Lexical representations in spoken language comprehension

LANGUAGE AND COGNITIVE PROCESSES, 1988,3(1) 1-16Lexical Representations in SpokenLanguage ComprehensionWilliam Marslen-WilsonMax-Planck Institute for Psycholinguistics, Nijmegen, The Netherlandsand M. R. C. Applied Psychology Unit, Cambridge, U. K.Colin M. BrownMax-Planck Institute for Psycholinguistics, Nijmegen, The NetherlandsLorraine Komisarjevsky TylerDepartment of Experimental Psychology, University of Cambridge,Cambridge, U. K.This study investigates the timing with which lexical representations aredeployed at different levels of the language system, contrasting linguisticaspects of verb argument frames with their consequences for interpretation inthe domain of non-linguistic, conceptual models. The experiment examinedmonitoring latencies to noun targets that were either normal with respect tothe preceding verb, or which violated either pragmatic, semantic, or categorialconstraints imposed by the verb’s argument frame and its associatedco-occurrence restrictions. The results show that syntactic and semanticconstraints derived from the verb have immediate effects on processing, andthat there is also a very rapid projection of the thematic properties of verbargument frames on to non-linguistic domains of interpretation and inference,involving the listener’s mental model of the discourse.I NTRODUCTI ONThe comprehension of a spoken utterance requires the listener to integratetogether a variety of different types of linguistic and non-linguistic knowledge.In particular, the linguistic properties of the utterance-its acoustic-phonetic,lexical, syntactic, and semantic properties-must in someway be mapped onto a mental model of the current discourse, taking intoRequests for reprints should be addressed to William Marslen-Wilson, M.R.C. AppliedPsychology Unit, 15 Chaucer Road, Cambridge CB2 ZEF, U.K.0 1988 Lawrence Erlbaum Associates Ltd. & V.S.P. Publications

2 MARSLEN-WILSON, BROWN, TYLERaccount the listener’s general non-linguistic knowledge of the world. It hasbecome increasingly evident , from developments in linguistics , psycholinguistics,and computational linguistics, that lexical representations play acentral role in this process of integration. Not only do lexical representationsprovide the basic bridge between sound and meaning, linking thephonological properties of word-forms with clusters of syntactic andsemantic attributes, but also they provide the basic structural framework interms of which the linguistic representation of the utterance is constructed,and in terms of which this linguistic representation is projected on to aninterpretation in a mental model.In the research reported here we exploit the diverse properties of theargument frames associated with verbs, in order to examine the basictiming with which lexical representations are deployed at different levels ofthe comprehension process, and in order to assess the relative importanceof these lexically based processes at each level. We begin with a briefsummary of the linguistic and psycholinguistic background to this research.BackgroundThe psycholinguistically most influential analysis of the properties oflexical representations is still Chomsky’s (1965) treatment of the cooccurrenceconstraints associated with a given lexical item. In a complexanalysis of how these constraints could be expressed as part of the basecomponent of a transformational grammar, he distinguished two majortypes of context-sensitive subcategorisation rules. Strict subcategorisationrules specify the categorial context in which lexical entries can occurstating,for example, that a verb like grow can be followed either by an NP,by an Adjectival form, or by nothing (in its intransitive use). Selectionalrules further subcategorise lexical items in terms of the syntactic features ofthe items that can fill the categorial possibilities specified by the subcategorisationrules. By syntactic features, Chomsky meant properties such as[*Animate] or [+.Abstract], which in later treatments have usually beenclassified as semantic in nature (e.g. Jackendoff, 1972).All subsequent analyses of the properties of lexical representations havepreserved, in one form or another, this fundamental distinction betweenthe categorial properties of the verb argument frame and the semantic andsyntactic properties of the items that can fill these argument slots. And,certainly, to develop a psycholinguistic model of how lexical representationsparticipate in language comprehension, it is necessary to capture insome way the processing implications of such a distinction. When averb-form is identified, what kinds of subcategorisation and selectionalconstraints are made available, and when and how do they affect theprocessing of the subsequent input? These are questions that have received

