Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Spoken word recognition as a function of lexical knowledge and language proficiency level in adult ESL… Barbour, Ross Patrick 1995

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


831-ubc_1995-982500.pdf [ 2.21MB ]
JSON: 831-1.0088823.json
JSON-LD: 831-1.0088823-ld.json
RDF/XML (Pretty): 831-1.0088823-rdf.xml
RDF/JSON: 831-1.0088823-rdf.json
Turtle: 831-1.0088823-turtle.txt
N-Triples: 831-1.0088823-rdf-ntriples.txt
Original Record: 831-1.0088823-source.json
Full Text

Full Text

SPOKEN WORD RECOGNiTIONAS A FUNCTION OF LEXICAL KNOWLEDGEAND LANGUAGE PROFICIENCY LEVELIN ADULT ESL LEARNERSbyRoss BarbourB.A., The University of British Columbia, 1968M.A., The University of British Columbia, 1983A THESIS SUBMiTTED IN PARTIAL FULFILMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinTHE FACULTY OF GRADUATE STUDIES(Department of Educational Psychology and Special Education)We accept this thesis as conformingto the required standardTHE UNiVERSiTY OF BRITISH COLUMBIAMarch 1995© Ross Barbou 1995In presenting this thesis in partial fulfilment of the requirements for an advanceddegree at the University of British Columbia, I agree that the Library shall make itfreely available for reference and study. I further agree that permission for extensivecopying of this thesis for scholarly purposes may be granted by the head of mydepartment or by his or her representatives. It is understood that copying orpublication of this thesis for financial gain shall not be allowed without my writtenpermission.(Signature)Department of E/3 tiC/- TI 0 /v4L /‘5 Y Ci g,c/7 S/’,C/P4 8/C,77O/’/The University of British ColumbiaVancouver, CanadaDate /,,. /2, /q 9/DE-6 (2/88)AbstractThis study assesses the usefulness of Marsien-Wilson’s (1989, 1987; MarsienWilson & Welsh, 1978) cohort model of spoken (first language) word recognition as amethod of explaining the high-speed, on-line processes involved in recognizing spokenwords while listening to a second language. Two important assumptions of the modelare: 1) syntactic and semantic properties of mental lexical entries can function to-facilitate spoken word recognition and 2) spoken word recognition is a function of thefrequency of exposure to words in the general language environment. Theseassumptions were tested in three functionally defined levels of language proficiency:Native Speakers of English, Fluent Users of ESL, and Advanced learners of ESL. Theirperformance was compared on a reading doze test and a spoken-word recognition taskin which there were five different levels of contextual richness prior to a target word,and two levels of word frequency.The doze results indicated that the three groups differed in their general Englishproficiency. Congruent with the cohort model, there was a significant overall effect ofsentence context and word frequency on recognition latency. Despite the difference indoze scores and immersion experience between the two ESL groups, there were noreliable differences in their recognition latencies or latency profiles across sentencecontexts or across word frequency. There was an interaction of ESL group, wordfrequency, and sentence context. This may be due to a reorganization of rules usedduring processing or a restructuring of lexical knowledge. There was also aninteresting non-linear relationship between recognition latency and language immersiontime. Spoken word recognition speed decreased in the early immersion experience, andUthen increased with further exposure.There was a significant difference in overall mean recognition latency betweenthe Native and the ESL speakers, with the ESL subjects responding on average 98 msecslower than the Native Speakers. However, there were no significant differences in theway Native Speakers and the ESL subjects used sentence context. In contrast with thecomparison across the sentential contexts, there was a significant difference in therecognition profiles of the Native English speakers and the ESL subjects across wordfrequency.UiTable of ContentsAbstractTable of ContentsList of TablesList of Figures.AcknowledgementChapter I. Research ProblemsChapter II. Background of Research ProblemsA. Speech Processing and the Lexicon1. Speech Processing2. The Lexicon11ivviviiviii12. Primary Research Paradigm: Discourse and Sentential ContextB. The Cohort Model and Research Paradigm1. The Cohort ModelChapter ifi. Extending the Model: Hypotheses and ExpectationsA. The Roles of Semantic and Syntactic Knowledge in the WordRecognition Process, Materials and Learner VariablesB. Word Familiarity and Word KnowledgeC. Second Language Proficiency1. Access2. Integration3. Lexicon4. Summary of Hypotheses and Predictable OutcomesChapter IV. MethodA. Subjects and DesignB. Language Materials1. Test sentences2. Filler material3. Example and Warm-up Items.C. Production of Language Stimuli.D. ProcedureE. Data Screening and Preparation ..525253535456565759Chapter V. ResultsA. Review: The Components of the Extended ModelB. Analyses111616161717213030333536394148626264iv1. Analysis of Response Latency.642. Supplementary Analyses 77Chapter VI. Discussion 82A. Summary of Results 82B. Discussion 841. Developmental differences between Users and Learners 852. Developmental differences between ESL and Native speakers 883. Processing Differences between the ESL and Native Speakers 90C. Factors Affecting Generalizability 96D. Implications for ESL Education and Directions for Future Research 98E. Condusion 103References 105Appendices 111Appendix A: Targets, lead-in sentences, and test sentences for EXACTmonitoring task 111Appendix B: Category based filler items 121Appendix C: Rhyming based fillers 124Appendix D: Exact Fillers 127Appendix E: Example items 129Appendix F: Warm-up items 130Appendix G: 50-Item Cloze Test 132VList of TablesTable 1Types of Context Derivable by Manipulating Verb Frame Features 26Table 2Research Design 52Table 3Examples of Test Sentences: Target word bubbles54Table 4Example Filler Material (target word underlined)55Table 5Cloze Test Score Means and Standard Deviations by Language Proficiency Group62Table 6Mean Response Latencies of Three Proficiency Levels in High and Low FrequencyWord Conditions 64Table 7Pearson Correlation Coefficients between Response Latency and Cloze Scores 79viList of FiguresFigure 1. Distribution of Cloze Scores by Proficiency Level 63Figure 2. Mean Response Latencies of Language Proficiency Levels 65Figure 3. Mean Response Latencies to High and Low Frequency Words 66Figure 4. Mean Response Latencies of All Subjects across Sentence Contexts 68Figure 5. Mean Response Latencies of Language Proficiency Levels acrossSentence Contexts 70Figure 6. Mean Response Latencies of Language Proficiency Levels to High andLow Frequency Words 72Figure 7. Mean Response Latencies of All Subjects to High and Low FrequencyWords across Sentence Contexts 73Figure 8. Mean Response Latencies of Three Language Proficiency Levels toHigh and Low Frequency Words across Sentence Contexts 74Figure 9. Combined ESL and Native English Speakers Response Latencies toHigh & Low Frequency Words across Sentence Contexts 76Figure 10. Mean Response Latencies of ESL Learners & Users to High and LowFrequency Words across Sentence Contexts 76Figure 11. Mean Response Latencies of ESL Learners & Users to High and LowFrequency Words across Sentence Contexts (Syntactic and RandomCombined) 77Figure 12. Distance Weighted Least Squares Smoothing for Mean Latency Time,Separately for Learners and Users, and Combined 81viiAcknowledgementThis project would not have been possible without the help, encouragement andparticipation of numerous people. In particular, I would like to thank:Atsuko, for all her help and support during the six long years that this enterprisetook. I know I could not, and would not have done it without her beside me all theway.Ian, who grew up while this was going on and became a young man before I hadtime to enjoy his childhood. I particularly want to thank him for the use of hiscomputer.Dr. Seong-Soo Lee, my steadfast advisor, who actually believed I could do it andbecause of that, made me do it well.The members of my committee, Dr. Ron Jarman and Dr. Rita Watson. Theirgermane and penetrating comments and suggestions ensured that I did not lose thefocus of the project.The students at UBC and KEC who unselfishly volunteered to take part in the study.The gang of seven, cohorts with whom I studied for six months, creating aneducational psychology course that surpassed anything any of us had experienced.Monique, my friend. We complemented each other and built a mutual supportsystem based on running, and blowing bubbles.Dr. & Mrs. Yamamoto, who sent care-packages from Japan and regualarly asked:“Rousu-san genki ka ne?’And finally, my good friend Janiie Patrie, bless his soul, who, in the face of thecoming darkness was the only one who really dared to ask, “What are you going to dowith PhD, anyways?”viiichapter i. Research ProblemsIn order to comprehend a spoken language, it is essential to recognize, quickly,correctly and automatically, which word is being spoken. Connected speech moves inreal time and any impediment to recognizing the individual words as they flow pastwill surely disrupt any attempt to create meaning from the spoken sentences. This is astrue for second-language comprehension as it is for first language. Although somework has been done in the general area of comprehension during spoken, second-language processing (McDonald, 198Th; McDonald & Heilenman, 1991; Sasaki, 1991),not much attention has been paid to the high-speed, on-line processes and knowledgeinvolved in spoken-word recognition during speech comprehension in a secondlanguage (see Hayashi, 1991).In research on the processes involved in speaking and comprehending a secondlanguage (McLaughlin, Rossman, & McLeod, 1983; Long & Sato, 1985; Sasaki, 1991), acommon approach has been the application of models or theories of first languageprocessing and acquisition to the area of second language (Bates & MacWhiriney,1981;Bialystok, 1990; Flynn, 1987; Flynn, & O’Neil 1988; Gass, 1984, 1987; Harrington, 1987;MacWhinney, 1987a). The present research follows this approach in order to examinethe fundamental processes of spoken word recognition in a second language. Inparticular, it evaluates the extension of Marsien-Wilson’s (1989, 1987; Marsien-Wilson &Welsh, 1978) cohort model of spoken (first language) word recognition to a secondlanguage context. It does so by (a) investigating the effects of three types of lexicalknowledge, syntactic, semantic and pragmatic, on the speed of spoken-word recognitionby second language users, (b) comparing, as a group, their word recognition-time1profile with that of native speakers of the target language, and (c) examining how twodifferent levels of second language proficiency interact with different types of lexicalknowledge to affect word-recognition speed.The research was motivated by the belief that having a theoretical model of thefundamental, high-speed processes involved in spoken, second-language word-recognition would provide a useful foundation for further theoretical and practicalresearch in the areas of second language learning and instruction. The cohort model ofspoken word recognition in particular was chosen because of the importance it placeson the role of lexical knowledge during the recognition process and because of thebelief that growth of lexical knowledge is an integral and essential part of growth insecond language proficiency.In the present study, lexical knowledge refers to information attached toindividual entries in an internal mental dictionary, information that will allow for quickanalysis of syntactic and semantic relationships among words that have been heard. AsHaegeman (1991) notes, in modern Chomskyan linguistics,We postulate that speakers of a language are equipped with an internal‘dictionary’, which we shall refer to as the mental lexicon, or lexicon, whichcontains all the information they have internalized concerning the words of theirlanguage. (p. 29).Importantly, she also notes a further assumption that “... the lexicon of a language islearnt [italics added] by each native speaker. The speaker learns [italics added] thewords of the language...” (p. 29). It seems reasonable to assume that second languagelearners also develop, through learning, a lexicon that is appropriate to the newlanguage and furthermore, that the knowledge stored in this lexicon plays somefunctional role during the processing of the language when it is spoken by others.2Given this assumption, to what extent is that role parallel to the role played by lexicalknowledge in first language processing?There are various models of spoken language processing for the first languagecontext (for overviews see Forster, 1989 and Klatt, 1989). One is Marsien-Wilson’s(1989, 1987; Marsien-Wilson & Welsh, 1978) cohort model of spoken word recognition.This model assumes a significant facilitative role for lexical knowledge (beyondphonological encoding) in the recognition process and thereby allows investigation intothe structure of lexical knowledge. In first language speech-processing, availableresearch findings indicate that sentential context has an influence on the speed withwhich words are recognized (Grosjean, 1980; Marsien-Wilson, 1985; Tyler & Wessels,1983). Because stored lexical knowledge can be linked to prior sentential context, it canbe used in a manner that will facilitate the recognition process as long as those sentencecontexts are syntactically, semantically and pragmatically normal. However, sentencecontexts that have syntactic disruptions or pragmatic or semantic anomalies in themeffectively block the use of certain aspects of lexical knowledge and consequently wordrecognition is slower in these contexts.Interestingly, these different contextual disruptions have different effects onword recognition time, suggesting different roles for different types of lexicalknowledge. In experiments by Marslen-Wilson, Brown & Tyler, (1988; Tyler, 1988,1985), word-recognition time under normal sentential context was compared torecognition time in contexts such as those in sentences 1, 2 and 3, (adapted from Tyler,1985) in which the target word is guitar. In sentence 1, the action of burying the guitaris unusual, although the sentence itself is otherwise correct. In their experiments, this3type of context has been labelled pragmatically anomalous. In sentence 2, there is asemantic anomaly in the relationship between the verb drink, which anticipates a direct1. The young man buried the guitar and2. The young man drank the guitar and3. The young man slept the guitar andobject that is ‘fluid’, and the actual direct object guitar, which does not meet thiscriterion. In sentence 3, there is a syntactic violation because sleep is an intransitive verband yet it appears to be taking a direct object. Consideration of sentence 3 also revealsthat there can be no structurally dictated semantic relationship between the verb andthe target noun that follows. Thus, in sentences which are disrupted in this mannerthere is neither syntactic nor semantic information which can be used to aid inrecognition of the target word.For speakers of Dutch as a first language, it was found that pragmatic anomaliesslowed down recognition time on average 28 ms compared to the normal context, whilethe delay due to semantic anomalies was 50 ms and that due to the syntactic disruptionwith its concomitant lack of semantic relation, 79 ms (Marslen-Wilson, Brown & Tyler,1988). Tyler (1985, 1988) found very similar results for her English speaking controls.That is to say, when recognition times to target words in these four (normal,pragmatically, semantically, or syntactically anomalous) contexts are compared acrossthe different experiments, there appears to be a consistent, increasing recognition-timeprofile which characterizes the differing contributions to the word recognition processof these different types of lexical knowledge. These studies were conducted in theframework of first language processing, but they can form the basis of a number of4questions that will enable a direct comparison of the role played by lexical knowledgein first and second language processing, and at the same time address the central aimof the present research: to assess an extension of the cohort model of spoken wordrecognition to the second language context.If it is assumed that second language learners do learn lexical knowledge relatedto the words of the new language, then certain questions arise naturally as aconsequence. Do second language learners become capable of using the new languagein a high-speed, on-line manner that is at all comparable to native speakers? In otherwords, would high-functioning second language users show a recognition-time profilesimilar to that of native speakers? More generally, does lexical knowledge, other thanthe obviously requisite phonological knowledge, enter into the word recognitionprocess at all during the process of comprehending a second language? Or, is it thecase that, in comprehending a second language, lexical information can only beprojected onto the developing mental representation of a utterance after the word hasbeen recognized? Finally, if lexical information does play a role during the wordrecognition phase of spoken second language processing, what are some specific aspectsof lexical knowledge that are instrumental in this process? These questions are, inessence, concerned with how close a second-language processing system comes toreplicating the functionality of a first language processing system. Some initialinformation on this can be brought to light by asking whether, under the particularconditions found in the above-cited research, the word-recognition profiles of firstlanguage users and fluent second language users are the same.Questions regarding how development of second language proficiency is related5to word recognition speed also arise. Second language users will develop a functionallexicon in the second language but the lexical knowledge of a fluent or expert user of asecond language and that of a learner or novice user of the language will be different,at least because the novice will not have as many or as complete lexical entries as theexpert, and perhaps for other reasons as well. For example, it is not obvious whether,over the course of learning a new language, different types of lexical knowledge willdevelop at the same rate. The adult learner may be able to make use of semantic orpragmatic conceptual knowledge acquired through the first language in order to fill outword entries in the developing second language lexicon. Bialystok (1990) argues thatadults in general need only to structure and analyze the linguistic domain as theybuild a new lexicon to serve the expressive functions of a conceptual structure alreadyconstructed through another language’ (p. 133).As an illustration, consider the target word and the verbs in sentences 2 and 3on page 3. It may be that learners quickly fill out some semantic features of the lexicalentries for guitar (e.g. that guitars are solid), drink (e.g. that drink is associated withliquids), and sleep (e.g. that inanimate things do not sleep), but will take longer to fixsyntactic information regarding the transitivity of drink or sleep. In this case, recognitiontimes to targets in sentences like 2 and 3 may be very similar, since, unlike in the caseof native speakers, neither verb would supply syntactic constraints on the followingwords. No doubt a number of other possibilities could also be proposed. However, ingeneral, the question is whether in second languages users, differences in the extent andpossibly the structure of lexical knowledge will result in different word-recognition-time profiles under such contexts as above. In other words, do proficiency-based lexical6differences in a second language have differing effects on the speed with which a wordis recognized, or wifi the recognition-time proffles of, say, fluent users and advancedlearners be parallel, though perhaps separate? If they are not parallel, will thedifferences in recognition time under different lexical knowledge constraints increase ordecrease as a person moves from novice to expert user of the second language? Thesequestions of an interactional nature define the final aim of the research: to examine howdiffering levels of language proficiency and different types of lexical knowledge interactto affect word-recognition speed.Although the above questions are interesting in and of themselves, the answersto them will be of greatest value when applied to the development of a model ofspoken-word recognition within second-language processing. Such a model couldsupply a conceptual basis that will generate questions for future study in a variety ofareas of second language learning research. For example, in the area of secondlanguage testing there has been an on-going thread of research and discussionregarding the construct validity of common test types. It has been questioned, forexample whether tests called reacting comprehension or listening comprehension really do,or even can measure different, specific language abilities as distinguished from a moreglobal, general language ability (Bachman & Palmer, 1982; Barbour, 1983, Powers, 1982;Oiler, 1981; Oiler, & Hinofotis, 1980; Buck 1992). Tn a recent multi-trait, multi-methodresearch report, Buck (1992) supplies some evidence of a separate listeningcomprehension trait. Tn his discussion of the results, Buck says,Finally, there is also a need for research into the nature of the listeningtrait to be related to current theories of listening. Having said this, itought to be pointed out that there is no such thing as a generally acceptedtheory of listening comprehension (italics added). (p. 350)7A detailed model of word recognition in a second language can make contributions tothe development of a comprehensive theory of listening comprehension in a secondlanguage by describing some of the central mechanisms that operate during on-lineprocessing of spoken language. This in turn can make suggestions relevant to theconstruction and analysis of tests that propose to test listening comprehension.The study of second language learning is another area which could benefit fromthe development of a valid second language spoken-word recognition model. Thefocus of the present research is lexical access during processing of a second language,not second language learning per se. However, as Ellis (1985) points out, ‘A little isknown about U phonology, but almost nothing about the acquisition of lexis” (p. 5).By looking at the effect of lexical knowledge on spoken word recognition acrossdifferent levels of language proficiency, aspects of second language acquisition can bestated in terms of the existence or completeness of lexical representations. Approachingthe topic from this conceptual point of view should stimulate a novel genre of empiricalinvestigations of second language acquisition. If clear effects of lexical knowledgecould be found in the recognition of words in spoken (second) language, then this factmay provide the foundation for some illuminating experimental designs in languageinstruction/learning research. That is to say, it should be possible to investigate therelationship between types of instructional methods and the growth of lexicalknowledge by experimentally manipulating instructional variables and then testing forchanges in word recognition effect by using one or another of the various experimentalparadigms that have been used in this area in first language research.Theoretical research into second language processing should also profit.8Throughout this chapter, I have referred to “first language” and “second language” asthough they were categories that had a fixed or constant relationship. This is not so.In fact, one area of intensive study in second language research is the attempt tocharacterize the differences in language processing strategies among the variouspossible first language--target language dyads that can exist (Kilbom & Cooreman,1987; McDonald, 1987a; McDonald & Heilenman, 1991; Sasaki, 1991). Most of thisinvestigation has been done in the context of the competition model of speechprocessing (Bates & MacWhinney, 1981, 1982). Tn the present research the firstlanguage is English and the second language is English, also, but as targeted by nativeCantonese speakers. However, the present research should demonstrate that the cohortmodel of spoken-word recognition and its associated research paradigms can also beextended to provide a new approach to investigating the role any first language playsin the processing of any second language.In summary, motivated by the questions posed earlier and the points justpresented, the present study extends Marslen-Wilson’s cohort model of wordrecognition to the context of second language processing and investigates the role oflexical knowledge in the word-recognition phase of spoken second-languagecomprehension. In doing so, the research attempts to determine whether lexicalknowledge is used at all during the word recognition phase of second-languageprocessing and if so, whether different types of lexical knowledge have greater or lesserfacilitative effect. Using the cohort model, the research compares native and secondlanguage word-recognition processes in order to compare the functional role of lexicalknowledge in these two systems. In addition, the research examines how differing9levels of second language proficiency interact with different types of lexical knowledgeto affect word-recognition reaction times. Most importantly, the over-riding aim of theresearch is to provide a basis on which to construct a model of second-language wordrecognition processes.10Chapter II. Background of Research ProblemsThe aim of the present research is to assess the extendability of Marsien-Wilson’s(1989, 1987; Marslen-Wilson & Welsh, 1978) cohort model of spoken word recognition,originally developed to account for processes in a first language, to the second languagecontext. In doing so it addresses several questions regarding the role of lexicalknowledge during spoken word recognition in a second language and the developmentof lexical structure with increased second language proficiency. The choice of thismodel is motivated by the apparent lack of research within the second language area onthe relationship between lexical knowledge and the high-speed, unconscious processesthat operate during spoken language comprehension.Within the area of second language research, the potential of the cohort modeland the work of Marslen-Wilson appears to be recognized, directly or indirectly. Buck(1992) notes that more of the type of research Marslen-Wilson has done into speechperception is needed before a theory of listening comprehension can be developed.Bates and MacWhinney (1981), in their seminal paper on the application of thecompetition model to second language research, point out that the type of researchdone, and the early model developed by Marsien-Wilson was compatible with theproposals that they were putting forward. Carroll (1992) explicitly