UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The effects of context on speech understanding in noise Howarth, Tanis Marie 1992

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

Item Metadata


831-ubc_1992_fall_howarth_tanis_marie.pdf [ 4.17MB ]
JSON: 831-1.0086560.json
JSON-LD: 831-1.0086560-ld.json
RDF/XML (Pretty): 831-1.0086560-rdf.xml
RDF/JSON: 831-1.0086560-rdf.json
Turtle: 831-1.0086560-turtle.txt
N-Triples: 831-1.0086560-rdf-ntriples.txt
Original Record: 831-1.0086560-source.json
Full Text

Full Text

THE EFFECTS OF CONTEXTON SPEECH UNDERSTANDING IN NOISEbyTANIS MARIE HOWARTHB.Sc., The University of Alberta, 1987A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFMASTER OF SCIENCEinTHE FACULTY OF GRADUATE STUDIES(School of Audiology and Speech Sciences)We accept this thesis as conformingto the required standardTHE UNIVERSITY OF BRITISH COLUMBIASeptember 1992© Tanis Marie Howarth, 1992In 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  Audiology and Speech SciencesThe University of British ColumbiaVancouver, CanadaDate October 13, 1992DE-6 (2/88)ABSTRACTTwelve normal-hearing subjects (22 to 29 years) listened to passages of text in the presence ofa twelve-speaker babble background noise. The passages were presented in a line-by-linefashion and subjects were asked to repeat each line as accurately as they could. The presentstudy investigated whether three text types differing in macrostructure (narrative, descripitive,and procedural) would differentially effect how speech is perceived in the presence ofbackground noise. Also, the number and types of errors (deletions, additions, substitutions,and exchanges) that ocurred as signal-to-noise conditions decreased from 0dB to -12dB wereexamined. The degree to which a subject's response matched the target at three linguistic levels(auditory, syntactic, and semantic) was analyzed as either a partial match, total match, or nomatch at all. It was found that when listening to narratives, significantly fewer errors weremade than when listening to descriptions. It was also shown that if only the number ofauditory errors are counted and used as a means by which to assess the listening abilities of anindividual in noise, then the robustness of syntactic and semantic information is notappreciated.ll111TABLE OF CONTENTSPageABSTRACT 	 iiTABLE of CONTENTS 	 iiiLIST of FIGURES	viACKNOWLEDGEMENT 	 viiChapter 1 INTRODUCTION	1Rationale	1Goals	2Hypotheses 	 3Chapter 2. REVIEW OF THE LITERATURE 	 5Speech Understanding 	 5Language Comprehension Model 	 5The Issue of Autonomy 	 6The Issue of Continuous vs. Discrete Processing 	 11Bottom-Up and Top-Down Processes 	 13Cognitive Considerations	16Coherence and Inference 	 16World Knowledge and Schemata	18Schemata and Prediction 	 21Working Memory 	 21Evaluation of Speech Understanding 	 25Everyday Communication 	 25Stimuli Used in the Evaluation of Speech Understanding 	 26Speech Signals 	 26ivWord-Level Speech Signals 	 26Sentence-Level Speech Signals 	 27Discourse-Level Speech Signals 	 29Text S tructure 	 30Competing Signals Used in the Evaluation of Speech Understanding 35Summary of the Shortcomings of Current Evaluation Stimuli 	 35Procedures Employed in the Evaluation of Speech Understanding 	 36Reception Tasks 	 36Tracking 	 36Summary of the Shortcomings of Current Evaluation Procedures 	 38Chapter 3. METHODS 	 40Participants 	 40Materials 	 40Procedures 	 42Hearing Screening 	 43Text in Noise Task 	 43Chapter 4. RESULTS 	 46Effect of S/N Ratio and Text Type on Number of Errors 	 46Effect of Familiarity 	 47Effect of S/N Ratio and Text Type on Error Type 	 48Classification of Error Types 	 48Effect of S/N Ratio on the Error Type 'Exchange' 	 49Effect of S/N Ratio on the Error Type 'Addition' 	 50Effect of S/N Ratio on the Error Type 'Substitution' 	 50Effect of S/N Ratio on the Error Type 'Deletion' 	 51Summary of the Effect of S/N Ratio on the Error Types	 52VEffect of S/N Ratio on Errors According to Linguistic Level &Degree of Error 	 53Error Classification by Linguistic Level 	 53Effect of S/N Ratio on the Auditory Level 	 55Effect of S/N Ratio on the Semantic Level (Partial Auditory Match) 	 55Effect of SIN Ratio on the Syntactic Level (Partial Auditory Match) 	 56Effect of S/N Ratio on the Semantic Level (No Auditory Match) 	 57Effect of S/N Ratio on the Syntactic Level (No Auditory Match) 	 58Preservation of Semantic and Syntactic Information 	 59Chapter 5. DISCUSSION	 64Text Type 	 64Familiarity and Text Type 	 65Error Type and Text Type 	 66Preservation of Auditory Information 	 68Preservation of Syntactic Information 	 69Degree of Syntactic Match & Partial Auditory Match 	 69Degree of Syntactic Match & Total Auditory Mismatch 	 69Preservation of Semantic Information 	 70Degree of Semantic Match: Partial Auditory & Total Auditory Mismatch 70Degree of Semantic Match & Partial Auditory Match 	 71Degree of Semantic Match & Total Auditory Mismatch 	 71The Preservation of Syntax in Relation to Semantics 	 72Future Directions	73REFERENCES	 74APPENDICES	 85LIST of FIGURESFigure 1 Language Comprehension Model	 7Figure 2. Total Number of Errors Observed for Each of the Text Types	 46Figure 3 Frequency of Errors Observed as a Function of Familiarity 	 48Figure 4. Frequency of the Error Type 'Exchange' for each of the Text Types	 49Figure 5. Frequency of the Error Type 'Addition' for each of the Text Types	 50Figure 6. Frequency of the Error Type 'Substitution' for each of the Text Types	 51Figure 7. Frequency of the Error Type Deletion' for each of the Text Types	 52Figure 8. Summary of the Error Types Across the Three Text Types	 53Figure 9. Effect of Signal-to-Noise Ratio on the Auditory Level	 54Figure 10. Effect of Signal-to-Noise Ratio on the Semantic Level (Partial Auditory Match)	 56Figure 11. Effect of Signal-to-Noise Ratio on the Syntactic Level (Partial Auditory Match)	 57Figure 12. Effect of Signal-to-Noise Ratio on the Semantic Level (No Auditory Match)	 58Figure 13. Effect of Signal-to-Noise Ratio on the Syntactic Level (No Auditory Match)	 59Figure 14. Substitution Errors (Partial Auditory Match) Preserving Some Syntactic or SomeSemantic Information 	 60Figure 15. Substitution Errors (Partial Auditory Match) Preserving Both Semantic andSyntactic, Only Syntactic, Only Semantic or Neither Level of Information	 61Figure 16. Substitution Errors (No Auditory Match) Preserving Some Syntactic or SomeSemantic Information 	 62Figure 17. Substitution Errors (No Auditory Match) Preserving Both Semantic andSyntactic, Only Syntactic, Only Semantic, or Neither Level of Information 	 63viviiACKNOWLEDGEMENTSI would like to express my sincere appreciation to everyone who supported me in thisendeavour. I would especially like to extend my thanks to Kathy Fuller for her guidancethroughout the year: Her enthusiasm for research was contagious. I would also like to thankJohn Gilbert for his input and editorial expertise. And last but certainly not least, without thelove and support of my husband, Guy Downie, and my family and friends this would not havebeen possible.1CHAPTER 1INTRODUCTIONRationaleIn the past audiologic evaluation and rehabilitation have largely been based on measures ofhow individuals hear single words or non-speech signals (Olsen & Matkin, 1991).Traditionally, when speech materials have been used, audiologists have assessedcommunication impairment by having their clients repeat single monosyllabic wordspresented in quiet. This type of evaluation, however, is not a valid means by which to assessan individual's ability to understand speech in everyday communication situations(c.f., CHABA, 1988; Noble, 1983). Ideal conditions for communication consist of two-talkers in a quiet environment using language that is familiar, simple, and clearly articulated.Unforunately, most situations are less than ideal. We often need to understand an unfamiliarmessage spoken with less than perfect articulation in a noisy environment amidst othercompeting distractions. To more accurately represent an individual's ability to understandspeech in real-life situations requires that we employ speech materials and test procedureswhich are more representative of listening in everyday situations. For example, a moreecologically valid speech assessment tool might employ connected discourse materialspresented in noisy, reverberant, binaural situations whose administration are under thecontrol of the client herself and which require the client to demonstrate her comprehension ofthe topic being presented (c.f., CHABA, 1988).Since everyday speech understanding typically involves the reception of continuousdiscourse which is supported by situational context (i.e., speaker, background activities,general setting), previous knowledge both about the topic, and of the underlying textstructure of the discourse, research needs to be directed at investigating the contribution that2each of these sources of information makes in speech comprehension. Different text typeshave been identified and shown to have differential effects on recall and readingcomprehension . It is the intention of this study to examine whether different text typeseffect the ease with which discourse is perceived in the presence of background noise.As long ago as 1955, Miller and Nicely demonstrated that the perceptual confusions ofconsonants in nonsense syllables presented in noise are not random errors. They tabled theirresults in confusion matrices that identified which phonemes were presented and whichphonemes were perceived in particular signal-to-noise ratios. This study was the first of itskind to perform a perceptual error analysis on the phonemic confusions which occur in arange of signal-to-noise conditions. It was the belief of those researchers that by describingthe types of errors that were made knowledge about the nature of speech perception would beincreased, with the possibility that communication could be improved. Following the logicof Miller and Nicely (1955), the present study is motivated by the belief that anunderstanding of the pattern of errors that occur when listening to discourse in noise willshed light on the process of speech understanding in difficult listening situations, and willalso provide insight for the development of effective communication strategies to teach inaural rehabilitation programs.GoalsThe first goal of the present study is to determine if text types which differ in macrostructuredifferentially effect how speech is perceived in the presence of background noise.The second goal of the present study is to determine the types and number of errors thatoccur when texts are listened to in different signal-to-noise conditions.3Hypotheses Hypothesis 1: The number and types of errors that occur when a listener repeats apassage of text with a predictable macrostructure (e.g., a narrative) will notbe significantly different from the number and types of errors that occurwhen a listener repeats a passage of text with a less predictablemacrostructure (e.g., a description or a procedure).Accompanying research hypotheses:The number of errors a listener makes when repeating a narrative will beless than the number of errors made when repeating a description (e.g.,restaurant review) or a procedural sequence (e.g., recipe).The types of errors a listener makes when repeating a narrative will besignificantly different from the types of errors made when repeating adescription (e.g., restaurant review) or a procedural sequence (e.g.,recipe).Hypothesis 2: As the signal to noise ratio decreases, and listening conditions becomeless favorable, the number of errors and the types of errors that occur whena listener repeats a passage of text will not be significantly different fromthe number and types of errors observed under more favorable signal-to-noise conditions.Accompanying research hypotheses:The number of errors a listener makes will increase as the listeningconditions become less favourable.4As signal-to-noise ratio decreases, the number of deletion errors willincrease, the number of substitution errors will increase, the number ofaddition errors will increase, and the number of exchange errors willincreaseAs signal-to-noise ratio decreases, the number of auditory errors willincrease, the number of semantic errors will increase, and the number ofsyntactic errors will increase with auditory errors out-numbering semanticerrors, and semantic errors out-numbering syntactic errors.5CHAPTER 2REVIEW OF THE LITERATURESpeech UnderstandingThe comprehension of spoken language requires the rapid construction of meaning from atransitory acoustic signal. Speech understanding is accomplished through the action andinteraction of 'bottom-up' and 'top-down' processes. 