LEXICAL REPRESENTATIONS IN SPOKEN LANGUAGE COMPREHENSION 3surprisingly little attention in psycholinguistic research. Scattered attemptswere made in the early literature (e.g. Downey & Hakes, 1968; Fodor,Garrett, & Bever, 1968; Hakes, 1971; Moore, 1972) to explore some of theprocessing consequences of Chomsky’s proposals for lexical representation,but the results of these studies were contradictory and inconclusive.Recent developments in the linguistic analysis of lexical representationsmake their processing implications all the more salient for an on-linetheory of language comprehension. Essentially every major initiative ingenerative linguistics over the past decade has served to move lexicalrepresentations into a central position in determining the syntactic andsemantic properties of a string-indeed, as Wasow (1985) remarks, clausestructure has come to be viewed as a projection of lexical semantics. Forexample, in both the Government-Binding approach (Chomsky, 1981; vanRiemsdijk & Williams, 1986) and in Lexical Functional Grammar (Bresnan,1982), lexical representations determine the basic relations betweenelements at both syntactic and semantic levels of the theory. Government-Binding achieves this through the combined effects of the Theta-Criterionand the Projection Principle, operating on the argument structures associatedwith verbs, whereas Lexical Functional Grammar assumes that lexicalentries have a dual specification, pairing predicate argument structureswith specifications of grammatical functions. The grammatical functionsassociated with lexical items map on to a syntactic constituent structure,whereas the lexical predicate argument structure determines the functionalstructure of the string, which in turn determines its semantic properties atthe appropriate level of representation.What is important about these developments is the closeness with whichthey link the subcategorisation properties of a lexical item with its argumentstructure in a semantic representation. The argument frames associatedwith lexical items specify not only how these arguments mightfunction in a purely syntactic representation of the string, but also in itssemantic interpretation. And it is this semantic interpretation that determines,in turn, how the string is projected on to a discourse model. Ineffect, on this type of enriched account of lexical representations, theargument frames associated with verbs have consequences not simply forthe linguistic analysis of an utterance, but also for the construction of aninterpretation in the non-linguistic, conceptual domain (c.f. Carlson &Tanenhaus, 1987).But even in this current linguistic climate of emphasis on the importanceof the lexicon, the on-line processing functions of lexical representationsremain relatively neglected-especially in the auditory domain. A numberof studies have begun to appear looking at lexical effects in the parsing ofwritten texts. These range from questionnaire studies (e.g. Ford, Bresnan,& Kaplan, 1982), to studies using more on-line measures (e.g. Clifton,

4 MARSLEN-WILSON, BROWN, TYLERFrazier, & Connine, 1984; Mitchell, 1987; Mitchell & Holmes, 1985;Tanenhaus, Stowe, & Carlson, 1985). All of these studies show thatlexical “preferences”-the relative salience of different subcategorisationframes associated with the same word-form-have an immediate effect onsubsequent structural processing. But in the auditory domain, as far as weare aware, there is no research which has looked explicitly at the on-linerecognition of verb-forms and the activation of their associated argumentframes.The research which seems most relevant to the current study is someresearch of our own which did not directly vary lexical variables at all. Thiswas an experiment (Marslen-Wilson & Tyler, 1980) designed to track thetime-course with which different types of processing information-globallydefined as syntactic and semantic-became available as listeners heard aspoken utterance. The subjects monitored for word-targets at differentserial positions in three kinds of prose material: Normal Prose, AnomalousProse, and Scrambled Prose. Anomalous Prose differed from NormalProse in having no semantic organisation while preserving syntactic andprosodic structure. Scrambled Prose differed from both in having neithersyntactic nor semantic organisation. These three types of test materialswere presented either as isolated sentences, or in the context provided by apreceding lead-in sentence. The pattern of monitoring responses acrossword positions for the three prose materials led to a view of sentenceprocessing in which, from the first word of an utterance, the syntactic andsemantic properties of the incoming speech are immediately interpreted inthe context of the current discourse model (Marslen-Wilson& Tyler, 1980;1981).These global claims about the properties of on-line language processinghave a rather direct interpretation in the lexical domain we have beendiscussing and, certainly, the enriched functions that are being associatedwith lexical representations make it easier to see how the speech input canbe interpreted as it is being heard. They imply that as a lexical entry isaccessed, this immediately makes available information about its argumentframe and places syntactic and semantic constraints on the types ofelements that can fill these slots. These argument frames are specified in away that has immediate consequences not only for the local syntacticrelations of items in the string, but also for their thematic or functionalroles. This means that the elements that fill these argument slots areassigned a value not only with respect to some putative level of syntacticorganisation, but also with respect to a semantic and ultimately pragmaticinterpretation. In the following section of the paper we describe anexperiment designed to establish how far lexical representations do makeavailable information or constraints that operate at these different levels,and the timing with which these come into operation.