relies on the cohortmodel in her development of a theory of cognates. Hayashi (1991) extends some of theresearch methods used by Marslen-Wilson and Tyler (1980) to the area of secondlanguage processing. However none of these studies addresses the questions posedearlier in the first chapter. Nor do they question whether the cohort model itself isapplicable to the second language situation.11Two of the salient features of the cohort model are its focus on the high-speed,real time processes involved in language comprehension (see below), and their relationto lexical knowledge. Within the second language research literature there arenumerous approaches to the treatment of the second language lexicon. Some, such asBialystok (1990), refer to it within the larger context of second language processing andproficiency, while others, such as Carroll (1992) and Hudson (1989), present theoreticaltreatments of more specific aspects of the lexicon. Where data is actually gathered, ittends to be based on responses to printed material (e.g. Singleton & Little, 1991), onwritten material produced by subjects in response to auditory stimulus (Kelly, 1991), oron written samples gathered from subjects’ compositions (Zoble, 1989).In second language research, perhaps the only model (and certainly the mostproductive one in terms of generating research articles) which professes to look atlanguage processing and performance in real time is the cornpetition model (Bates &MacWhinney (1982, 1981; MacWhinney, 198Th; MacWhinney, Bates, & Kliegl, 1984).The competition model has been variously described as “a general model of languageprocessing and acquisition” (MacWhinney, Leinbach, Taraban, & McDonald, 1989 p.256.), “a probabilistic model of speech processing” (Harrington, 1987, p. 352), “aprobabilistic theory of grammatical processing” (Kilbom and Cooreman, 1987, p. 417),and “a general psycholinguistic model” (MacWhinney, 1987a, p. 317). In terms of theempirical methods used, it is “a model of sentence processing” (MacWhinney, Bates &Kliegl, 1984, p. 128), for the predominant approach has been to use sentenceinterpretation studies (though see MacWhinney et al., 1989, for an expression and testof the model in a connectionist architecture). It has as its basis a functionalist12perspective (see Bates & MacWhinney, 1982) to the analysis of language. Initially thismodel was developed in the context of research on the pragmatic and semanticinfluences on first language grammar in children and adults. However, much of theresearch has been cross-linguistic and, prompted by Bates and MacWhinney (1981;MacWhinney 1987a), there has been a logical extension of its application to the area ofbilingualism and second language research (Gass, 1987; Harrington, 1987; Kilborn, &Cooreman, 1987; McDonald, 1987a, 198Th; McDonald, & Heilenman, 1991; Miao, 1981;Sasaki, 1991; Wulfeck, Juarez, Bates, & Kilborn, 1986).The competition model is primarily interested in the processes involved inconstructing meaningful interpretations of the spoken (or sometimes written) sentence.It recognizes that there is a significant contribution to these processes from, inparticular, the animacy/inanimacy knowledge associated with stored lexical items. Italso attempts to evaluate the relative weights of this contribution among differentlanguages and to evaluate the change in weights, as a learner increases in proficiency ina second language. Research in the model’s framework provides compelling evidenceof the influence of first-language cue-interpretation strategies on the processing of cuesin the second language. However, the model is not concerned with word recognitionor assessing the implicit use of lexical knowledge during word recognition.In fact, it might be argued that, although response latencies are occasionally usedas dependent measures, the method used in most of the studies falls short of directlytapping on-line processes at all. Mean decision latencies, when reported, are in theorder of 1.5 to 2 seconds after the end of the entire utterance (e.g. Harrington, 1987), ascompared to the much faster times of approximately 300-400 ms after word onset which13are obtained in the word monitoring research of the cohort model. Furthermore, insome of the research (e.g. Gass, 1987), subjects were given a maximum of 20 seconds inwhich to respond; or, in some cases, the subjects actually acted out their conceptions ofthe actions. Both of these tasks allow plenty of time to perform a retrospective,conscious analysis of the sentence. Thus, although Bates and MacWhinney speak of itas “a performance grammar that can account for the rapid and simultaneous integrationof many aspects of discourse during sentence comprehension and production” (Bates &MacWhinney, 1981, p. 191), it seems, at least in the context of its predominantly-usedparadigm of sentence interpretation, the competition model has yet to be used to assessthe contribution of lexical knowledge to the high-speed and apparently unconsciousprocess of word recognition during speech processing.Concern with real time processing is found in other areas of second languageresearch. In Effis’s (1994) review of foreigner talk (language used by native speakerstoward non-native speakers of a language), he notes that there is some evidence tosuggest that native speakers of language will adjust their speech rate in accordance withthe level of the person spoken to. Investigating the effect of speech rate oncomprehension, Conrad (1989) presented native speakers, and high-level and mediumlevel skill non-native speakers with sentences that were time-compressed to 40%, 56%,71%, 83%, and 91% of the original speed. These times corresponded to 450, 320, 253,216 and 196 words per minute (wpm). The native speakers were able to achieve almost100% accuracy in recall at 56% compression (320 wpm). The high-level and mediumlevel non-native speakers only achieved 72% and 44% accuracy respectively at theslowest rate of 91% compression (196 wpm). Griffiths (1991) reviews a number of other14studies involving the relation between speech rate and ‘input” to second languagelearners and finds the majority of these studies either deficient in data to support thevarious claims made or containing methodological errors that undermine theirconclusions. In a study of his own, Griffiths (1990) used passages recorded atapproximately 100 wpm, 150 wpm, and 200 wpm to assess speech rate differences onnon-native speakers of English. Comprehension, as measured by a true-false testfollowing the presentation, was marginally lower in the moderately fast presentation(200 wpm) but there was no difference between the average (150 wpm and the 100wpm). Studies such as these are concerned with real-time processing of speech, butthey are more concerned with overall comprehension than with the roles that wordsand spoken word recognition play within the comprehension process.By turning to the cohort model, it may be possible to remedy this apparent lackof research llnking the second language lexicon to the fast processes that operate duringspoken language comprehension. Before discussing the model, though, it would beuseful to consider the concept of word recognition within the larger domain of spokenlanguage comprehension by presenting a brief overview of speech processing and thelexicon. Following this, I wifi outline the essential details of the cohort model ofspoken word recognition and describe in more detail the basic research paradigmadopted from Marsien-Wilson, Brown and Tyler’s (1988) study in first-language wordrecognition and used in the present study.15A. Speech Processing and the Lexicon1. Speech ProcessingAn overview of speech processing will include a number of key functions (formore extensive overviews see Forster, 1989 and Klatt, 1989). First, a stream of acousticenergy must be received by the ear and be transformed to some other physical, andultimately mental, representation of the phonemic, phonetic or spectral structure of theoriginal sound image. This new representation must be such that a match can be madewith an internally existing pattern or mental representation so that, indeed, a word canbe “re-cognized”. As these words are accessed, the knowledge that they symbolize iscombined to produce propositions and sentences. These are then combined andintegrated with the immediate discourse context and world knowledge of the listener.In other words, the comprehension process spans the reception of physical stimulithrough to integration with ongoing discourse. Clearly these functions must be fulfilledregardless of whether the listener is trying to comprehend a first or a second language.In a very important sense the lexicon, and the process of lexical access, are at the cuspof this comprehension process. On the one side is the computation of some sort ofrepresentation of the physical input. On the other side is computation of meaning(Marsien-Wilson, 1989). In the centre is the access of meaningful units in the lexicon.2. The LexiconIn a narrow sense, the lexicon can be thought of as a mental dictionary in whichthere are entries that contain representations of the meanings of words (cf. JohnsonLaird, 1987). But considering the lexicon only in this manner may obscure the fact thatother types of knowledge must necessarily be part of a lexical entry. Miller (1978; see16also Butterworth, 1983) lists a variety of information that the lexicon must minimallycontain. Of particular relevance to the present research is the assertion that lexicalentries must contain information on syntactic categorization (the entry is a Noun, Verb,or Adjective etc.), syntactic subcategorization (the entry is transitive or intransitive), andsemantic and pragmatic relations to other concepts such that there are restrictions onthe contexts in which the entry might appear. Certainly, modern linguistic andpsycholinguistic theory assigns a central role to the concept of the lexicon. Theassumption that syntactic structure is determined to a large extent by lexicalinformation forms an important basis of Government and Binding Theory (Chomsky,1965; Haegeman, 1991). Lexical Functional Grammar, which strives for a unification oflinguistic and psycholinguistic research, puts even more emphasis on the lexicalcomponent (Bresnan & Kaplan, 1982; Bresnan, 1982) in processing models.B. The Cohort Model and Research Paradigm1. The Cohort ModelThe cohort model of spoken word recognition evolved from an analysis ofMorton’s logogen model and Forster’s “bin” model (Marslen-Wilson, 1987) and hasundergone some modification over its development. While the present research willfollow the version that is described in the 1987 and 1989 papers by Marslen-Wilson andthe 1989 paper by Tyler, some essential points have remained unchanged. The modelitself is concerned with what Marslen-Wilson and Tyler (1981) call the central processesof speech comprehension. They define these as• . .the set of automatic and obligatory mental processes that are triggered when aspeech input is heard, and which carry the analysis through to its message-level17interpretation. These on-line processes are not open to conscious awareness orconscious control, and form the basis for the normal, effortless comprehension ofan utterance in context by a normal adult listener. (p. 108)In the model, these processes are ‘distributed’ and parallel. They are distributedin that each lexical entry is a computationally active device, and parallel in that accessto the lexicon and assessment of the fit of an entry to both the phonological data andthe higher-level representation of the discourse are occurring at the same time.Marslen-Wilson (1989) summarizes the important properties of the model as follows:• It assumes discrete, computationally independent recognition elements for eachlexical unit, where each such unit represents the functional coordination of thebundle of phonological, morphological, syntactic, and semantic propertiesdefining a given lexical entry.• Each recognition element can be directly and independently activated by theappropriate patterns in the sensory input.• The level of activation of each element increases as a function of the goodnessof fit of the input pattern to the form specifications for each element. When theinput pattern fails to match, the level of activation immediately starts to decay.(p. 6-7)In the model, word recognition is a product of two fundamental processes:access and assessment. Access is the contact of the computed representation of thespeech input with the lexicon. That is to say, access occurs when, after there has beenan acoustic-phonetic analysis of the speech input, “a representation of the input inthese terms ... is projected onto the mental lexicon.” (Marsien-Wilson, 1987, p. 72).Fundamental to the cohort model is the notion of multiple access: all word forms whichmatch or closely fit the initial speech input (the first 100-150 ms of the word) areaccessed and activated by that input. Each of the lexical units activated by that initialinput continues to monitor the incoming sensory data and remains active as long asthere is a satisfactory match with it. These activated word-forms constitute the cohort of18the cohort model.Also fundamental to the model is multiple assessment. At the same time that therepresentation of a particular string of input sounds is accessing a cohort of words inthe lexicon, a higher-level representation of the continuing utterance is beingconstructed based on prior input and other contextual information. Under the principleof multiple assessment, all members of the cohort are assessed for lexical propertieswhich match the contextual open spaces or locations within the current higher-levelrepresentation of the utterance and discourse. According to Marsien-Wilson (1987),Once the appropriate senses associated with a given word-form have been bound tothese locations in the representation, then we can say that recognition has taken place”(p. 98) and thus “it is at this point ... that the output of the system becomes perceptuallyavailable” (footnote, p. 98).In earlier versions of the model, a candidate was defined on the basis ofphonemic information and simply dropped out of the cohort if there was a mismatchbetween it and either the sensory data or the linguistic context. However, severalproblems became apparent with this account. First, research (reported in MarslenWilson, 1987) indicated that word frequency needed to be accounted for. Under certainconditions, within the same sentential context, higher frequency words were recognizedsooner than lower frequency words. To overcome this shortcoming, the concept oflevel, or strength, of activation, which can be construed as a “metaphor to represent thegoodness of fit of a given candidate to the bottom-up input” (Marsien-Wilson, 1987, p.99), was introduced to the model. According to this, there is a differential response tothe sensory input between high and low frequency lexical elements. As a consequence,19the level of activation of high frequency words will rise more rapidly and take longer todrop than the activation of low frequency words.A second problem was that the model could not explain how the system couldcope with noisy input. For example, in some research, deliberately-introducedmispronunciations would often be overlooked by listeners (Cole, 1973; Marsien-Wilson& Welsh, 1978). To deal with this, the assumptions about the input to the word-recognition process were changed. Instead of conceptualizing it as a string of phonemiclabels, it was re-specified in terms of a set of feature values. For example, /p/ and /b/would both be activated along all features except voicing, in which they differ.The final problem was that the earlier version of the model could not explain therather obvious fact that if a word is spoken under normal conditions, it will beperceived regardless of whether it matches the specifications of the context or not. Toaccount for this, the model assumes that the higher-level representation of the utterancedoes not operate on the activation levels of the different candidates. There is no top-down inhibition of the level of activation of the lexical units so a member of a cohortwill not ‘drop out’ of the cohort simply because it does not fit the context of the currentdiscourse or utterance.Under the cohort model, then, the process of word recognition is the process ofselecting a single candidate out of the (possibly large) cohort that is activated by thespeech signal. When a word is spoken with no context or in a meaningless context,such as a list of unrelated words, it will only be selected when enough of the signal hasarrived to uniquely identify it from the competing members of the cohort.But when the signal is heard in context — and note that normal context isfluent conversational speech — there need be no explicit form-based20recognition decision. Selection — viewed as the decision that oneparticular word-form rather than another has been heard — becomes abyproduct of the primary process of mapping word-senses into higher-level representations. (Marsien-Wilson, 1987, pp. 98-99)How quickly the recognition process takes place will depend on two variables: a) theextent to which the bottom-up input differentiates one candidate from its competitorsand b) the extent to which the match of lexical properties of a candidate with thediscourse context similarly differentiates it.2. Primary Research Paradigm: Discourse and Sentential ContextResearch within the cohort model has used a number of different researchparadigms, for example, spoken word monitoring, speech shadowing, gating, auditorylexical decision, and cross modal lexical decision. Each of these can get at differentaspects of spoken language processing, and each presents potential for investigating thefit of the cohort model to the second language context. However, the spoken-wordmonitoring tasks used by Marslen-Wilson, Brown & Tyler, (1988; Tyler, 1988, 1985)motivated, and can directly address, the research questions put forward in this paper.Those tasks occur in real-time, involve word recognition, and can explicitly manipulatecontextual constraints. In the monitoring task paradigm, subjects are required to listenfor the occurrence of a particular word or a type of word. Upon hearing the words,subjects press a button or speak in order to indicate that the word has been perceived.Cues to these target words can, of course, be specified in a number of ways. Forexample, it could be the identical word itself, a rhyming word (e.g. cue: lead, target:bread) or a category to which the target word belongs (e.g. cue: metal, target: lead).However, specifications other than identical-word monitoring involve processes otherthan those strictly involved in recognition (Marsien-Wilson & Tyler, 1980).21In identical-word monitoring, the context of the target words can be manipulatedin order to assess the effect of different kinds of contextual constraints on the speedwith which the words are recognized. Roughly speaking, in this paradigm, context canbe categorized as discourse level, where meaning and context are seen as developingacross sentence boundaries, or sentential level, where within-sentence contextualconstraints are of interest.a. Discourse Level ContextThe effect of a broader, discourse level context on word recognition speed hasbeen investigated and reported with mixed findings. Marslen-Wilson and Tyler (1980)contrasted the effect of the presence or absence of a sentence which preceded (led into)the test sentence which contained the target word and found that there was afacilitation of word recognition in the condition with a lead-in sentence. On the otherhand, Marslen-Wilson, Brown and Tyler (1988) compared the different effects of normaland incongruous lead-in sentences (which they termed discourse incongruity) on wordrecognition in the test sentences which followed. In this study, each stimulus was apair of sentences, a lead-in sentence and a test sentence which contained the targetword. In half the cases, the lead-in sentences had no discourse relation to the targetbearing test sentence. That is to say, there was an incongruity between the twosentences. In the other half of the cases, the lead-in sentence and the test sentenceformed a normal, connected pair of sentences. They found no effect of discoursecongruity.One explanation was that a differential advantage coming from the discoursecontext disappeared as a result of increasing within-sentence constraints. Support for22this explanation is drawn from the earlier experiment (Marsien-Wilson & Tyler, 1980),in which the effect of target-word location was also investigated. They found adecrease in the effect of discourse congruency/incongruency as the target word wasmoved further from the beginning of the test sentence.Zwitserlood (1989), using a lexical decision task, came to the conclusion that evensentential context with a strong semantic bias did not pre-select contextuallyappropriate words (i.e. activate appropriate words before any word specific sensoryinformation is available), that both contextually appropriate and contextuallyinappropriate words were activated during lexical access, and that contextual effects arelocated after word-initial access but at a point in time when the sensory data is stillinsufficient by itself to disambiguate the word.A possible interpretation of these results is that there are really two quitedifferent mechanisms at work. In the Marslen-Wilson and Tyler (1980) research, acondition of a discourse-level context was compared with a condition of no discourse-level context. In the Marslen-Wilson et al. (1988) study, a condition with anincongruous discourse-level context was compared with a condition with a congruousdiscourse-level context. It may be the case that lexical access is not at peak efficiencyuntil after some, relatively short, period of activity. Then, once this short ‘warm-up’phase is completed, contextual effects on word recognition speed are dominated bylocal sentential constraints. Assuming this to be the case, it is clearly important toensure that if the focus of investigation is the effect of local constraints on wordrecognition, then these local conditions must be situated far enough from the beginningof an experimental cue sentence to ensure that lexical warm-up can not be a factor in23the recognition process.b. Sentential Level ContextMarsien-Wilson and Tyler (1980) presented subjects with three different types ofprose context, which they termed Normal Prose, Syntactic Prose, and Random Word-Order Prose. The sentence context in Normal Prose is syntactically and semanticallynormal. The example they provide (target-word emphasized) is:(1) The church was broken into last night. Some thieves stole most of thelead off the roof. (p. 8)To produce Syntactic Prose sentences, all content-words (except the target-words) in theNormal Prose sentences were pseudo-randomly replaced “with new words of the sameform-class and word-frequency, such that the resulting sentences had no coherentsemantic interpretation” (p. 17). Sentence two is the example they give:(2) The power was located into great water. No bums puzzle some in the leadoff the text. (p. 8)Random Word-Order Prose sentences were created by scrambling the order of thewords in the Syntactic Prose sentences, as in the following:(3) Into was power water the great located. Some the no puzzle buns in leadtext the off. (p. 8)The differences between recognition times to targets in these contexts weresignificant. However, while Normal Prose and Random Word-Order Prose clearlydefine the two limiting cases of normal lexical constraints and no lexical constraints atall, Syntactic Prose does not specify what lexical information is or is not available to therecognition process, nor does it allow investigation of specific lexical items.In response to these problems, Marsien-Wilson, Brown and Tyler (1988)investigated word recognition in contexts where there were local rather than the more24global violations of the Syntactic Prose. The linguistic background to this approach tomanipulating sentential context is covered well in that study, but a short review of it isin order here. They turned to Chomsky’s (1965) analysis of the properties of lexicalrepresentations, but the principles remain the same in more recent treatments ofGovernment and Binding Theory (Haegeman, 1991; van Riemsdijk & Williams, 1989).Essentially, the lexicon is a repository of all information that a speaker of a languagehas learned about the words of that language. Each lexical entry will contain categorialinformation about the entry; that is, information about whether it is a noun, verb,preposition, and so on. Furthermore, the entry will contain details on what subcategorythe particular entry belongs to. For example, in the case of verbs, it is necessary toindicate whether the lexical entry is intransitive, transitive or ditransitive (takes twoobjects). This subcategorial information is expressed in distributional frames. Twoexamples given by Haegeman (1991, p. 34), are:1. meet: V, [ NP]2. dither: V, [While the verb meet must appear in front of an NP (noun phrase), the verb dither takesno complement. In Haegeman’s words, “meet subcategories for or selects an NP (p.34).” Tn addition to this syntactic information on verb subcategorization frames in thelexicon, there is semantic information that limits what kind of NP can be used in theverb’s frame. Chomsky (1965) makes such suggestions as +Animate or -Human and soon as selectional features which restrict what can be inserted in these frames.By manipulating violations of these subcategorization and selection restrictionconstraints, four types of different sentential contexts can be created (see Table 1)25Table 1Types of Context Derivable by Manipulating Verb Frame Features1. normal sentences2. pragmatically anomalous sentences3. semantically anomalous sentences4. syntactically anomalous sentences(Marsien-Wilson, Brown and Tyler, 1988; Tyler, 1988, 1985). Tn brief, Type 1 sentencesare the syntactically correct sentences in which a transitive verb is followed by a directobject (the target word) that is both semantically and syntactically congruent with theverb. For example:1. She watched the bubbles floating in the air.Sentences such as these define a linguistically correct and plausible context and thusallow the listener to utilize any relevant linguistic knowledge during the recognitionprocess. Recognition times to target words in these sentences provide a base-linemeasurement for word recognition in the normal context of fluent conversationalspeech, which presumably is based on linguistic knowledge.The normal sentences are also of a structural pattern that allows for a controlledand systematic investigation of the interaction of certain aspects of linguistic knowledgeand linguistic context. The initial part of the sentence, in particular the verb, establishesconstraints, via selection restriction rules, on the latter part, the word in the object slotof the verb argument frame. By deliberately violating the verb-argument relations inthe utterance, and thus removing the constraints imposed by the lexical knowledge26associated with the verb, it is possible to prevent a particular kind of knowledge frombeing used during the recognition process. A more detailed description of the otherconditions will illustrate this.Type 2 sentence, which is labelled “pragmatically anomalous sentences” by bothMarsien-Wilson et al. (1988) and Tyler (1985, 1988), presents the subjects with sentencesin which a transitive verb is followed by a direct object which, to some extent, is atodds with the every-day experiences generated by the context, particularly the verb andits meaning, up to the point of the target word. For example:2. She ate the bubbles floating in the air.While eating bubbles is undoubtedly not impossible, and in fact is an activity that onemight pursue under certain circumstances, it is not a highly likely or plausiblecontinuation to the sentence after the verb. Furthermore, in this particular paradigm,test sentences can be preceded by a lead-in sentence. These lead-in sentences, while notsupplying a highly predictable situation, can be composed so as to tend to ground thesubsequent test sentence within a real world context. To the extent that the recognitionprocess makes use of lexical knowledge related to the plausibility of the relationshipbetween the verb and the word in the argument frame (in combination with thedeveloping phonological representation of the word), the word-recognition process inthese sentences will be slowed down. Marsien-Wilson et al. (1988) found a 28 msecslow-down in this context compared to the normal context, the smallest effect of thethree anomalous contexts. On the other hand, Tyler (1985), in a study that involved anagramatic patient, found that violations of this kind caused the greatest slowdown inword recognition in this person. The patient’s response profile had its peak in this27context.Type 3 sentences, “semantically anomalous sentences’, arise from the violationof a selection restriction rule of the verb-argument relation. Following the pattern ofType 1 and 2 sentences, Type 3, also contains a transitive verb followed by a directobject target word. In this context though, the verb has semantic features that are notcongruent with the target word. For example:3. She interviewed the bubbles floating in the air.Words in the direct object argument slot of the verb “interview” are semanticallyrestricted to having features which indicate that they are capable of being interviewed.This might be represented as [+sentient] or at least [+capable of language]. In this typeof sentence then, the semantic properties of the verb preceding the target word imposesselection constraints on the items that can ff1 the verb argument frame. The targetwords in these contexts do not conform to these constraints. To the extent that theword recognition processes make use of these constraints, the recognition process willbe disrupted and reaction time will slow down.Type 4 sentences, “syntactically anomalous sentences”, differ from Types 1 to 3,in that the stimulus word strings contain an intransitive verb and a target word that islocated as though it were the direct object of the verb. For example:4. She crawled the bubbles floating in the air.