using different kinds of linguistic construc- is typical of negative quanti®ers. tion. The present paper is concerned with the (2) Not many of the fans went to the game. claim that quanti®ers (e.g., some, all, a few, .2؅) They watched it with enthusiasm) few) can be differentiated in terms of a property (2Љ) They watched it on television. of focus and that this can in¯uence the interpretation that is assigned to an anaphoric pronoun The referent of They in (2Љ) is argued by during production or comprehension (Moxey & Moxey and Sanford (1987; Sanford et al., Sanford, 1987; Sanford, Moxey, & Paterson, 1996) to be the Complement set. When sub- 1994, 1996). We report two experiments that jects are invited to write continuations starting examine the resolution of anaphors that have a with the pronoun They in response to positive quanti®ed antecedent. and negative quanti®ed sentences, the results Investigations of the patterns of focus in- show just the kinds of bias described above. duced by various quanti®ers demonstrate dif- Positives almost invariably give rise to contin- ferences between those which are negative uations in which They refers to the Reference and those which are positive (see Moxey & set. With negatives, in a high proportion of Sanford, 1993a, for a discussion of negativ- continuations They is coreferential with the ity). A difference in focus patterns is indicated Complement set. There is, however, some- by the patterns of pronominal reference which thing of an asymmetry, in that the focus ef- seem to be licensed by negative and positive fects for the negatives are weaker. In the San- quanti®ers. In the simple case of a sentence ford et al. (1996) data, for instance, where a quanti®ed with the positive quanti®er some x, total of 10 positive and 10 negative quanti®ers such as Some of the fans went to the game, a were used, the Complement set rate for nega- satisfactory logical representation of the sen- tives was 62%, with 21% Reference set; the tence requires a mental representation of the Reference set rate for positives was 90%, with (necessary) set of fans who did go to the game, only 0.5% Complement set. Thus, the focus (which we shall call the Reference set), the pattern is more diffuse in the case of negatives (possible) set of fans who did not go (which than positives. we shall call the Complement set), and the Although there are these asymmetries in the possibility of things other than fans who might strength of focus effects revealed in continua- have gone (see, e.g., Johnson-Laird, 1983). tions, Sanford et al. (1996) found that focus Consideration of which of these sets might be effects for both negative and positive quanti- referred to by a subsequent pronoun shows ®ers had an equivalent effect on the ease with that the set of fans who went (the Reference which sentences were interpreted during read- Set) is licensed as a referent, as in (1؅), while ing when they contained references back to the set who did not go (the Complement Set) Reference set and Complement set. Subjects seems to be ruled out, as in (1Љ): read a series of three-sentence passages in which the second sentence was either quanti- (1) Some of the fans went to the game. ®ed by the positive expressions a few or many .1؅) They watched it with enthusiasm) or the negative expressions few or not many (1Љ) They watched it on television. (as shown in Table 1). The third sentence be- Moxey and Sanford (1987; Sanford, Moxey, gan with an anaphoric plural possessive noun & Paterson, 1996) showed that for negative phrase which referred to a property that was

AID JML 2561 / a015$$$$22 07-06-98 15:26:20 jmla AP: JML 292 PATERSON ET AL.

TABLE 1 ence set comes to be represented. The standard Example Material from Sanford et al. (1996)a argument (Kamp & Reyle, 1993) rests on the unacceptability of (3؅) as a continuation of A public meeting (3), which demonstrates that readers do not Local MPs were invited to take part in a public enquiry employ a set-subtraction operation to derive about proposals to build a new nuclear power station. # A few/Few of the MPs attended the meeting. # the Complement set. Readers cannot interpret Their presence/absence# allowed the meeting to run the pronoun They in (3؅) as referring to the more smoothly. two marbles that are not in the bag.

a Slashes (/) denote alternatives and hashes (#) denote (3) Eight of the ten marbles are in the bag. .region divisions used in Experiment 1. (3؅) They are under the sofa It is easy to show that They is typically used to refer to the marbles which are in the bag, -either consistent with the Reference set (i.e., as we argued for (1؅). However, this demon their presence) or the Complement set (i.e., stration of a failure of Complement set refer- their absence). ence is restricted to positive quanti®ers. There Readers spent more time on the sentence are two ways in which one might try to explain containing the anaphor when it referred to a away Complement set phenomena: demon- property of the mismatched set. That is to say, strate that Complement set reference is not they spent longer reading a sentence that re- really a reference to the Complement set, but ferred to a property of the Complement set rather a reference to the general set (e.g., Corb- following quanti®cation by either a few or lin, 1997) or show that Complement set refer- many and longer reading a sentence that re- ence relies on pragmatic factors rather than ferred to a property of the Reference set fol- being strict anaphora (e.g., Dowty, 1994). lowing quanti®cation by either few or not Since Corblin's approach cannot rule out all many. This shows that readers found it dif®- cases of Complement set anaphora (Sanford cult to integrate the anaphoric sentence with et al., 1996; see also the General Discussion), their understanding of the preceding text when we shall concentrate more on Dowty's sugges- the anaphor referred to a property of the unfo- tion, which is also partially compatible with cused set. Thus focus effects occur in compre- DRT. This view is that positive quanti®ers hension as well as in production. Furthermore, license Reference set anaphora (in DRT they these data suggest that during comprehension make the reference set available in the listen- the Complement set is strongly preferred as er's representation of the discourse). In con- the referent when the sentence is negatively trast, Dowty suggests that negative quanti®ers quanti®ed. do not license anaphora, but rather, if a pro- The observation of Complement set refer- nominal reference is made following a nega- ence is at ®rst sight at odds with formal lin- tive quanti®er, then a suitable referent set is guistic theories, such as discourse representa- determined pragmatically. This corresponds to tion theory (DRT; Kamp & Reyle, 1993). One the notion that Complement set reference may of the central tenets regarding quanti®cation be a form of ``mental deixis.'' So on this view, within this framework is that Complement set the issue is not so much whether Complement reference per se cannot occur. DRT supposes set reference is possible; rather, it is how the that only supersets and explicitly introduced pragmatic selection of a referent set for Com- subsets are represented, this being suf®cient plement set reference is established. for a truth±functional semantic account of Sanford et al. (1996) considered two expla- quanti®ers (e.g., Corblin, 1997; Geurts, 1997; nations of how readers process sentences that Percus, Gibson, & Tunstall, 1997). There is refer to the Reference and Complement sets no operation corresponding to ``set-subtrac- of positively and negatively quanti®ed state- tion'' in which the superset minus the Refer- ments. According to one of these accounts,