LEXICAL REPRESENTATIONS IN SPOKEN LANGUAGE COMPREHENSION 5Local Violation of Lexical ConstraintsIn an earlier study (Marslen-Wilson & Tyler, 1980), the availability ofdifferent types of processing information was evaluated using global distortionsof complete texts-as, for example, in the overall contrast betweenNormal Prose and Anomalous Prose. The problem with this technique is,first, that it rules out the possibility of looking at the processing informationmade available by any specific lexical item and, secondly, that it makesit difficult to decide exactly what are the types of information of which aglobal distortion is in fact depriving the listener (cf. Cowart, 1982; Marslen-Wilson& Tyler, 1983). Marslen-Wilson & Tyler (1980) presented thecontrast between Normal and Anomalous Prose as a difference in semanticinterpretability, but without systematically dissociating this from pragmaticinterpretability. Furthermore, as Cowart (1982) points out, AnomalousProse may have contained violations of syntactic constraints as well as ofsemantic constraints.These problems are avoided by looking at the local violation of lexicalconstraints in otherwise normal sentence contexts. Specifically, by holdingthe sentence constant, and varying the properties of the verb, we candetermine how and when different aspects of the verb’s argument frameaffect the processing of a spoken sentence. Take as an example thefollowing set:la. John carried the guitar.lb. John buried the guitar.lc. John drank the guitar.Id. John slept the guitar.These are four identical strings, all with the same monitoring target(guitar), where the only variation is in the verb, and in the argumentframes associated with the verb. In so far as these variations involveproperties of lexical representations that are significant for on-line processing,then they will affect the listener’s response to the monitoring target.In sentence (la) the relationship between the verb (carried) and thetarget (guitar) is fully acceptable on syntactic, semantic, and pragmaticgrounds. The subcategorisation requirements of the verb allow for anounphrase as direct object, a guitar has the appropriate semantics for theaction of carrying, and carrying a guitar is a perfectly reasonable activity inthe context of a standard model of the world. Response times to targets inNormal contexts like (la) form the baseline condition for the experiment.Sentences (lb) and (lc) illustrate two grades of potential violation of thelexical representations evoked by the verb. In both cases the target NP(guitar) remains categorially appropriate-the verbs are transitive and

6 MARSLEN-WILSON, BROWN, TYLERaccept a nounphrase as direct object. Sentence (lb), however, constituteswhat we will label a pragmatic anomaly, and contrasts with (lc), whichconstitutes a semantic anomaly. This, in effect, is the distinction betweenthe linguistic and the non-linguistic aspects of the lexical representation ofa verb. The anomaly-or the “oddness”-of “John buried the guitar”cannot be part of the linguistic specification of the semantics of the lexicalitems involved. In fact, it is not something that is likely to be pre-stored atall. The pragmatic oddness of burying a guitar is something that we have toinfer, given our knowledge of the world, and given what we know aboutguitars, the likely effects of burying them, and so on.The first question we ask here, therefore, is whether such pragmaticviolations affect the on-line response to the monitoring target-that is,whether responses are slower than in the normal condition. This will onlybe the case if the response in the normal condition reflects, at least in part,the pragmatic normality of the actions or events involved. And this, inturn, can only hold if the lexical representations associated with the verbhave already begun to have consequences for an interpretation in a mentalmodel of the discourse. This requires, we should remember, not only theconstruction of a linguistic argument frame, realising the structural constraintson the relationship between the verb and its potential arguments,but also the projection of the thematic consequences of these relations intothe non-linguistic domain.The second type of violation, in (lc), can plausibly (though not necessarily)be treated as a violation of the linguistically specifiable properties ofthe lexical representations associated with a given verb-a violation, inChomskian terms, of the selection restrictions on the semantic propertiesof the items that can fill the argument slots made available by the verb’ssubcategorisation properties. It is plausible that the linguistic specificationof drinking (that it involves liquids) and of guitars (that they are solidobjects) is sufficient to make guitar anomalous following drink withouthaving to invoke knowledge or operations outside the linguistic domain.That is why we refer to this as a semantic rather than a pragmatic violation.Our second question, therefore, is twofold: Are semantically anomaloustargets responded to more slowly than normal targets, and are they alsoresponded to more slowly than pragmatically anomalous targets? If thefirst of these holds, then this reflects the timing with which the semanticrestrictions associated with argument frames start to become available, andbegin to be integrated with the semantic properties of potential fillers forthese slots. If the second of these holds, and semantic anomalies are moredisruptive than pragmatic anomalies, then this tells us something about therelative significance, for the on-line analysis process, of these two types ofdisruptions (we will return later in the paper to the question of how farsemantic and pragmatic anomalies can be kept qualitatively distinct).