'Bottom-up' processing has beenreferred to as the processing of an acoustic signal from the level of the acoustic waveform tothe recognition of word elements and words to the level of semantic representations(e.g., Brown & Yule, 1989; CHABA, 1988; Marslen-Wilson & Tyler, 1981; Stine, Wingfield& Poon, 1989). In contrast, 'top-down' processing has been used to mean the influence ofworld knowledge and context in the interpretation and comprehension of the speech signal( e.g., Brown & Yule, 1989; Marslen-Wilson & Tyler, 1981; Stine, Wingfield & Poon, 1989).To gain an appreciation for how top-down and bottom-up processes contribute to languagecomprehension, it is useful to consider a model of language comprehension. One such modelwill be presented and then the operation of top-down and bottom-up processes will bediscussed.Language Comprehension Model In describing a model of speech processing, we are faced with three questions: First, islinguistic processing autonomous or is it part of an individual's general cognitive processingabilities; second, how do the phonetic, lexical, structural, and semantic components of alanguage processor interact ; and finally, is the speech signal analyzed continuously from themoment acoustic input impinges on the peripheral hearing system or is it processed in finitechunks ?6The Issue of Autonomy The autonomous versus non-autonomous positions on language processing have generatedgreat debate. Those who assert an autonomous view believe that a linguistic processing systemexists which is distinct from more general cognitive systems (e.g., Fodor, Bever, & Garrett,1974; Forster, 1979). The global processing position purports that linguistic processes are notseparate and are operated on by a central system which operates on all kinds of information(e.g., Marslen-Wilson & Tyler, 1981). This global model allows world knowledge toinfluence and assist in linguistic processing. In their extreme, these earlier views are mutuallyexclusive and one allows no room for the other.In a more recent model, Cairns (1984) proposes a model of language comprehension thatappears to strike a compromise. In her model, Cairns (1984) describes two sets of processorswhich operate in an independent fashion. One set of processors includes a lexical processorand a structural processor. Each of these sub-processors has access to the lexicon but not toworld knowledge. Cairns (1984) proposes that these processors analyze only the purelylinguistic properties of the speech input and respond only to linguistic information 1 . Thelexical processor retrieves lexical items from the lexicon using only phonological informationand is involved in the process of checking the contents of the lexical entry against the speechinput to assess the accuracy of the retrieval operation. Along with phonological information,each lexical entry has associated with it its form class and the sentence frames into which it canbe inserted. The structural processor creates a formal structural analysis of the speech input. Itcan only use syntactic information to create its analyses, and can perform no operation whichcannot be described in purely formal syntactic terms. In this way, it cannot access worldknowledge or perform any operations for which world knowledge is required. The collectiveoutput from these two subprocessors depicts the structural organization of individual lexical1 Cairns (1984) takes the position that semantics and pragmatics are not pure linguisticforms and thus are not part of the linguistic system proper.Conceptual Representation7c00xInterpretiveProcessorAStructuralProcessorALexical	)0' ProcessorPhonetic InputFigure 1. Language comprehension model after Cairns (1984).items and serves as input to the interpretive processor.The interpretive processor receives input separately from the lexicon and in turn from thestructural processor. In addition, it has simultaneous access to real-world knowledge, therebyhaving the capability of generating inferences and assessing the plausability of relationshipsbetween and within sentences. According to Cairns (1984), it is through the actions of thisprocessor that both linguistic (i.e., syntactic and phonological) and nonlinguistic informationare used to generate a coherent, meaningful representation of incoming discourse (denoted inFigure 1. as 'conceptual representation').Cairns (1984) assigns the interpretive processor two primary functions: First, the construction8of a nonlinguistic conceptual representation of the current speech input through the process ofintegrating information from the individual lexical items (via input from the lexicon) based ontheir structural organization (via input from the structural processor) together with inferencesbased on real-world knowledge; and second, the integration of individual sentences into aconceptual representation incorporating the linguistic and nonlinguistic contexts in which theyoccur. Since semantics and pragmatics are typically considered to be not merely a part ofgeneral world knowledge, it seems to me that the operation of the semantic and pragmaticcomponents of language processing is inherent in the functions ascribed to the interpretiveprocessor.While Cairns (1984) states that her model is an autonomous one, this is only true if oneexcludes semantics and pragmatics as part of the linguistic system. If, however, one includesthese components within the realm of the linguistic system then the model presents as a hybridbetween a modular language processing system and a global language processing system. It ismodular in the sense that initial lexical retrieval is dependent solely upon phonologicalinformation and does not involve world knowledge (see Fodor, 1983). It is also modular inthat the assignment of a formal syntactic organization of the lexical items is achieved withoutinput from world knowledge (see Fodor, 1983). The model is global in that it allows forlanguage processing to be a part of a broader general cognitive processing scheme.If one considers semantics and pragmatics to be a part of the linguistic system then this modelcannot be said to describe an autonomous view of language processing because the semanticand pragmatic components interact with world knowledge. In this way, Cairns' (1984) modelof linguistic processing can be said to be autonomous prior to the involvement of semanticsand at the level of semantics, language processing ceases to be autonomous.9As one would expect, the way in which lexical access is achieved is dependent on whether amodular or non-modular approach is taken. The non-modularists believe that the influence ofcontext and world knowledge results in only one lexical item being retrieved, whereas themodularists believe that an exhaustive retrieval is made of all lexical items which arephonologically appropriate. The correct item is selected when context and world knowledgeparticipate in a post-lexical access stage. Numerous experimental studies have been citedwhich support the former (see Marslen-Wilson, & Tyler, 1981) as well as the latter (seeCairns, 1984). The critical factor that differentiates these two approaches is the point in time atwhich world knowledge is introduced into the system. This is perhaps best highlighted by anexamination of how both perspectives use cohort theory as the primary mechanism of initiallexical access and describe the generation of similar end-products, but rely on differentoperatonal pathways to achieve the final output.Both Cairns (1984) and Marslen-Wilson and Tyler (1981) discuss lexical access within theframework of cohort theory. Cohort theory states that incoming phonological informationactivates a cohort of lexical items, the initial syllables of which phonetically.match the input.Members of the cohort deactivate as further phonetic information produces a mismatch. In theabsence of context or world knowledge, a lexical item will be selected as soon as it is uniquelydetermined by the incoming information. In this theory, contextual information can berecruited to deactivate semantically ineligible members of the cohort after they have beenactivated by phonetic information.Marslen-Wilson and Tyler (1981) discuss the operation of the word-initial cohort within ageneral cognitive processing framework. They refer to a series of experiments (Marslen-Wilson, 1973, 1975, 1976; Marslen-Wilson and Tyler, 1980) which employed shadowingtasks as a means to evaluate the speed of on-line speech recognition. Subjects were required to10repeat back continuous speech as it was heard. Repetition delays were then measured betweenthe onset of the target word in the material they were listening to and the onset time of therepetition of that word. After the application of a correction factor for the time taken to actuallymake the response, word recognition time was estimated to be 200 milliseconds. Thismeasurement indicates that words were being recognized before the listener could havedetermined what the word was solely on the basis of the acoustic-phonetic information. Withregards to cohort theory, Marslen-Wilson and Tyler (1981) propose that since spoken wordscan be recognized when many words are still compatible with the available sensory input, thenit must be the influence of context which aids in the selection process. To accomplish thisMarslen-Wilson and Tyler (1981) state that it is critical that the active subset of word initialcohorts be sensitive to not only the continuing sensory input but also to the compatability of thewords they represent within the available structural and interpretive context. A mismatch witheither source of constraint would cause the lexical item in question to be excluded from the poolof potential lexical selections.Cairns (1984) states that since cohort theory postulates activation processes which are sensitiveto only phonetic information and selection processes that are sensitive to other types ofinformation then "[such a] theory is not necessarily nonautonomous (p.221)". She suggeststhe need for a cohort theory containing a passive lexicon and selection processes that activelyemploy other types of information (i.e., syntactic, semantic, and world knowledge). In thisway the modularity of lexical access is maintained without losing the facilitory actions ofcontext.It seems to me that each of these approaches defines the same set of processes with the criticaldifference being the specific time when lexical access stops and post access selection andverification begins. Within the processing system of Marslen-Wilson and Tyler (1981),11context is said to operate at the level of lexical access, whereas within the processing system ofCairns (1984) context operates at a post-access stage. In the former model the lexical item isavailable to further processing during the access stage and therefore only one item is part of theoutput into the post-access stage, whereas in the latter model further processing can not occuruntil the item has been retrieved. As a consequence, numerous lexical items are pulled into thepost-access stage for further analyses before any one item is singled out. Thus, how thecohorts are operated on once they become available seem not to differ.Deciding whether lexical selection processes are autonomous or not is dependent onexperimental research which elaborates the kinds of information which are necessary for lexicalaccess and the time frame within which processes operate on the information. As statedpreviously, Marslen-Wilson and Tyler (1981) have found word recognition to occur by 200milliseconds. These results have been confirmed by a series of experiments in which subjectswere presented with the first 50 msec of a word and asked to identify it (Grosjean, 1980). Ifthe subject was unable to do so then they were given the first 100msec, then the first 150 msecand so on until they identified the word. Grosjean (1980) found that words in context could berecognized within 175 to 200 msec of onset. He also found that words out of context took onan average 333 msec to identify. Cairns (1984), however, reports that previous research byherself and her colleagues has found retrieval and post-retrieval processes to be completedwithin 500 milliseconds. Based on these findings, it is evident that while both views recognizelexical access as time-dependent, they differ on the time window within which these processesoccur.The Issue of Continuous vs. Discrete ProcessingMarslen-Wilson and Tyler (1975) present evidence that speech input is continuously analysedby the listener with respect to its implications for a message-level interpretation. They12conducted an experiment in which three types of prose materials were used: normal prosewhich contained normal utterance strings; anomolous strings which were syntactically correctin form but contained no coherent semantic interpretation; and scrambled strings which wereboth semantically and syntactically anomalous. Subjects were presented these strings in eitherthe presence or absence of a preceding context sentence. The task of the subject was to listento the test materials for a target word specified in advance, and respond as rapidly as possiblewhen they detected it. The idea underlying this approach was that the different types ofprocessing (e.g., syntactic, semantic) would interact with the word recognition processsesupon which the target words depended and by measuring response times at different pointsacross the test materials it was possible to determine when syntactic and semantic informationbecame available. Results showed that normal sentences yielded significantly faster responsetimes than anomolous and scrambled strings. This was said to indicate that the extra-processing information that normal prose provides is developed by the listener from the earliestmoments of speech input. It was found, however, that this early advantage of normalsentences over the other word strings was present only when a context relevent sentencepreceded the sentence of interest. Based on this finding, it was then said that the effectsobserved with the context sentence were evidence that the first words of an utterence wereevaluated in relation to their discourse context. In this way it was argued that the listenerattempts to construct a discourse relevant representation of the incoming speech signal as soonas it is received.The model proposed by Cairns (1984) is compatible with the notion proposed by Marslen-Wilson and Tyler (1981) that the speech signal is processed continuously and not in discreteunits. The operation of the processors within Cairns' (1984) model is assumed to beindependent but not necessarily serial. An example of a serial operation within this modelwould be that before the structural processor can begin its analysis it must have information13about the form class of the items it will be processing therefore it must wait for lexical retrievalto occur. An example of parallel processing within this model would be that the interpretiveprocessor could be developing expectancies based on world knowledge, that has been recruitedon the basis of preceding linguistic input, at the same time the lexical and structural processingis proceeding. These expectancies in turn play a role in the integration of the most recentlyprocessed speech input into the developing conceptual representation. The concept thatmessage level interpretation of the speech input is initiated right from the beginning of theutterance (i.e., with the utterance of the first phonemes) differs significantly from the originalobservations and proposals which have emphasized syntactic or functional clauses as units ofspeech perception (Fodor, Bever, & Garrett, 1974). Perhaps the most contemporary view putforth for these processing mechanisms is that of a temporally continuous, on-line analysis ofthe speech stream which involves a knowledge-driven search for the completion of functionalsemantic relationships (e.g., Stine, Wingfield, & Poon, 1989). This type of processing hasbeen supported by by Marslen-Wilson, & Seidenberg (1978, as cited in Cairns, 1984), whodemonstrated that the ideal segmentation units are not only functionally complete but alsoinformationally complete. It has been in this way that the 'clause-effect' was explained to bethe natural outcome of higher level processes integrating semantically cohesive units on apreferential basis (Cairns, 1984). If informational completeness underlies higher-levelprocessing, then the role of discourse structure to drive higher cognitive processes to constructcoherent representations of incoming discourse becomes apparent.Bottom-up and Top-down Processes The initial set of analyses performed on the speech input are assumed to be 'bottom-up'. Thatis, the members of the word-initial cohort are determined by properties