AID JML 2561 / a015$$$$22 07-06-98 15:26:20 jmla AP: JML POLARITY AND FOCUS DURING READING 293 which we will refer to as the Set-Driven ac- ments of the Reference set or by a generaliza- count, the quanti®er causes one of the sets tion about the superset (e.g., Corblin, 1997). to become the focus of processing attention. Our claim is that these answers are most likely Positive quanti®ers cause the Reference set to to be provided by elements of the Comple- become the focus of processing attention, ment set and that this explains the preponder- while negatives cause the Complement set to ance, but not exclusive production, of Com- become the focus of processing attention. Sub- plement set continuations in production stud- sequent anaphora are evaluated with respect ies (Moxey & Sanford, 1993a; Sanford et al., to the focused subset. Consequently, subjects 1996). The Inference-Driven account differs perceived anaphoric sentences to be anoma- from the Set-Driven account in proposing that lous, resulting in a longer reading time, when negatives do not cause the Complement set they were incongruent with the focused sub- to become available on reading the quanti®ed set. In the Set-Driven account, it was proposed statement, but that a suitable Complement set that the focus pattern is selected when a quan- can be inferred during integration of the ana- ti®er is encountered; that is, the processor is phoric sentence. tuned to accept linguistic input which relates A pure Inference-Driven account is sustain- to the Reference set in the case of positive able if positive quanti®ers also serve to raise quanti®ers and to accept Complement set ma- a virtual question in the reader's mind. An terial in the case of negative quanti®ers. The analysis of continuations to positive quanti®- Set-Driven account captures the spirit in ers shows two types to predominate: ®rst, sim- which DRT deals with positive quanti®ers. It ple continuations of the type ``what happened also suggests that negative quanti®ers are next,'' and second, reasons, so that given X treated in the same way. A process of set- (a subset) did Y, the continuation is about why subtraction causes the Complement set to be X did Y. It may be that positives cause the mentally represented and preferred as the ref- reader to anticipate continuations that explain erent of a subsequent pronoun. why a particular state of affairs occurred or An alternative account, which we will refer describe the consequences of that state of af- to as the Inference-Driven account, was intro- fairs taking place. These continuations would duced by Sanford et al. (1996) to account for be consistent with focus on the Reference set. negative quanti®ers. According to this ac- However, the Inference-Driven account re- count, negative quanti®ers do not directly iso- quires that positive quanti®ers do not cause late the Complement set as a focused anteced- the Reference set to become available on read- ent. Rather, it takes as a starting point the idea ing the quanti®ed statement, but that it is in- that a negative quanti®er, like other negatives, ferred during integration of the anaphoric sen- asserts a denial of a supposition (cf. Clark, tence. 1976; Wason, 1965). Thus, given the sentence Following Dowty (1994), it is also possible Not many fans went to the game, the statement that anaphora are treated differently in posi- effectively says ``Someone expected ``many'' tive and negative quanti®ed contexts. Dowty's fans to go, and less went'', a claim which has claims are consistent with a processing ac- direct empirical support (Moxey & Sanford, count in which positives are treated according 1993b). The argument here is that given a to the Set-Driven account and negatives are denial, a virtual question is induced in the treated according to the Inference-Driven ac- mind of the reader/listener which is ``Why did count. That is, negative quanti®ers do not not more people go?''. The answer to this cause the Complement set to be mentally rep- question will often be found in the reasons resented and therefore available as the referent why ``they'' could not go, which is focused of a subsequent pronoun. Rather, negative on elements in the Complement set. However, quanti®ers are treated in the manner proposed it is important to note that an answer to the by the Inference-Driven account, such that virtual question could be provided by ele- readers must infer a referent set on reading

AID JML 2561 / a015$$$$22 07-06-98 15:26:20 jmla AP: JML 294 PATERSON ET AL. the anaphoric sentence. The nature of this set tent with the focus patterns generated by either (i.e., Reference set or Complement set) will positive or negative quanti®ed statements. depend on the content of the sentence con- The critical question is whether these mis- taining the pronoun and how this content an- match effects show the same pattern for posi- swers the virtual question raised by the negatives and negatives or if they differ. If Dowty tive quanti®er. is correct, then we would expect to ®nd rela- The Set-Driven account proposes that read- tively early mismatch effects when positives ers process positives and negatives in the same are used, since according to the Set-Driven way. Dowty proposes that readers process account, positives cause the reader to instanti- positives and negatives differently. Therefore, ate a Reference set and evaluate subsequent teasing apart these alternative accounts will anaphora with respect to it. In contrast, we depend on either demonstrating that negative would expect to ®nd relatively late mismatch and positive quanti®ers are treated in the same effects for negatives, since the Inference- manner or that they are treated differently. Driven account depends on the reader estab- The present paper reports two eye-movement lishing a ®t between the content of the ana- experiments which further examine the in¯u- phoric sentence and the virtual question raised ence of negative quanti®ers on comprehen- by the negative quanti®ed statement. Impor- sion. Previous comprehension studies (San- tantly, we would also expect mismatches to ford et al., 1996) examined data for sentence cause a different magnitude of disruption in reading times and these show equivalent ef- positive and negative contexts. For positives, fects for negatives and positives. The rationale pronominal reference to the Complement set here is that eye-tracking data will provide a should be perceived as anomalous and cause ®ner instrument for detecting the appropriate massive disruption to sentence processing. For distinctions. negatives, references to either the Reference There is considerable evidence that an anal- or Complement sets will be acceptable if they ysis of eye movements can provide a detailed provide an answer to the virtual question; account of the sequence of processing events therefore sentence processing should not be that occur during reading and are sensitive to disrupted when the anaphoric sentence ini- short-lived effects that occur when processing tially appears to contain reference to the Ref- anaphoric expressions (e.g., Ehrlich & Ray- erence set. Rather, sentence processing dif®- ner, 1983; Duffy & Rayner, 1990; Garrod, culty should be experienced at a later point in Freudenthal, & Boyle, 1994; Rayner, Ra- the reading process, when the reader ®nds it ney, & Pollatsek, 1995). In particular, Garrod dif®cult to infer a congruent relationship be- et al. (1994) found that subjects had a longer tween the virtual question raised by the quan- ®rst pass reading time (i.e., had a longer read- ti®ed statement and an anaphoric sentence that ing time on the ®rst encounter) for a pronoun describes a property of the Reference set. and verb when this was inconsistent with a character who was the protagonist of a text EXPERIMENT 1 but consistent with another character taking a Method subordinate role. These results suggest that, under some conditions at least, readers will Materials and design. We used the 24 pas- immediately interpret a pronoun as referring sages from Experiment 3 of Sanford et al. to a focused antecedent and experience pro- (1996). An example is shown in Table 1, with cessing dif®culty at the ®rst point that this other examples in Appendix 1. Each passage assignment appears anomalous. There are had a title and began with a context-setting therefore good reasons to expect that an analy- sentence. The following sentence began with sis of eye movements will enable us to detect a noun phrase that was quanti®ed by either a any mismatch effects that occur on encounter- few or few. The ®nal sentence began with a ing an anaphoric expression that is inconsis- plural anaphoric noun phrase that described a