LEXICAL REPRESENTATIONS IN SPOKEN LANGUAGE COMPREHENSION 7Finally, we can also predict that if the original difference between NormalProse and Anomalous Prose in Marslen-Wilson and Tyler (1980) wasindeed equivalent, as originally intended, to violations of semantic cooccurrencerestrictions, then the reaction time difference between Normaland Semantic targets here should be of the order of 55-60 msec.The third type of sentence, exemplified in (Id), differs from the othertwo in violating the subcategorisation frame associated with a given verb.A verb-form like sleep is subcategorised as an intransitive verb, and has nosubcategorised argument slot into which a nounphrase like guitar can fit.This categorial violation should be highly disruptive of monitoring performance.The target nounphrase will be heard in the context of a verbargument frame to which it cannot be attached. This means that thelistener cannot construct the appropriately configured structural object forprojection on to any other domain of interpretation. The collocation ofsleep and guitar simply has no interpretation, either semantically or pragmatically.This means that the monitoring decision that the word guitar ispresent can be based only on the bottom-up sensory input. If the distinctionhere between semantic and categorial anomalies is equivalent to thedistinction in the earlier study (Marslen-Wilson & Tyler, 1980) betweenAnomalous and Scrambled Prose, then reaction times in the categorialcondition should be on the order of 25-30 msec slower than in the semanticcondition, and about 90-100 msec slower than targets in the normalsentences.In a final manipulation, over and above the four primary contrastsalready described, we will investigate further the role of the discoursemodel in these lexical activation processes, by comparing responses tonormal targets (as in la) when the test sentence coheres normally to itspreceding discourse context as opposed to cases where there is no coherentrelationship. When the utterance containing the target word cannot bereadily mapped onto the scenario established in the preceding sentence,does this slow down responses? These are not, strictly speaking, localviolations, but we include them here in order to maintain comparabilitywith the earlier study, where we did find effects of the presence or absenceof a discourse context for target-positions early in the test sentence.MaterialsMETHODThe test materials consisted of 40 sets of sentence pairs. Each pair withinthese sets was made up of a lead-in sentence and a continuation sentence.The monitoring target always occurred in the second sentence, placed inobject position, immediately following the verb. Four kinds of continua-

8 MARSLEN-WILSON, BROWN, TYLERtion sentences were constructed by varying the relationship between theverb and the target noun in the embedded verb-noun sequence. Withineach item set the target noun was kept constant. This gave the followingfour anomaly conditions, together with one extra condition constructed byvarying the congruence of the lead-in sentence with the continuationsentence:1. Normal: Here the verb and the target noun form a natural sequence instandard Dutch subject-verb-object order (e.g. “The boy held theguitar”, where “guitar” is the target).2. Pragmatic: Here the verb in combination with the target noun forms apossible, but pragmatically implausible real-world situation (e .g. “Theboy buried the guitar”).3. Semantic: Here the semantic properties of the verb are incompatiblewith the semantic properties of the noun (e.g. “The boy drank theguitar”).4. Categorial: Here the verb forms a syntactically illegal combination withthe target noun. This was done by choosing intransitive verbs that couldnot be followed by a noun in direct object position (e.g. “The boy sleptthe guitar”).5. Discourse Congruence: A fifth manipulation applied to the Normalcontinuation sentences only. Here the lead-in sentence was varied togive either a natural or an unnatural discourse linkage with the continuationsentence.The following example shows one complete stimulus set, with thesubjects hearing either lead-in sentence (1) plus one of the four lead-insentences (a)-(d), or lead-in sentence (2) plus continuation sentence (a).Note that the original materials were in Dutch (the set given here is forillustration only):1. The nurses walk to their work each morning.2. Christmas falls on a Friday this year.a. They puss the beach on their way to the hospital.b. They measure the beach on their way to the hospital.c. They chew the beach on their way to the hospital.d. They yawn the beach on their way to the hospital.Pre-testsThe 40 sets of test materials were selected from a larger initial pool of 60sets, on the basis of three types of pre-tests.1. A predictability pre-test was run to ensure that the target nouns were