“Sentences” such as these set up violations of strict subcategorisation rules. The verb issubcategorised in the lexicon as having no argument frame, but by placing a nounphrase immediately after the verb, it appears that the sentence contains a direct object.Because of this kind of violation, the verb is not able to supply productive constraints28on the target word. As a result the listener is only able to use the phonologicalrepresentation to recognize the word bubbles. Thus, recognition time to target words inthis type of sentence should approximate recognition time under a condition wheretarget words are presented in a meaningless context such as a list of unrelated words.29Chapter III. Extending the Model: Hypotheses and ExpectationsA. The Roles of Semantic and Syntactic Knowledge in the Word Recognition Process,Materials and Learner VariablesExtending the cohort model of spoken word recognition to second languageprocessing wifi need to take into account two significant differences between a secondlanguage context and the adult, first language context in which the model wasdeveloped. One is language-based variabffity. In any application of the model to asecond language there will be an existing first language processing system which maybe working in concert or in conflict with the second language processing system.Different first/second language dyads may result in important differences in the way inwhich the second language is processed. While these potential differences are the mainfocus for study in cross linguistic research (see competition model references above),they might result in uncontrolled variance if subjects from different linguisticbackgrounds were included in a single study. This can be avoided by limiting a studyto one first and one second language.However, even when the research scope is narrowed to a single first languageand a single second language, a major difference between contexts remains; namely,that development or learning takes place in the second language and thus there can bemajor individual differences in language proficiency. The cohort model is a modelwhich was constructed within the context of a fully developed system of adult, firstlanguage and, as a result, one can assume stability in the system. Adult secondlanguage learning, on the other hand, can encompass states of linguistic developmentthat range from an absolute neophyte through an advanced learner to an expert who30can function effortlessly in a broad range of linguistic situations. By focussing on thisdifference, the present study (and future ones) should provide information on bothsecond language speech processing and second language learning.To review how this is so, consider two of the salient points of the cohort model.First, the lexicon and lexical access are central to the cohort model and, according to thecohort model, lexical entries are defined by bundles “of phonological, morphological,syntactic, and semantic properties’ (Marslen-Wilson, 1989, p. 6). These properties arelearned. Thus, a defining aspect of increased proficiency will be a larger, morecomprehensive lexicon with richer, more complete entries. Second is the function ofthese properties in word recognition. During speech comprehension, the first 100-150ms of an incoming word activates the set of all lexical entries that match its word-initialphonological form. Prior to word recognition, the syntactic and semantic properties ofall of the members of this cohort become available for assessment as to fit in the higherlevel context. As the phonological input continues to arrive and be processed, theincreasing data further restricts the possibffities for membership in the cohort.When the semantic and/or syntactic properties of a particular member of theactivated cohort fit the higher-level mental representation of the context, that word isintegrated effortlessly into the ongoing discourse, often before the end of the word isreached and sometimes before enough phonological information has been provided todistinguish it from other competitors. At this point, the word has been recognized.However, when the context is disrupted in such a way that there are no contextualopen spaces or locations within the current higher-level representation of the utterancethat match the semantic and syntactic properties of any members of the cohort of31activated word-candidates, the recognition process must wait until sufficientphonological data has arrived to identify the word. Thus, the speed of wordrecognition slows down.However, in the case of a person listening to speech in a second language, thesyntactic and/or the semantic properties of a spoken word may not be available to thelistener because the relevant properties have not yet been learned and included in the“bundles that constitute the entry in the listener’s mental lexicon. In this case, if wordrecognition takes place at all, it will be based entirely on phonological data. In otherwords, word-recognition speed in a second language will be a function, in part at least,of the existence or not within the listener’s mental lexicon of the particular, context-relevant lexical properties. This has an important implication. If the word recognitionspeed of a second language user is relatively slow even though syntactic or semanticconstraints are available in the message, then the relevant lexical properties must bemissing from the pertinent mental lexical entry or entries of that person.Thus, by looking at the recognition-time proffle across sets of contexts wheredifferent syntactic and semantic constraints are or are not available, it should bepossible to ascertain which types of properties are functioning in the second languagerecognition process. Also, by comparing recognition-time profiles of different levels oflanguage proficiency, it should be possible to get some idea of how these propertiesdevelop functionally, relative to each other. Finally, since highly familiar words willhave better defined lexical entries than less familiar ones, an even finer-grainedassessment of this functional development should be obtainable by looking at wordrecognition time differences within speakers but between selections of famffiar and less32familiar words. This will be developed in detail in the next section.B. Word Familiarity and Word KnowledgeAlthough early versions of the cohort model (Marsien-Wilson and Welsh, 1978;Marsien-Wilson and Tyler, 1980) made no mention of word frequency or activation, thishas changed in later versions (Marsien-Wilson, 1987). Part of the motivation formoving to a strength of activation metaphor was the finding that word-frequency was afunctional variable in the recognition process. Marsien-Wilson (1987) reports strongfacilitation to lexical decisions on the higher frequency member of pairs of high/lowfrequency words which were matched according to recognition point (the point in aword where it “can be discriminated from the other members of its word-initial cohort,taking into account both contextual and sensory constraints, Marsien-Wilson, 1987, p.80). Other evidence from research using lexical decision tasks (Taft & Hambly 1986),cross-modal priming (Marslen-Wilson (1987), and gating tasks (Tyler, 1984) supports theidea that word frequency wifi have an effect on speed of word recognition. To accountfor this, it was proposed that high frequency words would initially have a higherrelative level of activation than low frequency words (Marslen-Wilson, 1987). Morerecent studies in printed (Allen, McNeal & Kvak, 1992) and auditory (Connine, Titone,& Wang, 1993) word recognition also indicate that word recognition speed is related toword frequency, and that higher frequency words will be responded to more quicklythan lower. There is controversy about whether these frequency effects reflectdifferences in initial activation levels of lexical entries or post-access bias (MarslenWilson, 1990), but Marslen-Wilson reports research findings that suggest these effects33occur early in the selection process (Marslen-Wilson, 1987, 1990).It is reasonable to expect that these activation effects will prevail in the secondlanguage context, too, regardless of their temporal location in the word recognitionprocess. That is, second language learners should also respond more quickly to morefrequently encountered words than to less frequently encountered ones. Thisexpectation will be referred to as the word familiarity: activation hypothesis to distinguishit from another effect that is predictable from concepts of word familiarity and growthin language proficiency. This effect will be outlined next.The effect of word frequency on recognition speed has not been directlyinvestigated in the word monitoring experiments conducted within the purview of thecohort model and will need to be assessed, but if the effect is experimentally observablethen it can be instrumental in the investigation of second language learning. Usually,word frequency is defined on the basis of data that summarizes a written corpus (e.g.Carroll, Davies & Richman, 1971; Francis, & Kucera, 1982) and can be seen as anestimate of actual frequency of exposure and consequently, of opportunity to learn andbecome familiar with the word. While this assumption would probably not be accuratefor beginning language learners since much of their exposure to the language may takeplace in restricted contexts such as classrooms or language textbooks, it becomes morereasonable with increased proficiency of the individual and longer immersion in thetarget language setting. In addition, it seems plausible that ongoing exposure to wordsin use in the global second language context would result to some extent in increasedword knowledge in that language. Because of learning resulting from greater exposurethen, entries for high frequency words should have more complex lexical structures34with richer, more complete sets of lexical properties than those for low frequencywords. If this is true, then word frequency can also be used as an indirect measure ofword knowledge and therefore provide an additional strategy for investigating secondlanguage growth. This word familiarity: learning hypothesis proposes that the effect ofdifferent levels of language proficiency on word recognition speed should be reflectedwithin a proficiency level by high and low frequency words.C. Second Language ProficiencyBefore the lexical characteristics of second language proficiency are discussed,one other aspect of differences in fluency must be reviewed. In the cohort model, thelexicon is central to the word recognition process, but the cohort model is comprised ofa number of sub-functions and assumes processes that are external to the model. Withsecond language learners, it is not necessarily the case that the different processes andstructures which underlie or influence the word recognition process will develop atequal rates. In particular, the cohort model divides the process of spoken word-recognition into three basic functions: access, the mapping of the sound signal onto thelexical representation of words; selection, the selection of the word-form that is closest inmatch to the input; and integration, the process of integrating the syntactic and semanticinformation of the word being recognized with the higher level representation of thecurrent utterance.Both access and integration involve processes that are not described in MarsienWilson’s model. In the case of access, there is the process of transforming the physicalinput signal to a mental representation appropriate for lexical access, and in the case ofintegration, there is the process (or processes) of constructing a higher, message-level35representation. In order for the cohort model to be of value in investigating thedevelopment of the second language lexicon, it is important to be able to argue that thelocus of the particular experimental effects that are to be examined is in the lexicon andnot the pre-access or post-integration stages of speech processing. Therefore, beforediscussing how the development of the lexicon might impact on word recognition, Iwifi discuss the relationship of increased proficiency to the access and integrationprocesses first.1. AccessPrior to access, the stream of acoustic energy must be received by the ear and betransformed to some other physical, and ultimately mental, representation of theoriginal sound image. This new representation must be such that matches can be madewith internally existing mental representations so that a cohort of words can beactivated and eventually a word be recognized. In the early discussions of the cohortmodel, this process of taking input in the form of a speech wave and generating someform of representation usable by the access process was outside the model proper. Theapproach was to assume that the peripheral auditory system provided the lexicalprocessing system with some form of acoustic-phonetic analysis of the speech which isprojected onto the mental lexicon (eg. Marslen-Wilson, 1989; Zwitserlood, 1989). Morerecently, the nature of the representations in the recognition lexicon (Lahiri andMarslen-Wilson, 1991; Marsien-Wilson, 1993) and of the input to the recognition lexicon(Marsien-Wilson and Warren, 1994) have been the focus of investigation.In an ordinary first language context, it is safe to assume that both the processesoperating at the speech input level and the word-form representations stored in the36lexicon have stabilized in the adult. It is not certain when the input processing systemstabilizes in native speakers of a language but there is evidence of a reorganizationfrom a ‘universal’ to a ‘language-specific’ phonetic perception occurring between 6 and12 months (Werker & Lalonde, 1988; Werker, 1993). When, if ever, does the adultsecond language learner’s peripheral auditory system and segmentation process beginto output consistent representations of the sound structure of the new language? Thereare indications that at least some new phonological information can be learned byadults in a relatively short time. In two one-hour laboratory sessions, McClaskey,Pisoni & Carrell, (1983) succeeded in training subjects to distinguish a stop intermediatebetween /p/ and /b/ by using voice onset time stimuli. Furthermore, thisdiscrimination ability was generalized to another new stop category, between /k/ and/g/. On the other hand, the very term ‘foreign accent’ suggests that there are secondlanguage learners who never seem to attain the ability to produce the differencebetween particular phonemes in a second language (or to use certain syntacticstructures) consistently. These errors have become fossilized (Selinker, 1972) in theirproduction systems and likely represent lack of discrimination in their auditory codingsystems, too. In a review of cross-language speech perception research, Werker (1993)concludes that “...although young infants are equally sensitive to both native and nonnative phonetic contrasts, adult perception is modified by language experience, and theimpact of experience is more profound for some non-native contrasts than it is forothers....’ (p. 61). What may be the case is that the output from the emerging auditorycoding system becomes internally consistent rather rapidly, fixing on the sub-set of thesignificant contrasts in the spoken sounds of the new language which are either still37available to the learner or readily re-learned.I would suggest that, under the cohort model, regardless of whether this codingsystem develops slowly, and moves toward native-speaker like discrimination, orquickly, and retains lack of discrimination among certain phonological features, themain effect of a less-than-native-like input system would be slower word recognitiontime and increased variability of these times. This is because of what Marslen-Wilson(1987) calls the contingency of perceptual choice, in which the identification of a worddepends not only on what words are present but also on what words are not present.A system which can not discriminate along certain features will retain more words inthe cohort longer than one which can. Thus, recognition will take longer. However,since this will only occur in word-cohorts in which a particular feature is functional indiscriminating among the members, the slow-down will not be universal andconsequently there will be greater variability in recognition times. Furthermore, unlessthese unmastered distinctive phonological features are correlated with syntactic orsemantic lexical features (unlikely in the case of the initial sounds of root words inEnglish), an increasing proficiency in a second language input coding system shouldnot interact with syntactic or semantic information so as to affect word recognitiontime.In summary, decreased proficiency in the auditory encoding system shouldincrease word recognition time overall, but should not interact with changes in thesyntactic or semantic structure of the lexicon. That is, word recognition will take longerfor those who are less proficient. This will be termed the delayed access hypothesis.382. IntegrationIn the cohort model, the integration function ‘concerns the relationship of therecognition process to the higher-level representation of the utterance” (Marsien-Wilson,1987, p. 72). Because the focus of the cohort model is on word recognition, not a greatdeal is said about how these higher-level representations are constructed. If they comeabout as a result of processes that are different from the assessment/integration whichoriginate in the lexicon, then it seems reasonable to ask whether increased proficiencyin a second language is primarily related to increased capabilities in these higher-levelprocesses. Relevant to the present study, members of the activated cohort are assessedagainst these higher level representations during the recognition process.Consequently, more accurate representations of the message, which result from greaterproficiency, may also contribute to increased word recognition speed. As a result, aninterpretation problem appears to arise if a difference in word recognition speed isobserved that is associated with a difference in language proficiency. Can thisdifference in speed be attributed to differences in lexical structure or differences incapability in the higher level discourse processes?However, this is not as great a problem as it may seem. First, when MarsienWilson and Tyler (1981) discuss higher-level representations, they propose two distincttypes of processes that contribute to the construction of these sentential and discourse-level representations. They characterize them as (1) those that are within the model (theautomatic and obligatory central processes I have described earlier) and (2) those thatare off-line, post-access processes which they say are “idiosyncratic and variable, andnot, we believe, central to the normal process of speech understanding” (p. 322). Tn39addition, relevant to the present study, Tyler (1989) stresses the immediacy with whichthe lexical representation of a verb imposes constraints on the processing of thesubsequent input. The verb becomes the dominant source of contextual constraint. Itwould appear, then, that the cohort model is at least adequate for explaining the effectof lexical knowledge related to verbs on the recognition of subsequent nouns.Let us return to the relationship between the processes involved in wordrecognition and the processes involved in constructing and interpreting higher-levelrepresentations. If language processing can be legitimately modelled as an informationprocessing system (cf. McLaughlin, Rossman & McLeod, 1983), then the likelihood thatthese two separate sets of processes must take place simultaneously should result in asignificant interaction of proficiency with word recognition speed under somecircumstances. It is reasonable to suppose that during the development of secondlanguage proficiency, the construction of the higher level representation will put a‘load on the processing system. As proficiency develops, this load will decrease. Inthe case of first language processing, this construction process would generally beeffortless. In a comparison, then, of two situations where recognition takes place underdifferent load conditions, native speakers of a language and second language learners(or even fluent second language users) may perform differently.Consider, for example, the situation when a target word is presented in arandom list of words and compare it to Type 4 sentence (syntactically anomaloussentences) above. Under both of these conditions, there are no syntactic or semanticconstraints to aid in word recognition. Where they differ, however, is that in Type 4sentences it is possible to construct a higher level representation right up to the point40where the anomalous noun phrase occurs. in the random list, it is not possible to dothis. For native speakers of a language this would make no difference. Wordrecognition would proceed according to phonological data alone. However, for a lessproficient second language user, the effort of constructing the representation may drawresources away from the processing of the phonological data. That is to say, if there isno attempt to construct a meaningful representation when a word is presented in arandom list, then target word recognition for the language learner under this conditionwill be faster than in the Type 4 sentences.In summary, this review of the integration processes of the cohort model hasidentified a possible source of interaction of language proficiency with word recognitionspeed. The information processing hypothesis proposes that, for second language users,because of a reduced load on the processing system, word recognition speed will befaster in the context of a random list than in a sentential context which would cueconstruction of a higher level representation but not impose syntactic or semanticconstraints on the target word. This should not be the case for first language users.Importantly, this effect will not obscure effects on word recognition speed related tolexical structure.3. LexiconOn the basis of the above arguments, it is reasonable to proceed with extendingthe cohort model to the second language context in a preliminary investigation of thestructure and development of the second language lexicon. Lexical representations“provide the basic bridge between sound and meaning, linking the phonologicalproperties of word-forms with clusters of syntactic and semantic properties.” (Marsien41Wilson, Brown & Tyler, 1988, p. 2). Also, the lexicon is, if not the central link inlanguage learning as suggested by Carroll (1992), one of the most importantcomponents in both first and second language learning, since the lexicon must be learntby both native speakers and second language speakers.But what is the nature of this lexical learning in a second language context?Two specffic questions arise: (1) What is the nature of the limiting case of the secondlanguage (i.e, highly proficient users’) lexical structure and word recognition? and, (2)What is the nature of growth toward that limiting case? First, I will sketch out threevery general limiting cases toward which the development of the second languagelexicon may move. Each of these possibilities predicts a characteristic effect on the useof contextual constraints in the word recognition process. Then I will present a moredetailed analysis of how different developmental sequences of lexical verb and nounentries might affect the speed of word recognition in a second language.The first possibility is that the underlying processes in the cohort model do notfunction in parallel in even proficient second language users. That is, in a secondlanguage, particularly one learned after puberty, phonological coding acts strictly as anaddress for the meaning of the particular word, and syntactic and semantic informationdo not become available for processing until after the word has been recognized. Thereis no mechanism for multiple assessment of candidates against the developing context.The process of integration would