AID JML 2561 / a015$$$$22 07-06-98 15:26:20 jmla AP: JML POLARITY AND FOCUS DURING READING 295 property of either the Reference set (e.g., their tions had ``yes'' answers and half had ``no'' presence) or Complement set (e.g., their ab- answers. Subjects responded to the compre- sence) of the preceding quanti®ed sentence. hension questions by pressing a ``yes'' or a Passages were single line-spaced. ``no'' response key. There was no feedback There were two experimental manipula- on their answers. tions. The quanti®ed sentence contained either The computer displayed each experimental a positive (a few) or negative ( few) quanti®er, list in a ®xed Latin Square order, together with whereas the immediately succeeding sentence 24 ®llers that were materials for an unrelated contained an anaphor that referred to a prop- experiment and an additional ®ve ®ller pas- erty of either the Reference or Complement sages that appeared at the beginning of the sets that represent the meaning of the quanti- experiment and following three rest periods. ®ed sentence. Participants. A total of 24 undergraduate Procedure. An SRI Dual Purkinje Genera- students from the University of Glasgow were tion 5.5 eye-tracker was used to monitor the paid £5 for their participation in this experi- gaze location and movement of subjects' right ment. All participants were native English eye during reading. The eye-tracker has an speakers and had normal and uncorrected vi- angular resolution of 10' arc. A PC displayed sion. materials on a VDU screen 60 cm from subjects' eyes. The tracker monitored subjects' Results gaze location every millisecond. The tracker's Regions. The test sentences were divided output was sampled to produce a sequence of into four scoring regions, as illustrated in Ta- eye ®xations, recorded as x and y character ble 1. Region 2 contained the quanti®ed sen- positions, with their start and ®nish times. tence. Region 3 was the critical region that Before the start of the experiment, subjects contained the anaphoric noun phrase (e.g. read an explanation of the eye-tracking proce- their presence or their absence), and Region dure and a set of instructions. They were in- 4 contained the remaining words in the ana- structed to read at their normal rate and to phoric sentence. read to comprehend the sentences as well as Analysis. An automatic procedure pooled they could. Subjects were then seated at the short contiguous ®xations. Fixations of less eye-tracker and placed on a chin rest and un- than 80 ms were incorporated into larger ®xa- der forehead restraint to minimize head move- tions found within one character, and ®xations ments. Subjects then completed a calibration of less than 40 ms that were not within three procedure. characters of another ®xation were deleted. Before each trial, a ®xation cross appeared Prior to analyzing the eye movement data, we near the upper-left-corner of the screen. Im- removed those trials where either subjects mediately subjects ®xated this cross, the com- failed to read the passage properly or where puter displayed a target sentence, with the ®rst there had been tracker loss. More speci®cally, character of this sentence replacing the ®xa- those trials were removed in which a zero ®rst tion cross. This also served as an automatic pass reading time was recorded for any of the calibration check, as the computer did not dis- text regions. This accounted for 11.6% of the play the text until it detected a stable ®xation data. on the cross. If subjects did not rapidly ®xate We report three measures of eye move- the cross, the experimenter recalibrated the ments: ®rst pass reading time, total reading eye-tracker. Once subjects had ®nished read- time, and ®rst pass regressions. First pass ing each sentence, they pressed a key, and the reading time was de®ned as a sum of the dura- computer either displayed a comprehension tion of ®xations made on ®rst entering a region question or proceeded to the next trial. Com- of text until an eye movement exits the region prehension questions followed 25% of the ex- to either the left or right, providing an indica- perimental and ®ller trials. Half of these question of the dif®culty experienced when ini-

AID JML 2561 / a015$$$$22 07-06-98 15:26:20 jmla AP: JML 296 PATERSON ET AL. tially processing a region of text. Total reading First pass reading times. At Region 2, time is the sum of the duration of all ®xations the quanti®ed sentence, there was no main made within a region and provides an indica- effect of quanti®er type (F1 , F2 õ 1), but tion of the overall amount of time spent pro- there was a main effect of reference type cessing text in that region. First pass and total (F1(1,23) Å 6.25, p õ 0.05; F2(1,23) Å 4.39, reading times are reported as ms/character as p õ 0.05), with a longer ®rst pass reading an adjustment for small differences in the time for this sentence when the subsequent character size of regions across experimental anaphor referred to a property of the Com- conditions. First pass regressions are a sum plement rather than the Reference set. We of regressive saccades made from the current consider this effect to be spurious, since it most rightward ®xation with a region of text, occurs prior to the reader encountering the indicating the degree to which left to right region of text that would give rise to such movement was disrupted during the ®rst an effect and has not been replicated in any sweep of the eyes through a region of text. other studies that we have conducted. There First pass and total reading times for Regions was no interaction of quanti®er and refer-

2, 3, and 4 and ®rst pass regressions from ence type (F1 , F2 õ 1). Regions 3 and 4 were subjected to two 2 There were no differences in the ®rst pass (quanti®er type) 1 2 (reference type) ANO- reading time for Region 3, the region con- VAs, one treating subjects as the random vari- taining the anaphoric expression, for any as- able and the other treating sentences as the pect of the experimental manipulation (all Fs random variable. Mean ms/character ®rst pass õ 1). In Region 4, the ®nal region, there was and total reading times and mean ®rst pass no signi®cant main effect of quanti®er type regressions are given in Table 2. (F1(1,23) Å 1.25, p ú 0.1; F2(1,23) Å 2.95,

TABLE 2 Mean First Pass and Total Reading Times (ms/Character) with Standard Errors for Regions 2, 3, and 4, and Mean First Pass Regressions with Standard Errors from Regions 3 and 4 of Materials Used in Experiment 1

Condition

A few / A few / Few / Few / Reference Complement Reference Complement Region Measure set set set set

2 First pass reading time 29 (1.6) 31 (1.5) 30 (1.4) 32 (1.5) (ms/character) 3 First pass reading time 24 (2.1) 25 (1.5) 25 (1.6) 25 (1.8) (ms/character) 4 First pass reading time 30 (1.9) 31 (2.3) 32 (1.8) 32 (1.9) (ms/character) 3 Mean ®rst pass 1.8 (0.3) 2.1 (0.2) 2.7 (0.4) 2.9 (0.5) regressions 4 Mean ®rst pass 8.2 (0.6) 10.4 (0.6) 8.6 (0.5) 8.4 (0.6) regressions 2 Total reading time 34 (1.4) 35 (1.4) 35 (1.2) 35 (1.4) (ms/character) 3 Total reading time 32 (2.6) 42 (2.6) 40 (2.8) 34 (2.5) (ms/character) 4 Total reading time 40 (1.7) 47 (2.6) 44 (2.7) 38 (1.6) (ms/character)

AID JML 2561 / a015$$$$22 07-06-98 15:26:20 jmla AP: JML POLARITY AND FOCUS DURING READING 297 p ú 0.05), no main effect of reference type indicated that readers did not experience dis-