LEXICAL REPRESENTATIONS IN SPOKEN LANGUAGE COMPREHENSION 9not easily predictable given the discourse sentences and the various pretargetverbs. Three sentence combinations were tested: the normal condition(la above), the discourse incongruence condition (2a above), and thepragmatic anomaly condition (lb above). A total of 36 subjects were testedin three groups, using a written cloze procedure, with items rotated acrossgroups such that no subject had to predict the same target word twice. Thesubjects’ responses were scored on a 4-point scale. A scale value of 4corresponded to responses unrelated to the target noun, 3 to responsesrelated to the target noun, 2 to responses that were synonyms of the targetnoun, and 1 to cases where subjects responded with the actual target noun.The overall mean predictability ratings for the 40 test sentences was verylow for all three conditions-3.86 for normal, 3.91 for discourse incongruence,and 3.87 for pragmatic anomaly.2. A subcategorisation pre-test was run to ensure that the intransitiveverbs used in the verb-noun sequences did indeed exclude the possibilityof inserting nouns in direct object position immediately following the verb.The same 60 sentence pairs from the predictability pre-test were used,along with 30 fillers. The fillers contained sentences that did allow for theuse of direct objects. Twelve subjects read the neutral lead-in sentencefollowed by the (d)-continuation sentence, up to and including the intransitiveverb, and were asked to write down a continuation. They were alsoasked to indicate on a 7-point scale whether or not they thought thesentence pair-including their own continuation-was a natural one, bothwith respect to grammar and meaning. Responses were checked for theoccurrence of direct objects. If direct objects were given along with anatural rating, the sentence pair was removed from the stimulus pool. Veryfew prospective stimuli failed this test.3. In a naturalness pre-test, we tested the naturalness of the discourselinkage between the lead-in and the continuation sentences (as well as theunnaturalness of the linkage for the discourse incongruence pairs). Twogroups of 12 subjects were given 60 typed sentence pairs containing anequal number of normal (la) and incongruent (2a) pairs. Subjects wererequired to give naturalness judgements on a 7-point scale, with value 1representing “very unnatural”, and value 7 representing “very natural”.The mean rating for the final stimulus set was 6.35 for the normal pairs and1.31 for the incongruent pairs.Filler MaterialsIn order to distract subjects from the specific construction of the testsentence pairs, 80 filler items were constructed. These also consisted of alead-in sentence and a continuation sentence. Three kinds of variationswere made on the filler items:

1 O MARSLEN-WILSON, BROWN, TYLER1. Target position: The standard construction of the test continuationsentences is subject-verb-target noun. In order to prevent subjects fromnoticing the uniform position of the target noun, and the fact that all testtargets were nouns, we vaned both the word type and the word position ofthe filler target. Filler targets could occur either in the lead-in sentence, inthe beginning of the continuation sentence, in the same position as the testsentences, or late in the continuation sentence.2. Anomaly of target: In the test sentence pairs, three out of the fivecombinations have an anomalous target noun (lb, lc, Id). This could leadsubjects to believe that every anomaly is a target. To prevent this, most ofthe fillers contained anomalous words which were not the monitoringtargets (as well as containing non-anomalous targets).3. Task: The experimental task for all test stimuli is identical monitoring.For the filler sentences two further tasks were used, namely rhymemonitoring and category monitoring, in order to make the test situationcomparable to that used by Marslen-Wilson and Tyler (1980), and in orderto reduce the possibility that the subjects would develop special listeningstrategies in carrying out the identical monitoring response. A total of 20fillers were run using identical monitoring, 30 fillers using rhyme monitoring,and 30 fillers using category monitoring. This task distribution broughtthe total number of items presented using identical monitoring up to 60 (40test and 20 filler items), balanced by 60 filler items on rhyme and categorymonitoring.Practice SentencesA total of 25 practice sentences were also constructed. These practice itemsreflected all conditions and variations used in the test and filler sentences.A further 10 filler items were constructed to serve as start-up sentences tostabilise subject responses following the breaks after the practice sessionand half-way through the test sequence.RecordingThe materials were recorded in a soundproofed booth, using a Revox A700reel-to-reel tape-recorder, by a female speaker who was naive with respectto the purpose and background of the experiment. All sentences were readwith a normal intonation pattern, without contrastive stress on either thetarget words or their preceding verbs.