take place after a combined recognition/selectionprocess, unlike in the cohort model where selection during nonnal conversation‘becomes a by-product of the primary process of mapping word-senses into higher levelrepresentations.” (Marsien-Wilson, 1987, p.98-99). The implication of this for word-42recognition under different message-level contextual constraints is that there would beno difference in recognition time. Only the phonological data could be used, Theresponse-time profile for even advanced users would be horizontal.This seems unlikely as a limiting case for proficiency in a second language.First, it would mean that even highly proficient second language users would always berunning a processing deficit that would increase with the length of the continuousdiscourse. There would be some time needed to integrate a word with the on-goingdiscourse once its semantic and syntactic properties became available for processing.While this may apply to learners, it does not seem to be a realistic characterization ofthe nature of proficient second language use. Second, Hayashi (1991) appears to haveafready demonstrated that in second language users there is an effect of context onword recognition speed. Since in Hayashi’s experiment the target words were obscuredto some extent with white noise, it is possible that their identity was arrived at throughsome kind of post-perceptual deductive strategy rather through normal recognitionprocesses. Thus, however unlikely this possibility is, it can not be ruled out on thebasis of prior research.Another possibility is that the phonological encoding acts strictly as an address,as just described, but that at any particular address there almost always remains, ratherthan a single candidate, a number of candidates. That is to say, the phonologicaladdress and the encoding process never reach such efficiency that words can easily beuniquely identified on the basis of bottom-up data. When a word is heard out ofcontext, the identification process would have to rely on some sort of phonological andsituational best-guess procedure to reach any decision. When the word is in a43meaningful context, then the selection process can use syntactic and semanticconstraints to make the final choice. In this case, the reaction time profiles of nativespeakers and fluent second language speakers under different contextual constraintswould be parallel but not coincident. In particular, the second language users’ profilewould lag behind by at least the difference between the native speakers’ recognitionpoint, often occurring in the middle of a word, and the offset of the word.The final, contrasting possibility is that the limiting case of the second language(i.e., highly proficient users’) lexical structure may be essentially the same as nativespeakers. In this case, for highly proficient second language users and native speakersof a language, the profiles of word-recognition times under different conditions ofcontextual constraints would be parallel, and in fact would coincide.Of course, defining the limiting case of second language lexical structure doesnot say anything about the developmental path taken to get there. One plausiblehypothesis of integrated lexical development regarding such a path is that the different typesof properties, including the phonological, are highly interdependent, and that they‘assemble’ uniformly and evenly with no particular property or attribute becomingmore well defined or more highly functional in the word recognition process thananother at any time. Once a word is ‘recognizable’ (i.e., has a stable phonologicalrepresentation), all other properties will be available to the recognition process. Animplication of the hypothesis suggests that, as a speaker moved from lesser to greaterproficiency, the pattern of recognition would result in a gradually decreasing number ofunrecognized words (missed targets), an increasingly significant use of availablecontextual constraints of all types, and a decreasing variance in word recognition speed.44This hypothesis also predicts that for all target words that are recognized, there will bean effect of contextual constraint.At first glance, this type of lexical development may seem unlikely in secondlanguage learning since one of the jobs of language instruction is teaching new or moreprecise meaning or usage for words that students already use or recognize. That is tosay, second language learners will often have a phonological representation stored inthe lexicon but will not have certain aspects of meaning or syntax attached to it.However, if word recognition is specifically understood as the process of fixing theidentity of a word through those central processes described by Marslen-Wilson andTyler (1981) and not as some sort of post-perceptual deductive processes, howeverquick they may be, then this interdependent model of lexical development should havesome validity.What is more probable, though, is that word recognition processes will make useof whatever lexical properties are available. Furthermore, lexical entries will grow,properties will accumulate, along those dimensions or within those categories which areeasiest to infer from the spoken, contextualized presentation of language which the usermeets with on a day to day basis. As I will argue below, in terms of syntactic orsemantic properties, it is more likely to be the semantic properties that are inferred firstand the semantic constraints that become functional soonest.When Marsien-Wilson et al. (1988) carried out their study on the effect ofsyntactic and semantic constraints on word recognition speeds, they exploited thefundamental distinction between the categorial properties of the verb argument frameand the semantic and syntactic properties of the items that can fill these argument slots’45(p. 2). It is possible to outline some fairly strong prerequisites for these specificconstraints to function in an identical manner in the second language word recognitionprocess. For an English verb to impose such constraints on a noun phrase that followsit, certain lexical conditions must be in effect. First, rather obviously, the ESL personmust know what the verb and the target noun sound like. That is to say, in thatperson’s mental lexicon there must be phonological representations for the entries ofboth the constraining verb and the target noun. Second, the verb’s lexical entry mustinclude the categorial information that it is, indeed, a verb. In addition, that same entrymust supply sub-categorial information about what type of verb it is. In particular, theverb will need to supply information that it is a transitive verb, for it is this propertythat imposes the syntactic constraints on the interpretation of subsequent input. Forthere to be semantic constraints as well, there must also be a well-constructed semanticrepresentation of the verb’s meaning. This semantic information about the verb definesrestrictions on the words that can be selected to fill the open verb-frame slot or slots(see Chapter II). Finally, for the syntactic constraints to act upon the noun, the noun,too, must be categorized within the second language lexicon as a noun, and for thesemantic constraints to come into play, the noun, too, has to have a well-constructedsemantic representation.This analysis appears to require that for this particular case (recognition of objectnouns constrained by the preceding verb), the syntactic properties must be functionallyoperative before there can be a semantic effect in word recognition. If this apparentlogical requirement is also a developmental sequence, then learners of a secondlanguage may not show any difference in speed of word recognition when presented46with sentences such as 1 and 3. This would happen if learners can develop1. She watched the bubbles floating in the air.3. She interviewed the bubbles floating in the air.representations of verbs which include categorial and sub-categorial information but donot have well differentiated semantic representations which can select the nouns thatsemantically fit in the argument slots.The problem with this analysis is that it is once again based on the fullydeveloped lexicon and does not take into account how lexical entries develop,something that must be done if the cohort model is to be applied to the secondlanguage situation. A more plausible hypothesis would be that of semantic dominance.This proposes that initial lexical entries are phonological form representations linked tosemantic representations which are syntactically uncategorized. Those that have somesort of predicate argument structure eventually become categorized as verbs or asnouns, adjectives or prepositions that have arguments (Haegeman, 1991). Tn theserelatively immature lexical entries, the syntactic properties of the argument frames arenot well specified though there may be semantic constraints on what could fill them.The verb in sentence 3 can serve as an example. The meaning of the verbinterview implies two participants, the interviewer and the interviewee. Similarly, thenoun interview also implies two participants, (e.g. ... the reporter’s interview of thevictims...). Where the noun and the verb differ is that when interview is a verb, fillingboth frames is syntactically mandatory whereas when it is a noun they are bothoptional. Furthermore, with respect to syntax, the verb arguments must be filled withnoun phrases, while one of the optional arguments of the noun must be a prepositional47phrase. It is likely the case that syntactic details such as these take longer to sort outand become part of the functional properties of the lexicon than it does to establish asemantic representation of the concept of interview that excludes such things as bubblesas the interviewee. If this is so, then in the case of the second language learner,response time to target words in sentences such as number 3 will be slower than thosein normal sentences because of the lack of semantic constraints within the message, butresponse time to target words in sentences such as number 4 will not differ from that to4. She crawled the bubbles floating in the air.number 3, because the learner is unable to make use of the syntactic constraints in thosekind of sentences.4. Summary of Hypotheses and Predictable OutcomesThe foregoing conceptual analysis of the application of the cohort model tosecond language word recognition allows us to express more clearly the particularquestions that were addressed in the present study, and to state some expectationsregarding the outcomes. The first question is:a). How are the word recognition systems of native speakers and fluentsecond language users different?Two hypotheses are proposed on particular areas where these systems will differ. First,the delayed access hypothesis proposes that decreased proficiency in the auditoryencoding system will increase word recognition time overall. Consequently, even fluentsecond language users will take longer than native speakers to recognize wordspresented in any context. Second, the information processing hypothesis proposes that, forsecond language users, there will be a reduced load on the processing system whenword recognition takes place in the context of a random list. Consequently, recognition48time will be shorter than in a sentential context which allows construction of a higherlevel representation but supplies no recognition aid in the form of syntactic or semanticconstraints on the target word. This will not be the case for first language users andmay not be the case for fluent second language users, especially with high frequencywords.Although each of the previous hypotheses predicts differences between theprocessing systems of native speakers and fluent second language speakers, they do notaddress whether the use of lexical knowledge is similar in both cases. The secondresearch question is then:b). Is the role of lexical knowledge during the spoken word recognitionprocesses of fluent second language users comparable to that of nativespeakers?Broadly speaking, there will be comparability in the two systems in that both will makeuse of some sort of lexical knowledge. As argued above, for fluent second languageusers the contrary case seems to impose such a heavy load on listening comprehensionas to be unrealistic. The integrated lexical development hypothesis proposes that thedifferent types of properties, including the phonological, are highly interdependent, thatthey ‘assemble’ uniformly and evenly with no particular property or attribute becomingmore well defined or more highly functional in the word recognition process thananother at any time. On the other hand, the semantic dominance hypothesis predicts thatnovices will, after quickly mastering sufficient phonological code, initially learnsemantic properties of a word and that these wifi play a leading role in the wordrecognition process. Thus, a learner’s systems will not be comparable to an expert’s inthat the specific types of lexical properties they can make use of during word49recognition will be different. The extent to which a second language word recognitionsystem will converge in functionality on a first language system may depend on age-dependent limits to second language learning. Johnson and Newport (1991) haveproduced evidence that adult learners of a second language do not master at least someaspects of universal grammar. Possibly, then, this dominance of semantic propertieswill persist even in fluent users but it will be decreased as word knowledge (andgeneral proficiency) increase.This interaction of differential use of lexical knowledge with proficiencyintroduces the final research question.3. What is the nature of the development of the word recognition system asit moves from learner to fluent user?Each of the specific hypotheses proposed under a) and b) above can be restated interms of interaction with different levels of proficiency and with different levels ofword knowledge. First, the delayed access hypothesis predicts that word recognition timewill be faster across conditions for fluent users than for advanced learners. Similarly,recognition time to high frequency words will be faster than recognition time to lowerfrequency words. Second, the information processing hypothesis predicts that asproficiency increases, the load on the system from constructing higher levelrepresentations will decrease. Consequently, the recognition time difference betweenthe load and no-load conditions will decrease with proficiency and will be smaller forhigh frequency words than for low frequency words. Finally, as suggested in thepreceding paragraph, the semantic dominance hypothesis predicts that learners of a secondlanguage will first show effects of semantic constraints on words. As proficiencyincreases toward fluency (or as word familiarity increases), though, syntactic constraints50will increasingly affect the speed of word recognition.51Chapter IV. MethodA. Subjects and DesignTable 2 presents the basic design of the research. Sixty students, twenty for eachof three language proficiency levels participated as subjects for the present study.Table 2Research DesignFamiliarity High Frequency Words Low Frequency WordsContext Type N P Se Sy RO N P Se Sy ROLanguage Native Eng.Proficiency SpeakerLevel — —Fluent ESLUserAdvancedESL Learner: normal context F: pragmatic anoma y Se; semantic anomalySy: syntactic anomaly RO: random orderThe overall age range was 17 to 45 years (mean age=24.9) with no significantdifferences among the language proficiency levels (Means: native English speaker, 25.3years, a range of 19 to 39 years; fluent ESL user, 26.8 years, a range of 19 to 45 years;advanced ESL learner; 22.6 years, a range of 17 to 32 years). Gender was evenlyrepresented from each language group. No subjects reported any hearing defects.Native English speakers (Native Speakers) were recruited from graduate orundergraduate level educational courses at the University of British Columbia (UBC).All ESL subjects spoke Cantonese as their first language. Subjects in the fluent ESL user(Fluent Users) level of proficiency were UBC students who were fully enroled in52content courses (i.e. not attending supplementary language courses). Subjects in theadvanced ESL learner (Advanced Learners) level were recruited from collegepreparatory English dasses, at a local community college. This college has an extensivein-house language testing program which is used to govern movement from one levelto the next of a ten-level system. Promotion from the highest level confers B.C.provincial credit for Grade 12 English and Grade 11 Social studies. Subjects werestudents in the last two terms of this program.B. Language Materials1. Test sentencesForty nouns were chosen from Francis & Kucera (1982). Because it wasnecessary for timing purposes to accurately determine the beginning of each of thetargets, only words beginning with a stop consonant (eg. lb!, It!, /d!, etc.) werechosen for the test sentences. Subject to this constraint, one half of them wererandomly selected from within the adjusted frequency rank range 300 to 1200, and theother half were selected from the range 3500 to 4500. The mean adjusted frequency ofthe high frequency words was 127.7, and the mean of the low frequency words was13.623, a significant difference, F(1,39) = 63.8, p <.001 (MSe=2091.820). Using each ofthese as a target repeated over the five different sentence types (see Table 3), forty setsof stimulus sentences were constructed, resulting in a 200-item corpus (40x5, seeAppendix A). (Strictly speaking, the test items of contexts 3, 4, and 5 are not sentences,but to simplify discussion they will be referred to as such.) As can be seen in Table 3,each stimulus sentence pair from contexts 1 to 4 starts with the same lead-in sentence53but ends with a different test sentence. The stimulus string in context 5 is a scrambledset of the words from both the lead-in sentence and test sentence of context 1 of eachset.Table 3Examples of Test Sentences: Target word bubblesSentential Lead-in TestContext Sentence Sentence1. Normal Lynda was playing in the garden. She watched the bubbles floating in the air.2. Pragmatic Lynda was playing in the garden. She ate the bubbles floating in the air.Anomaly3. Semantic Lynda was playing in the garden. She interviewed the bubbles floating in the air.Anomaly4. Syntactic Lynda was playing in the garden. She crawled the bubbles floating in the air.Anomaly5. Random floating the Lynda watched the in was she bubbles playing the garden in airOrder2. Filler materialExcept for context 5, the test sentences follow a very regular pattern and thiscould result in subjects developing conscious or unconscious response sets. Forexample, the task in all test materials is to monitor for the exact word. In addition, thetarget words are all nouns beginning with stop consonants and the location of thetarget word is usually about 9 to 14 words from the beginning of the stimulus. Mostimportant, the target word is always the second word after the main verb in the testsentence. In order to decrease the likeithood of subjects developing guessing strategiesbased on these regularities, eighty distracter sentences were constructed to add to the54set of test sentences. To divert attention from the use of exact-word monitoring in thetest sentences, two other tasks are introduced, category and rhyme monitoring(Marsien-Wilson et al. 1988; Marsien-Wilson & Tyler, 1980) (see Table 4).Table 4Example Filler Material (target word underlined)Category: A Kitchen UtensilThe large spoon is often misplaced. It should be kept in the drawer.Rhyme: trueShe couldn’t undo the water. She left it on the shelf.In the category monitoring task, subjects listen for an example of the specifiedcategory (e.g. Kitchen Utensil--spoon). The material for this task was taken from Battig& Montague (1969) and 30 items were constructed (see Appendix B). In the Rhymingtask, subjects listen for a word which rhymes with the given cue word (eg undo: true).Inspiration for rhyming fillers was from Holofcener (1960), and Johnson (1957) and 30items were constructed (see Appendix C). Following Marslen-Wilson et al. (1988), wordtype and word position have also been varied in the distracter sentences. In addition,so has the type of initial sound. As in that research, the target words in three out ofthe five contexts are anomalous. To conceal this regularity, the distracter sentencescontain anomalous words and other oddities that are not targets. Again followingMarsien-Wilson et al. (1988), twenty filler items using the exact-word task but otherwisefollowing the above criteria were also constructed (see Appendix D). The sole purpose55of the filler items was to discourage any guessing strategies on target words. Theywere not the source of any data.3. Example and Warm-up ItemsIn addition to the test sentences and the distracter items, 9 example (seeAppendix E) and 18 warm-up items (see Appendix F) were constructed. The exampleitems presented each task three times with the target word appearing in a variety oflocations and in a representative set of contexts. The warm-up items were alsorepresentative of both the filler and the test material but were presented in a morerandomized fashion.C. Production of Language Stimuli and Computer ProgramThe materials were recorded on a reel-to-reel tape-recorder at 7 1/2 ips. Thespeaker was a professional actress who was not aware of the purpose of theexperiment. The target words were not indicated in the recording script. The materialswere then digitized at 22Khz using Voice Editor II (Creative Labs, 1991) and eachstimulus sentence was saved in an individual sound file. For test sentences of types 1to 4, the target word and all words following it were excised from the file, One of theexcised sections was duplicated and appended to each of the truncated test sentences.In this manner presentation of the target word was identical across the members of eachset of test sentences.A display/timing program was developed so as to control the presentation ofthe stimulus materials and time the subjects’ responses. Using a sound-file editor, adigital marker was placed within each test item file at the onset of the target word.This marker was used by the display/timing program to start a counting loop within56the program. On the two 33MHz, 386 DOS computers used in the present study, morethan 17,000 iterations of the ioop were equal to one second. That is, one iteration wasless than 0.1 msec. The two computers were calibrated using 10 different one-secondsound clips with flags at the beginning and end of each. The mean number ofiterations for the respective computer was used to calibrate reaction latencies in msecs.D. ProcedureAs each subject arrived at the experiment room, s/he was asked to write a 50-item doze test (see Appendix G) and to fill in a questionnaire form for languagebackground and biographical information. Subjects were allowed a maximum of 40minutes to complete the doze test. Cloze tests tend to correlate well with othermeasures of language proficiency and it was felt that such a measure would be helpfulin illustrating the global language ability differences among the three functionallydefined language proficiency levels.Furthermore, the doze test as a test type is typical of a general class of “off-line’,retrospective type of language proficiency measures. If the skills measured in such atype of test are related to the on-line, high speed processes which we wish to measurein the word-recognition paradigm, then scores on this test should allow for a moreprecise look at relationship between language proficiency and word recognition. Onthe other hand, if the skills measured in the two tasks (spoken word recognition andthe doze test) are different, then we can not expect to find a significant relationshipbetween the two.The material used by the display/timing program is cued by a separate data file57containing a list of the stimulus files in the order that they are to be presented. Five ofthese cue files were created in order to counterbalance presentation of the words overthe five sentence contexts. Each file presented all of the example and warm-up materialand 40 of the 200 test-sentences in random order mixed with all of the 80 filler items.Each of the cue ifies presented eight different target words, four high frequency andfour low frequency, under each of the five contexts. Target words were heard once andonly once by each subject, but were presented under all five contexts. Four subjects,two male and two female, from each of the three language proficiency levels weretested on each of the five cue files for a total of 60 subjects (=4x3x5).The subjects were tested in a quiet room using a personal computer equippedwith a sound-card, and closed-ear headphones. Before each session, subjects wereintroduced to each task type through the example items. They were told to respond asquickly as possible, to avoid guessing ahead or trying to anticipate the location of thetarget word and not to push the button more than once during a trial. When they hadcompleted the example items, subjects began the session proper, starting with thewarm-up items and then moving through the items as arranged in the cue file.The computerized display/timing program mentioned earlier was used todisplay target words and to time responses. Subjects faced the video screen andpressed the space bar to begin a cue sequence. The program displayed the task type(exact, category, or rhyme) in the middle of the screen. Below this, in the samelettering, it displayed the cue word. After a three-second pause, it began playing thevoice ifie containing the appropriate material. If the material was filler, the programplayed the voice ifie and then waited for the subject to signal readiness for the next58item. No data were recorded.If the material was a test item, the program started the voice file, waited for themarker of the target-word onset and then began timing. When the subject pressed thespace bar, the timing stopped. If the subject pressed the bar before the onset of thetarget word, the value for that trial was set at -1. If the subject failed to press the bar atany time during the presentation the sound file, the value was set at -2. This was donein order to make a distinction, if necessary, between missed (unrecognized) targetwords and over-anticipation of a target word.At the end of each trial, the program wrote the appropriate data to a file openedfor that subject. The data consisted of a code for the particular target word, a code forthe context it was presented under, and a value for the reaction time. The latter was-1, -2, or the total number of iterations of the loop from word onset to bar press. Wheneach session task was completed, the subject’s participation was acknowledged withappreciation.E. Data Screening and PreparationPeriodically during data-gathering, subjects’ response files were reviewed withthe intent of finding unusual or unacceptable cases. As a result of this, two atypicalcases were identified. In one case, the subject had an exceptionally low score on thedoze test. This subject was in the Fluent User pooi but a review of her biographicalinformation