of the sensory input.Once the system receives the output of this initial 'bottom-up' processing, a characteristic set ofoperations ensue. Marslen-Wilson, and Tyler (1980) demonstrated this when they asked14subjects to focus their attention on only the acoustic-phonetic properties of the input and foundthat the subjects were not able to avoid identifying the words involved. These results implythat the kind of processing involved in spoken-word recognition are mediated by automaticprocesses (Marslen-Wilson, & Tyler 1981). Based on this work, it has been proposed that thissame type of automatic processing holds true for operations throughout the languageprocessing system (Marslen-Wilson, & Tyler, 1981). Such a position subsequently impliesthat 'top-down' processing (i.e., involvement of contextual expectations) is not involved ininitial or normal 'first pass' processing. Marslen-Wilson and Tyler (1981) propose that theapparent interaction of word recognition processes with top-down influences are an illusion.They state that what is really occurring is the incorporation of the correct word, based onphonetic input and syntactic constraints, into the interpretive representation of the utterance.This is similarly demonstrated by Cairns' (1984) model in which the output of the lexicon andstructural process feed directly into the interpretive processor. The benefit of a systemoperating in this manner is that it ensures that the analysis of the input will always be developedas far and as fast as it can. If one subscribes to this type of obligatory bottom-up processingthen one is forced to allow the recruitment of top-down processes only after the automatic flowof the bottom-up proceses have been interrupted. Within this perspective top-down processesare recruited in the 'second pass' after bottom up processes have failed to achieve completecomprehension on the first pass.Another way of differentiating bottom-up and top-down processing is to say that bottom-upprocessing involves only the perceptual processes that result in the activation of thephonologically appropriate lexical items in the lexicon and that all subsequent processes aretop-down because they involve structures and processes that are integral components ofhigher-level human cognition. Within this perspective first pass and second pass processing isa moot issue.15For our present purposes it is important to be aware that where bottom-up processes end andwhere top-down processes begin have not been universally agreed upon, however, it seems tobe unanimous that the initial speech input is acted upon by bottom-up processes and that top-down processes do involve the recruitment of context and world knowledge in achievingcomprehension. It is beyond the scope of this discussion to review all the experimentalresearch that has accumulated on this topic, but since the current experiment manipulates thequality of the acoustic-phonetic signal, it is relevant to discuss some of the studies which haveexamined the the facilitory effects of top-down processing on the perception of lower-levelinformation when the acoustic signal is degraded.In 1963, Pollack and Pickett conducted an experiment in which they isolated words from taperecorded samples of normal converstaional speech and presented them individually. Theyfound that the words were not intelligible in isolation, but were readily understood whenreinserted into the fluent speech. In isolation the articulatory precision of each of the individualwords was extremely poor and, as a consequence, could not be understood. Also in 1963,Miller and Isaard demonstrated that a word imbedded in noise is more intelligible if it is framedby a normal sentence then if it is framed by a semantically anomolous sentence, and is leastintelligible when it is framed by a syntactically anomolous string of words. This result wasinterpreted to mean that the intelligibilty of strings of words depends partially on theirconformity to the linguistic rules of the listener.The phonemic-restoration effect as described by Warren (1970) also demonstrates the operationof top-down processes. This effect was found when subjects heard a speech signal as normalwhen a phoneme in the speech signal had been replaced by white noise and even when theywere told in advance that they would be listening to a speech signal in which a phoneme hadbeen replaced.16Wingfield, Poon, Lombardi, and Lowe (1985) have demonstrated the facilitory effects ofcontext on the speech signal when it has been compressed in time. Using normal sentences,semantically anomolous sentences, and syntactically random strings they showed that whenspeech was speeded up to 425 words per minute (e.g., more than twice the rate at whichnormal conversations occur) young and old subjects showed excellent recall with normalsentences and significantly poorer recall with syntactically anomolous strings.Together the studies described above demonstrate that when bottom-up processing at theperceptual level is less than ideal, top-down processing is recruited to facilitate construction ofa coherent message. This point is relevant to this experiment since it is the intention of thisstudy to perform an error analysis on the types of responses people produce when theyexperience listening conditions in which the signal is systematically degraded. By evaluatingthe responses in relation to the semantic, syntactic, and phonetic characteristics of the targetmessage insight into the actions of bottom-up and top-down processsing may be gained.Cognitive Considerations The effectiveness of top-down processing is a function of the higher-level cognitve structuresand processes that comprise it. For the purposes of this study, it is relevant to discuss some ofthe relevant cognitive structures and processes: coherence, inference, world knowledge andschemata. It is also important to consider how working memory affects comprehension. Thisis the cognitive space where these processes interact and produce the final output of thecomprehension process.Coherence and InferenceAfter the speech signal impinges on the comprehension system of a listener, during processing17each word, phrase, clause, and/or sentence is integrated into a conceptual representation. Theproperty which confers a sense of relatedness on a passage of discourse and whichdifferentiates a text from a sequence of unrelated sentences or ideas is known as coherence(Brown & Yule, 1989). Coherence is computed by the listener with the aid of a collection ofknowledge sources: knowledge of the prevailing situation, or the situation being referred to;knowledge of the causal and logical relations among the described events; knowledge and useof the explicit cues provided by the discourse to guide the integrative processes; and knowledgeof the discourse structure (Just & Carpenter, 1987).Brown and Yule (1989) identify two principles of interpretation that are available to the listenerand which form the basis of his or her assumption of coherence. These are the principle oflocal interpretation and the principle of analogy. The principle of local interpretation dictatesthat the listener not construct any text larger than he or she needs to arrive at an interpretation.It is the listeners knowledge of the world which constrains his or her local interpretation. Thisis related to the notion that it is an individual's experience of past events that will generateexpectations about what are likely to be relevant aspects of context. The principle of analogy states that the listener will assume that everything will remain as it was before unless specificinformation is given otherwise. It is this principle that underlies the idea that human beings tryto interpret any text they encounter (Brown & Yule, 1989).If the discourse or text does not explicitly indicate how the clauses or sentences are related toone another then the listener must engage a second strategy that aids in generating anunderstandable and interrelated interpretation of what they are hearing or reading: Inference.According to Just and Carpenter (1987) the integration of sentences of a text is particularlydependent on inferences. Inferences are based on the comprehender's knowledge of theworld, knowledge of his or her language, and knowledge of the preceding portions of18discourse already received.World Knowledge & SchemataIn a discussion of speech and text processing it is necessary to consider the knowledge that areader/listener contributes to the comprehension process. To any listening situation the listenerbrings with him or her a body of knowledge which has been acquired through either directexperience with objects, people, and events in the real world or through indirect experiencewith a world described by a text or another person. This prior knowledge influences andshapes the comprehender's interpretation of every aspect of his or her experience in the world,including his or her experiences with text (see de Beaugrande, 1980).The issue of how 'prior knowledge' functions in language comprehension is at the heart ofschemata based theories of comprehension. Schema theories have focused on outlining theinteraction of old knowledge with new incoming knowledge. A schema is a stereotypedinternal representation of elements and their relations (Mandler & Johnson, 1977) andsummarizes what is known about a variety of cases that differ in their detail (Anderson &Pearson, 1984). A schema provides a means by which incoming information can becomeintegrated into existing knowledge structures. These highly organized generic knowledgestructures contain a set of slots, where each slot defines a particular type of information.These slots are instantiated with the content of incoming information (Anderson, Spiro, &Anderson, 1978). Different schema-constructs have been identified: Schemas (Bartlett,1932), scripts (Schank & Abelson, 1977), and frames (Minsky, 1975). Each of these containshighly organized structures and conventionalized information about a set of elements and theirrelations. Perhaps the most frequently cited example of a schema is that of going to a restaurantto eat. Individual occurrences have much in common but may differ in specific details. Thereis a common goal structure (i.e., getting something to eat for a price), there is a common19setting (i.e., a place of business which earns money through the preparation and sale of food),there is a common set of roles and instruments which tend to be present (i.e., customer, waiter,kitchen, tables, menus, plates) and the actions are standard (i.e., ordering, eating, paying).What the waiters are wearing, what is on the menu, how expensive the food is are some of thedetails that are likely to vary from restaurant experience to restaurant experience. In theinterpretation of a particular situation the elements in the situation are matched with the elementsthat are part of the generic characterizations in the schematic knowledge structure itself. It is inthis way that the slots of the schema are instantiated with the semantic content of a particularsituation.Evidence has been presented which supports the use of schemata in the comprehension andmemory of text materials (see Kintsch, 1977; Mandler and Johnson, 1977; Rumelhart, 1977;Thorndyke, 1977). These researchers have focused on demonstrating that text comprehension,encoding and retrieval processes rely heavily on the activation and use of schemata. For eachpassage of text a comprehender encounters, a schema consisting of a frame with labelled butunfilled slots is evoked/activated. When a schema is activated its slots are subsequentlyinstantiated with the contents of the incoming information.The literature on schema recognizes that, in addition to schemata for routine events (i.e., goingto the proverbial restaurant), there are also schemata for structural aspects of text (Just &Carpenter, 1987; Minsky, 1975; Rumelhart & Ortony, 1977). Just & Carpenter (1987)describe a schema for a fable as containing slots for setting and characters as well as for aninitial event, a complication, a resolution, and a moral. They suggest that a schema wouldindicate aspects of the temporal relations among the components as wellWhen a highly structured schema is evoked highly organized macrostructures are generated.20Well organized macrostructures facilitate comprehension processes of text, thereby, makingthem more efficient and effective. It is the contention of Kintsch & Young (1984) that sometexts invite the construction of efficient well-defined macrostructures and others do not,depending on the schema that is activated by a particular text. The finding that narratives arerecalled better than other texts can be interpreted within this framework: When listening to orreading a narrative the comprehender brings with him or her a well worked out schematicstructure with which to organize the text during comprehension. This permits the developmentof a stable macrostructure which consists of elements in the text that are essential for theunderstanding of the text. Kintsch & Young (1984) contrast comprehension and recallperformance of narratives with that of expository texts. They suggest that because expositorytexts lack a strong inherent organizational structure, the schemata that are evoked are poorlydefined and as a consequence the macrostructures which are formed by the reader/listener areloosely constrained. In their 1984 study Kintsch and Young found that recall was better for anarrative text type than for an expository-descriptive text type. Similarly, Kintsch andYarbrough (1982) observed that subjects were better able to answer questions for texts thatwere clearly organized according to a familiar rhetorical form than for terxts which containedidentical content but lacked organozation. Thereby supporting the idea that the generation ofwell defined macrostructures leads to better performance in comprehension measures.Also, Just and Carpenter (1987) cite that when a story does not permit a schema to be evokedreaders rate the story as difficult to comprehend and demonstrate poor recall. Findings such asthese demonstrate that comprehenders rely on schemata to organize the information containedin a text. When a familiar text is encountered, either because of its content or structure, theschemata which the comprehender possesses facilitates the speed of interrelating sentences,formulating expectations and drawing inferences. In this way comprehension is facilitatedbecause the schema helps the reader impose an organization on the incoming information,21providing her with a ready made structure into which she can insert the new information sheacquires from the textSchemata & Prediction Schemata guide the interpretation of explicit input, the generation of inferences, and