LEXICAL REPRESENTATIONS IN SPOKEN LANGUAGE COMPREHENSION 1 1Design and ProcedureA total of 40 test sentence pairs were used under five conditions, along with80 filler items, 10 start-up fillers, and 25 practice items. The 40 testsentence pairs were rotated by conditions over five experimental versions,with each experimental version having eight items in each condition. Eachtarget word occurred only once in each version, and each subject heard anequal number of targets in each of the five test conditions. The fiveexperimental versions were made by cross-recording from a master tape.ProcedureThe 40 subjects were tested in groups of 4. They were seated facing aprojection screen on to which was projected the slide specifying themonitoring task for the next trial, and the appropriate cue word. Only the“identical monitoring” task was used for the test stimuli, so that the cueword was always the target itself. Thus, if the target was “beach”, thesubject would see the words “IDENTICAL” and “BEACH’. All the testmaterials were presented over closed-ear headphones.Each trial began with a warning tone that also triggered the projection ofthe relevant slide. The sentence pair followed 3 sec later. The subjectspressed a response button as soon as they detected the monitoring target.They were instructed to respond as quickly as possible, to avoid guessing inadvance what the target word could be, and not to try to correct anymistakes they heard in the sentences.A complete experimental session, including instructions and practice,lasted approximately 1 hour. Subjects were given a short break half-waythrough the test sequence.SubjectsA total of 40 paid student subjects participated in the experiment, of whom20 were male and 20 female, and all were native speakers of Dutch.RESULTS AND DISCUSSIONEight subjects were run in each condition. There were 28 missing valuesdue to machine and subject error. A further 26 extreme values were set tozero because they were well outside the range of the other data points inthe condition in which they occurred. These 54 missing and extreme values(totalling 3.4% of the data) were replaced‘by the mean for the item in thecondition in which they occurred. Table 1 gives the main results for thecorrected data.

~~12 MARSLEN-WILSON, BROWN, TYLERTABLE 1Item Means by Conditions (msec)Normal Incongruous Pragmatic Semantic Categorial241 235 268 291 320Two separate one-way analyses of variance were computed, on subjectsand on items. The main effect of conditions was highly significant (minF’(4,46)=16.678, P

LEXICAL REPRESENTATIONS IN SPOKEN LANGUAGE COMPREHENSION 13pended on non-linguistic aspects of interpretation and anomalies that didnot. A pragmatic anomaly-someone, for example, burying a guitarcouldonly exist as an anomaly on the basis of some form of inferentialcomputation, involving not only the lexical semantics of terms like guitarand bury, but also more general knowledge of the world. A semanticanomaly-someone, for example, drinking a guitar-could be instantiatedsimply as a function of the stored linguistic properties of terms like guitarand drink, without involving non-linguistic systems of knowledge at all.The empirical question we asked, then, was whether pragmatic anomaliesdid significantly affect the on-line response to the monitoring target, orwhether the effect was restricted only to the semantic anomalies. Theresults show clear effects of each type of anomaly, with pragmatic anomalies(at 268 msec) and semantic anomalies (at 291 msec) both significantlyslower than responses to normal targets (241 msec). The effect for pragmaticanomalies means that the lexical representations associated with theverb must be being interpreted on-line, not only within a linguistic systemof analysis, but also with respect to the listener’s mental model of theutterance-that is, with respect to a level of representation which incorporatesinferential processes operating on non-linguistic as well as linguisticknowledge. This in turn is consistent with a view of lexical processingwhich stresses the immediacy with which the thematic consequences of agiven lexical argument frame can be projected on to non-linguistic domainsof interpretation.It is also worth pointing out that these conclusions hold even if one takesthe view of monitoring performance argued for by Tanenhaus, Carlson,and Seidenberg (1985), in which the sorts of context effects described inMarslen-Wilson and Tyler (1980) are re-analysed as a form of sophisticatedguessing. Even if this were the way that context effects are mediated, thedifference between normal and pragmatic conditions still requires theon-line computation of the pragmatic plausibility that the target will appearas the object of the verb in the two conditions. The listener’s estimate, onthe guessing account, of the likelihood that guitar will be the object ofbury, can only differ from the subjective likelihood following curry on thebasis of some form of inference process, dependent on the properties of theactual verb being heard. These inferences would require the projection ofverb properties onto a non-linguistic domain, so that the pattern ofresponses still remains as evidence for the immediacy with which thisprojection takes place.Turning to the effects for semantic anomalies, these confirm, first of all,that semantic aspects of lexical representations are also immediatelyactivated in on-line processing. Secondly, the comparability between thesize of the effect for these stimuli (at 50 msec) and the effect for thecontrast between Normal Prose and Anomalous Prose in Marslen-Wilson