revealed that she was a post-doctoral student. There are no languagerequirements for admission of post-doctoral students in UBC, and it appears that hergeneral language proficiency was probably lower than that of any of the advanced59learners. The second case was a subject who apparently did not understand theinstructions or was very tired during the session. This subject did not respond to 11 ofthe 40 target words. These two subjects were rejected and substitutes were found forthem.Missing values occurred in the data if subjects either pressed the bar before theonset of the target word or not at all during the trial. In the final set of 2400 responses,there were 26 missing values. There was no discernible pattern to these data. Theywere replaced with the mean of the remaining replications for the subject within theparticular cell of Word Frequency by Context.In order to identify and replace extreme outliers, a series of 40 regressions wasrun, one for each of the four repetitions within each of the 10 cells of Word Frequency(2) by Sentence Context (5) conditions. In these regressions, the dependent variable wasresponse latency, while language level was a categorical predictor, and target-wordlength was used as a covariate. All raw data points with studentized residuals greaterthan 2.5 were replaced with the regression estimates. Thirty-two values were identifiedin this manner. These procedures resulted in a total of 58 (2.4%) values out of theentire 2400 which were estimated rather than measured.Subjects responded to four words under each of the ten context by wordfrequency conditions. The mean of these four replications was used as the unit ofanalysis of the response latency data. That is to say, each subject’s 40 responses werereduced to 10 mean values, thus reducing the original 2400 responses to 600 meanresponses. The research design called for analysis of these means using a multivariateanalysis of the 60 subjects’ 10 repeated measurements. Consequently, these60measurements were treated as 10 variables and further investigated for outliers anddepartures from parametric assumptions. Inspection of density plots, maximum andminimum standardized scores, and skewness estimates on each variable by languagelevel indicated that some of the within cell distributions of the Native speakers tendedto be slightly positively skewed. However, analyses of the residuals from regressions ofeach of the 10 latency variables onto language level and word length did not indicatethat a transformation of the data would be necessary. The Bartlett’s test forhomogeneity of group variances did not suggest there were any major violations of thisassumption. Mahanalobis distances did not reveal any multivariate outliers. Afterthese screening procedures were completed, analysis of the data in terms of thehypotheses was commenced.61Chapter V. ResultsA. Review: The Components of the Extended ModelThe present research is concerned with the extension of the cohort model ofword recognition to second language comprehension and learning. Two importantassumptions of the model as outlined by Marsien-Wilson (1989) are: 1) that syntacticand semantic properties of a mental lexical entry can function to facilitate spoken wordrecognition and 2) that spoken word recognition is a function of the initial level ofactivation of a word, which in turn is a function of the frequency of exposure to it inthe general language environment. In order to extend the cohort model to account forsecond language processing, it is necessary to interpret the consequences of theseassumptions within the context of differences in language proficiency. Before dataanalyses are conducted, the assumptions underlying the definition of the languageproficiency level of subjects in this research need to be examined.In this research, language proficiency level was functionally defined and threeproficiency groups were investigated: native speakers of English (Native Speakers),fluent ESL users (Fluent Users), and advanced learners of ESL (Advanced Learners).Table 5Cloze Test Score Means and Standard Deviations by Language Proficiency GroupAdvanced Fluent NativeLearners Users SpeakersMean 17.471 25.150 32.65Stand. Dev. 3.99 5.86 4.94562The results from the doze exercise support this functional defirition, suggesting thethree samples differed significantly in their global or general English proficiency,F(2,55’)=41.76, p<.OO1 (MSe = 24.93). Although the group means are well separated(see Table 5), there is overlap (see Figure 1) in scores even between the Native Speakersand the Advanced Learners, suggesting the doze scores may be a fruitful avenue offurther investigation into the relationship between language proficiency and wordrecognition speed.Score50140 11 51 41 330 1 5 21 21 3 21 620 655 2110Advanced Fluent NativeLearners Users SpeakersNumbers indicate frequency of doze score.Figure 1. Distribution of Cloze Scores by Proficiency LevelThe two ESL groups were also different in the amount of time they reportedhaving spent in an English immersion context. The mean of the Advanced Learnerswas 14.8 months (range = 33), while the mean of the Fluent Users was 61.6 months(range = 132). There were, however, no significant differences among the mean ages ofthe three groups, F(2,57)= 2.62, p> .05 (MSe=4.226).1 Two doze scores were missing from the data set.63B. Analyses1. Analysis of Response LatencyThe means of subjects’ responses to the replications within cell were analyzedusing a multivariate analysis of the 60 subjects’ 10 repeated measurements. Thebetween-subjects factor was language proficiency level and the two within-subjectsfactors were word frequency and sentence context. Table 6 presents a summary of themean response latencies under all conditions.Table 6Mean Response Latencies of Three Proficiency Levels in High and Low FrequencyWord ConditionsCONTEXTLanguageWord_________________________________________Group Frequency Normal Pragmatic Semantic Syntactic ListAdvanced Low 393.2 437.9 429.6 460.8 475.9Learner High 356.8 392.6 396.3 384.7 400.1Fluent Low 420.9 445.2 439.6 448.2 449.5User High 337.3 376.9 388.3 400.4 431.0Native Low 278.3 322.8 346.9 352.3 349.3Speaker High 267.9 286.4 313.2 314.4 323.0Note: Times rounded to nearest .1 msec.a. Differences among the Language Proficiency GroupsThe delayed access hypothesis predicts that word recognition will take longer forsubjects who are less proficient. In the present study, that would suggest that NativeSpeakers should respond more quickly to target words than ESL subjects in general,64450400350,250200150100500Figure 2. Mean Response Latencies of Language Proficiency Levelsand that Fluent Users should recognize spoken words more quickly than AdvancedLearners. Figure 2 summarizes the data related to this hypothesis. A contrast of meansmade between the mean of the Native Speakers (M = 315.5) and the combined means ofthe ESL groups (Fluent Users, M = 413.7; Advanced Learners, M = 412.8; combined, M= 413.3) was significant, F(1,57)=29.55, p<.OO1, MSe=43159.05. However, as might beexpected from looking at Figure 2, the contrast between the means of the Fluent Usersand Advanced Learners was not significant. The delayed access hypothesis predictsresults that would be analogous to those of the Cloze test, where the distinct differencesbetween each of the three proficiency groups reflected the differences in functionalproficiency of the groups. The results here do not completely conform to this, since thetwo ESL groups showed no reliable difference in response latency. This suggests thateither the hypothesis itself is incorrect or that the doze task and the word recognition65Learners Users NativesLanguage Proficiency Leveltask assess two quite different aspects of language proficiency.b. Word FamiliarityThe cohort model predicts that high frequency words will be recognized fasterthan low frequency words because the higher frequency ones are at higher state ofinitial activation. For the purpose of the present research this prediction has beenlabelled the word familiarity: activation hypothesis. The difference in response timesbetween the high (M=358.0) and low (M=403.4) frequency word conditions (see Figure3) was significant, F(1,57)= 144.32 p < .001 (MSe=2142.98), which provides tentativesupport to the extension of the cohort model from the population of first languageusers to the larger, combined population of both first and second language users.500450400350300‘250200150100500Figure 3. Mean Response Latencies to High and Low Frequency WordsHowever, further analysis of the data revealed a source of variation in response latencythat was not directly related to the fundamental cohort model, but which couldLow HighWord Frequency66influence the interpretation of results. This was the length of the spoken target word(as measured from onset to offset in hundredths of a second).Although it may not necessarily be the case that longer words will take longer torecognize than shorter words, in this data set there was a positive relationship betweenthe length of each of the 41 recorded words and the mean response time of all thesubjects to that word, T(=41) = .79 p < .001. This effect of word length has been reportedin previous research (Grosjean, 1980; Marsien-Wilson and Welsh, 1978; Marsien-Wilsonand Tyler, 1980). Furthermore, it was found that the length of the spoken target wordsin the high frequency group of words (M = 367 ms) was significantly different from thelength of the low frequency words (M = 435 ms), F(1,39) = 7.523, p <.01. The potentialeffect of this difference in spoken target word length is in the same direction as thedifference predicted by the word frequency effect, and therefore in this research context,the effect of word frequency is inherently contingent on spoken word length.To investigate the word length effect further, an additional analysis wasundertaken. Individual regressions were run for each subject’s 40 responses, with wordlength as a predictor. The ten within-cell means of the residuals from these 60regressions were then calculated and analyzed using the same repeated measures modelas described above. In this analysis, the effect of word frequency on response latencyremained significant, F(1,57)=15.05, p<.00l, MSe=1,541 .98, supplying additional supportfor the word familiarity: activation hypothesis.As a result of the finding that target-word length appeared to influence responselatencies, the remaining analyses used word length as a covariate to achieve a statisticalcontrol over word length when assessing the effect of word frequency.67c. ContextThe cohort model predicts that increased word recognition time will result fromdecreasing the amount of pragmatic, semantic, and syntactic lexical information in thespoken message that links a target word to the preceding context. The profilepresented in Figure 4 tends to support this. The test of the within-subjects factor ofsentence context was significant, univ. F(4, 227) = 33.24, p < .001, MSe=2028.75); PillaiTrace = 0.77501, multiv. F(4,50) = 43.058, p < .001, indicating differing use of contextualinformation during the recognition process.A more fine-grained analysis of the effect of context can be achieved by looking500300250 I I- o>. c0cE aEo o ---oz cc aContextsFigure 4. Mean Response Latencies of All Subjects across Sentence Contextsat how progressively increasing contextual information in the spoken message affectsthe recognition time. Target words in both the syntactically anomalous context and therandom list context have no contextual information which can facilitate the recognition68process. Thus, there should be no real difference between them in recognition time.This is borne out in the analysis, there being no significant difference between the meanresponses in these contexts, F(1,53) = 3.238, p>.O5, MSe=2379.720.While the syntactically anomalous context and the random list context have nocontextual links to the target word, the semantically anomalous context does imposesyntactic constraints on the target. Adding this syntactic constraint to the messageshould decrease response time. When the mean response under the semanticallyanomalous context is compared to the mean of the combined syntactically anomalouscontext and the random list context, there is, as predicted, a significant difference(estimated to be 13.48 ms, F(1,53) = 7.845, p<.Ol, MSe=1849.775). Similarly, adding thesemantic constraint which is inherent in the pragmatically anomalous context should alsoreduce response time. When the mean of the pragmatically anomalous context iscompared to the mean of combined semantic, syntactic and random contexts, there isanother significant drop in response time (estimated to be 17.69 ms, F(1,53)=15.820,p<.OOl, MSe=1779.680). The most significant difference in mean response times, though,is between the mean of those in normal conditions and the mean of those in thecombined anomalous contexts (estimated to be 47.80 ms, F(1,53)=143.355, p<.OOl, MSe1529.849). These results supply further support to the contention that the explanatoryscope of the cohort model can be broadened to include word recognition in a secondlanguage.d. Language Proficiency and Context, InteractionIt was hypothesized that, when the cohort model is extended to the secondlanguage context, language proficiency will interact with sentence context to affect word69500450-400 - - -- LearnerE35cJ —-User:: INativeO E Eo‘-O ÷-0Z CD0 C C Co 9< < cContextFigure 5. Mean Response Latencies of Language Proficiency Levels across SentenceContextsrecognition time. Specifically, the semantic dominance hypothesis and the informationprocessing hypothesis predict that certain sections of the different groups’ profiles acrossthe contexts will not be parallel. The semantic dominance hypothesis proposes thatsemantic aspects of lexical attributes will be learned more quickly by second languagelearners than syntactic aspects. This would result in the semantic attributes becomingfunctional more quickly than syntactic ones in the word recognition processes of secondlanguage learners. The information processing hypothesis proposes that the extra load puton novice word recognition systems when constructing high level representations of thespoken message will interfere with word recognition. Indeed, Figure 5 suggests thereare some differences in the way the language groups make use of available contextualinformation, particularly between the Native Speakers and the ESL subjects as a whole.70However, neither a comparison of the ESL subjects with the Native Speakers nor acomparison of the Fluent Users with the Advanced Learners was significant. Thisfinding supports neither the semantic dominance hypothesis nor the information processinghypothesis. It does, however, concur with the integrated lexical development hypothesis.e. Language Proficiency and Word Frequency, InteractionThe word familiarity: activation hypothesis predicts that all three languageproficiency groups will recognize higher frequency words faster than low frequencywords because in all cases greater exposure to words leads to a higher initial level ofactivation. This has already been supported in the above analysis. The word familiarity:learning hypothesis, on the other hand, argues that at lower levels of languageproficiency, the lexical entries for low frequency items are not as likely to be as fullydeveloped as those for high frequency items. At higher levels of proficiency, evenlower frequency words are more or less completely learned. Consequently, thishypothesis predicts that the difference between mean response time to high and lowfrequency words will be greater for subjects with lower proficiency.In Figure 6, the relative slopes of the three lines present a graphical method ofassessing this hypothesis. Apparently, the two ESL groups are almost identical in theway word frequency affects their word recognition times, there being no significantdifference between the differences between response times in the high and low wordfrequency contexts of these two groups, F(1,57)<.O1, p>.9, MSe=2142.98. On the otherhand, as can be seen by comparing, in Figure 6, the slope of the Native Speakers withthe almost coincident slopes of the ESL speakers, there was a difference between theway the ESL speakers and the Native Speakers were affected by differences in target71500450[4c0 I--- -- Learner.E35o-UserNative250 IHigh LowFrequencyFigure 6. Mean Response Latencies of Language Proficiency Levels to High andLow Frequency Wordsword frequency, F(1,57)=9.50, p<.004, MSe=2142.98. These results indicate that exposureto different levels of word frequency results in differential word recognition speedsbetween the ESL subjects and Native Speakers. It does not, however, lend support tothe word familiarity: learning hypothesis, since, despite an almost four-year difference inimmersion time (i.e. learning time), the two ESL groups show no overall difference ineffect of word frequency.f. Context by Word FrequencyThe word familiarity: activation hypothesis predicts that there will be no interactionof context and word frequency. The initial level of activation of a word is a function ofa subject’s general exposure to it, and this initial level of activation defines a base ormean which context effects will add on to or subtract from. On the other hand, word725c0iE300250 I Ie> c> c2 E—-4-_ -+-2 00 c0 oc oE cE 0.o --o C—iz g cCI)< Cf)<ContextFigure 7. Mean Response Latencies of All Subjects to High and Low FrequencyWords across Sentence Contextsfamiliarity: learning predicts that there will be an interaction. This hypothesis states that,to some extent, learning of a word is facilitated by exposure to it. The pragmatic,semantic and semantic attributes of low frequency words are therefore less likely to befully developed than those of high frequency words. Consequently, low frequencywords are less likely to show a context effect since subjects will not be able to usecontextual information to recognize a word as they will not have the relevant attributeswell defined in their lexical entry for that word. The data do not support the learninghypothesis (see Figure 7), as there was no significant difference in segment slopes alongthe two lines.g. Three-way Interaction: Word Frequency by Context by Language ProficiencyThe three-way interaction term is of particular interest because it relates to both73500450 — - - A -- Learners Low--_--;A - -. - - - -. - - - - Learners HighE- —A- Users Low3(X) — - Users High250 A Native Low200 I I I I Native High-ö c> c) c- Sf 4- — -I- rc -I-- rC 00 C 0ES 02 aE osC—iz oO sg cg o< -< cContextFigure 8. Mean Response Latencies of Three Language Proficiency Levels to Highand Low Frequency Words across Sentence Contextsthe semantic dominance hypothesis and the information processing hypothesis. Both of thesepredict three-way interaction. However, inspection of Figure 8 does not support eitherof these specific hypotheses.In the syntactically anomalous items, there are no syntactic contextual links tothe target word. In the semantically anomalous items there are, but according to thesemantic dominance hypothesis, novice ESL speakers will not have learned enough of alow frequency word’s syntactic attributes to make use of them during the recognitionprocess. Thus, this hypothesis predicts that the semantic to syntactic segment of theAdvanced Learner low frequency profile will be flat or relatively so. On the other hand,Native Speakers will “lose “functional syntactic information when faced with thesyntactically anomalous items as compared with the semantically anomalous items and74so this segment of the Native Speaker high frequency profile should show a slopewhich rises from semantic to syntactic. In the present research, these two segments areactually reversed from these hypothesized aspects.The information processing hypothesis states that construction of higher levelrepresentations will draw on limited processing resources. In the case of items in thesyntactic anomaly context, it is possible to construct a meaningful message, at least upto the target word which is not linked at all to the prior context. In the case of items inthe random list, there is no message and so no processing resources will be expendedin an attempt to build one. The information processing hypothesis predicts that lowproficiency ESL subjects will have slower recognition times in the syntacticallyanomalous items than in the random list because of this drain” of processing resources.Higher proficiency ESL users and native speakers will not experience this because theirlanguage processes are more highly automatized. As a result, the information processinghypothesis predicts that the slope of all syntactic to random-list segments would either befalling (for subjects with lower proficiency) or flat (for subjects with greater proficiency).All but the Native-Speaker, low-frequency segments rise, and this one falls, which iscontrary to its hypothesized flatness.However, there is a significant three-way interaction present. While acomparison between the combined ESL speakers and Native Speakers shows nodifferences in the four slopes (see Figure 9), a comparison between the two ESL groupsdoes (see Figure 10), Pillai Trace = 0.30589, multiv. F(4,50) = 5.51, p < .002. Inspectionof Figure 10 indicates that the centre of this interaction appears to be the change incontext from a condition where there is at least some syntactic contextual link to the75500-- A— — A - - Combined ESL-. Lowp400- Combined ESL0 - HighE350 .A Native Low300• Native High250 I I I —1- Q> 0>.- O 22az< cContextFigure 9. Combined ESL and Native English Speakers Response Latencies to High& Low Frequency Words across Sentence Contexts-A460—-A40A420400 A , - - -. — - - - - A- - Learners Low380E50 ,‘ / - - - - Learners High340 — A - Users Low320—- Users High300 I I- Q>.- O> o 2o 22 cE cE o.z O 20 -0c—iccContextFigure 10. Mean Response Latencies of ESL Learners & Users to High and LowFrequency Words across Sentence Contextstarget word (semantic anomaly) to one where there is none (syntactic anomaly andrandom list). When the syntactic anomaly and random list contexts are combined intoa single category, defined as items having no syntactic link between the prior contextand the target (“No Context”, see Figure 11), this becomes more apparent. This76interaction effect appears to be as follows. When there is any prior contextual link atall to the target word, the effect of word frequency is greater on the Fluent Users thanon the Advanced Learners. When there is no such contextual link, this is reversed. Apost hoc test comparing the semantically anomalous items and the combined ‘NoSyntax items was significant, univ. F(1,56) = 6.37, p < .02. This kind of interactionseems to suggest more of a re-structuring of rules or knowledge than a learning of newknowledge.480460 -_A440 -A- --- -A - A - - Learners Low420 -E - • - - - - Learners HighE -, A- A - Users Low360—- Users High320300 I.5 .Q>.. .c2 5 o o.22 zZ OC a)cn<ContextFigure 11. Mean Response Latencies of ESL Learners & Users to High and LowFrequency Words across Sentence Contexts (Syntactic and Random Combined)2. Supplementary AnalysesThe three language proficiency subject groups differed by design on the natureof their language proficiency. The Native Speakers spoke English as their first77language. The two ESL groups differed on the functional academic use to which theywere applying their English. The Advanced Learners were in full- or half-time ESLclasses. The Fluent Users were attending university dasses taught in English. Thesethree groups incidentally differed on two other characteristics, their mean doze testscore and, in the case of the ESL groups, on the amount of time they had spent in anEnglish immersion context. Further analyses were carried out to determine whetherthese particular aspects of the groups were related in any significant manner to theirword recognition processes.a. Cloze ScoreAll three language proficiency groups were included in the series of analyseswhich looked at the relationship between doze score and word recognition latency.Adding syntactic and semantic contextual information to the cues decreased subjects’response time to target words (see above). Syntactic and semantic knowledge also playa part in deducing answers to the doze exercise. It would seem likely, therefore, that ifthere is a relationship between the doze score and response latency it should bestrongest in the contexts where the context is richest. That is to say, in the normal,pragmatic and perhaps semantic contexts. However, for the purpose of comparing theresults, the analysis was repeated in each of the contexts. The first analysis tested tosee if the regression lines of the doze score were parallel in the language groups. Therewas some evidence that this may not be the case. In the normal context and thesyntactically anomalous context, the interaction terms were marginally significant(F(2,54)=3.334, p<.O5, MSe=14,953 and F(2,54)=3.014, p<.O6, MSe=16,710, respectively).To investigate this further, Pearson correlation coefficients between doze score and78response latency were calculated within each level, across each of the contexts (seeTable 7).Table 7Pearson Correlation Coefficients between Response Latency and Cloze ScoresAdvanced Fluent NativeLearners Users Speakersn=20* n=20 n=20Normal -0.377 0.297 0.414Pragmatic -0.223 0.413 0.175Semantic -0.441 0.172 0.140Syntactic -0.345 0.358 0.242Random -0.207 0.302 0.015* 2 missing scores were replaced with group meanNone of the individual coefficients in Table 7 is reliably different from zero.However, the pattern of negative and positive coefficients suggests that the relationshipbetween the doze test scores and response latency may be different in the AdvancedLearners than in the Fluent Users and Native Speakers. It may be the case that, withinthe Advanced Learners, low scores on the doze are more indicative of generally lowproficiency than low scores in the other two groups. These low scores would beassociated with higher latency times and thus account for the negative correlations.Such an explanation is speculative of course.b. Time in an English Immersion ContextTime spent in an English immersion context is, of course, relevant only to theESL groups so the investigation of the relation of this measure to spoken wordrecognition was limited to the Advanced Learners and the Fluent Users. The initial79analysis strategy was similar to that used for the doze score. First, the analysis lookedat a repeated measures model which included language level, time in the immersioncontext, and a term for the interaction between these two. In this analysis, theparameter for this interaction term was significantly different from zero F(1,36)=7.510,p<.Ol, MSe=38,195. The implication of this is that the regression slopes of the “time in”variable are different in the two groups.In order to determine the nature of these unequal regression slopes, the data wasre-analyzed. From the range and the variance of each of the two groups (Users:range=132, var=1422.45; Learners: range=33, var= 93.139), it is fairly apparent that thetwo groups differ a great deal in their dispersion. A test of this