theformation of expectations (Anderson & Pearson, 1984; Kozminsky, Kintsch & Bourne,1981). Over time and experience the individual comprehender recognizes regularities andgeneralizes over his or her experience. In this way it becomes possible for the individual torecognize incoming experiences as being of a particular type (e.g., a job interview, a lecture, anarrative, a recipe). This process of drawing generalizations is what underlies the developmentof schemata. Equipped with templates for particular events and/or for particular linguisticstructures it becomes possible to predict what is likely to happen, and what are and are notlikely to be relevant features of a situation. In fact, schemata play such a strong role in thedevelopment of our expectations and abilities to predict what is a probable component in agiven situation that the phrase "structures of expectation" has been used to characterize theinfluence of schemata on our thinking (Tannen, 1979 as cited in Brown & Yule, 1989).According to den Uyl and van Oostendorp (1980 ) an expectation is any kind of readiness toaccept one course of events as more likely, more natural than another. Research studies haveshown that readers do use a schema as a set of expectations for predicting the structural aspectsof text (Mandler & Johnson, 1977, Kintsch & Van Dijk, 1978; Stein & Glenn, 1979).Working Memory When we read a book or listen to a lecture or conversation, our cognitive systems mustintegrate the incoming information in such a way that coherent semantic units areconstructed which permit the passage or lecture to be understood. For the most part, theinput an individual receives when reading or listening is serial in nature. As the input isreceived it is related to and integrated with concepts that have just preceded it and priorknowledge housed in long term memory. In the process of manipulating the incominginformation both partial and final products of comprehension are generated. It is aconsequence of the sequential nature of language input that, in order for thesecomputational products to be integrated with one another to effect successfulcomprehension, the comprehender must be able to temporarily store information from earlyparts of a sequence in order to be able to integrate and relate them to later parts. What istypically referred to as 'understanding' is in essence the processing, integration, andtemporary storage of interrelated concepts.The site where information can be both processed and temporarily stored during complexcognitive tasks such as language comprehension has been identified as a single centralsystem called working memory (WM) (Baddeley & Hitch, 1974; Carpenter & Just, 1989;Just & Carpenter, 1987). Through the allocation of its processing and storage capabilitiesit is the role of working memory to manage incoming information from perceptual encodingprocesses, retrieval from long term memory, and/or from the output of a comprehensionprocess (i.e., a partial or final product) (Daneman & Carpenter, 1980).In a complex cognitve task, such as language comprehension, an individual receives andmust process information across time. For example, when listening to a lecture an effectivelistener must 'take in' the topic of the lecture, the representation of the ideas presented, themajor propositions from preceding sentences and their interdependencies, as well as arunning multi-level representation of the most current statement. It is information of thistype that WM must manage and allocate its storage and processing resources to. For anidea to be be housed in WM it must be actively maintained if successful integration is tooccur (Carpenter & Just, 1989). It has been proposed that information is kept active in22working memory through its participation in the comprehension process (Daneman andCarpenter,1980). That is, if a piece of information is not actively integrated into theexisting "chunks" of structure that are being maintained in working memory andsubsequently stored within that chunk in working memory, then that concept is lost fromworking memory.Working memory is assumed to have a limited capacity (Miller, 1956). The limits ofworking memory have been demonstrated to be seven chunks (Miller, 1956), out of thisresearch finding has arisen the question of how large a chunk is (Simon, 1974): Is a chunka single item (e.g., one digit) or is it a hierarchical structure (e.g., a propositional network).Contemporary thought is that the set of seven chunks refers to seven pointers that indexrelational data structures (Carpenter & Just, 1989). By possessing "chunks" which can bequite large it becomes possible to more fully understand how people can reason andunderstand language in spite of the limitations of WM.Even though complex cognitive processes can be carried out, WM is not infinite and boththe processing and storage functions must share this limited resource (Baddeley & Hitch,1974; Daneman & Carpenter, 1980). As a consequence of this sharing, tradeoffs occurbetween the storage and processing components of WM; it is assumed that the processingfunction takes priority over the storage function. Daneman and Carpenter (1980) designeda comprehension task to demonstrate the tradeoffs between the storage and processingcomponents of WM. In the task, a subject is presented with sets containing two to sevenunrelated sentences. The subject is required to read each of the sentences in a set aloud andat the end of each set of sentences recall the final words of each sentence. The largest setsize from which all of the sentence final words can be recalled is defined as thatindividual's reading span. Based on the hypotheses that information is kept active through23its participation in the comprehension process and that maintaining even a small number ofextraneous unrelated words in WM places considerable burden on WM because theirmaintenance is not a natural part of the comprehension proceses, together with the notionthat processing takes precedence over storage, Daneman and Carpenter (1980) reasonedthat individuals with more efficient processing would have more storage space to dedicateto the maintenance of the extraneous word than would indiviudals with less efficientprocessing, and would therefore demonstrate greater WM spans. Their results supportedthis hypothesis: subjects with greater WM spans were better readers than those with smallerWM spans (Daneman & Carpenter, 1980). Other research findings have supported thishypothesis: It has been demonstrated that individuals with larger WM spans are moreefficient readers or listeners than those with smaller WM spans (Daneman & Carpenter,1980; Daneman & Green, 1986).It is important to note that WM does not possess a set capacity. Rather the capacity of WMvaries as a function of how efficient the individual is at the specific processes demanded bythe task to which WM is being applied (Daneman & Green, 1986). Also, storage capacityis only traded away to provide additional processing resources in those situations in whichan individual is close to or above his or her WM span (Carpenter & Just, 1989). Thus,exactly when processing becomes a burden is different for different memory loadconditions and is different for different individuals.WM capacity is a crucial source of individual differences in language comprehension. Thecapacity of WM is determined by the efficiency with which the processing and storagefunctions are coordinated with one another. Individuals with less efficient processes mustdedicate more of their WM resources to processing functions and as a consequence are leftwith a functionally smaller storage space (Daneman & Carpenter, 1980).2425A discussion of WM is relevant to the current study because the constraints imposed byWM may account for individual differences in listening to speech in noise. A listener withless efficient WM processes may demonstrate poorer abilities to understand speech-in-noisethan an individual with more efficient WM processes. In very poor signal-to-noiseconditions the resources dedicated to perceptual (or bottom-up processing) are greater thanin ideal listening conditions. This may have the effect of significantly reducing bottom-upprocessing. Using this line of reasoning, individual's with more efficient WM systemswill have more available resources to dedicate to higher-level cognitve processsing andtherefore will be more successful in difficult listening situations than those individuals withless efficient WM systems.If WM is the site where bottom-up meets top-down processing, then the interactionsbetween these two processing systems will be constrained by the capacity and limitations ofthe WM system. As previously mentioned, processing demands take precedence overstorage demands. Therefore if a given task requires a great deal of processing, then storagecapacity will be reduced. If storage is seriously degraded, then an individual will have amore difficult time integrating and interrelating incoming information. Consequentlycomprehension will become very difficult. In the case of listening to speech in noise , ifthe noise is so great that almost all the resources of WM are devoted to processing, thenstorage will be reduced and it becomes difficult to construct a coherent message.Evaluation of Speech UnderstandingEveryday CommunicationIdeal conditions for commuication consist of two-talkers in a quiet environment using languagethat is familiar, simple, and clearly articulated. Everyday listening , however, typicallyinvolves situations that are less than ideal. We often need to understand an unfamiliar message26spoken with less than perfect articulation in a noisy environment amidst other competingdistractions. The stimuli and procedures used in traditional audiological evaluations are notdesigned to assess individuals in everyday listening conditions.Stimuli Used in Communication EvaluationsResults from pure-tone tests do not provide an accurate means by which to assess anindividual's ability to understand speech in everyday communication situations (e.g.,Elliott,1963; Marshall & Bacon, 1981; Young & Gibbons, 1962). As a consequence, standardaudiologic procedures have employed suprathreshold speech stimuli to evaluate speechunderstanding (see Penrod, 1985). These stimuli have included words and sentences,primarily.Speech Signals Used in the Evaluation of Speech UnderstandingWord-Level Speech SignalsMonosyllabic word tests are perhaps the most widely employed audiological means ofassessing an individual's speech understanding. In North America, the CID Auditory Test W-22 (Hirsh et al., 1952) and the Northwestern University Auditory Test No. 6 (Tillman &Carhart, 1966 as cited in Penrod, 1985) are two such tests which receive widespread use.They contain lists of fifty single-syllable words that are balanced for word familiarity andeither phonetic or phonemic distribution. The primary function of monosyllabic word tests isthe assessment of an individual's ability to make correct phonemic judgements based onacoustic information (see Brewer & Resnick, 1983). Performance on these tests is oftenconsidered as a means by which to assess daily communication effectiveness and anindividual's social adequacy in hearing (e.g., Hodgson, 1980; Penrod, 1985). Such tests,however, do not incorporate the syntactic, semantic, pragmatic, contextual, suprasegmentaland cognitive information that are readily available in everyday speech: they assess only the27perceptual and word-level hearing abilities of an individual . For these reasons, performanceon monosyllabic word tests clearly falls short of being an adequate reflection of an individual'sfunctional communication abilities.Sentence-Level Speech SignalsIn an effort to ensure the individual could access cues normally available in an everydaylistening environment, suprathreshold sentence tests were developed. Silverman and Hirsh(1955) created a set of sentences which consisted of ten lists of ten sentences with fifty keywords contained in each list. In the design of these sentence-level tests, word familiarity,phonetic balance, word length, free variation in syllabic stress, phonetic distribution withineach sentence, number of words per sentence, redundancy and abstraction were all considered.Also in an effort to overcome some of the shortcomings of the monosyllabic word tests,Speaks & Jerger (1965) introduced the test of Synthetic Sentence Identification (SSI). Thistest used sentences which followed the rules of English syntax but were semanticallyanomolous. It was proposed that although the SSI is a closed-set identification test, thesematerials were a better approximation of everyday speech understanding than monosyllabictests because intonation patterns were preserved and responses were reliant on a closed set ofinformation (i.e., knowledge of syntax and limited response set).The speech perception in noise (SPIN) test was developed by Kalikow, Stevens, and Elliott(1977) as a means to assess a listener's ability to understand everyday speech. It was theircontention that since speech understanding depends upon both the function of the peripheralhearing system (i.e., bottom-up processing) and the employment of cognitive processes (i.e.,top-down processing) then any test of speech understanding necessitates that both of thesekinds of processing be assessed. To accomplish this, they created a test which was sensitive totwo types of information contained in the speech signal: the acoustic phonetic and the28syntactic-semantic (i.e., context dependent information). In the SPIN test, each test item is inthe form of a sentence. The listener is required to report the sentence-final word after hearingeach sentence. The original test was comprised of ten forms of fifty sentences each. Each formcontained twenty-five high-predictability and twenty-five low-predictability sentences. Asentence-final word is considered high in predictability if it can be identified through the use ofsyntactic, semantic and prosodic cues available in the sentence (e.g., A zebra has black andwhite stripes). A sentence is defined as being low in predictability when the only means ofidentifying the last word in the sentence is through the recognition of the acoustical propertiesof that word (e.g., The man talked about the stripes). On the basis of studies of normal-hearing subjects, Kalikow et al. (1977), concluded that each of their forms was equivalent toone another with respect to phonetic content, average intelligibility, predictability of key words,and word familiarity. Instead of using a sample of normal-hearing subjects, Bilger, Neutzel,Rabinowitz, and Rzezkowski (1984) conducted a study to standardize the SPIN test on hearingimpaired-listeners. They found the SPIN test as devised by