(F1 , F2 õ 1), nor an interaction of quanti®er ruption at Region 4 when the negative quanti- and reference type (F1 , F2 õ 1). ®er was used. First pass regressions. For regressions orig- Total reading time. At Region 2, the quanti- inating in Region 3, there was a signi®cant ®ed sentence, there was no main effect of main effect of quanti®er type (F1(1,23) Å quanti®er type (F1 õ 1, F2 õ 1), no signi®cant 9.75, p õ 0.01; F2(1,23) Å 8.11, p õ 0.01), main effect of reference type (F1(1,23) Å 2.64, such that more regressions were made follow- p ú 0.05; F2(1,23) Å 2.01, p ú 0.05), and no ing quanti®cation by few. There was no main interaction of quanti®er and reference type effect of reference type (F1 õ 1, F2 õ 1.5) or (F1 , F2 õ 1). No main effects were found at an interaction of these factors (F1 , F2 õ 1). Region 3 (Fs õ 1). However, there was a It thus appears that there may be processing signi®cant interaction of quanti®er and refer- dif®culty associated with the integration of ence type (F1(1,23) Å 20.58, p õ 0.001; any information with the representation set up F2(1,23) Å 7.33, p õ 0.05). An analysis of the by a statement quanti®ed by few (as distinct simple effects established that subjects spent a from a few) which is detectable through ®rst signi®cantly longer amount of time in this re- pass regressions. gion when it described a property of the Com- Analysis of regressions originating in re- plement as compared to the Reference set fol- gion 4 showed a main effect of quanti®er that lowing quanti®cation by a few (F1(1,23) Å was signi®cant by items only (F1(1,23) Å 16.54, p õ 0.001; F2(1,23) Å 5.60, p õ 0.05). 2.54, p ú 0.05; Å F2(1,23) Å 4.50, p õ 0.05), There was also a numerical difference in the a main effect of reference type that was mar- amount of time spent in this region when it ginal by items (F1(1,23) Å 1.98, p ú 0.05; described a property of the Reference as com-

F2(1,23) Å 3.97, p õ 0.06), and a signi®cant pared to the Complement set following quanti- interaction of these factors (F1(1,23) Å 6.00, ®cation by few; however, this difference was p õ 0.05; F2(1,23) Å 5.30, p õ 0.05). Further only signi®cant on the subjects analysis simple effects analyses were conducted in or- (F1(1,23) Å 5.51, p õ 0.05; F2(1,23) Å 2.14, der to determine the nature of this interaction p ú 0.05). This suggests that while subjects and revealed that more regressions were made experienced dif®culty in processing this por- when the noun-phrase anaphor in Region 3 tion of the anaphoric sentence when it mis- was congruent with the Reference set rather matched with the pattern of focus set up by than the Complement set of a statement quan- the quanti®er, the disruption was only robust ti®ed by a few (F1(1,23) Å 8.92, p õ 0.01; in the case of the positive quanti®er. F2(1,23) Å 7.80, p õ 0.05). However, there At Region 4, there were no main effects of was no difference in the number of regressions either quanti®er (F1 õ 1.7, F2 õ 1.3) or refer- made in Region 4 when the anaphor was con- ence type (F1 , F2 õ 1). However, there was gruent with either the Reference or Comple- a signi®cant interaction of these two factors ment set following few (F1 , F2 õ 1). So, al- (F1(1,23) Å 30.92, p õ 0.001; F2(1,23) Å though there is some evidence of detection of 14.71, p õ 0.001). An analysis of simple ef- quanti®er focus mismatches during the ®rst fects established that more time was spent on pass of the anaphoric sentence, this appears the region when the preceding region had de- to be restricted to the case where the quanti®er scribed a property of the Complement as com- is a few. A further simple effects analysis was pared to the Reference set following quanti®- conducted in order to determine if more re- cation by a few (F1(1,23) Å 21.51, p õ 0.001; gressions were made following the use of few, F2(1,23) Å 10.72, p õ 0.01) and when the as compared to the a few condition when the preceding region had described a property of anaphor was congruent with the Reference set. the Reference as compared to the Complement

There was no difference between these means set following quanti®cation by few (F1(1,23) (F1 , F2 õ 1). Therefore, ®rst pass regressions Å 10.66, p õ 0.01; F2(1,23) Å 4.62, p õ

AID JML 2561 / a015$$$$23 07-06-98 15:26:20 jmla AP: JML 298 PATERSON ET AL.

0.05). This is consistent with the Sanford et rent sentence. Clearly, negative (few) and pos- al. (1996) ®ndings, which used global self- itive (a few) quanti®ed statements do not pro- paced reading time as a measure of integration duce an identical pattern of effects. dif®culty. Sanford et al. (1996) found that global reading times showed a symmetrical focus mis- Discussion match effect for few and a few. In the present In this experiment, we monitored eye move- experiment, the total reading times for Re- ments during reading to examine the resolu- gions 3 and 4 showed that readers had detected tion of a noun-phrase anaphor that had a quan- the mismatch between the set described by the ti®ed antecedent. Subjects read a series of anaphor and those licensed by the quanti®ers short passages that contained either positive and that this disrupted the reading process. (a few) or negative ( few) quanti®ed state- Readers spent more time on Regions 3 and 4 ments and was followed by a sentence that when the anaphor described the Complement began with a plural pronoun anaphor. The ana- as compared to the Reference set of a state- phor either referred to a property of the Refer- ment quanti®ed by a few and more time on ence set (i.e., their presence) or Complement Region 4 when the anaphor described the Ref- set (i.e., their absence) of the quanti®ed state- erence as compared to the Complement set of ment. We found no difference in the ®rst pass a statement quanti®ed by few. This demon- reading time for either the region of text con- strates that although it is more dif®cult to pro- taining the critical anaphoric expression (i.e., cess anaphoric sentences when they refer to a their presence or their absence) or the region property of the unfocused set, this disruption of text that contained the remainder of this was found in both regions of the anaphoric sentence. So on the basis of this measure, it sentence when positives were used, but only appears that readers found it as easy to initially in the ®nal region when negatives were used. process the anaphor and to initially process In summary, we have found evidence for the remainder of the anaphoric sentence, re- two things. First, there is no immediate effect gardless of whether it matched or mismatched of mismatches for either quanti®er (at Region the Reference or Complement sets and regard- 3). Second, there is evidence for a stronger less of the form of prior quanti®cation. effect of anomaly with the positive quanti®er, First pass regressions showed an interesting evident in both ®rst pass regressions and in pattern. In Region 3, more regressions oc- the distribution of total reading time differ- curred when the preceding sentence had been ences over both regions of the anaphoric sen- quanti®ed by few than by a few, regardless of tence. In contrast, the mismatch effect was the form of the current sentence. This suggests restricted to total reading time for the ®nal that the representation set up by a sentence region of the anaphoric sentence when the quanti®ed by few may be more dif®cult to negative quanti®er was used. process and that this dif®culty is detectable Before relating these data to theory, we re- during the initial sweep of the eyes through port Experiment 2. While the design used in the anaphoric sentence. In Region 4, the evi- Experiment 1 was suf®ciently powerful to de- dence was for more regressions when the tect dif®culty in integrating the anaphoric sen- quanti®er of the previous sentence had been tence with an understanding of the preceding a few rather than few and the content of the text, we could not be certain that it was power- current sentence mismatched with the Com- ful enough to detect early mismatch effects. plement set. This suggests that focus mis- In Experiment 2 we used manipulations that matches between an antecedent based on a should promote the early detection of mis- few and the current sentence is detected during match effects. First, we conjoined the quanti- the initial analysis of the sentence containing ®ed and anaphoric sentences using the causal the anaphor, but that this is not true for mis- connective so. This connective takes a state matches between the focus of few and the cur- of affairs and makes the second state of affairs