confirmed that thevariances were not homogenous (Bartlett’s Chi-sq(1, N40) = 27.151, p<.OO1). Beforecontinuing, the time-in-an-immersion-context variable was log transformed and theresult used in subsequent analyses. Using the overall mean response latency for eachsubject as the dependent variable and log time in an English immersion context as theindependent variable, linear regressions were done within each of the two languagelevels. The slopes in these regressions differed in aspect (Fluent Users: b=-62.273,t(19)=-2.697, p<.O2; Advanced Learners: b42.102, t(19)=2.220, p<.O4), with wordrecognition time for Advanced Learners apparently increasing the longer they havebeen immersed, and decreasing under the same conditions for Fluent Users.This suggests a curvilinear relationship with an inflection point somewherewithin the middle of the combined group. For illustrative purposes, a distanceweighted least squares smoothing algorithm was used to determine a curve whichpassed through the data points. This was first done for the Advanced Learners and the80Fluent Users separately, then applied to the combined data set. The results areUser (*)— —— Learner (+)CombinedEnglish Immersion Time (log months)Figure 12. Distance Weighted Least Squares Smoothing for Mean Latency Time,Separately for Learners and Users, and Combined.displayed in Figure 12.6005O040Oa,C03002001 2 3 4 5 681Chapter VI. DiscussionIn order to facilitate discussion of the results and their implications, it will beuseful to present a brief review of the study. The primary objective of the presentstudy was to assess the usefulness of the cohort model of spoken (first language) wordrecognition as a method of explaining the processes involved in recognizing spokenwords during the comprehension of a spoken second language. The strategies used toachieve this objective were 1) to attempt to demarcate differences in the spoken wordrecognition systems of native English speakers and second language users, 2) tocompare the role of lexical knowledge during the spoken word recognition processes ofnative English speakers and second language users, and 3) to investigate the nature ofthe development of the word recognition system as it moves from learner to fluentuser. Two groups of ESL subjects and one group of native speakers of English (NativeSpeakers) served as subjects for the present study. The ESL subjects were characterizedas Advanced Learners, who were studying ESL at a local community college inpreparation for entrance to higher education courses and Fluent Users, who wereattending full-time academic classes taught in English at the University of BritishColumbia. The performance of these three groups was compared on a reading dozetest and a spoken-word recognition task in which there were five different levels ofcontextual richness prior to a target word. In order to find how the different subjectgroups handle recognition materials of different familiarity, two levels of target wordfrequency were presented.A. Summary of ResultsThe main findings can be summarized briefly by looking at 1) the overall results,822) the results from a comparison of the two ESL groups, and 3) the results from acomparison of the Native Speakers with the ESL speakers. Overall, the three groupsdiffered significantly in the way they performed on the doze test. The results werecongruent with the assumption that the Advanced Learners were not as proficient inEnglish as the Fluent Users, who in turn were not as proficient as the Native Speakers.Second, as the extended cohort model predicts, there was a significant overall effect ofsentence context on word recognition latency. Cues with greater contextual linkage tothe target word resulted in faster recognition latencies. Finally, again in accordancewith the predictions of the extended cohort model, there was a significant overall effectof word frequency on recognition latency. Higher frequency words were responded tomore quickly than lower frequency words, as expected.Between the two ESL groups, there were no significant differences in recognitionlatencies or recognition latency profiles across sentence contexts or across wordfrequency. This finding does not conform to the delayed access hypothesis, whichpredicted a difference in recognition time because of the putative overall difference inlanguage proficiency between the two ESL groups. The basis for the attribution of anoverall language proficiency difference was, in addition to their statistically differentmean scores on the 50 item doze test, the difference in their academic use of English.There was a three-way interaction of ESL group, word frequency, and sentence context.The pattern of this interaction did not conform to the patterns predicted by the semanticdominance hypothesis or the information processing hypothesis. It did indicate that the effectof word frequency on ESL group changed, depending on whether contextual links totarget words were available or not. In addition, there was a very interesting curvilinear83interaction of recognition latency with time spent in an immersion context. Thisinteraction clearly indicates that spoken word recognition speed actually decreases for aperiod of time, after the immersion experience begins, and then begins to increase withfurther exposure.Between the Native English speakers and the ESL subjects there was a significantdifference in overall mean recognition latency. This is in accord with the delayed accesshypothesis. There were no significant differences in the recognition profiles of theNative English speakers and the ESL subjects across sentence contexts. This findingdoes not support the semantic dominance hypothesis, which was based on the argumentthat semantic lexical attributes would be learned in a second language faster thansyntactic lexical attributes. Nor does it support the information processing hypothesiswhich argued that Advanced Learners would have different recognition profiles acrossContexts four (Syntactic anomaly) and five (Random list) than Native Speakers (andperhaps the Users) because of the higher processing load (up to the target) in thesyntactically anomalous context as opposed to the random list context. In contrast withthe comparison across the five sentential contexts, there was a significant difference inthe recognition proffles of the Native English speakers and the ESL subjects across wordfrequency.B. DiscussionAn examination of these individual findings in light of their theoreticalimplications, and the aims and strategies of the study itself is in order. First, the resultswill be interpreted with respect to their theoretical implications regarding thedevelopment of second language proficiency. Following that, the apparent differences84between the ESL and the native speaker word recognition systems will be considered.1. Developmental differences between Users and LearnersOne of the strategies used in this study was to attempt to investigate the natureof the development of the word recognition system as it moves from learners to fluentusers. In this respect, the similarity of the performance of the two ESL groups on theword recognition task was remarkable. There was less than 1 msec difference betweenthose groups’ respective overall mean response latencies. In addition, their responseprofiles across word frequencies and across sentence contexts were almost coincident.On the assumption that these groups represented different language proficiency levelsand thus had lexicons at different stages of development, the delayed access hypothesispredicted there would be a significant difference between the two groups. If these werethe only data available for analysis, one would be tempted to conclude that they weresamples from the same population and that they did not represent different proficiencygroups, to say the least, in terms of recognizing incoming spoken words.However, they do come from distinct subject pools. One group was stillinvolved in studying ESL, while the members of the other group were using theirEnglish to pursue higher educational academic goals. Furthermore, as groups, theydiffered significantly on the doze exercise. Finally, the Advanced Learners, for themost part, had not been immersed in the ESL context for as long as the Fluent Users.Only two Advanced Learners reported having lived in an immersion context longerthan the minimum time reported by the Fluent Users. Thus it would not be a validinterpretation to suggest that the two groups are from the same population.What the data clearly show is that, compared to the Native Speakers, the85Advanced Learners have a pervading inefficiency in their spoken-word recognitionprocesses, and that, despite almost four years of further second language immersionand experience, the Fluent Users have not perceptibly improved their word recognitionspeed, either overall or within the contexts.At the same time, though, it is not possible to say that the two groups haveidentical word recognition processes or strategies. The reason is that the two ESLgroups did differ in two respects, although their overall word recognition latencies, andword recognition latency profiles across word frequencies and across sentence contextswere very similar. In the analysis of the relationship between recognition latency andlength of time in an English immersion context, it was found that the regression slopesof the two groups had different aspects. By combining the two groups into one, andusing a non-linear method of analysis, it became clear that there was a curvilinearrelationship between these two variables. Word recognition latency apparentlyincreases for the first two to three years of immersion and decreases after that. Onepossible explanation for this is found in Effis’s (1994) discussion of U-shaped behaviour:Learners may sometimes pass through an early stage of development where theymanifest correct use of a target-language feature if this feature corresponds to anLi feature and then, subsequently, replace it with a developmental U featurebefore finally returning to the correct target-language feature. In such a case, thefacilitative effect is evident in the early stages of acquisition, before the learner is‘read’ to construct a developmental rule. The ‘re-learning’ of the correct target-language rule occurs when learners abandon the developmental rule as theycome to notice that it is incompatible with the input.While the data seem to fit this pattern, it is difficult to know what the ‘feature’ is that isundergoing this re-structuring. Looking at the other difference between the two ESLgroups sheds some light on this.In the analysis of the three-way interaction of ESL level, sentence context and86word frequency, it was found that the location of the interaction appeared to be wherethe sentential context changes from containing at least some syntactic contextual link tothe target word (semantic anomaly) to one where there was none (syntactic anomalyand random list). Ellis’ U-shaped behaviour can be simplified to three stages, 1)application of Li principle to U, 2) development of an interlanguage principle, 3)mastery of U principle. Under this simplification, performance in stage one would besimilar to that in stage three, while stage two would be different from both. If theAdvanced Learners are indeed at a lower developmental level than the Fluent Users,then a logical progression in terms of ESL level, and high and low frequency wordswould be for Advanced Learners to be in stage one for the low frequency words, andstage two for the high frequency words. The Fluent Users would be at stage two in thelow frequency and stage three in the high frequency words. If this is correct, we wouldexpect the performance of the Advanced Learners in low frequency words (stage one)to be similar to Fluent Users in high frequency words (stage three), and theperformance of Advanced Learners in high frequency words to be similar to FluentUsers in low frequency words (both stage 2). This indeed seems to describe the results.Thus, it appears that the use of both phonological and contextual information in thespeech signal is undergoing some kind of transition or restructuring over thedevelopmental period encompassed by these two groups.In summary, the data from the present research suggests three points about therelative development of the two ESL groups, Advanced Learners and Fluent Users.First, despite almost four years difference in immersion experience, there are no overalldifferences in spoken-word recognition speed between the groups. Second, from the87Advanced Learner to the Fluent User there appears to be a re-organization of the wayin which the incoming data, both phonological and contextual, is actually used duringthe recognition process. Third, there appears to be, in the Advanced Learner group, aninitial decrease in overall recognition speed with the passage of time, which is reversedin the Fluent Users by an gradual increase in overall recognition speed. The data donot allow extrapolation to a period of second language development prior to that of theAdvanced Learners or subsequent to that of the Fluent Users. In fact it may be quitemisleading to conceive of word recognition speed as being a continuous function thatcan be extrapolated much in either direction. Clearly, a word cannot be recognizeduntil adequate phonological data have been received and processed (see 3. ProcessingDifferences... below, for further discussion of this point). It may also be the case that, ifa word has not been recognized by the time most of its phonological data has arrived, itwill not be recognized at all. In other words, aside from variation due to individualdifferences in processing speed and attention, it may be that the time needed torecognize any particular word is constrained by specific ‘endpoint’ parameters, and theprimary learning job of the system, once it is capable of actually recognizing words, isto increase efficiency within those parameters.2. Developmental differences between ESL and Native speakers.In addition to emphasizing differences between the Fluent Users and theAdvanced Learners, the use of high and low frequency target words revealed adifference between the ESL groups and native speakers. The present study took anapproach to the word frequency effect that was different from that of the originalcohort model. For the second language learner, word frequency was hypothesized to88produce two effects. The first one, which was subsumed under the word familiarity:activation hypothesis, is equivalent to the effect of activation attributed to word frequencyunder the cohort model. In this case, increased exposure to words raises either theinitial level of their activation in the lexicon or else their susceptibility to activation.This heightened activation is not related to the completeness of the lexical entry. In asense, this effect can be thought of as a result of differences in language-basedenvironmental priming of words. The second word frequency effect was predicted bythe word familiarity: learning hypothesis. Under this hypothesis, increased exposure towords will result in the lexical information becoming more complete through learning.This learning should affect all properties of a lexical entry. Jnterestingly, MarsienWilson (1990) touches on the relationship between learning and activation, saying “inthis era of learning models it becomes increasingly implausible to suppose thatfrequency of experience is not somehow reflected in the basic organization of the waythe system responds to sensory inputs...’ (p. 149). It does not appear, though, that asimilar learning vs. activation distinction is made by him.In the present study, no strong evidence was found to support this distinction.The difference in frequency of the two sets of words had a greater effect on the meanresponse latency of the ESL subjects than on that of the native speakers by an estimated25 msec. Under a combined activation! learning model, this difference would beexplained by attributing the relatively slower word recognition speed of the ESLsubjects within the low frequency context to less-well defined lexical entries for thesewords. However the activation! learning model also predicts a difference between thetwo ESL groups (Fluent Users and Advanced Learners), as well as a difference between89the ESL speakers and the Native speakers and there was none. Unless some thirdfactor is introduced, such as a relation between on-going word exposure level andretention of word knowledge, the model does not fit the data of the present study.Even a simple activation model is inadequate, though, without an additionalexplanation as to why the “language-based environmental priming” of low frequencywords is not as effective for ESL speakers as it is for Native Speakers of English. Twobroad categories of such explanations would be learner-internal or learner-external. Inthe first category, the explanation must appeal to inherent differences in the actualprocess of activation of words within speakers of first and second languages. In thesecond category, an environmental factor is appealed to. One such explanation is thatpeople who speak a second language are in general faced with different wordfrequency levels than native speakers simply because, as second language users, theyevoke different speech uses and patterns with their native language speakinginterlocutors. The input frequency hypothesis (Ellis, 1994) invokes a similar concept toexplain the acquisition orders of grammatical morphemes and syntactic structures.Deciding among these various solutions will have to be left for future research.3. Processing Differences between the ESL and Native SpeakersCharacterizing the nature of differences in the spoken word recognition systemsof ESL and Native speakers was one of the goals of the present study. Chapter ifioutlined three hypothetical limiting cases toward which the development of the secondlanguage lexicon may move. The first proposed limiting case was that there would beno access to non-phonological lexical knowledge during word recognition, and that90word recognition times would be the same under the different contexts. This can berejected on the basis of the results found in this study. Increasing the strength of thelink between the prior context and the target word by supplying syntactic, semantic orpragmatic relationships between the two resulted in decreased mean word-recognitionlatencies. Similarly, a strict interpretation of the third proposal can also be rejectedbecause it predicted both parallel and coincident proffles for the ESL and native subjects.While the profiles were not significantly different in shape, the mean response for theESL groups was almost 98 msec. slower than that of the Native Speakers, indicatingsome inefficiency in the ESL system.This strongly suggests that the ESL subjects were making use of the pragmatic,semantic, and syntactic cues in the context but leaves open the question of where, in therecognition process, those properties of a word candidate become accessible forassessment against the higher level representation. In second language processing, dothese properties become available for assessment against the higher level representationas soon as a lexical entry becomes a member of the cohort of word candidates, as in thecohort model? Or, do the word recognition processes exhaust all phonological inputbefore relying, if necessary, on other lexical properties to recognize a word? Thedifference between the two cases, as mentioned, is where in the temporal presentation ofthe word the recognition takes place. This, however, introduces the question of what‘where’ means in relation to the general concept of word.One possibility is word offset, in which case, we are comparing response latencyto spoken word length. It is important to note that spoken word length is not atheoretical concept. It is the actual, measured length (in msec., for example) from word91onset to word offset. In the present study, this value was held constant for each wordacross the five different contexts by excerpting a single sound segment containing theword from one of the five recordings and then splicing it in the appropriate place in theremaining four cues. Word length varied from word to word, of course; however,under other experimental conditions, it would be possible to vary the length ofindividual spoken words using either speech compression (Conrad, 1989) or increasingthe rate of speech, (Griffiths, 1990).Another instance of the “where” of word recognition is the concept of recognitionpoint which Marsien-Wilson (1987) defines as the point in a word where it “can bediscriminated from the other members of its word-initial cohort, taking into accountboth contextual and sensory constraints”. He notes that this point can “occur wellbefore the end of the word”, (p.80). The recognition point can be estimated throughword monitoring and gating tasks. A third, related location, in the word recognitionprocess is the uniqueness point (Marslen-Wilson, 1990) or isolation point (Grosjean, 1980)which Grosjean defines within the gating task as:the point at which the listener has isolated a candidate but may still feel quiteunsure about it. He or she will therefore continue to monitor the acoustic-phonetic information until some criterion level of confidence is reached and theword is accepted or recognized.” (Grosjean, 1980, p.273).The difference between the recognition point and isolation point is the level of certaintyin the mind of the person who is monitoring the word. Marslen-Wilson (1984) alsodiscusses an optimal discrimination point which “can be defined, for spoken words heardin isolation, as the point at which a particular word becomes uniquely distinguishablefrom any other word in the language beginning with the same sound sequence.” (p.141). Lists of phonetically transcribed words are necessary to determine such a point,92the issue of which goes beyond the present data.For the purpose of the present study, which is to compare the systems of nativeand second language speakers of English, it is instructive to look at several of thesepoints within the two distinct groups (ESL, Native English). The mean response latencyof the ESL group under the fully contextualized (normal) cues was estimated atapproximately 377 msec. The mean spoken word length in this study, as measuredfrom word on-set to word off-set, was estimated to be about 399 msec. if it is assumedthat there is some finite amount of time that passes between when a subject makes adecision that the present word is the target word and when that subject actually pressesthe button to indicate recognition (Marslen-Wilson and Tyler, 1980, estimate responseexecution takes about 50-75 msec), then the ESL speakers’ recognition points are, underthe fully contextualized cues, somewhat before target word offset.However, the mean response latency of the Native Speakers in the twouncontextualized cue conditions (syntactic anomaly and random list) was estimated tobe about 335 msec. This can be taken as an estimate of the mean recognition point forthe word-list in the present study, under conditions where it could reasonably beassumed to be a direct function of the individual optimal discrimination points of thetarget words. We can see that the ESL group as a whole is recognizing words generallybefore word offset but after the point in the words where there is adequate phonologicaldata (for the native speakers) to distinguish the target word from the rest of thecompetitors in the cohort. On the other hand, 30% (12 out of 40) of the ESL subjectsrecorded mean response latencies in the normal context that were faster than the 335msec mentioned above. This seems to indicate that, at least in some cases, second93language word recognition processes can, like first language word recognitionprocesses, make rapid use of relevant lexical knowledge that is activated while a word isbeing heard. Also, it seems to indicate that the observed limiting point of efficiency inthe ESL groups’ second language spoken word recognition processes is at leastconsistent with the multiple access and multiple assessment processes that arefundamental to the cohort model (Marsien-Wilson, 1987).While it may be true that the ESL word recognition system can be effectivelymodelled by the cohort model, it is also true that the word recognition systems of theseESL subjects are not as efficient as those of native speakers. The cohort model may alsohelp explain this, at least to the extent of suggesting where the greatest inefficiency liesin the word recognition process. The major problem does not appear to be in theconstruction of that part of the higher level representation of the discourse upon whichthe assessment process relies to facilitate integration. If the problem resided here, wewould expect to see a difference in the recognition latency profiles of the two languagegroups across Normal speech and Random list, but we don’t. Tn other words, the majorsource of the relative ‘slowdown’ of the ESL group recognition latencies appears to beindependent of the process(es) involved in constructing the higher level representationsagainst which lexical information is compared during recognition. This suggests thatthe selection process is slowed down in the ESL word recognition process, perhapsbecause more phonological data is needed by the ESL subjects.The cohort model can be used to make a more specific analysis of what might begoing on here. As Marsien-Wilson (1993) points out, ‘Listeners know what the wordsin their language sound like. This knowledge, constituting what we can call the94recognition lexicon, defines the perceptual targets of the access process.” (p.187). Prior tolexical access, the physical sound is analyzed and transformed to create the input to thisrecognition lexicon. This input representation and the representation of lexical form in therecognition lexicon may be central to the relative inefficiency of the second languagelistener. As mentioned in Chapter II, the input to the word-recognition processes is notconceptualized as a string of phonemic labels, but rather as a set of feature values.Furthermore, in the cohort model, it is assumed “that the representations of lexical formin the mental recognition lexicon are structured arrays of features, where thespecification of the lexical item abstracts away from the detailed phonetic properties ofthe surface form of the word.” (Marslen-Wilson & Warren, 1994, p. 653). Warren andMarslen-Wilson (1987, 1988) showed that these features play an important role innarrowing the choices of potential target words. These features, and other modulationsof the speech signal, are “tracked in detail by the processes responsible for lexical accessand selection.” (Lahiri & Marslen-Wilson, 1991, p. 256). For example, the nasalisation ofa vowel, which for a language like English indicates that the following consonant willbe nasal, has an early effect on lexical choice, limiting the selection of potential wordsto those that fit this cue (Warren & Marsien-Wilson, 1987).Of particular relevance to the efficiency problem of the ESL subjects is the studyby Lahiri and Marsien-Wilson (1991). They exploited the