Kalikow et al. (1977) to containone weakness: the ten forms demonstrated statistically significant differences in theirequivalence. Based on this finding, they revised the lists to be equivalent for hearing-impairedlisteners. By this process they reduced the number of lists from ten to eight, but maintained theratio of high to low predictability sentences.According to Martin (1986), sentence tests have not gained a popular following because "theirvery construct enables the better guesser to make more meaning of a sentence than anotherpatient with a similiar speech discrimination problem " (p.135). While it is true that the use ofcarefully balanced and well standardized sentences containing no inherent predictability willdifferentiate good discriminators from poor discriminators these tests will not differentiateeffective from ineffective communicators.29To date, the SPIN sentences are one of the best attempt at measuring speech understandingThey not only assess the perceptual abilities of the peripheral hearing mechanism, but they alsoconsider the contribution of higher-level linguistic/cognitive processing to speechunderstanding. Although the advent of this test moves standardized suprathreshold speechtesting in a direction which allows for a more valid evaluation of speech understanding thantraditional speech discrimination tests, it does not account for the roles of cohesion, coherence,inference, prior knowledge, and memory in everyday speech understanding behavior.Discourse or text materials would involve these components and therefore would be the mostappropriate material to use in assessing functional speech understanding abilities.Discourse-Level Speech SignalsIn an effort to create a test of speech understanding that possessed a high degree of facevalidity, the Connected Speech Test (CST) was developed (Cox, Alexander, & Gilmore1987). Underlying this test was the rationale that since continuous discourse is the type ofspeech typically encountered in everyday listening situations, then tests of speechunderstanding should use continuous discourse. The test consists of forty-eight passages ofconnected speech, each with twenty-five scoring words. Each passage is about a familiar topicand consists of ten sentences and word describing the topic is presented prior to thepresentation of a passage. The sentences are presented to the listener one at a time and thelistener is required to repeat each sentence exactly as it was heard. Performance is defined interms of the percentage of scoring words correctly repeated. A multi-talker babble, whoselevels are adjustable depending on the goals of a given evaluation, is used as a competingsignal. The presence of speechreading cues is optional, again depending on the goals of theevaluation (Cox, Alexander, Gilmore, & Pusakulich, 1989).Although this test incorporates continuous discourse, thereby involving the processes of30cohesion, coherence, inference, prior knowledge, and memory, its potential for capturing andquantifying how effectively an individual really understands speech has not yet been exploited.Performance on the CST is based on the percentage of correct word repetitions. This type ofmeasure does not evaluate speech understanding, rather it is yet another test which appears tobe assessing speech understanding but is not. The CST was primarily developed as a tool withwhich to evaluate hearing aid benefit: The hearing aid yielding the best CST score was selectedas the hearing aid providing the greatest degree of communication benefit (i.e., speechunderstanding). When speech understanding is evaluated in this manner, it is not possible toevaluate what processes (i.e., bottom-up or top-down) are engaged or how well the message iscomprehended. If we are to evaluate speech understanding in a way that allows us to makejudgements about communication abilities, then the assessment of each of these processes mustbe included.In the typical clinical setting, discourse or text materials are not employed in the assessment ofcommunicative impairment. The traditional focus on rigid test standardization and the need toeliminate useful redundancies in the test signal have created tests which are efficient atidentifying perceptual difficulties but have prevented the evolution of clinical materials whichare ecologically representative of everyday speech understanding. Before such clinicalmaterials can be developed, a knowledge of different discourse/text structures and anunderstanding of how different text structures interact with speech understanding arenecessary.Text Structure Text structure, text genre, text type, and rhetorical form are a few of the terms used to refer tothe supra-organization that unites a text. This global organization is not dependent on syntacticrelationships but rather on underlying semantic relations which yield the overall interpretaion of31a text. Kintsch and van Dijk (1978) characterize the semantic structure of discourse at twolevels: The microstructural and macrostructural. The microstructure of a text refers thestructure of the individual propositions and their relations, whereas the macrostructure is moreglobal and characterizes the text as a whole. The macrostructure of a text expresses the gist,overall organization, and main points of a text (Kintsch and Yarbrough, 1982). It is themacrostructure that provides a text with overall coherence and it is this facet of text structurethat is relevant to this discussion.Standard text structures are thought to underlie all texts (Fayol, 1991; Hiebert, Englert, &Brennan, 1983; Horowitz, 1985; Kintsch & Yarbrough, 1982; Kintsch & Young, 1984;Meyer, 1975) and a number of researchers have sought to describe and classify each of thesetemplates. Mandler and Johnson (1977) and Rumelhart (1977) describe the structure ofsimple stories. Thorndyke (1977) characterizes single-goal, single-protagonost narratives.Kintsch and van Dijk (1978), Meyer (1975,1977) and, Meyer and Freedle (1984) consider thetext structure of expository writing.Even from this short listing it can be seen that a text structure is typically categorized as either anarration or an exposition. Narrative structures consist of more temporal and causal relations,whereas expository text is more loosely defined in its semantic relations. Once a text has beenidentified as either narrative or expository, it is then often analysed further using a storygrammar (Mandler & Johnson, 1977; Stein & Glenn, 1979) or an analysis technique specific toexpository texts. The best known of these macrostructure analysis techniques are those ofMeyer (1975) and Kintsch and van Dijk (1978). Analysis of rhetorical structure is not a goalof this study and therefore a detailed discussion of text analysis will not be covered here.Several researchers have identified and defined categories of text types which involve more32than a simple assignment of a text to either just the narrative or expository category. Meyer(1975) organizes text types into five basic categories: Collection, description, causation,problem/solution, and comparison. Kintsch and Yarbrough (1982) describe four differentrhetorical forms: Argument, comparison and contrast, definition, and procedural descriptions.Horowitz (1985) identifies five types of text types found in school texts: temporal order,attribution, adversative, covariance, and response.Despite the differences in nomenclature each of these classification systems encompasses textgenres of a relatively similar nature. Meyer's collection structure refers to a list of elementsassociated in some manner. Meyer's description structure has only one organizationalcomponent: elements within the text are subordinate to the topic. Horowitz's category ofattribution is somewhat akin to both these organizational structures; whereas Kintsch andYarbrough's category of definition is more similar to Meyer's description structure. Kintschand Yarbrough and Horowitz each describe explicit categories for events associated with aspecific temporal order: procedural description and temporal order. Meyer does not have sucha category, although through the sequencing of either a collection or a description a similarcategory of text type is defined. The categories of causation and covariance are equatable, andthe structure type they represent is more organized than either the collection or descriptionforms. Elements in a causation structure are grouped within a time frame and are causallyrelated. Meyer's comparison structure, Horowitz's adversative structure, and Kintsch andYarbrough's comparison-contrast structure are all representative of the same category: the textis organized on the basis of similarities and differences, not on time or causalities. Thecategories of problem/solution, and response are also equatable. The text type they represent isperhaps the most organized of the schemes. According to Meyer (1975) it has all theorganizational components of the causation structure with the addition of overlapping content33between propositions in the problem and the solution. Since the text analysis systems used byeach of the above researchers is based on expository rather than narrative writings, they do notspecifically categorize the structure of stories. Meyer and Freedle (1984), however, state thatstories contain a combination of these categories and as an example cites folktales ascontaining description, causation,  and collection within an overall problem/soultionorganization where the protagonist confronts and resolves a problem.Each of these discourse types can be found to occur in a variety of familiar contexts: Politicalspeeches (comparison); grocery lists (collection); recipes (procedural descriptions); and stories(any number or combination of the categories). Kintsch and Yarbrough (1982) and Meyer andFreedle (1984) describe most texts as containing more than one of these basic organizations.It may in fact be the case that pure text types do not occur in everyday reading or listening. It ismore typically the case that individual pieces of text possess a combinatiuon of text typeswithin which a particular text type may or may not dominate. In fact all the variouscombinations and permutations of text types within a piece of discourse may be representativeof separate and individual text types, which to date have not yet been fully delineated becauseof the complexity of such a task.It has been demonstrated that the text types have differential effects on recall andcomprehension. Recall for passages organized as descriptions have been shown to be poorerthan recall for passages organized within a problem-solution structure (Meyer & Freedle,1984;Meyer & McConkie, 1973), comparison or causation structure (Meyer & Freedle, 1984).Narrative texts have been shown to be more effective in facilitating comprehension and recallthan expository texts (Graesser, 1981; Graesser, Hauft-Smith, Cohen, & Pyles, 1980; Kintsch& Young, 1984; Kozminsky, 1977; Tun, 1989). It is thought that the the additionalorganizational components found in the the comparison, causation, problem-solution,  and34narrative structures facilitate encoding, efficient use of memory, and retrieval processes(e.g., Kintsch & Yarbrough, 1982; Meyer & Freedle, 1984).Within each of the three clasification systems described above it becomes apparent that there isa continuum across which the organization of a text varies. Meyer and Freedle (1984)express this contimuum with description at the least organized end and problem/solution at themost organized end of the continuum. Similarily, Kintsch and Young (1984) identifynarratives as possessing more organizational cues than descriptions.Some text types possess a conventional ordering, which results in their associatedmacro structures being highly predictable: In a procedural description the order in which aseries of steps are to be executed prescribes the order in which they are to be given; in adefinition the most general information precedes more detailed attributes; and in a story theproblem to be confronted and resolved is presented before the solution. For other text typescues to organization are much looser, such that there is no pre-ordained sequence to which theinformation contained in a text must conform ; for example, in a simple description theinformation presented is not dependent on a particular position in the text, or on its positionrelevant to that of another piece of information.It is the intention of this study to capitalize on the inherent predictability conferred on certaintext types in examining how different text structures affect an individual's ability to listen todiscourse in noise. It is expected that when listening to a passage of text whose organizationdictates a macrostructure of a highly predictable form, performance in noise will be superior tothat of listening to a passage with a less predictable macrostructure. This hypothesis is basedon the idea that when a text structure is well-defined, predictable and organized in a manner thatpromotes top-down processing (see Mandler & Johnson, 1977), then performance in noise35will be optimal, since reliance on perceptual properties of the incoming speech stimulus are lessbeneficial than engaging higher level cognitive processing.Competing Signals used in the Evaluation of Speech UnderstandingWe typically listen to speech in the presence of background noise yet, with the exception of theSSI and SPIN tests described above, suprathreshold speech materials were initially designed tobe presented in quiet. Testing in quiet does not accurately reflect the discrimination problemsencountered in everyday life. If we are to evaluate the everyday speech understanding of anindividual then we must consider their performance in the presence of noise. In listening tospeech, the auditory system must separate the information bearing components of the acousticalsignal from the unwanted background noise.Low-pass filtered noise (Cohen & Keith, 1976), a single speaker (Speaks & Jerger, 1965), amixture of noise plus one or two speakers (Carhart, Tillman & Greetis, 1969), a babble of oneto three speakers (Carhart, Johnson & Goodman, 1975 as cited in Martin, 1986) , or a babbleof two and four talkers (Young, Parker & Carhart, 1975) have all been used as competingsignals in speech discrimination testing. For young normal-hearing subjects it has been shownthat speech recognition thresholds are lower when the competing signal is speech, than whenthe competing signal is noise with the same long-term average spectrum as the speech stimuliitself (Duquesnoy, 1983).Summary of the Shortcomings of Current Evaluation Stimuli Speech discrimination tests are valuable tools in determining the extent of an individual'sdiscriminatory difficulties in a very