AID JML 2561 / a015$$$$23 07-06-98 15:26:20 jmla AP: JML POLARITY AND FOCUS DURING READING 299 result from it. In addition, Moxey and Sanford phrase (intransitive verb and adverb) that was (1987; 1993a) found that the use of causal congruent with one of the two situations de- connectives reinforced Complement set focus scribed by the quanti®ed sentence and incon- in negative quanti®ers. Second, we localized gruent with the other. For example, in Table the potential mismatch to an intransitive verb 3 the verb phrase gambled recklessly is con- phrase rather than to a noun phrase. Previous gruent with a situation in which the men were work has shown that the main verb is a good careless with their money, but incongruent site for detecting the results of pronoun assign- with a situation in which they were careful ments (e.g., Garrod et al., 1994). Taking these with their money. For half the experimental changes into account, an example of a mis- materials the verb phrase was congruent with match material might be A few of the men were the unmarked version of the quanti®ed state- careful with their winnings, so they gambled ment and for other half the verb phrase was recklessly....Finally, the target sentence congruent with the marked version of the was arranged to be the same over all four quanti®ed statement. The sentence was com- experimental conditions, so that comparisons pleted by a temporal prepositional phrase. could be made over identical regions. This had a similar construction across all of the experimental materials. The quanti®ed and EXPERIMENT 2 anaphoric sentences were conjoined using the Method causal conjunction so in order to mark a causal Materials and design. We constructed 32 relation between the two sentences. The pas- sets of passages such as those in Table 3; other sages were double line-spaced. examples are shown in Appendix 2. Each pas- There were two experimental manipula- sage had a title and began with a context- tions. The quanti®ed sentence contained either setting sentence. The following sentence be- a positive (a few) or a negative ( few) quanti- gan with a noun phrase which was quanti®ed ®er, while the verb phrase of the quanti®ed by either a few or few and had a verb phrase sentence either matched or mismatched with which described one of two contradictory situ- the verb phrase of the subsequent sentence. ations. For example, in Table 3 the verb phrase This produced a fully crossed experimental describes a situation in which the men were design. either careful or careless with money. In many Participants. A total of 36 undergraduate of the materials, the distinction between the students from the University of Glasgow were two situations was achieved by using morpho- paid £5 for their participation in this experi- logically marked verbs. That is, one verb was ment. All participants were native English marked as the negative of the other. The ®nal speakers and had normal and uncorrected vi- sentence of each passage began with a plural sion. No one who had participated in Experi- pronoun that referred to the preceding quanti- ment 1 also participated in Experiment 2. ®ed noun phrase and was followed by a verb Procedure. We followed the same procedure as Experiment 1, except that a bite-bar was used in place of a chin rest in order to TABLE 3 restrict subjects' head movements. An Example Material from Experiment 2a Results

At the casino Regions and analysis. The test sentences A group of men won a lot of money on the roulette were divided into four scoring regions, as il- wheel.# A few/Few of the men were careful/careless lustrated in Table 3. Region 2 contained the with their winnings, so# they gambled recklessly# quanti®ed sentence and the causal connective until the money was gone. so. Region 3 contained the anaphor (they) and a Slashes (/) denote alternatives and hashes (#) denote the critical intransitive verb phrase (e.g., gam- region divisions. bled recklessly) at the start of the anaphoric

AID JML 2561 / a015$$$$23 07-06-98 15:26:20 jmla AP: JML 300 PATERSON ET AL. sentence, and Region 4 contained the temporal First pass reading time. At Region 2, the prepositional phrase which completed the ana- quanti®ed sentence, there was no main effect phoric sentence. We followed the same pre- of quanti®er type (F1 õ 1.2; F2 õ 1.4), no liminary analysis procedure as described in main effect of reference type (F1 , F2 õ 1), Experiment 1. Prior to analyzing the eye- and no interaction of quanti®er and reference movement data, we removed trials in which type (F1 , F2 õ 1). At Region 3, the critical either subjects failed to read the passage prop- region, subjects had a signi®cantly longer erly or where there had been tracker loss. This reading time when the verb phrase matched was de®ned as those trials in which a zero the Complement as compared to the Reference

®rst pass reading time was recorded for any set of the preceding sentence (F1(1,35) Å 6.14, of the text regions. This accounted for 3.5% p õ 0.05; F2(1,31) Å 4.38, p õ 0.05). There of the data. were no other effects (all Fs õ 1). At Region As in Experiment 1, ®rst pass and total 4, there was a main effect of quanti®er type reading times for Regions 2, 3, and 4 and the (F1(1,35) Å 4.79, p õ 0.05; F2(1,31) Å 3.88, ®rst pass regressions from Regions 3 and 4 p õ 0.06), with a longer ®rst pass reading were subjected to two 2 (quanti®er type) 1 2 time for this region following quanti®cation (reference type) ANOVAs, one treating sub- by a few. There was no main effect of refer- jects as the random variable and the other ence type (F1 , F2 õ 1) and no signi®cant inter- treating sentences as the random variable. action of quanti®er and reference type

Reading times are reported as ms/character to (F1(1,35) Å 2.38, p ú 0.05; F2(1,31) Å 2.04, enable comparison with those reading times p ú 0.05). reported for Experiment 1. Mean ms/character First pass regressions. At Region 3, there ®rst pass and total reading times and ®rst pass were no main effects of quanti®er type regressions are given in Table 4. (F1(1,35) Å 2.17, p ú 0.05; F2 õ 1), reference

TABLE 4 Mean First Pass and Total Reading Times (ms/Character) with Standard Errors for Regions 2, 3, and 4, and Mean First Pass Regressions with Standard Errors from Regions 3 and 4 of Materials Used in Experiment 2

Condition

A few / A few / Few / Few / Reference Complement Reference Complement Region Measure set set set set

2 First pass reading time 28 (1.4) 28 (1.0) 29 (1.1) 29 (1.5) (ms/character) 3 First pass reading time 30 (1.5) 32 (1.5) 30 (1.5) 32 (1.9) (ms/character) 4 First pass reading time 31 (1.5) 32 (1.6) 31 (1.8) 29 (1.5) (ms/character) 3 Mean ®rst pass 5.8 (0.5) 6.5 (0.6) 6.1 (0.5) 6.4 (0.6) regressions 4 Mean ®rst pass 9.1 (0.6) 10.7 (0.7) 10.1 (0.6) 9.3 (0.7) regressions 2 Total reading time 36 (2.0) 40 (2.1) 41 (2.3) 40 (2.2) (ms/character) 3 Total reading time 37 (2.0) 43 (2.6) 43 (2.6) 41 (2.8) (ms/character) 4 Total reading time 38 (2.2) 42 (2.4) 39 (2.4) 36 (2.1) (ms/character)

AID JML 2561 / a015$$$$23 07-06-98 15:26:20 jmla AP: JML POLARITY AND FOCUS DURING READING 301

type (F1 , F2 õ 1), or an interaction of these time when the verb phrase matched the Refer- factors (F1 , F2 õ 1). At Region 4, there were ence or Complement sets following quanti®- no differences in the number of regressions cation by few (F1 , F2 õ 1). due to either a main effect of quanti®er type At Region 4 there was a main effect of