fact that in Bengali bothvowels and consonants contrast in nasality whereas in English, only consonants do so.They were able to show that English subjects listening to English language cues wereable to anticipate an upcoming nasal consonant on the basis of its nasalising effect onthe preceding vowel. In contrast, Bengali subjects listening to Bengali cues interpreted95nasalised vowels which preceded nasal consonants as nasal vowels, and consequentlydid not begin to constrain the word choices until the beginning of the consonant. Inthe Lahiri and Marsien-Wilson cross-linguistic study, there was no investigation of firstlanguage! second language recognition interaction. That is, native Bengali speakerswere not tested with English cues, or vice versa for the native English speakers.However, it is reasonable to conjecture that, given the correct cue material, a differentialuse of the nasalized vowel cue could be found in native Bengali and English speakersrecognizing English and Bengali second language words. It is also reasonable toconjecture that, in the present study, the Cantonese speaking ESL subjects may alsohave been using a variety of word recognition cue strategies that were transferred fromtheir first language (see also B, part 1 above). Such speculation should be brought toempirical verification of course, but the point of making such a conjecture is todemonstrate the inherent value of adopting the cohort model of spoken wordrecognition as a method of generating hypotheses about spoken second languageprocessing and, eventually, second language acquisition.C. Factors Affecting GeneralizabilityA number of factors affect the generalizability and interpretability of the findingsof this study. Perhaps the most important is the educational background of the ESLsubjects. In order to avoid variation that might have arisen because of differences infirst language, only speakers of the Cantonese dialect of Chinese were invited to jointhe study. As a result, most of the ESL subjects had taken their early schooling (K-12)in Hong Kong, where most of the local Cantonese-speaking immigrant population96comes from. Many Hong Kong schools have an English as-a-foreign-languagecomponent, and, in fact, all of the subjects had studied English for at least a few hoursa week for two or three years, and many had studied it from kindergarten through toForm six (Grade 12-13). A few even reported attending classes where it was used asthe language of instruction. This background in English study may explain why theperformance of the two ESL groups on the word recognition task tended to be sosimilar despite the difference in their length of time in the immersion context. The twoESL groups did differ from each other in one aspect of their educational background.While several of the Fluent Users reported having attended two to three years of highschool in Canada, none of the Advanced Learners did. This may be related to the twosignificant interactions (ESL group by context by word frequency and ESL group bytime in immersion context) found in the comparison of these two groups, sinceotherwise the performance of the two groups seemed almost identical.The syntactic anomaly word monitoring results in this study were not consistentwith results from previous research. Marslen-Wilson, Brown, and Tyler (1988) reportedthe successive differences in mean response for their normal, pragmatic, semantic andsyntactic items to be -28 msec, -22 msec, and -29 msec, respectively. In the presentstudy, for the native speakers, the comparable estimated values were -31 msec, -26msec, and -3 msec respectively. The first two values in the two studies correspondwell, but the difference between response latencies of the semantic and syntacticmaterial in the present study is not comparable to that in the Marsien-Wilson et The importance of this is that the present research results did not really allowfor an assessment of the semantic dominance hypothesis. In order to do that, it will be97necessary to show that the supposed differences between the items in the semantic andsyntactic categories consistently produce response differences in native speakers over awide range of materials which conform to the criteria which define them.D. Implications for ESL Education and Directions for Future ResearchThe results of the present study brought forth some implications for ESLeducation and establish a starting point and a number of routes for future research.The study implicated input representation and the representation of lexical form aspossible sources of inefficiency in the word recognition process. This suggests thatthere is potential for improvement in word recognition speed, and consequentlycomprehension, by somehow overcoming these inefficiencies. The cohort modelassumes that the lexicon is accessed through structured arrays of features (MarsienWilson & Warren, 1994). Although the model is not specific about the nature of thesefeatures, possible examples of contrasting features would be voiced or unvoicedconsonants, and nasal or non-nasal vowels etc. Research has shown that first languageexperience modifies adult perception and that the ability to perceive these contrasts in asecond language is decreased and, in some cases, lost (Clifton, 1993; Werker, 1993). Inthe cohort model it is assumed that these features are abstracted from the surface form.If adult learners are to regain, or redefine the relevant features and contrasts, they mustbe supplied with adequate data to do so. It is important to point out that this does notmean pronunciation instruction on a word by word, or phoneme by phoneme basis.The surface representations of sounds of words vary a great deal, depending onnumerous factors ranging from the shape of the vocal tract, through dialect, to the98specific phonological contexts under which they are uttered (Lahiri & Marsien-Wilson,1991). Unless this variation is well exemplified in second language instructionalsituations by dealing with words in spoken discourse context and by presenting avariety of speakers dealing with the same material, it will not be possible for thestudent decide which aspects and variations of the surface sounds are idiosyncratic andirrelevant or redundant, and which are not. Without such instructional material, it isunlikely that students would ever be able to construct an efficient and fast wordrecognition system.It is also clear from both the cohort model and the results of the present studythat fully developed mental lexical entries include a variety of pragmatic, semantic andsyntactic information, which can be made available for exploitation by the wordrecognition system even before the word is identified. Unless all of these aspects ofwords are addressed during instruction, the ESL learner will not be able to developcomprehensive entries, and word recognition will not be as fast as it should be for easy,on-line comprehension of speech. Again, in order to allow students to extract therelationships among the various lexical features of different words, learning new wordsshould involve hearing them in meaningful context. The study also seems to suggestthat increased exposure to words has a positive effect on their learning and recognition.The message here is that students have to be encouraged to truly immerse themselvesin the spoken (and perhaps the written) language in order to actually become functionalinit.The present study suggests diverse routes for future research. One direction willbe the on-going study of word recognition processes and the fit of the cohort model to99the second language context. One particular study will be another look at the semanticdominance hypothesis. The present study did not support the semantic-syntacticdistinction found by Marsien-Wilson et al (1988) in the response patterns of the nativespeakers. Unless such differences can be consistently found in materials which conformto their criteria, it is difficult to judge whether this hypothesis is tenable or not.The study should also be extended to lower levels of second languageproficiency. The main reason that this was not done in the present study was that thedistracter items proved rather difficult for the less advanced learners on whom thematerial was pilot tested. For one thing, many students were unfamiliar with themeaning of the word “rhymet’. It should be possible, though, to precede the actualtesting session with a training program that presented both the category and rhymetasks more completely. It would be feasible to set a mastery criterion in this programso that subjects demonstrate a full grasp of the tasks before moving into the testingphase. In addition, a future study would profit greatly if more complete data (responselatencies, errors, number of trials to criterion, etc.) were gathered on the distracteritems. Through graded selection of the category targets, it should be possible to getsome estimate of the size of subjects’ passive, listening vocabulary. Such a studyshould be expanded to include other types of tests, particularly listening comprehensiontests, in order to lay ground-work for the development of a much-needed theory oflistening comprehension (Buck, 1992).The present study was concerned with a very specific aspect of speechprocessing in a second language, word recognition. Other studies have also beeninterested in the speed of processing in a second language (e.g. Conrad, 1989; Griffiths,1001990) but have focussed more on listening comprehension in a second language. It is notpossible to compare the level of English proficiency of the subjects in these studies withthat of the subjects in the present study. However, it is possible to make a roughcomparison of the speech rates. Based on the mean length of the spoken target word(399 msec), the mean speech rate would be approximately 150 wpm. As estimated by asample of ten of the cue sentences, it would be about 180 wpm. Both of these arewithin the range looked at by the Conrad and Griffiths studies. Comparing all threestudies helps focus on a major difference among them and suggests a fruitful avenuefor research on the relationship between spoken word recognition and listeningcomprehension.On the basis of the analysis of the data from the present study (see above),increasing speech rate (either through compression or through an actual increase inspeaking rate) should not change the relative word recognition latencies of native andnon-native speakers. According to the analysis, non-native speakers need more data,not necessarily more time, in order to recognize words. That is, increasing the rate offlow of the speech input data should increase the rate with which the data arrives atlexical access, and therefore decrease recognition latency proportionally for both nativeand non-native speakers. Based on her analysis of patterns of missed words, Conrad(1989) attributed the low comprehension rates of the non-native speakers to the fact thattheir processing capacities were overloaded “beyond where these participants’ syntacticexpectations could operate” (p.13). In the context of the present study, this might beequivalent to locating the major inefficiency in second language spoken-wordprocessing at the integration stage of the cohort model or even beyond this, within the101processes involved in constructing higher level representations, rather than at the accessstage.While Conrad’s conclusion appears to be at odds with an analysis made on thebasis of word recognition data, it is not in fact. There is an inherent feedback betweenaccess and comprehension under conditions of increased speech rate. If, for somereason, words are not identified properly because of increased speech rate, thencomprehension will suffer. When comprehension breaks down, the mentalrepresentation of the message will be disrupted and, as for example, in the case oftarget words in the pragmatic anomaly context, there will be a need during access formore data for word identification. However, since the precipitating problem in thissequence of events was lack of time for processing, the entire system will undergo evenfurther disruption, and in all likelihood become dysfunctional. Similarly, if the processof constructing higher level representations can not keep up with the speech rate, it willtake longer to integrate the accessed words. The higher level representation will befurther disrupted, requiring even more time at the access stage and once again thespeech processing system will break down.The analysis under the present data suggests that the breakdown of the system isnot precipitated from within the word recognition side of the process, but from thecomprehension side. It is possible, of course, that adult second language learners,because of maturational constraints on their language development Oohnson &Newport, 1989; Long 1990), have an inherently, and pervasively sluggish secondlanguage processing ‘platform’. Regardless, the source of the collapse of the systemcould be readily be located by combining a speech-compressed word recognition study102with a speech-compressed listening comprehension study.E. ConclusionThe purpose of the present study was to evaluate the feasibility of extending ofMarslen-Wilson’s (1989, 1987; Marsien-Wilson & Welsh, 1978) cohort model of spoken(first language) word recognition to a second language context. This model is atheoretical description of the fundamental, high-speed processes involved in spokenword-recognition. The study was motivated by the belief that such a model wouldprovide a useful foundation for theoretical and practical research in the areas of secondlanguage learning and instruction. This expectation has been fully upheld by theresults of the study.One of the most significant substantive findings was the similarity in overallresponse latency times and response latency patterns across context and wordfrequency between the two ESL groups in this study. Despite an almost four yeardifference in second-language immersion experience, these two groups showed littledifference in these areas. When this is combined with the finding that the ESL groupswere, on average 98 msec. slower than the Native speakers, the implication of the datais that there is a performance limit on word recognition speed in a second languagethat is below that of native speakers and which is not responsive to extended exposure tothe language. Related to this is the finding that differences in rate of exposure, asestimated by different levels of word frequency, did result in differences in wordrecognition speed in both of the ESL groups as well as the Native speakers.The study also revealed fundamental similarities between first and secondlanguage processing systems. In both first and second language word recognition,103increasing contextual information that is linked to a target word will increase the speedwith which that spoken word is recognized. Furthermore, there do not appear to bemajor differences in the way different types of contextual information is used. In otherwords, during spoken language processing, second language speakers can make use ofa variety of types of stored lexical knowledge in a high-speed, online manner similar toa native speaker.The findings of this study demonstrate clearly the validity of extending thecohort model to the second language processing context. The similarities between firstand second language processing systems that have been brought out in the presentstudy indicate that the model itself can be fitted to the processes involved inrecognizing words spoken in a second language. Furthermore, where differences havebeen identified, the model has supplied enough analytic power to suggest specificsources of the differences. Overall, the present study has demonstrated that the cohortmodel and its associated research paradigms, are useful, powerful tools for formulatingand testing hypotheses regarding the word recognition processes of second languagelearners, and by extension, the process of comprehending speech in a second language.104ReferencesAllen, P.A., McMeal, M. & Kvak, D. (1992). Perhaps the lexicon is coded as a functionof word frequency. Journal of Memory and Language, 31, 826-844.Bachman, L.F., & Palmer, A.S. (1982). The construct validity of some components ofcommunicative proficiency. TESOL Quarterly, 16, 449-465.Barbour, R.P. (1983). An exploratory study of the hypothesis of divisible versus unitarycompetence in second language proficiency. Unpublished master’s thesis, Universityof British Columbia.Bates, E. & MacWhinney, B. (1981). Second-language acquisition from a functionalistperspective: Pragmatic, semantic, and perceptual strategies. In H. Winitz (Ed.),Annals of the New York Academy of Science Conference on native and foreign languageacquisition (pp. 190-214). New York: New York Academy of Sciences.Bates, E., & MacWhinney, B. (1982). Functional approaches to grammar. In E. Wanner& L.R. Gleitman (Eds). Language acquisition (pp. 173-218). Cambridge: Cambridge.Battig, W.F. & Montague, W.E., (1969) Category norms for verbal items in 56Categories: A replication and extension of the Connecticut category norms.Journal of Experimental Psychology Monograph, 80 (3, part 2).Bialystok, E. (1990). Communication strategies. Cambridge, Mass.: Basil Blackwell.Bresnan, J. (1982). Control and complementation. In J. Bresnan (Ed.). The mentalrepresentation of grammatical relations (pp. 282-390). Cambridge, Mass.: MIT Press.Bresnan, J. & Kaplan, R. M. (1982). Introduction: Grammars as mental representationsof Language. In J. Bresnan (Ed.). The mental representation of grammatical relations(pp. xvii-lii). Cambridge, Mass.: MiT Press.Buck, G. (1992). Listening comprehension: Construct validity and trait characteristics.Language Learning, 42, 313-357.Butterworth, B. (1983). Lexical representation. In B. Butterworth (Ed.). Languageproduction (Vol. 2, pp. 257-294). London: Academic.Carroll, J.B., Davies, P., & Richman, B. (1971). The American heritage word frequency book.New York: American Heritage.Carroll, S.E. (1992). On cognates. Second Language Research, 8, 93-119.Chomsky, N. (1965). Aspects of the theory of syntax. Cambridge, Mass.: MIT Press.105Cole, R.A. (1973). Listening for mispronunciations: A measure of what we hear duringspeech. Perception and Psychophysics, 13, 153-156.Connine, C.M., Titone, D., & Wang, J. (1993). Auditory word recognition: Extrinsic andintrinsic effects of word frequency. Journal of Experimental Psychology: Learning,Memory, and Cognition, 19, 81-94.Conrad, L., (1989). The effects of time-compressed speech on native and EFL listeningcomprehension. Studies in Second Language Acquisition, 11, 1-16.Creative Labs, Inc. (1991). Soundblaster pro user reference manual.Ellis, R. (1985). Understanding second language acquisition. Oxford: Oxford UniversityPress.Ellis, R. (1994). The study of second language acquisition. Oxford: Oxford Press.Flynn, S. (1987). A parameter-setting model of L2 acquisition. Dordrecht, Holland: D.Reidel.Flynn, S., & O’Neil, W. (1988). Introduction. In S. Flynn and W. O’Neil (Eds.), Linguistictheory in second language acquisition (pp.1-24). Dordrecht, Holland: Kiuwer.Forster, K.I. (1989). Basic issues in lexical processing. In William Marslen-Wilson, (Ed.)Lexical representation and process, (pp. 75-107). Cambridge: Bradford.Francis, W. N. & Kucera, H. (1982). Frequency analysis of English usage: Lexicon andgrammar. Boston: Houghton Mifflin.Gass, S.M., (1984). Development of speech perception and speech production abilitiesin adult second language learners. Applied Psycholinguistics, 5, 51-74.Gass, S.M., (1987). The resolution of conflicts among competing systems: A bidirectionalperspective. Applied Psycholinguistics, 8, 329-350.Griffiths, R., (1990). Speech rate and NNS comprehension: A preliminary study in time-benefit analysis. Language Learning, 3, 311-336.Griffiths, R., (1991). Pausological research in an U context: A rationale, and review ofselected studies. Applied Linguistics, 12, 345-364.Grosjean, F. (1980). Spoken word recognition processes and the gating paradigm.Perception and Psychophysics, 28, 267-283.Haegeman, L. (1991). Introduction to government and binding theory. Cambridge, Mass.:Blackwell.106Harrington, M. (1987). Processing transfer: Language-specific processing strategies as asource of interlanguage variation. Applied Psycholinguistics, 8, 351-377.Hayashi, T. (1991). Interactive processing of words in connected speech in Li and U.International Review of Applied Linguistics, 29, 151-160.Holofcener, L., (1960). A practical dictionary of rhymes. New York: Bonanza Books.Hudson, W. (1989). Semantic theory and U lexical development. In S. M. Gass & J.Schachter (Eds.), Linguistic perspectives on second language acquisition (pp. 222-238). Cambridge: Cambridge Press.Johnson, B., (1957). New rhyming dictionary and poet’s handbook. New York: Harper &Row.Johnson, J.S., & Newport, E.L. (1989). Critical period effects in second languagelearning: The influence of maturational state on the acquisition of English as asecond language. Cognitive Psychology, 21, 60-99.Johnson, J.S., & Newport, E.L. (1991). Critical period effects on universal properties oflanguage: The status of subjacency in the acquisition of a second language.Cognition, 39, 215-258.Johnson-Laird, P.N. (1987). The mental representation of the meaning of words.Cognition, 25, 189-211.Kelly, P. (1991). Lexical ignorance: The main obstacle to listening comprehension withadvanced foreign language learners. IRAL, 29, 135-149.Kilborn, K., & Cooreman, A., (1987). Sentence interpretation strategies in adult Dutch--English bilinguals. Applied Psycholinguistics, 8, 415-431.Klatt, D.H. (1989). Review of selected models of speech perception. In William MarslenWilson, (Ed.) Lexical representation and process, (pp. 169-226). Cambridge:Bradford.Lahiri, A. & Marslen-Wilson, W. (1991). The mental representation of lexical form: Aphonological approach to the recognition lexicon. Cognition, 38, 245-295.Long, M. (1990). Maturational constraints on language development. Studies in SecondLanguage Acquisition, 12, 251-286.Long, M. & Sato, C. (1985). Methodological issues in interlanguage studies: Aninteractionist perspective. In A. Davies, C. Criper, & A. Howatt (Eds.),Interlanguage (pp. 253-279). Edinburgh: Edinburgh University Press.107MacWhinney, B. (1987a). Applying the competition model to bilingualism. AppliedPsycholinguistics, 8, 315-327.MacWhinney, B. (198Th). The competition model. In Brian MacWhinney (Ed.),Mechanisms of language acquisition. Hi]lsdale, NJ: Eribaum.MacWhinney, B., Bates, E., & Kliegl, R. (1984). Cue validity and sentence interpretationin English, German, and Italian. Journal of Verbal Learning and Verbal Behavior, 23,249-308.MacWhinney, B., Leinbach, J., Taraban, R. & McDonald, J. (1989). Language learning:Cues or rules? Journal of Memory and Language, 28, 255-277.Marslen-Wilson, W.D. (1985). Speech shadowing and speech comprehension. SpeechCommunication, 4, 55-73.Marsien-Wilson, W.D. (1987). Functional parallelism in spoken word-recognition.Cognition, 25, 71-102.Marslen-Wilson, W.D. (1989). Access and integration: Projecting sound onto meaning. InWilliam Marsien-Wilson, (Ed.) Lexical representation and process (pp. 3-24).Cambridge:Bradford.Marslen-Wilson, W.D. (1990). Activation, Competition, and Frequency in LexicalAccess. In G. T. Altmann (Ed.). Cognitive Models of Speech Processing (pp.148-178).Cambridge: MIT Press.Marslen-Wilson, W.D. (1993). Issues of Process and Representation in Lexical Access.In G. T. M. Altmann & R. Shilicock (Eds.), Cognitive models of speech processing:The second Spelonga meeting (pp. 187-210). Hove, England: Lawrence Eribaum.Marslen-Wilson, W.D., Brown, C., St Tyler, L.K., (1988). Lexical representations inspoken language comprehension. Language and Cognitive Processes, 3, 1-16.Marslen-Wilson, W.D. & Tyler, L.K., (1980). The temporal structure of spoken languageunderstanding. Cognition, 8, 1-71.Marslen-Wilson, W.D. & Tyler, L.K., (1981). Central processes in speech understanding.Philosophical Transactions of the Royal Society, B, 295, 317-332.Marsien-Wilson, W.D., & Warren, P. (1994). Levels of perceptual representation andprocess in lexical access: Words, phonemes and features. Psychological Review,101, 653-675.108Marsien-Wilson, W.D. & Warren, P. (1994). Levels of Perceptual Representation andProcess in Lexical Access: Words, Phonemes, and Features. Psychological Review,101, (pp. 653-675).Marsien-Wilson, W.D., & Welsh, A. (1978). Processing interactions and lexical accessduring word recognition in continuous speech. Cognitive Psychology, 10, 29-63.McClaskey, C.L., Pisoni, D.B., & Carrell, T.D. (1983). Transfer of training of a newlinguistic contrast in voicing. Perception & Psychophysics, 34, 323-330.McDonald, J. L. (1987a). Assigning linguistic roles: The influence of conflicting cues.Journal of Memory and Language, 26, 100-117.McDonald, J. L. (198Th). Sentence interpretation in bilingual speakers of English andDutch. Applied Psycholinguistics 8, 379-413.McDonald, J.L., & Heilenman, L.K. (1991). Determinants of cue strength in adult firstand second language speakers of French. Applied Psycholinguistics, 313-348.McLaughlin, B., Rossman, T., & McLeod, B. (1983). Second language learning: aninformation-processing perspective. Language Learning, 33, 135-158.Miao, X., (1981). Word order and semantic strategies in Chinese sentencecomprehension. International Journal of Psycholinguistics, 8(3), 109-122.Miller, G. A. (1978). Semantic relations among words. In M. Halle, J. Bresnan & G. A.Miller (Eds.), Linguistic theory and psychological reality (pp. 60-118). Cambridge,Mass: MIT Press.Oller, J.W. Jr., (1981). Language testing research (1979-1980). in R.B. Kaplan, R.L. Jones,& G.R. Tucker (Eds.), Annual Review of Applied Linguistics 1980 (pp.124-150).Rowley, Mass.: Newbury House.Oiler, J.W. Jr. & Hinofotis, F.B., (1980). Two mutually exclusive hypotheses aboutsecond language ability: Indivisible or partially divisible competence. In John W.Oller Jr., and Kyle Perkins, (Eds.), Research in Language Testing (pp. 13-23).Rowley, Mass.: Newbury House.Powers, D.E., (1982). Selecting Samples for Testing the Hypothesis of Divisible VersusUnitary Competence in Language Proficiency. Language Learning, 32, 331-335.Sasaki, Y. (1991). English and Japanese interlanguage comprehension strategies: Ananalysis based on the competition model. Applied Psycholinguistics, 12, 47-73.Selinker, L. (1972). Interlanguage. International Review of Applied Linguistics in LanguageTeaching, 10, 209-231.109Singleton, D.S., & Little, D. (1991). The second language lexicon: some evidence fromuniversity-level learners of French and German. Second Language Research, 7, 61-81.Taft, M., & Hambly, G., (1986). Exploring the Cohort Model of spoken wordrecognition. Cognition, 22, 259-282.Tyler, L.K. (1984). The structure of the initial cohort: evidence from gating. Perceptionand Psychophysics, 36, 415-427.Tyler, L.K. (1985). Real-time comprehension processes in agrammatism: A case study.Brain and Language, 26, 259-275.Tyler, L.K. (1988). Spoken language comprehension in a fluent aphasic patient. CognitiveNeuropsychology, 5, 375-400.Tyler, L.K. (1989). The role of lexical representations