controlled setting and this is an initial step in evaluating anindividual's overall hearing function, but there is a need to recognize the difference between anevalution of the discriminatory abilities of the peripheral hearing mechanism and an evaluation36of functional communication abilities.The speech signal is highly redundant: Acoustic, phonemic, linguistic, contextual, andsuprasegmental cues often overlap in the information they convey. As a consequence, areduction in one or several of these may not significantly impair the effective communicationof a message. It is the ability of a listener to use higher-order contextual structures to frameand interpret lower-order phonetic information that enables a listener to contend with adegraded speech signal and understand a conversation. Given this case, simply testing anindividual's perceptual abilities with single word tests or testing an individual's abilities toreceive sentences in isolation or in quiet are not adequate tools on which to base judgements ofcommunication impairment or effectiveness.Procedures Employed in the Evalution of Speech UnderstandingReception TasksIn suprathreshold tests the required response is often a verbal or written repetition of thestimulus presented (e.g., a word). If the listener is not provided with a specific set of itemsfrom which to select the response, the test is said to be 'open-ended' and the task required ofthe subject is one of recognition. If the listener is provided with a visually displayed list fromwhich to select a response, then the test is said to be 'closed-response' and the task required ofthe subject is one of identification. In either of these cases, the task required of the subject is areceptive one: a word is presented and the subject responds. Everyday communicationinteractions involve at least two people, therefore an evaluation method which employs aunidirectional flow of speech can not adequately represent communication effectiveness.TrackingIn 1978, DeFilippo & Scott introduced a method with which to evaluate the reception of37ongoing speech. This method, known as tracking, was aimed at overcoming two limitations ofmore traditional speech procedures: It was developed around the use of connected speech andit aimed to address the issue of correct speech reception (rather than having only a singleopportunity to respond). In this task a speaker transmits connected discourse to a listener byreading short segments of a text. The listener is then required to repeat back verbatim whatwas read. If the listener repeats the segment back correctly, the speaker moves onto the nextsegment. When the listener fails to provide a verbatim response, the speaker proceeds to useone or more of a series of strategies to help the listener achieve 100% recognition. The speakerdetermines the segment length presented; based on her decision of what a logical linguisticsegment should be and what segment length will not exceed memory constraints. The listeneris also free to interrupt the speaker at any time, therby shortening the segment to be repeated.Thus a two-way, albeit somewhat restricted, interaction is achieved. The goal in tracking is toelicit a repetition that matches the text verbatim without the use of non-speech related cues(e.g., gesture or facial expression). A typical tracking session lasts from five to ten minutesand the session is scored by calculating the average number of words correctly repeated perminute.Although the tracking procedure involves the use of discourse materials in an interactivefashion between the sender and the receiver, it is still no more a reflection of normalconversation than sentence-level tests. (Tye-Murray & Tyler, 1988). Tracking requires that thereceiver repeat words verbatim and this requirement is no different than sentence tests whichalso require verbatim responses. Comprehension of a conversation requires that the receivergo beyond identification of the words being spoken. Comprehension involves 'getting thegist' of a message as opposed to each word (van Dijk & Kintsch, 1983), and it also requiresthe listener to make inferences because all the information conveyed in a conversation is notalways stated explicitly. Although an individual may not be able to identify or remember all the38words presented, he or she may still comprehend the meaning of the message. Tye-Murray &Tyler (1988) pointed out that in the case of slow tracking, the procedure actually possesses lessface validity than sentence tests because a slow rate involves multiple repetitions of syllablesand words by both the sender and the receiver, which is not at all like normal communication.Summary of the Shortcomings of Speech Evaluation Procedures If the goal of assessing speech understanding is to obtain a measure of how well an individualfunctions in everyday life, then it is imperative that assessment protocols employ procedureswhich approximate day-to-day communication situations. For this reason reception tasks andtracking are not adequate assessment tools. Reception tasks exclude the speaker from thecommunication process entirely. And while, tracking does require the interaction of thespeaker with the listener, the interaction is limited and not natural. Erber (1988) advocates theconversational process itself as the vehicle for communication training. In his model (Erber,1988),depicts the interaction between the two communication partners in such a way thatneither the beginning nor the end of the communication interchange is specified. In moretraditional models of communication, the partners are identified as the 'listener' and the'speaker' thereby implying that the communication process is unidirectional. This of course isnot representative of real-life. As Erber (1988) makes clear, real communicators take turns,give and/or get informtion, and exchange ideas and feelings. These aspects of thecommunication process need to be considered in speech evaluation procedures.In the procedures described above, the listener is not required to understand what has beenspoken. There is a need for the scoring of speech evaluation measures to be based not on thenumber of words accurately repeated but rather on the degree to which the meaning of themessage has been communicated. Without a requirement for a demonstration ofcomprehension, suprathreshold speech tests makes no measure of speech understanding. In39order to assess the degree to which the message was understood, the use of a scoringprocedure which analyzes the semantic match of a response to its target is desirable. As well,the use of questionnaires which include questions that require the listener to demonstrate bothexplicit and implicit knowledge about the discourse (or conversation) would provideinformation about an individual's ability to understand speech.CHAPTER 3MethodsParticipants Twelve participants, eight women and four men between the ages of 22 and 29, were recruitedfor this study. All subjects presented with normal hearing as demonstrated by a hearingscreening test; with normal sensitivity defined as thresholds lower than 25dBHL from 500 to3000 Hz (see Appendix A). The number of years spent by the participants in post-secondaryeducation ranged from 5 to 9 years with a mean of 6.3 years. All participants were nativespeakers of english. One subject reported previous experience with listening to speech innoise in an experimental situation and the remaining eleven reported no such prior experience.MaterialsIn the text in noise portion of this study nine passages were employed. Each of the ninepassages was recorded by a female speaker of English with intonation patterns appropriate foreach of the text types at an average rate of 170 words/minute or 4.1 syllables per second. Reallife everyday speech is produced in rapidly articulated strings without much attention toarticulatory production patterns. In traditional speech tasks the speech employed is typicallyspoken with unnatural stress, intonation and articulatory precision. In the current study,approximating natural speech was a goal.Using the NeXT Soundworks software program, each of the passages was digitally recordedin its entirety into a single sound file. Simultaneously, a twelve speaker babble was recordedfrom audiocassette onto the second channel of the same sound file. Each of these mastersoundfiles was subsequently edited into smaller soundfiles which were then fed into a4041computer program developed locally to run these materials in the fashion to be describedbelow. Recording of the speech signal was conducted in a sound attenuating booth using aMadsen Electronics microphone.The nine passages consisted of three text types (i.e., stories, restaurant reviews, and recipes)(adapted from Davies, 1982; Schultz, 1985; Schwartz, 1985; & Wadden, 1990) that werechosen to be representative of three different text structures or genres (i.e., narrative,description, and sequence) (see Appendix B).The passages were equated for readability (Flesch, 1948). Each of these texts was judged tobe at a grade seven level using the Grammatik III software program which bases its operationson Flesch's formula for readability (see Appendix C). This formula does not take vocabularyinto account but, rather, analyses sentences and syllable length and uses these measures as anindex to judge pieces of writing. This level was chosen so that materials would be appropriatefor individuals with a wide range of educational backgrounds. The passages were also equatedfor overall number of syllables. Across the nine passages the number of syllables ranged from502 to 587, and from 502 to 553, 533 to 587 and 529 to 570 for the stories, restaurant reviewsand recipes respectively. The number of lines per passage (see below for the definition of aline) ranged from 41 to 48, and from 45 to 55, and 54 to 62 for the stories, restaurant reviewsand recipes respectively. The passages were not equated for vocabulary. Foreign words werenot used unless explicitly defined in the passage.Each of the passages was divided into clauses, noun phrases, and/or verb phrases, using thefollowing criteria: The ideal division was to split the passages into clauses as per theexperimental evidence of Wingfield and his colleagues (Wingfield, Lahar, & Stine, 1989;Wingfield & Nolan, 1980) who have shown that subjects most frequently divide passages at42clause boundaries when given the opportunity. To divide the passages only at clauseboundaries is too limiting a rule and is not likely an accurate representation of how we processspeech (Cairns, 1984). This is also demonstrated by Wingfield and Nolan's 1980 data in thattheir subjects also divided passages into noun, verb, and prepositional phrases some of thetime. For this reason portions of the passage that appeared awkward and unnatural if placedinto clauses were broken down into noun phrases or verb phrases. Consideration was alsogiven to the intonation patterns that accompanied a given sentence, clause, noun or verbphrase. If dividing a sentence into a clause, or a clause into a noun or verb phrase appeared todisrupt the intonation pattern then the sentence was not subdivided. In light of the fact thatthese divisions were made on the intuitions of the experimenter, a pilot subject was asked tolisten to each of the passages in their entirety and divide them into manageable parts (seeWingfield & Nolan, 1980). The pilot subject introduced divisions at the identical boundariesas the experimenter for almost all lines. In those instances in which the pilot subject introduceda boundary where the experimenter did not, the boundary was changed to match that of thepilot subject when the division was inter-sentential or inter-clausal, but was not changed whenthe pilot subject's boundary fell to include an extra clause, noun, or verb phrase (see AppendixD for examples).A breakdown of the number of lines per passgae and a frequency count of the words per lineacross text is provided in Appendix E. The number of words in a given line range from 1 to 19across all the passages.ProceduresThe study required each participant to attend three sessions. In the first session a hearingscreening test and the first of three text in noise tasks was administered. In the second sessiona second text in noise task and in the third session the final text in noise task were43administered. A hearing and language case history was also obtained.Hearing ScreeningA hearing screening was conducted to assure that each participant presented with normalhearing. Hearing sensitivity was defined as normal if hearing thresholds were 25dBHL orbetter from 500 to 3000 Hz. Speech recetion thresholds, speech discrimination thresholds, anda babble threshold were also obtained. Standard audiometric measurements and procedureswere employed.Text in Noise Task The text in noise task required each participant to listen to a passage of text in twelve speakerbabble noise presented monoaurally to the right ear. The signal was presented at 50dB abovethe babble threshold for the right ear. Each participant listened to three exemplars of threedifferent text text for a total of nine passages of text: three narratives, three restaurant reviews,and three recipes. In a given session a participant listened to three exemplars of only one texttype (e.g., narrative). Three text types can be presented in six different orderings. Twosubjects listened to each of the six possible orderings of the text types. Within a session, thefirst exemplar was always presented at 0dB S:N, the second at -6dB S:N, and the third at-12dB S:N. These S/N values were selected on the basis of pilot experiment in which it wasdetermined that most subjetcs performed nearly perfectly at 0dB S:N and had extreme difficultyperceiving discourse at -12dB S:N. This rigid S/N presentation regime provided participantswith the opportunity to become familiar with the task in the best of listening conditions first. Inthis way, it could be said that some of the errors observed in the 0dB S/N condition are in parta consequence of learning a new task.For this task each participant was seated in a sound attenuating IAC booth and was fitted withTDH-39P headphones. The participants were asked to repeat back as accurately as possible44each line of a passge they were about to hear and were encouraged to revise their responses atany time during the course of listening. They were also instructed to provide multipleresponses if they were unsure of what they heard (see Appendix F).In the report by CHABA (1988), it was recommended that a subject's response serve as thecatalyst to trigger presentation of subsequent speech materials. In this way pacing isappropriate in that the subject is provided with the necessary time to organize and make aresponse. To accomodate this a participant signalled his or her request for subsequent lines ofa passage through the press of a button.Using a NeXT computer and an OB 802 clinical audiometer, the experimenter presented eachof the passages in a line by line fashion to the participant. After each line was presented, andthe participant repeated back as accurately as possible what they has just heard and/or took theopportunity to make revisions and/or additions to previous responses, the experimetner codedthe response and/or revision as correct and/or incorrect. If correct the experimenter went on topresent the next line of the passage and if incorrect the experimenter entered the error and/orrevision and then presented the next line of the passage.After the presentation of the passage, the participant was required to answer ten multiple choicequestions about the passage (see Appendix G). Five of which were requests for explicitinformation and five were requests for implicit information. Participants were made aware inthe initial instructions that they would be required to answer questions following each passage.It was thought that if participants were aware that they would be required to answer questions,then they would be more likely to engage in deeper comprehension processes more often thanif they thought the task was simply to repeat back each line as they heard it withoutconsideration of the contents of the passage as a whole (Craik, 1983; Craik & Lockhart, 1972).45Also, after the presentation of each passage, the participant was required to rate their familiaritywith recipes, restaurant reviews, or stories in terms of their personal experience with these(Appendix H).46CHAPTER 4RESULTSEffect of Signal-to-Noise Ratio and Text Type on Number of ErrorsThe total number of errors observed for all subjects for each of the three text types (narrative,restaurant review, and recipe) for each of the three signal-to-noise (S:N) conditions (0 dB,-6dB, -12dB) are plotted in Figure 2. 