14 MARSLEN-WILSON, BROWN, TYLER& Tyler (1980)-where responses to Anomalous Prose were 58 msecslower-suggests that the effects in this earlier study were also due toviolations of semantic restrictions on co-occurrence relations. Thirdly,there is also a tendency for the semantic anomalies to have a moredisruptive effect on responses than the pragmatic anomalies. Although themean difference of 22 msec is not significant using a min F’ based statistic,it is significant on the Neuman-Keuls using error terms derived either fromthe subject or the item analyses alone.This indicates that an anomaly based on violations of semantic selectiona1restrictions-as classically defined-will tend to be more disruptive thanan anomaly based solely on violations of contingent plausibility. This doesnot, however, allow us to conclude that there is a qualitative differencebetween the types of representation involved-that semantic anomaliesinvolve only disruptions within a strictly linguistic system of semanticrepresentation and analysis, whereas pragmatic anomalies necessarily takeus outside this domain. This is because the semantic anomalies are anomalousnot just with respect to putatively language internal co-occurrencerestrictions, but also with respect to considerations of real-word plausibility.If it is unusual and implausible to bury a guitar, it is far more unusual todrink one. And while one can easily construct a context in which burying aguitar is perfectly plausible, it is far more difficult to construct a context inwhich the act of drinking a guitar becomes possible or plausible. Theadditional increase in response time for the semantic anomalies may reflectthese additional difficulties in the domain of pragmatic inference, ratherthan difficulties at some other level of the system. The present researchdoes not allow us to discriminate between these possibilities.The final aspect of the results concerns the categorial violations, wherethe target noun appears in a position in the sentence that violates thesubcategorisation frame associated with the preceding verb. This has thestrongest effect on response time, with responses being significantly slowerthan to the semantic anomalies, and some 79 msec slower than responsesto the same targets in normal contexts. We predicted these strong effects inthe Introduction, on the argument that this type of violation of subcategorisationconstraints means that the listener is presented with a structurallyuninterpretable combination of lexical items. The target nounphrase isheard in the context of a verb subcategorisation frame which contains noargument slot into which it can be attached. This means that the listenercannot construct the appropriate structural formation for projection on tonormal domains of interpretation. For the purposes of the monitoring task,then, the listener can only have access to the monitoring target by treatingit as a word heard in isolation.This interpretation is supported by the similarity in the size of the effectof categorial violations here to the difference (of 85 msec) between the