in language comprehension. InWilliam Marslen-Wilson, (Ed.) Lexical representation and process, (pp. 439-462).Cambridge:Bradford.Tyler, L.K., & Wessels, J. (1983). Quantifying contextual contributions to wordrecognition process. Perception and Psychophysics, 34, 409-420.van Riemsdijk, H. & Williams, F. (1989). Introduction to the theory of grammar.Cambridge, Mass: MIT Press.Warren, P., & Marslen-Wislon, W. D. (1987). Continuous uptake of acoustic cues inspoken word recognition. Perception and Psychophysics, 41, 262-275.Warren, P., & Marslen-Wislon, W. D. (1988). Cues to lexical choice: Discriminatingplace and voice. Perception and Psychophysics, 43, 21-30.Werker, J.F., & Lalonde, C.E. (1988). Cross-language speech perception: Initialcapabilities and developmental change. Developmental Psychology, 24, 672-683.Wulfeck, B.B., Juarez, L., Bates, E., & Kilborn, K. (1986). Sentence interpretationstrategies in healthy and aphasic bilingual adults. In J. Vaid (Ed.), Languageprocessing in bilinguals: Psycholinguistic and neuropsychological perspectives (pp. 199-219). Hilisdale, NJ: Erlbaum.Zobl, H. (1989). Canonical typological structures and ergativity in English L2acquisition. In S. M. Gass & J. Schachter (Eds.), Linguistic perspectives on secondlanguage acquisition (pp. 203-221. Cambridge: Cambridge Press.Zwitserlood, P. (1989). The locus of the effects of sentential- semantic context in spokenword processing. Cognition, 32, 25-64.110AppendicesAppendix A: Targets, lead-in sentences, and test sentences for E)(ACT monitoring task.Codes: Cue group and frequency level, sequence number, target word, actualfrequency, adjusted frequency.TG1.1H 1144 bridge 117 76.71The light on the shore was fading.Cl John quietly faced the bridge then walked away.C2 John quietly grabbed the bridge then walked away.C3 John quietly delighted the bridge then walked away.C4 John quietly hesitated the bridge then walked away.C5 shore walked away then on John fading the light the bridge the was facedquietlyTG1.1L 3507 bubbles 25 15.44Lynda was playing in the garden.Cl She chased the bubbles floating in the air.C2 She ate the bubbles floating in the air.C3 She interviewed the bubbles floating in the air.C4 She crawled the bubbles floating in the air.C5 the garden Lynda the in playing she floating the bubbles was in chased airTG1.2H 344 book 292 256.92The office was very quiet.Cl The secretary placed the book on the desk.C2 The secretary twisted the book on the desk.C3 The secretary cured the book on the desk.C4 The secretary consented the book on the desk.C5 quiet the was office very the book placed secretary the on desk theTG1.2L 3577 prisoner 31 14.79The sky over the building was dark.Cl No-one saw the prisoner when he left.C2 No-one obeyed the prisoner when he left.C3 No-one postponed the prisoner when he left.C4 No-one happened the prisoner when he left.C5 building the saw sky was he no-one the prisoner over the dark when left111Appendix A, ContinuedTG1.3H 1065 columns 107 83.02Several workmen were tidying up the site.Cl One of them washed the columns of the building.C2 One of them buried the columns of the building.C3 One of them swam the columns of the building.C4 One of them ambled the columns of the building.C5 of several up site the tidying washed were them the columns of building theworkmen oneTG1.3L 3837 candle 23 12.92Audrey had practically finished decorating.Cl She took a candle from the centre of the mantlepiece.C2 She swallowed a candle from the centre of the mantlepiece.C3 She unified a candle from the centre of the mantlepiece.C4 She aspired a candle from the centre of the mantlepiece.C5 decorating Audrey the practically mantlepiece centre a candle took the from shehad of finishedTG1.4H 790 bed 139 104.79The truck drove up to the front of the house.Cl Two men set the bed on the lawn.C2 Two men chopped up the bed on the lawn.C3 Two men wore the bed on the lawn.C4 Two men vanished the bed on the lawn.C5 the house of two the front up set men the bed the drove to truck lawn on theTG1.4L 3621 boot 30 14.44Edward was enjoying shopping.Cl The salesman held the boot in his hand.C2 The salesman tasted the boot in his hand.C3 The salesman sipped the boot in his hand.C4 The salesman chuckled the boot in his hand.CS salesman was held Edward hand the boot in the shopping his enjoying112Appendix A, ContinuedTG2.1H 703 tree 160 125.26The students often walk to school.Cl On the way, they pass the tree in front of the library.C2 On the way, they kiss the tree in front of the library.C3 On the way, they breathe the tree in front of the library.C4 On the way, they appear the tree in front of the library.C5 the front of the on walk often they school pass way library the tree to in thestudentsTG2.1L 3557 blade 26 14.98The machinery came to a stop.Cl Paul detached the blade carefully.C2 Paul licked the blade carefully.C3 Paul astonished the blade carefully.C4 Paul proceeded the blade carefully.CS to detached came carefully a Paul the stop machinery the bladeTG2.2H 699 price 164 125.89The small store was busy.Cl Mary liked the price of the blue dress.C2 Mary drew the price of the blue dress.C3 Mary operated the price of the blue dress.C4 Mary abstained the price of the blue dress.C5 the small was blue the price Mary dress store busy the of likedTG2.2L 3753 drum 26 13.46The audience was very noisy.Cl A young boy moved the drum on the stage.C2 A young boy smashed the drum on the stage.C3 A young boy surprised the drum on the stage.C4 A young boy behaved the drum on the stage.C5 very the audience was a moved young boy the on noisy stage the drum113Appendix A, ContinuedTG2.3H 968 gun 142 91.30There were many interesting items in the room.Cl Louise removed the gun on the wall.C2 Louise bounced the gun on the wall.C3 Louise toasted the gun on the wall.C4 Louise yawned the gun on the wall.C5 items the wall there interesting on removed room the gun Louise were the manyinTG2.3L 3574 tire 31 14.82Children often play on the beach.Cl They roll the tire on the sand.C2 They polish the tire on the sand.C3 They frighten the tire on the sand.C4 They sleep the tire on the sand.C5 beach sand roll on play the they the on children the tire oftenTG2.4H 744 dog 147 117.24The Smiths walk to the hill every evening.Cl They frequently take the dog with them.C2 They frequently carry the dog with them.C3 They frequently quote the dog with them.C4 They frequently gossip the dog with them.C5 the them with they to hill frequently Smiths the dog take evening walk the everyTG2.4L 3883 chapel 22 12.59Religious tourists enjoy visiting that part of town.Cl Most of them enjoy the chapel by the river.C2 Most of them clean the chapel by the river.C3 Most of them amaze the chapel by the river.C4 Most of them ache the chapel by the river.C5 of enjoy most them visiting that town enjoy tourists of by the river the chapelpart114Appendix A, ContinuedTG3.1H 949 page 102 93.57The magazine was open on the shelf.Cl Sandra touched the page that faced her.C2 Sandra sketched the page that faced her.C3 Sandra stimulated the page that faced her.C4 Sandra blushed the page that faced her.C5 Sandra that on open magazine was the shelf touched her the faced the pageTG3.1L 3579 cook 22 14.77The customers were talking noisily.Cl A woman waved at the cook in the kitchen.C2 A woman woke the cook in the kitchen.C3 A woman mixed the cook in the kitchen.C4 A woman coughed the cook in the kitchen.C5 kitchen woman customers the cook in a talking waved noisily the at the wereTG3.2H 373 door 384 240.33Most of the students were in the class.Cl The teacher shut the door quietly.C2 The teacher painted the door quietly.C3 The teacher attracted the door quietly.C4 The teacher arose the door quietly.C5 the class were of the shut teacher the students in quietly the door mostTG3.2L 3851 tickets 30 12.78The buses run every hour.Cl Most people get the tickets ahead of time.C2 Most people borrow the tickets ahead of time.C3 Most people pour the tickets ahead of time.C4 Most people faint the tickets ahead of time.C5 every ahead the people most get of hour time buses the tickets run115Appendix A, ContinuedTG3.3H 411 table 242 215.50The display is very charming.Cl Lots of people admire the table near the pool.C2 Lots of people scratch the table near the pooi.C3 Lots of people wake the table near the pooi.C4 Lots of people come the table near the pool.C5 is the people lots display admire the table the pooi very near charming ofTG3.3L 3894 dome 25 12.54The morning view is spectacular.Cl The sunlight catches the dome at the very top.C2 The sunlight melts the dome at the very top.C3 The sunlight prolongs the dome at the very top.C4 The sunlight occurs the dome at the very top.C5 morning is view sunlight catches top the dome spectacular the very the at theTG3.4H 309 doctor 349 285.99John visits his mother every day.Cl He often meets the doctor in the room.C2 He often tickles the doctor in the room.C3 He often sharpens the doctor in the room.C4 He often competes the doctor in the room.C5 John room his mother meets the day he every in the doctor visits oftenTG3.4L 3698 battery 22 13.90The engine stopped suddenly.Cl Anne pulled the battery away from the stand.C2 Anne chewed the battery away from the stand.C3 Anne tempted the battery away from the stand.C4 Anne chatted the battery away from the stand.C5 Anne away engine the battery the pulled from stand stopped the suddenly116Appendix A, ContinuedTG4.1H 1131 ball 123 77.53The game lasted over an hour.Cl The players put the ball in the bag.C2 The players slit the ball in the bag.C3 The players quantified the ball in the bag.C4 The players shivered the ball in the bag.C5 the hour lasted the put players in over an bag the game the ballTG4.1L 3591 dawn 26 14.66It was still very early.Cl Joan watched the dawn from her window.C2 Joan cursed the dawn from her window.C3 Joan raided the dawn from her window.C4 Joan quarrelled the dawn from her window.C5 watched still window the dawn it Joan from was early very herTG4.2H 942 poet 144 93.92The atmosphere was electric.Cl The audience cheered the poet when he stepped on stage.C2 The audience fed the poet when he stepped on stage.C3 The audience steamed the poet when he stepped on stage.C4 The audience progressed the poet when he stepped on stage.C5 cheered the audience the on he stage when was stepped the poet electricatmosphereTG4.2L 759 dollar 144 121.43The small shop wasn’t busy.Cl John deposited the dollar on the counter.C2 John bent the dollar on the counter.C3 John quickened the dollar on the counter.C4 John arrived the dollar on the counter.C5 the John busy shop on the dollar small counter the deposited wasn’t117Appendix A, ContinuedTG4.3H 731 corner 134 120.34Traffic was not very heavy.Cl Louise regarded the corner of the street.C2 Louise sprayed the corner of the street.C3 Louise separated the corner of the street.C4 Louise sneezed the corner of the street.C5 traffic not Louise the heavy the street corner of regarded very wasTG4.3L 3973 peak 24 12.06The mountain is beautiful on rainy daysCl Everyone loves the peak in the mist.C2 Everyone dries the peak in the mist.C3 Everyone invites the peak in the mist.C4 Everyone seems the peak in the mist.C5 days the in mountain mist the rainy the peak on beautiful loves is everyoneTG4.4H 825 blood 122 107.90The clinic was great success.Cl The technicians saved the blood in plastic bags.C2 The technicians served the blood in plastic bags.C3 The technicians rattled the blood in plastic bags.C4 The technicians chattered the blood in plastic bags.C5 bags the in saved plastic the blood was clinic technicians great the successTG4.4L 3854 bear 24 12.76The tourists stood taking pictures of the animals.Cl Two children eyed the bear in the meadow.C2 Two children squirted the bear in the meadow.C3 Two children repeated the bear in the meadow.C4 Two children dozed the bear in the meadow.C5 in children eyed tourists pictures taking of meadow the stood the two animalsthe bear the sleep118Appendix A, ContinuedTG5.1H 1100 park 111 79.59The workers are always busy.Cl Two men sweep the park every morning.C2 Two men burn the park every morning.C3 Two men drink the park every morning.C4 Two men disappear the park every morning.C5 morning two every always the park are busy men the workers sweepTG5.1L 3528 cafe 25 15.23John and Fred wanted a new business.Cl They opened the cafe on East Broadway.C2 They flooded the cafe on East Broadway.C3 They embarrassed the cafe on East Broadway.C4 They relied the cafe on East Broadway.CS a Fred business wanted John opened Broadway the cafe and on new East theyTG5.2H 1153 boat 123 75.92The waves were not large.Cl Sheila shoved the boat away from the dock.C2 Sheila blew the boat away from the dock.C3 Sheila shamed the boat away from the dock.C4 Sheila commented the boat away from the dock.C5 the away shoved large the boat sheila dock were not from the wavesTG5.2L 3956 treaty 24 12.15Delegates from both countries were very happy.Cl They accepted the treaty quickly.C2 They rinsed the treaty quickly.C3 They payed the treaty quickly.C4 They coexisted the treaty quickly.C5 both country very delegates happy they quickly from were the treaty accepted119Appendix A, ContinuedTG5.3H 819 truth 130 108.37The two scientists were excited.Cl They had found the truth about the strange event.C2 They had obscured the truth about the strange event.C3 They had filled the truth about the strange event.C4 They had collaborated the truth about the strange event.C5 excited strange the scientists the truth were they about event found the two hadTG5.3L 3925 patrol 24 12.31The soldiers moved carefully.Cl They heard the patrol in the dark.C2 They vacuumed the patrol in the dark.C3 They edited the patrol in the dark.C4 They brooded the patrol in the dark.C5 the heard soldiers dark in carefully the patrol moved the theyTG5.4H 1083 block 98 81.17The sculpture would be large.Cl John and Louise pushed the block into place.C2 John and Louise ran the block into place.C3 John and Louise argued the block into place.C4 John and Louise marvelled the block into place.CS Louise into be John the would large pushed and the block sculpture placeTG5.4L 3778 costume 28 13.30After the party, Maggie undressed.Cl She draped the costume over a chair.C2 She squeezed the costume over a chair.C3 She resumed the costume over a chair.C4 She emerged the costume over a chair.C5 after she party the costume a the undressed Maggie draped over chair120Appendix B: Category based filler items1. A Precious Stonechanging the on RUBY light was road ran Fred quietly looked beach the then away2. A Relativeoffice the was quiet UNCLE very the secretary comic the on desk the dreamed3. A Kind of Cloththe parked the side the WOOL put refrigerator men the disk on road at of the trucktwo4. A Type of Reading Materialsurgeons twelve jogging up were MAGAZINE the site of them the washed wall of thefew5. A Colourwas drinking he the chopping held YELLOW in hand pianisthis the6. A Four-Footed Animalover had centre searching almost she a HORSE the Andrew of the took from train7. A Metalthe saw dark bus over the someone was the when he left STEEL building8. A Unit of Timewas meadow playing Lynda the stared she in the rising wind SECOND woods drain9. A Kitchen UtensilThe large SPOON is often misplaced. It should be kept in the drawer.10. A Part of SpeechThe second VERB comes too soon. Move it to the end of the sentence.11. A FruitThe meal was not very good. The APPLE pie was delicious, however.12. A Part of the Human BodyThe little restaurant was quiet. Mary’s LEG was very itchy.13. A WeaponThe valuable SWORD danced in the display case. John watched it carefully.12114. A Type of Human DwellingClimbing the APARTMENT stairs was difficult. Bill hopped up the slope.15. An Alcoholic BeverageApparently the dinner was full. WINE was the beverage of choice.16. A CrimeThe newspaper report rambles the news. THEFT is being hit by inflation.17. A Substance for Flavoring FoodMaking SALT pork is easy if you have all the correct parts.18. A Type of FuelThe old Q!J stoves were good but contributed to pollution.19. A SportAlthough SWIMMING uses a lot of strangeness, it is still a very good exercise.20. (A Name Applied to a Person to Indicate His) Occupation or ProfessionWhen the new LAWYER coughed the bible, the members of the jury laughed.21. A Carpenter’s ToolJim dropped his HAMMER and began to recite his toolbox.22. A Natural Earth FormationNot far behind the MOUNTAIN there is a small town.23. A Weather PhenomenonIn northern Canada, SNOW starts falling in early Autumn.24. A Girl’s First NameAfter she had raged the poster, MARY sat down to eat a large dinner.25. A Musical InstrumentEd replaced the window blind with a new VIOLIN that he had stolen.26. A BirdThe work in the fields is boring. A SPARROW sings incessantly.27. A Nonalcoholic BeverageThe menu was dreaming under the counter. MILK was the cheapest drink.28. An Article of ClothingGetting dressed wasn’t exactly a dance. His SHIRT was the easiest thing to find.12229. A FlowerThe plants and shrubbery scurried the shade. A ROSE grew near the edge of thegarden.30. A diseaseThe young nurse scolded the hospital blanket. Ben’s MEASLES were very itchy.123Appendix C: Rhyming based fillers1. ablazecrowd the appraise active very an was old the moved lady microphone on square the2. fiercethe under floor pierce the was ceiling everyone dark the helped mechanic he whenworked3. facewomen several the up were the walking all road of them window car of the4. navigatesome there the unusual were on liberate thing the Margaret roof drawer took beaterfrom5. matchend nearly of the Wendy completed a she hatch vacuuming candy had from removedcouch the6. stricttranscribed had the Joan nearly revisions predict she a from tape the machine oldremoved7. halfdeparted the train one from east the two of girls village laugh the delivered drain onspeaker8. aimimproving waved was coach the Richard running the flag in blame the9. needsleep kittens often on rug they the string throw proceed the in10. screenJohnsons along the every run the afternoon trail them cat with they often green bringthe11. chanceThe people dance a lot. It is part of the culture of that area.12. goodLots of people stood near the gateway. The restaurant was full.13. cheekyKay was surprised at the mess. The leaky pot had dripped on the floor.12414. creamyThe room was very uncomfortable. It was hot and steamy inside.15. horsesif he forces the soldiers to leave, he wifi be in trouble.16. trueShe couldn’t undo the water. She left it on the shelf.17. coinIt is a wonderful athletic club. To jj is very easy, also.18. movedMartha did not fill the cloud. Her friends approved but the teacher didn’t.19. meshThe fresh peaches tasted very good with cream.20. dangerousA day ago, a curious event occurred near the drugstore. No one actually saw it.21. ladderThe little boy was sadder than he could dream.22. mudEach spring, rains flood the river valley and several houses are damaged.23. huntJohn didn’t concern the murder. A blunt instrument had been used.24. dignifyI can’t smirk the last part. Please clarify what you mean there.25. paintThe village remains several houses. The quaint one belongs to an old man.26. hourIn the forest, the old moose wandered, nibbling sour berries and munching old roots.27. bornHe looked carefully at the map. The paper was. the pamphlet.28. neglectedThe bandits had sneezed the bank. The young policeman inspected the damage.29. presumePick up the world, my friend. There are still many things to consume before we go.12530. daggerAt seven o’clock, the bar closes all the people and the customers stagger to their cars.126Appendix D: Exact FillersThe material for the Exact Fillers was constructed according to the following generalcriteria:Modal Verbs Adj AdvGerundList 1 1 1 1Othersentences 3 3 3 3 3Exact, List, Filler1. standingtwo standing the men the division from down the were slope technical2. narrowhis on buildings narrow was street a long mainly old apartment with3. handedthe man bushes in little handed subway boy a would the stub he go the away so4. hardlyneeded it any complicated with calculations wind drift of force or tide and anyonecharts hardly race to convince that.5. mostlyother the half of the was window hidden tall mostly pot by plantsExact, Sentence, Filler6. nextHe was standing there in the evening but not the day.7. hurtShe had a bad cut in her leg. It hurt so she couldn’t walk.8. quietlyArmed with a weapon, she felt better. She quietly checked the bolts on the frontentrance.9. afraidHe was supposed to work in a machine shop but he was afraid to go in.12710. fromIt was a wonderful evening. Janet gazed at the moon from her porch.11. strangeThe small engine wasn’t drinking. John tossed the wrench on the strange counter.12. almostThe investigation was almost finished, and very successfully, too, the detectivesdecided.13. couldSky-diving could be dangerous. Especially on days like today.14. largeHumour doesn’t go through. The large movie will be interrogated.15. siftedThe plants looked good. He knelt again and gently sifted the soil with his glovedhands.16. fearfullyThe creatures persisted the forest. They would live fearfully in the dark undergrowth.17. swiftlyEdwind swiftly hid his old friends. They persevered the jam on toast.18. smallThe computer card didn’t fit. The salesman lurked the small part into place.19. sweetThe box of candy pleased Joan. Sweet things were her favorite stigma.20. nearHis complaint about the truck landed near the feet of the driver128Appendix E: Example itemsExact1. bucketlowered farmer the into well bucket he up the brought the some water2. shovedThe shoppers were very angry. Several shoved their way into the store.3. refrigeratorThe incident coincided the dinner. John went to the refrigerator to get some beer.Rhyme4. stankGeorge the although a lot had of he wanted go to still bank tomoney5. thinkThe dress wifi shrink if you wash it in hot water, I am sure.6. floorThe boy cried the flute. More music didn’t help at all.Category7. A Treethe near oak road an spread provided branches and it’s pedestrians shade for the8. AfishThe lack of rain made the river shallow. The salmon would not enter to spawn.9. A Type of FootgearPrivate Smith felt uncomfortable. The rifle misfired the sandals on his feet.129Appendix F: Warm-up itemsExact1. lampturned Elizabeth the sleep to lamp and out went2. worebought Brenda new a skirt and it to with the wore her movie friend3. becauseThey are going to go to California because they want to visit a warm place.4. jelloFor dessert, the guests were surprised by in an ice-cream bucket.5. radicalThe golfers militated the radical elements from the club before they could take-over6. sourThe cherry cobbler misfired the waiter. The result was a group of sour-facedcustomers.Category1. A relativewhile himself father to talking Ed’s away the put camping carefully equipment2. A kind of clothsuit Frank not wanted had a could but he mind up his he went wool so make home3. Religious buildingSunday morning was warm and clear. Most of the church members arrived early.4. An Elected position, presidentSeveral other presidents had run the meetings in the same way that Sean had.5. A countryMy niece spent last summer bicycling her feet through France.6. A member of clergyAs the sermon digressed the priest, the congregation became more tolerant.130Rhyme1. fendI please need lend some dollars me five help2. makemoonlight off the was reflecting beautiful and very lake the romantic3. investigatorWe wifi need a navigator for our sailing trip to the Bahamas.4. teacherInterested spectators gathered around and speculated as to what the creature might be.5. compareDespite the excellent repair work, the ship couldn’t swim as well as before.6. grumbleDon’t stumble. The ground is pretty heavy around here.131Appendix G: 50-Item Cloze TestFULL NAME_________________Read the following passage and fill in each blank with one word which fits both themeaning of the whole passage and the grammar of the sentence that it is in.Please Print neatly.There are several ways of shortening words and phrases to create new wordsfor old ones. Acronyms are the result of forming a word from thefirst letter or letters of each word in a phrase, often the title of anorganization. The American Society of Composers, Authors, and Publishers isusually referred to as ASCAP; the United Nations Children’s Fundis called UNICEF; a solider who is absent withoutleave is AWOL; and the Students Opposed to Grading Systems verylikely selected the name of their organization __carefully , so that it wouldyield the memorable acronym SOGS . Most acronyms are temporarylexical items, going out of use quickly, as do the organizations orsituations which they describe, but a few become permanent entriesin the lexicon of a language. Scuba is an acronym forsef-contained underwater breathing apparatus.Closely related to acronyms is another type of shorteningcommonly used in English -- the use of the initial letters of thewords in a phrase, with the letters pronounced individually,rather than as a single word. These strings of letters arewords, not just abbreviations; we often use them without evenknowing the phrase that led to their creation. For example, the GOPis the Republican Party and the source of the letters apparently wasGrand Old Party. But many people use the word GOP long before132they learn the full phrase. Other examples of this type of word creationare HEW ( Health Education, and Welfare), TLC (tender loving care), andLSD (in the United States, usually the hallucinogenic drug lysergicacid diethylamide; in Great Britain, librae, solidi, denarii a Latinphrase meaning ‘pounds, shilling, pence’; and the military phrase landingship dock).Shortening is also at work in the creation of words by backformation, a popular formative analysis of long words. In backformation , a short word is created from a longer one on thebasis of similarities between the longer word and other words in thelanguage. For example, the word editor existed in the lexicon ofEnglish long before the word edit. Comparing their word editorto other words such as writer, singer and worker, speakersof English assumed that, just as writer, singer, and worker consistedof two formatives each, editor must also have two formatives — asuffix or (in pronunciation identical to the suffix er), meaning ‘one whoperforms the action described by the verb’, and a verbedit . Thus, the new word edit entered the lexicon ofEnglish. At first, some speakers objected that “there was no suchword,” but today edit is completely acceptable In more recent times, backformation has resulted in the creation of enthuse from an earlierform, enthusiasm . Enthuse provokes strong negative reactions from somespeakers who object that “it isn’t a word.” In onesense , such people are correct, for enthuse is a relatively newword. But, on the other hand, back formation is a perfectly proper, normalprocess for creating words . Making such objections, in an attempt tokeep the language from changing, wifi most likely have no moreeffect than earlier objections to the word edit, or to the words televise and donate, alsocreated by back formation.133


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items