500400300200100  —o-- NarrativeReview—o— Recipe    0'	-18 •  -12 	 -6 0 6Signal-to-Noise Ratio (dB)Figure 2. Total number of errors observed for each of the text types in eachsignal-to-noise condition.The number of errors increased as the S:N ratio decreased, with the fewest errors being madeat 0dB S:N, more errors at -6dB S:N, and even more at -12dB S:N. Fewer errors were madein the text type narrative than in either the text types restaurant review or recipe across allthree S:N conditions. An analysis of variance confirmed this description with a significanteffect of S:N condition, F(2,22) = 270.05 , g<.01, a significant effect of text type, F(2,22) =5.23, g<.01, but no significant interaction of S:N condition with text type, F(2,22) = .862,47p>.05. This analysis of variance was conducted after the data were subjected to a square-roottransformation in order to reduce heterogeneity of variance, thereby satisfying the assumptionof homogeneity of variance underlying the test (Hartley's F ax(9,12) = 3.66, g <.01) (Kirk,1982). A Student-Newman-Keuls test of multiple comparisons confirmed that the three S:Nconditions differed significantly from each other, with errors increasing as S:N ratiodecreased from 0 to -6dB, ((<.05), and with an even greater increase in errors as S:N ratiodecreased further from -6 to -12dB S:N, ((<.01). A second Student-Newman-Keuls test ofmultiple comparisons indicated a significant difference between the text type narrative andthe other two text types, (p<..05), but no significant difference was indicated between the texttypes recipe and restaurant review, (p>.1)Effect of Familiarity The relationship between a subject's sense of familiarity with a text type and the number oferrors observed was considered. Subjects who rated themselves as very familiar with a texttype were separated from those who did not rate themselves as very familiar with a particulartext type. The mean error frequency for each of the text types are plotted in Figure 3according to whether the text type was rated as very familiar or not familiar. The number oferrors observed within each S:N condition seems not to be affected by a subject's sense offamiliarity with the text types narrative and restaurant review . Furthermore, the number oferrors observed at 0dB S:N for the text type recipe is not related to a subject's sense offamiliarity . In contrast, the number of errors observed at -6 and -12dB S:N for the text typerecipe appears to be related to the subject's familiarity rating. It is worth noting, that thenumber of subjects reporting that they were familiar with recipes (n=5), was approximatelyequal to the number reporting that they were unfamiliar (n=7), suggesting that these resultscannot be attributed to unusual behavior by a small number of subjects. Furthermore, whileit is difficult to know how subjects interpreted "familiarity", it seems that this notion wasZ 80a) 6040coa)M 2048more easily understood with respect to recipes, a rather specialized text type, in comparisonto the more general narrative and description types. Further discussion of the possiblereasons for this pattern of results will be dicussed later.narFAM narUNFAM revFAM revUNFAM recFAM recUNFAMRating of Text Type FamiliarityFigure 3. Frequency of errors observed for each of the text types in each of thesignal-to-noise conditions as a functin of subjects' familiarity ratings for each texttype.Effect of Signal-to-Noise Ratio and Text Type on Error TypeClassification of Response Error TypesIn an effort to understand the nature of the errors observed, each error was categorized as adeletion, addition , substitution, or exchange. A deletion was defined as the omission of anypart of the target (bound morpheme, word, phrase, clause, or sentence). An addition wasdefined as the addition of any information that was not present in the target (boundmorpheme, word, phrase, clause, or sentence). A substitution was defined as information thatwas used in place of information that was given in the target. An exchange referred either to200 —0--- exchangeNARexchangeREV-0-- exchangeREC 60-18 	 -12 	 -6 	 0Signal to Noise Ratio (dB)49metathesis of portions of the target or to instances where a portion of the target (boundmorpheme, word, phrase,clause, or sentence) was repeated correctly but not in the originaltarget position.Effect of Signal-to-Noise Ratio on the Error Type 'Exchange' The number of exchange errors that occurred in each of the text type conditions in each S:Ncondition are plotted in Figure 4. Overall, few errors of this type were observed, and thenumber of errors did not vary with either S:N or text type condition. A Chi-square testconfirmed this description with no significant effect of S:N condition, X 2 (df=2) = .19,> .05, no significant effect of text type, X 2 (df=2) = 4.71, g > .05, and no significantinteraction of S:N condition with text type, X 2 (df=4) = .71, 12 > .05.Figure 4. Frequency of the error type exchange for each of the text types as afunction of signal-to-noise ratio.0-18 	 -12 	 -6 	 0Signal to Noise Ratio (dB)650Effect of Signal-to-Noise Ratio on the Error Type 'Addition' The number of addition errors that occurred in each of the three text type conditions in eachS:N condition are plotted in Figure 5. More addition errors were made then exchange errors.For addition errors, the number of errors does increased as the S:N ratio decreased but it didnot vary much with text type. A Chi-square test confirmed that the three S:N conditionsdiffered significantly from one another, X 2 (df=2) = 65.08,12 < .005, with errors increasing asS:N decreased. There was no significant difference between the three text types, X2 (df=2)= 2.05,12 > .05 and no significant interaction, X 2 (df=4) = 4.88, 12 > .05>, 200 	 —o-- additionNAR2	 additionREVa) 	 —0— additionRECzEE100Figure 5. Frequency of the error type addition for each of the text types as afunction of signal-to-noise ratio.Effect of Signal-to-Noise Ratio on the Error Type 'Substitution' The number of substitution errors that occurred in each of the three text type conditions ineach S:N condition are plotted in Figure 6. Far more substitution errors were made thanaddition or exchange errors. The number of errors increased as S:N ratio decreased, with agradual increase in errors from 0 to -6dB S:N and a much steeper rise in errors from -6 to-12dB S:N. The number of errors does not appear to vary with text type except that there—o-- substitutionNARsubstitutionREV—o-- substitutionREC•020010051were slightly fewer errors for narratives than for the other two text types at 0dB S:N. AChi-square test confirmed the first part of this description, with a significant effect of S:Ncondition, X2 (df=2) = 484.73, p <.005. However, there was no significant effect of texttype, X2 (df=2) = 2.59, p > .05., and no significant interaction, X 2 (df=2) = 5.99, p > .05.-18 	 -12 	 -6 	 0 	 6Signal to Noise Ratio (dB)Figure 6. Frequency of the error type substitution for each of the text types as afunction of signal-to-noise ratio.Effect of Signal-to-Noise Ratio on the Error Type 'Deletion' The number of deletion errors that were observed in each of the three text type conditions ineach S:N condition are plotted in Figure 7. In comparison to substitution errors,  deletion errors were about as numerous and also increased markedly as S:N ratio decreased. Incontrast to substitution errors, deletion errors did vary with text type. While the number ofdeletion errors observed for each of the text types at 0dB appears to be similar, at -6 and-12dB S:N, differences were evident. At both -6 and -12 dB S:N, the number of errors for52the text type narrative were less than the number of errors for the other two text types.Furthermore, at -6 dB S:N, the number of deletion errors for the restaurant review were lessthan those for the recipe , and at -12 dB S:N this relationship reversed. A Chi square testconfirmed this description with a significant difference between text types, X 2 (df=2) =24.04, p <.005, a significant difference between S:N conditions, X 2 (df=2) = 433.47, g <.005,and a significant interaction between text type and S:N condition, X2 (df=2) = 9.98, g < .05.—0--- deletionNAR—0-- deletion REV>. 200 --a-- deletionRECC0•1121000-18	 -12 	 -6 	 0 	 6Signal-to-Noise Ratio (dB)Figure 7. Frequency of the error type deletion for each of the text types as afunction of signal-to-noise ratio.Summary of the Effect of Signal-to-Noise Ratio on the Error TypesThe number of exchange, addition, deletion, and substitution errors that were observed acrossall three text types in each of the S:N conditions are summarized in Figure 8. Exchangeerrors were fewest and occurred in similar proportions in each of the three S:N conditions.Addition errors occurred more often than exchange , errors, but less often than either deletion or substitution errors. The number of addition errors increased at a steady but gradual rate asthe S:N ratio decreased. Substitution and deletion errors occurred in near-equal numbers in600500400a)cr3002001000-18 	 -12 	 -6 	 0Signal to Noise (dB)6—o-- exchangeadditiondeletion—o-- substitution53each of the S:N conditions. From 0 to -6dB, the number of substitution and deletion errorsincreased at a rate similar to the addition errors, but from -6 to -12dB S:N the number ofsubstitution and deletion errors increased at a much greater rate than the rate of increase ofaddition errors. A Chi-square test confirmed this description with a significant differencebetween the types of errors, X 2 (df=2) = 969.35, R<.005, a significant difference betweenS:N conditions, X2 (df=2) = 917.35, ja <.005, and a significant interaction between error typeand S:N condition, X 2 (df=6) = 52.93, p <.005.Figure 8. Summary of error frequencies across text types as a function of signal-to-noise ratio.Effect of Signal-to-Noise Ratio on Errors According to Linguistic Level and Degree of ErrorError Classification by Linguistic Level Subsequent to classifying errors as additions, deletions, substitutions, or exchanges, eacherror was coded as a total, partial, or no match to the target at each of three linguistic levels(auditory, syntactic, or semantic). For example, if the following target and response—*-- Partial auditory matchNo auditory match1000800600400200054occurred:(1) Target: 	 "and bounced downstairs to the basement"Response: "and bounced downstairs to the cellar"then the response error was classified as a substitution which was not an auditory match,butwas a total semantic match, and a total syntactic match. For another example,(2) Target: 	 "and bounced downstairs to the basement"Response: "and bounced down the stairs to the basement"the error would be classified as a substitution (of a phrase for a lexical item), which wasfurther classified as a partial auditory match, a total semantic match, and no syntactic match"(see Appendix I for more examples).-18	-12 	 -6 	 0	6Signal-to-Noise Ratio (dB)Figure 9. Frequency of response errors which were a partial or complete auditorymatches to the target.55Effect of Signal-to-Noise Ratio on the Auditory Level By definition, errors were always responses which failed to match the target auditorily (eitherpartially or totally). Errors classified as partial or no matches have been plotted in Figure 9.Across all three S:N conditions, there were fewer partial matches than no matches. In bothcases, errors increased with decreasing S:N ratio. A Chi-square test confirmed thisdescription with a significant difference between the two degrees of auditory match X 2(df=1) = 430.89, p < .005, and a significant difference S:N conditions, X 2 (df=2) = 901.31, g< .005, but no significant interaction between degree of match and S:N condition, X 2 (df=2)= 4.48, g >.05.Effect of Signal-to-Noise Ratio on the Semantic Level (Partial Auditory Match)Addition and substitution errors that were classified as partial auditory matches were thencategorized according to the degree (partial, total, no) of semantic match between responseand target (see Figure 10). (It was only meaningful to evaluate addition and substitutionerrors because, by definition, for deletion errors, no semantic or syntactic information waspresent, and for exchange s synactic and semantic information was perfectly preserved). Mostof the responses did not semantically match the target, with the percentage of no matchesbeing similar at 0 and -6dB S:N, and slightly greater at -12dB S:N. While total semanticmatches slightly exceeded partial semantic matches at 0dB S:N, the reverse was true at -6dBS:N, and at -12 dB S:N, partial semantic matches far exceeded total semantic matches. AChi-square test confirmed this description with a significant effect for degree of semanticmatch, X2 (df=2) = 166.92, g < .005, a significant effect for S:N condition, X 2 (df=2) =301.99, g < .005, and a significant interaction between degree of semantic match and S:Ncondition, X2 (df=2) = 10.62, g < .05.10080a) 60acTwe 40a.20	•• —0-- Total semantic matchPartial semantic matchNo semantic matchh560 	 '-18 	 -12 	 -6 	 0 	 6Signal-to-Noise Ratio (dB)Figure 10. Percentage of addition and substitution errors (partial auditory matchto target) classified as total, partial, or no semantic match as a function of signal-to-noise conditionEffect of Signal-to-Noise Ratio on the Syntactic Level (Partial Auditory Match)Addition and substitution errors that were classified as partial auditory matches to the targetwere also categorized for degree (total, partial, no) of syntactic match (see Figure 11). Ineach of the three S:N conditions, most responses that partially matched the target auditorilytotally matched the target syntactically. The percentage of total syntactic matches droppedsteadily as S:N ratio decreased. Furthermore, partial matches out-numbered no matches at0dB S:N and -6dB S:N. However, at -12dB S:N, there were equal percentages of partialand no matches. A Chi-square test indicated a significant effect for degree of match, X2(df=2) = 69.60, 2 < .005, a significant effect for S:N condition, X2 (df=2) = 298.46, 2 < .005,and a significant interaction between the degree of syntactic match and S:N condition, X 2(df=4) = 30.99, g <.005. 