(F1 , F2 õ 1) or reference type (F1 õ 1.6, quanti®er type (F1(1,35) Å 8.81, p õ 0.01; F2 õ 1.7). However, there was a signi®cant F2(1,31) Å 7.51, p õ 0.01), but no main effect interaction of these factors (F1(1,35) Å 7.49, of reference type (F1 , F2 õ 1). However, there p õ 0.01; F2(1,31) Å 6.84, p õ 0.05). There was a signi®cant interaction of these two fac- were signi®cantly more regressions when the tors F1(1,35) Å 10.74, p õ 0.01; F2(1,31) Å verb phrase of the anaphoric sentence matched 12.84, p õ 0.01). An analysis of simple effects the Complement set as compared to the Refer- showed that more time was spent on this re- ence set following quanti®cation by a few gion when the preceding verb phrase matched

(F1(1,35) Å 6.66, p õ 0.05; F2(1,31) Å 6.54, the Complement as compared to the Reference p õ 0.05), but no difference following quanti- set following quanti®cation by a few (F1(1,35) ®cation by few (F1 õ 1.7, F2 õ 1.2). Å 7.40, p õ 0.01; F2(1,31) Å 8.81, p õ 0.01). Total reading time. At Region 2, there was More time was also spent on this region when a main effect of quanti®er type (F1(1,35) Å the preceding verb phrase matched the Refer- 7.94, p õ 0.01; F2(1,31) Å 7.00, p õ 0.05). ence as compared to the Complement set fol- There was no signi®cant main effect of refer- lowing quanti®cation by few (F1(1,35) Å 3.67, ence type (F1(1,35) Å 2.66, p ú 0.05; F2(1,31) p õ 0.07; F2(1,31) Å 4.41, p õ 0.01). Å 1.05, p ú 0.05). There was a signi®cant interaction of quanti®er and reference type Discussion

(F1(1,35) Å 8.02, p õ 0.01; F2(1,31) Å 5.70, In this experiment we compared reading p õ 0.05). An analysis of the simple effects times for regions of a text containing a pro- showed that there was a signi®cantly longer noun and verb phrase that was congruent with reading time for this region when the verb an action expected of either the Reference or phrase of the anaphoric sentence matched the Complement set of a preceding quanti®ed Complement set as opposed to the Reference statement. A similar manipulation has been set following quanti®cation by a few (F1(1,35) used to successfully demonstrate the early de-

Å 9.29, p õ 0.01; F2(1,31) Å 7.06, p õ 0.05). tection of focus mismatches in previous eye- However, there was no difference in total tracking studies (Garrod et al., 1994). We ob- reading time when the verb phrase matched tained a robust effect on ®rst pass reading time either the Reference or Complement sets fol- for the region containing the pronoun and in- lowing quanti®cation by few (F1 , F2 õ 1). transitive verb phrase, such that more time At Region 3, the critical region, there was was spent initially reading this region of text no main effect of quanti®er type (F1 õ 1.8; when the verb phrase was congruent with the

F2 õ 1.2), but there was a main effect of Complement set. Therefore, it appears that reference type (F1(1,35) Å 5.46, p õ 0.05; readers found it dif®cult to initially process

F2(1,31) Å 5.41, p õ 0.05). There was also a text that referred to the Complement set of the signi®cant interaction of quanti®er and refer- quanti®ed statement, regardless of the quanti- ence type (F1(1,35) Å 10.91, p õ 0.01; ®er used. This was an important ®nding be- F2(1,31) Å 7.35, p õ 0.01). An analysis of cause it demonstrated that the experimental the simple effects established that there was manipulation could give rise to processing a signi®cantly longer time after reading for preferences at the critical region. This ®nding this region when the verb phrase matched the also suggests that the focus effects associated Complement as compared to the Reference set with negative quanti®ers did not license an following quanti®cation by a few (F1(1,35) Å initial interpretation of the pronoun as refer- 14.68, p õ 0.01; F2(1,31) Å 10.53, p õ 0.01). ring to the Complement set. Rather, it is con- However, there was no difference in reading sistent with an account in which the pronoun

AID JML 2561 / a015$$$$23 07-06-98 15:26:20 jmla AP: JML 302 PATERSON ET AL. is initially attached with a referent that is de- positives and negatives during comprehen- scribed in the preceding text (i.e., the Refer- sion. Subjects found sentences that referred ence set). This possibility is expanded upon to the Complement set of positive quanti®ed in the General Discussion. statements as dif®cult to process as sentences As with Experiment 1, there was evidence that referred to the Reference set of negative in the ®rst pass regression pattern for detection quanti®ed statements. The results suggest that of a focus mismatch in Region 4 when the positives and negatives have an equal but op- quanti®er was a few, but not when it had been posite effect on the comprehension of senten- few. Furthermore, the total reading times ces containing plural anaphora. showed that in Regions 2 and 3 there was a The experiments reported in the present pa- mismatch effect with a few, but none with few, per further examined the comprehension of and a mismatch effect with both a few and plural anaphora in positive and negative quan- few in Region 4. These effects again suggest ti®ed contexts. We monitored subjects' eye that focus mismatch caused a greater disrup- movements to obtain a more detailed account tion of sentence processing when a few was of the reading process and so enable us to used than when few was used. detect ®ne-grained differences in the processing of texts containing an anaphor that GENERAL DISCUSSION refers to the Reference or Complement sets of Although there has been considerable work a preceding quanti®ed statement. The results on the processing of simple anaphora, there seem clear. There was a systematic difference has been relatively little on more complex plu- between the pattern of results obtained when ral anaphora, including those cases in which the context contained the positive quanti®er a the textual antecedent is within the scope of few and the negative quanti®er few. Subjects quanti®cation. The present paper explored experienced sentence processing dif®culty how focus effects associated with positive and when the anaphor matched the Complement negative quanti®ers in¯uence the comprehen- set of positive quanti®ed statements and the sion of sentences containing plural anaphora. Reference set of negative quanti®ed state- The examples we presented and the review ments. However, a few yielded a more disrup- of the work by Moxey and Sanford (1987; tive mismatch effect in the sense of producing Sanford et al., 1996) demonstrate that positive detectable differences in total reading time for and negative quanti®ers can be differentiated Regions 2, 3, and 4 (disruption was restricted in terms of a property of focus and that this to Region 4 for few), and ®rst pass regressions has consequences for the interpretation of sub- to mismatches in Region 4 (but none for few) sequent plural anaphora. However, different for Experiment 2. A comparable pattern of patterns of results have been obtained on pro- results was found in Experiment 1. There was duction and comprehension tasks. According evidence of a more disruptive mismatch effect to production data, positives (e.g., a few, for a few, with differences in the total reading many) have a restricted pattern of focus and time for Regions 3 and 4 (again disruption due only license the Reference set as the referent to a mismatch effect with few was restricted to of subsequent plural anaphora. In contrast, Region 4), and evidence of ®rst pass regres- negatives (e.g., few, not many) have a more sions from Region 4 in response to the mis- diffuse pattern of focus, such that both the match. There was no evidence that regressive Reference and Complement sets are possible saccades were made in response to mismatch referents of subsequent plural anaphora, al- when few was used, but more regressions were though there is a statistical preference for pro- made from Region 3 when the preceding sen- ducing sentence completions in which the ana- tence contained few as opposed to a few, phor refers to the Complement set. However, which most likely re¯ected general dif®culties self-paced reading data reported by Sanford in sentence integration. Clearly, focus mis- et al. (1996) indicated symmetrical effects for matches caused greater disruption to sentence