LEXICAL REPRESENTATIONS IN SPOKEN LANGUAGE COMPREHENSION 15Normal Prose contexts and the Scrambled Prose contexts in the Marslen-Wilson and Tyler (1980) study. In Scrambled Prose the subjects heardtargets in unstructured word-lists, constructed by scrambling the order ofAnomalous Prose sentences. Evidently, subcategorial violations have acomparable effect, by depriving the target nounphrase of any kind ofsystematic sentential context of interpretation.The goals of this study, in conclusion, were to gather some basicinformation about the timing with which different aspects of lexical representationscould begin to have consequences for the processing of subsequentitems in a spoken utterance. The results, looking at the effects ofviolating the pragmatic, semantic, and structural restrictions deriving fromthe argument frames associated with transitive and intransitive verbs, showimmediate effects both at strictly linguistic levels of interpretation (asclassically defined) and at levels of interpretation and inference relevant tothe listener’s construction of a mental model of the current utterance.These are results that are consistent with approaches to language comprehensionwhich stress its continuous and incremental nature (e.g. Marslen-Wilson & Tyler, 1987; Steedman, 1988; Tyler, 1988), and which emphasisethe role of lexically derived thematic roles in constructing these higherlevelinterpretations (e.g. Tanenhaus & Carlson, 1988).Manuscript received May 1987Revised manuscript received February 1988REFERENCESBresnan, J. (1982). The menial representation of grammatical relations. Cambridge, Mass. :MIT Press.Carlson, G. & Tanenhaus, M. (1987). Thematic roles and language comprehension. In W.Wilkins (Ed.), Thematic relations. London and San Diego: Academic Press.Chomsky, N. (1965). Aspects of the theory ofsyntux. Cambridge, Mass.: MITPress.Chomsky, N. (1981). Lectures on Government and Binding. Dordrecht: Foris.Clark, H. (1973). The language-as-fixed-effect fallacy: A critique of language statistics inpsychological research. Journal of Verbal Learning and Verbal Behavior, 12,335-359.Clifton, C., Frazier, L., & Connine, C. (1984). Lexical expectations in sentence comprehension.Journal of Memory and Language, 23,696-708.Cowart, W. (1982). Autonomy and interaction in the language processing system: A reply toMarslen-Wilson and Tyler. Cognition, 12,109-117.Downey, R. G. & Hakes, D. T. (1968). Some psychological effects of violating linguisticrules. Journal of Verbal Learning and Verbal Behavior, 7,158-161.Fodor, J. A,, Garrett, M., & Bever, T. G. (1968). Some syntactic determinants of sententialcomplexity, 11: Verb structure. Perception and Psychophysics, 3,453-461.Ford, M., Bresnan, J. W., & Kaplan, R. M. (1982). A competence-based theoryof syntacticclosure. In J. W. Bresnan (Ed.), The mental representation of grammatical relations.Cambridge, Mass.: MIT Press.

16 MARSLEN-WILSON, BROWN, TYLERHakes, D. T. (1971). Does verb structure affect sentence comprehension? Perception andPsychophysics, 10,229-232.Jackendoff, R. (1972). Semantic interpretation in generative grammar. Cambridge, Mass. :MIT Press.Marslen-Wilson, W. D. & Tyler, L. K. (1980). The temporal structure of spoken languageunderstanding. Cognition, 8,l-71.Marslen-Wilson, W. D. & Tyler, L. K. (1981). Central processes in speech understanding.Philosophical Transactions of the Royal Society, 8295,317-332.Marslen-Wilson, W. D. & Tyler, L. K. (1983). Reply to Cowart. Cognition, 15,227-235.Marslen-Wilson, W. D. & Tyler, L. K. (1987). Against modularity. In J. Garfield (Ed.),Modularity in knowledge representation and natural-language understanding. Cambridge,Mass.: Bradford Books.Mitchell, D. (1987). Lexical guidance in human parsing: Locus and processing characteristics.In M. Coltheart (Ed.), The psychology of reading. London: Lawrence ErlbaumAssociates Ltd.Mitchell, D. & Holmes, V. (1985). The role of specific information about the verb in parsingsentences with local structural ambiguity. Journal of Memory and Language, 24,542-559.Moore, T. E. (1972). Speeded recognition of ungrammaticality. Journal of Verbal Learningand Verbal Behavior, 11,550-560.Steedman, M. (1988). Grammar, interpretation, and processing from the lexicon. In W. D.Marslen-Wilson (Ed.), Lexical representation andprocess. Cambridge, Mass.: MIT Press.Tanenhaus, M. & Carlson, G. (1988). Lexical structure and language comprehension. InW. D. Marslen-Wilson (Ed.), Lexical representation andprocess. Cambridge, Mass.: MITPress.Tanenhaus, M., Carlson, G., & Seidenberg, M. (1985). Do listeners compute linguisticrepresentations? In D. Dowty, L. Karttunen, & A. Zwicky (Eds), Natural languageparsing. Cambridge: Cambridge University Press.Tanenhaus, M., Stowe, L., & Carlson, G. (1985). The interaction of lexical expectationand pragmatics in parsing filler-gap constructions. In Proceedings of the Seventh AnnualCognitive Science Society Meetings. Hillsdale, N.J.: Lawrence Erlbaum Associates Inc.Tyler, L. K. (1988). The role of lexical representation in language comprehension. In W. D.Marslen-Wilson (Ed,), Lexical representation andprocess. Cambridge, Mass.: MIT Press.Van Riemsdijk, H. & Williams, E. (1986). Zntroduction to the theory of grammar. Cambridge,Mass.: MIT Press.Wasow, T. (1985). Postscript. In P. Sells (Ed.), Lectures on contemporary syntactic theories.Stanford: CSLI.