AID JML 2561 / a015$$$$23 07-06-98 15:26:20 jmla AP: JML POLARITY AND FOCUS DURING READING 303 processing in positive than negative quanti®er action that was congruent with the Reference contexts. set, regardless of the polarity of the quanti®er The results are consistent with theoretical used. This suggested that there was a default accounts in which there is an asymmetry in the preference for Reference set perspectives. processing of positive and negative quanti®ed This is an important ®nding, because it dem- statements. Thus, the results are not consistent onstrated that subjects did attempt to rapidly with the Set-Driven account, according to evaluate the congruency of an anaphor and which positives cause the reader to focus on potential referents mentioned the preceding the Reference set and negatives cause the text, but that focus effects associated with the reader to focus on the Complement set. Fol- negative quanti®er did not in¯uence this pro- lowing this account, we expected to obtain cess. Finding a ®rst pass processing advantage symmetrical mismatch effects for positives for anaphors that refer to the Reference set is and negatives when anaphora were interpreted consistent with an account proposing asym- with respect to the focused subset. The results metric effects of positive and negative quanti- are consistent with an alternative account in ®ers. This ®nding is not consistent with the which positives make available the Reference Set-Driven account, which claims that nega- set as the referent of subsequent anaphora, but tives cause the Complement set to be made negatives do not license strict anaphora. That available as the referent of a subsequent ana- is to say, negatives do not cause the Comple- phor. ment set to become available as the referent The short-lived processing advantage for of a subsequent anaphor, but are used to deny Reference set anaphors is best explained in that a particular state of affairs in true. On terms of a distinction between processes this view, the reader's principle task is not to which establish an immediate bonding be- compute a coreferential relationship between tween an anaphor and potential referent in the the anaphor and sets made available by the preceding text and processes contributing to quanti®ed statement, but to determine the con- the ultimate resolution of the anaphor (San- gruency of the anaphoric sentence and the de- ford & Garrod, 1989; Garrod & Sanford, nial asserted by the negative quanti®ed state- 1994). During the initial bonding process, ment. Moxey and Sanford (1987; Sanford et readers attempt to form a link between the al., 1996) have shown that when asked to pro- anaphor and a focused referent on the basis of duce a continuation to negative quanti®ed low-level information, including gender and statements, subjects tend to provide reasons number information. The ultimate resolution for asserting the denial. These continuations of the anaphor, however, depends on estab- predominantly concern the Complement set, lishing a semantic match with the potential therefore anaphoric sentences which refer to referent, taking account of information pro- the Complement set are most likely to appear vided by the discourse context. Importantly, congruent with the quanti®ed statement. How- the immediate bonding may be short-lived, ever, subjects do not provide continuations and the potential referent selected for immedi- that exclusively concern the Complement set. ate bonding may not be the same as the one Both the Complement and Reference sets can that is ultimately assigned to the anaphor. In in principle provide reasons for asserting the the present case, an immediate bonding is es- denial, and therefore initial references to either tablished between the pronoun and Reference of these sets should not result in an immediate set of the quanti®ed statement. This immediate mismatch effect. Thus, the alternative account bonding is then evaluated in terms of the ®t predicts an asymmetrical pattern of results for between the verb of the anaphoric sentence positives and negatives. and Reference set properties, resulting in a In Experiment 2, we obtained a short-lived reading time advantage when these are con- ®rst pass processing advantage when the ana- gruent. It appears that the focusing properties phor and following verb phrase described an of the negative quanti®er do not in¯uence the

AID JML 2561 / a015$$$$23 07-06-98 15:26:20 jmla AP: JML 304 PATERSON ET AL. immediate bonding process, but do in¯uence 20 cows did not give milk in (6), but that the ultimate resolution of the anaphor. might nonetheless mean that the herd of cows Several researchers have questioned the re- was unproductive. This is illustrated in (6) ality of Complement set focus and argued that above where the expression Not quite all indi- those continuations which appear to refer to cates that there was reason to expect all would the Complement set are in fact generalizations give milk. Percus et al. argued that apparent about the superset (Corblin, 1997; Geurts, reference to a property of the Complement 1997; Percus et al., 1997). The generalization set can be explained as cases of collective claim can be illustrated by the example below. predication. They used this to claim that quan- -Given (4), (4؅) could be considered a general- ti®ers never make the Complement set avail ization about the set of fans: able as an antecedent. While we are certain that collective predication will occur under (4) Few of the fans went to the game. some conditions, and in itself is not inconsis- .4؅) They watched it on TV instead) tent with the Inference-Driven account of neg- .The argument is that (4؅) is about fans in gen- atives, it cannot account for all of the data eral, as in They mostly watched it on TV in- Critically, both the generalization and collec- stead. Sanford et al. (1996; Moxey & Sanford, tive predicate arguments predict that subjects in press) dismissed this argument, since it can- should judge anaphora that appear to refer to not hold for the case where the Complement the Complement set as actually referring to set is very small compared to the Reference the superset, or the set in general. Yet, subjects set. Had (4) been (5) instead, then (4؅) could rarely provide such a response when asked to not be a generalization: make such a judgment in production experiments, but instead judge such anaphora to con- (5) Not quite all of the fans went to the stitute reference to the Complement set game. (Moxey & Sanford, 1987; Sanford et al., Sanford et al. found that many continuations 1996). Sanford et al. found that at most sub- of a complement set type were made to quanti- jects perceived their responses as generaliza- ®ed statements in which the Complement set tions on only 11% of occasions. Furthermore, was very near to zero. So we would argue that in almost all cases where there was a possible Complement set reference is a reality. We do Complement set reference, subjects judged not argue that generalizations never occur, their responses to be Complement set refer- however. ences and not generalizations. Therefore it ap- A related and more sophisticated version pears that while generalizations about the su- of the generalization argument is restricted to perset are an option in response to a negative predicates which can take a collective reading quanti®er, genuine Complement set reference (Percus et al, 1997; Geurts, 1997). Percus et predominates. al. considered the following type of example: For some linguists, these alternatives are considered important in order to preserve the (6) Not quite all of the cows gave milk. rule that the Complement Set is not singled Their unproductivity was a bad omen. out as an entity to support reference. We think They argued that in this example unproductiv- that as a knock-down argument, this maneuver ity is a collective property of the herd of cows, fails to be convincing. But in terms of our meaning that while each and every cow need Inference-Driven account of negatives, we not be deemed unproductive, they are unpro- have no dif®culty in accepting that each of ductive when considered as a group. This dif- these possibilities may occur. Our present ex- fers from the generalization argument in that periments demonstrate that plural anaphora the property (i.e., unproductivity) need not be are processed differently when they appear in true of the majority of the superset. For in- negative rather than positive quanti®ed con- stance, it may be the case that only 2 out of texts. We have argued that negatives do not

AID JML 2561 / a015$$$$24 07-06-98 15:26:20 jmla AP: JML POLARITY AND FOCUS DURING READING 305 directly license a coreferential relationship be- tem installed over the Christmas holidays. tween an anaphor and sets that represent the A few/Few of the secretaries had previously meaning of a quanti®ed statement. Instead we used the system. have argued that negatives are used to deny Their experience/inexperience made the oth- that a particular state of affairs is true, and ers feel more relaxed. subsequent anaphoric sentences are evaluated in terms of their congruency with the quanti- A public meeting. ®ed statement. This entails a more diffuse set Local MP's were invited to take part in a pub- of referential possibilities than in the case of lic inquiry about proposals to build a nu- positives and may include generalizations or clear power station. A few/Few of the MP's reference to the Complement set. attended the meeting. Their presence/absence helped the meeting APPENDIX 1 run more smoothly. Examples of Materials Used in Experiment 1 APPENDIX 2 In a hospital. Examples of Materials Used The doctor needed permission from some of in Experiment 2 the patients before she tested a new drug on them. A few/Few of the patients agreed In the classroom. to act as guinea pigs. The teacher lectured the entire class about be- Their consent/refusal was noted by the hospi- having properly. tal registrar. Q of the children were scolded/praised by the teacher Testing job applicants. so they wept tearfully until their parents ar- Prospective air traf®c controllers had to ®ll in rived. a personality questionnaire then sit a series of aptitude tests. A few/Few of the appli- At the casino. cants passed. Their success/failure con- A group of men won a lot of money on the ®rmed the organizer's expectations. roulette wheel. A few/few of the men were careful/careless In the court. with their winnings, Some local youths were arrested during a po- so they gambled recklessly until the money lice raid at a party and accused of drug- was gone. dealing. A few/Few of the youths were found guilty of the crime. After the accident. Their acquittal/conviction was a relief to the Poor driving caused a major motorway pile- whole neighbourhood. up. A few/few of the drivers admitted/denied re- At the gym. sponsibility for the accident, Some weight-lifters from the gym competed so they apologized profusely when the police to see who could lift a heavy dumbbell. A arrived. few/Few of the weight-lifters managed to lift it off the ground. A student house. Their strength/weakness surprised their A group of students moved into a new house. friends. A few/few of students felt cold/warm in the house, In the of®ce. so they shivered icily until the heating came The insurance of®ce had a new computer sys- on.

AID JML 2561 / a015$$$$24 07-06-98 15:26:20 jmla AP: JML 306 PATERSON ET AL.

At the football match. ford, (1993) Communicating quantities. Journal of The local football fans found it dif®cult to get Semantics, 18, 87±94. Haviland, S. E., & Clark, H. H. (1974). What's new? Ac- tickets for the cup ®nal. quiring new information as a process in comprehen- A few/few of the fans were present at/absent sion. Journal of Verbal Learning and Verbal Behav- from the game, ior, 13, 512±521. so they cheered excitedly when the home team Johnson-Laird, P. N. (1983). Mental models. Cambridge, scored. UK: Cambridge University Press. Kamp, H., & Reyle, U. (1993). From discourse to logic: Introduction to model-theoretic semantics of natural After a hijacking. language, formal logic and discourse representation The hijackers were trapped in the aircraft and theory. Dordrecht: Kluwer Academic. surrounded by police. Moxey, L. M., & Sanford, A. J. (1987). Quanti®ers and A few/few of the hijackers decided to con- focus. Journal of Semantics, 5, 189±206. Moxey, L. M., & Sanford, A. J. (1993a). Communicating cede/resist defeat, quantities: A psychological perspective. Hove, UK: so they surrendered unconditionally before the Erlbaum. police attacked. Moxey, L. M., & Sanford, A. J. (1993b). Prior expectation and the interpretation of natural language quanti®ers. REFERENCES European Journal of Cognitive Psychology, 5, 73± 91. Clark, H. H. (1976). Semantics and comprehension. The Moxey, L. M., & Sanford, A. J. (in press) Focus effects Hague: Mouton. with negative quanti®ers. In M. Crocker, M. Picker- Corblin, F. (1997). Quanti®cation et anaphore discursive: ing, & C. Clifton (Eds.), Proceedings of ®rst confer- la reference aux complementaires. Langages, 123, ence on Architectures and Mechanisms of Language 51±74. Processing (AMLaP). Cambridge: Cambridge Uni- Dowty, D. (1994). The role of negative polarity and con- versity Press. cord marking in natural language reasoning. In M. O'Brien, E. J., Shank, D. M., Myers, J. L., & Rayner, K. Harvey and L. Santelmann (Eds.), Proceedings of the (1988). Elaborative inferences during reading: Do 4th annual conference on semantics and linguistic they occur on-line? Journal of Experimental Psy- theory, (Vol. 4, pp. 114±144). Cornell University, chology: Learning, Memory and Cognition, 14, 410± Department of Linguistics. 420. Duffy, S. A., & Rayner, K. (1990). Eye-movements and Percus, O., Gibson, T., & Tunstall, S. (1997). Anteced- anaphor resolution: Effects of antecedent distance enthood and the evaluation of quanti®ers. Poster pre- and typicality. Language and Speech, 33, 103±119. sented at the 10th CUNY conference, Santa Monica, Ehrlich, K., & Rayner, K. (1983). Pronoun assignment CA, March 20±22. and semantic integration during reading: Eye-move- Rayner, K., Raney, G. E., & Pollatsek, A. (1995). Eye ments and immediacy of processing. Journal of Ver- movements and discourse processing. In R. F. bal Learning and Verbal Behavior, 22, 75±87. Lorch & E. J. O'Brien (Eds.), Sources of coherence Garrod, S., Freudenthal, D., & Boyle, E. (1994). The role in reading, Hillsdale, NJ: LEA. of different types of anaphor in the on-line resolution Sanford, A. J., & Garrod, S. (1989). What, when and of sentences in a discourse, Journal of Memory and how?: Questions of immediacy in anaphoric refer- Language, 33, 39±68. ence resolution. Language and Cognitive Processes, Garrod, S., O'Brien, E. J., Morris, R. K., & Rayner, K. 4, 235±262. (1990). Elaborative inferencing as an active or pas- Sanford, A. J., Moxey, L. M., & Paterson, K. B. (1994). sive process. Journal of Experimental Psychology: Psychological studies of quanti®ers, Journal of Se- Learning, Memory and Cognition, 16, 250±257. mantics, 10, 153±170. Garrod, S., & Sanford, A. J. (1977). Interpreting ana- Sanford, A. J., Moxey, L. M., & Paterson, K. B. (1996). phoric relations: The integration of semantic infor- Attentional focusing with quanti®ers in production mation during reading. Journal of Verbal Learning and comprehension. Memory and Cognition, 24, and Verbal Behavior, 16, 77±90. 144±155. Garrod, S., & Sanford, A. J. (1994). Resolving sentences Wason, P. C. (1965). The contexts of plausible denial. in a discourse context: How discourse representation Journal of Verbal Learning and Verbal Behavior, 4, affects language understanding. In M. Gernsbacher 7±11. (Ed.), Handbook of psycholinguistics. San Diego: Academic Press. (Received May 27, 1997) Geurts, B. (1997). Review of L. M. Moxey and A. J. San- (Revision received December 9, 1997)

AID JML 2561 / a015$$$$24 07-06-98 15:26:20 jmla AP: JML