10080604020—o— Total syntactic match--e-- Partial syntactic matchNo syntactic match57      0-18 	 -12 	 -6 	 0 	 6Signal-to-Noise Ratio (dB)Figure 11. Percentage of addition and substitution errors (partial auditory matchedtarget) classified as total, partial, or no syntactic match as a function of signal-to-noisecondition.Effect of Signal-to-Noise Ratio on the Semantic Level (No Auditory Match) Addition and substitution errors that were classified as not matching the target auditorilywere categorized for degree of semantic match (see Figure 12). In all three S:N conditionsvirtually all errors failed to match the target semantically. A Chi-square test indicated asignificant effect of degree of semantic match, X2 (df=2) = 378.34, < .005, significanteffect of S:N condition, X2 (df=2) = 301.99, 12 < .005, but no significant interaction betweendegree of semantic match and S:N condition, X2 (df=4) = 10.62, 2 > .05.10080rncts 60 —0— Total semantic match8 40Partial semantic matchNo semantic matcha_200-18 	 -12 	 -6 	 0 	 6Signal-to-Noise Ratio (dB)Figure 12. Percentage of addition and substitution errors (no auditory match)classified as total, partial, or no semantic match as a function of signal-to-noisecondition.Effect of Signal-to-Noise Ratio on the Syntactic Level (No Auditory Match) Addition and substitution errors that were classified as failing to match the target auditorilywere categorized for degree(total, partial, no) of syntactic match (see Figure 13). In each ofthe three S:N conditions, the vast majority of errors failed to match the target syntacticallywith most of the remaining errors totally matching the target syntactically. A Chi-square testindicated a significant effect of degree of syntactic match, X 2 (df=2) = 142.962, 2 < .005,significant effect of S:N condition, X 2 (df=2) = 213.31, 2 < .005, and a significant interactionbetween the degree of syntactic match and S:N condition, X2 (df=4) = 19.03, 2 < .005.58—o— Total syntactic match100 —o— Partial syntactic matchNo syntactic match806040200' •-18 	 -12 	 -6 	 0 	 6Signal-to-Noise Ratio (dB)Figure 13. Percentage of addition and substitution errors (no auditory match)classified as total, partial, or no syntactic match as a function of signal-to-noisecondition.Preservation of Semantic and Syntactic InformationFor substitution errors, the responses classified as total or partial semantic matches werecollapsed into one category called some semantic match likewise the responses classified astotal or partial syntactic matches were collapsed into one category called some syntactic match. (Addition errors were excluded from this analysis because, by definition, they wouldnever match the target either syntactically or semantically.) This re-grouping facilitated anevaluation of how the preservation of semantic information was, if at all, related to thepreservation of syntactic information.59	••Partial auditory matchSome semantic matchSome syntactic match400 r.602,cpco I- 300c)—u) 0a) •-co asM 200(7)-15CD CDc2o 113) 1004,-.0 wEEc2 	0 	-18 -12 	 -6 	 0 	 6Signal-to-Noise Ratio (dB)Figure 14. Substitution errors (partial auditory match) preserving some syntactic orsome semantic information.For errors that partially matched the target auditorily, the new categories are plotted in Figure14. More syntactic information was preserved than semantic information. As a function ofS:N ratio, there was a slight decrease in the percentage of errors where semantic and syntacticinformation were preserved. A Chi-square test confirmed this description with a significantdifference between the category of Some semantic match and the category of some syntactic match, X2 (df=1) = 54.87, lZ <.005 and a significant difference at each of the three S:Nconditions, X 2 (df=2) = 263.56, 12 <.005. No significant interaction was observed, X2 (df=2)_.14,12 >.05As can be seen in Figure 15, errors preserving syntactic information alone wereapproximately equal in number to errors preserving both syntactic and semantic information.That is, in virtually half of the cases in which syntactic information was preserved, semanticinformation was also preserved. There were almost no cases in which semantic informationwas preserved without preservation of syntactic information. Errors where neither syntactic61nor semantic information was preserved were few in number except at the -12dB S:Ncondition. Overall, even when there was only a partial auditory match to the target, there wasgreat preservation of syntactic information and considerable preservation of semanticinformation.400Z g 300cu „l–c/2 0 0a Eco 0 .2200co cocc z,5 2 cp.0 CD .2)Ez ED! 4,6) 100Z CL0-18• • • Partial auditory matchPreserved semantic&syntacticPreserved semantic—0-- Preserved syntactic—0--- Neither preserved-12 	 -6	0	6Signal-to-Noise Ratio (dB)Figure 15. Substitution errors (partial auditory match) preserving both semanticand syntactic, only syntactic, only semantic, or neither level of information.For substitution errors that failed to match the target auditorily, the new categories, somesemantic match and some syntactic match, are plotted in Figure 16. Again syntacticinformation was better preserved than semantic information. As a function of S:N ratio againthere was a slight decrease in the percentage of errors where semantic and syntacticinformation were preserved. A Chi square test confirmed this description with a significantdifference between the category of some semantic match and the category some syntactic match, X2 (df=1) = 21.14, p <.005 and a significant difference at each of the three S:N	 No auditory matchSome semantic matchSome syntactic match62conditions, X2 (df=2) = 162.51, 12, <.005, no significant interaction between S:N and degreeof semantic match was observed, X2 (df=2) = 75, 2 >.05.400CDC (L)CD CD(i):12 0 300o_cnN ..-u)Cg M 200(Y)a) a)cr• g 100agoEE 2 e2 	 0 	-18Signal-to-Noise Ratio (dB)Figure 16. Substitution errors (no auditory match) preserving some syntactic orsome semantic information.As can be seen in Figure 17, in all S:N conditions, substitution errors preserving bothsyntactic and semantic information slightly out-numbered substitution errors preserving onlysyntactic information. That is, in over half the cases in which syntactic information waspreserved semantic information was also preserved. Again there were almost no cases inwhich semantic information was preserved without preservation of syntactic information.And agin, errors where neither syntactic nor semantic information was preserved were few innumber, except in the -12dB S:N condition. Even when the response completely failed tomatch the target auditorily, neverthelesss there was great preservation of syntacticinformation and semantic preservation.-12	 -6 	 0	663z0 2) 300■-c o oa 4)cc wECD 12 s 200"6 -68 2 a).0 0)E 	 100ti)3 112z a. o400—0--No auditory matchPreserved semantic&syntacticPreserved semanticPreserved syntacticNeither preserved 0-18	 -12 	 -6Signal-to-Noise Ratio0	6Figure 17. Substitution errors (no auditory match) preserving both semantic andsyntactic, only syntactic, only semantic, or neither level of information64CHAPTER 5DISCUSSIONIn this study two hypotheses have been tested:(1) the number and types of errors a listener makes when repeating a narrative will be nodifferent than the number and types of errors made when repeating a restaurant review or arecipe; and(2) as the signal to noise ratio decreases and listening conditions become less favorable, thenumber of errors and the types of errors that occur will not be significantly different from thenumber and types of errors observed under more favorable signal-to-noise conditions. Thefollowing discussion will address the results as related to each of these hypotheses.Text Type:.The results indicated a significant difference in the number errors observed between the texttype narrative and the text types restaurant review (representative of the text type description)and recipe (representative of text type procedural). This supports the initial research hypothesiswhich stated that the number of errors a listener makes when repeating a narrative will be lessthan the number of errors made when repeating a restaurant review or a recipe. Recall that therationale for employing these three text types (i.e., narrative, description, and procedural) wasbased on the experimental literature which has described different text types as possessingdifferent types of organizational structures which yield different degrees of predictability. Theresultant outcome of these differences is that text structures with higher predictability are able toeffect more top-down processsing than text types which are less predictable. The importanceand relevance of this to the current study is as follows: When a speech signal is degraded by65noise, bottom-up processing yields a poorer product than when a speech signal is heard in theabsence of background noise. In the context of a first-pass/second-pass approach toprocesssing it could be said that on the first pass, processing fails to achieve comprehension inan automatic fashion and as a consequence more effortful top-down processes are recruited tofacilitate comprehension in the second-pass. Top-down processes facilitate the generation of acoherent message more when a text has a high degree of predictability than when it has lessstructral organization. In the former case, the listener can fall back on using a macrostructure amacrostructure which prescribes a constrained set of possibilites, whereas in the latter case, themacrostructure prescribes a much wider range of choices. It is for these reasons that in thepresent experiment, we see subjects better able to perceive narratives than descriptions (i.e.,restaurant reviews) in noise.Familiarity and Text Type: Significant differences were found between the text type narrative and the text type recipe at allthree S:N conditions. The differences between the text type recipe and the text type restaurantreview are minimal at 0 and -12 dB S:N, with performance being poorer at -6dB for the texttype recipe. It was thought at the outset that performance on recipes would be more similar tonarratives than to descriptions because recipes are by their very nature procedural; that isperforming one set of actions in a recipe is very closely tied to the actions that precede andfollow it, and because the possible events within a recipe are constrained relative to adescription, it was thought that at the outset that performance on recipes would be more similarto narratives than decriptions. A measure of familiarity with each of the text types wasincluded as part of the protocol of this experiment. This measure was made at the outsetbecause it was thought that familiarity might play a role in an individual's performance becauseonly subjects familiar with the recipe schema would be expected to use it to advantage. Afterthe recipe data were found to be more similar to descriptions than narratives, suggesting that66the participants were not using the inherent predictable structure of recipes, the error data wereregrouped based on subject's familiarity ratings. The analysis revealed that in fact thosesubjects who had knowledge and familiarity with recipes made fewer errors at -6 and -12 dBS:N. This result suggests that individuals who are familiar with recipes have probablydeveloped a schema for them and as a consequence demonstrate a similar error rate to thatfound for narratives. In contrast, those individuals who did not rate themselves as veryfamiliar with recipes demonstrate error rates similar to those found for restaurant reviewsbecause for them recipes are not a logical series of steps, but more like the collection ofstatements found in a description. This result suggests that familiarity with a text typeimproves an individual's ability to communicate in a noisy situation.Error Type and Text Type: The results indicated no significant difference in the number of exchange, addition, andsubstitution errors between each of the text types in each of the S:N conditions A significantdifference in the number of deletion errors between each of the text types was found.Therefore, except for the error type deletion, text type did not influence frequency of errortype. Across all three S:N conditions, the number of deletion errors is fewest for the text typenarrative. One could speculate that a text type with a highly predictable format facilitates thereception of the speech signal such that less information is lost or alternatively, such that moreof the missing information is provided by top-down processing. Also, the number of deletionsincreased as S:N ratio decreased. This is not surprising. One would expect deletions toincreases because as S:N decreases the signal becomes degraded and portions of the textbecome inaudible.The error type exchange was rare and its occurrence did not vary with text type or S:N ratio.These results would appear to suggest that the act/process of exchanging words within a67sentence (i.e., metathesis) or shifting a word from one location to another is a phenomenon thatis present in speech and its frequency is not related to listening conditions.Addition errors were more common then exchange errors and did vary with S:N ratio but notwith text type. This might be indicative of a need to impose coherency on a text that becomesless perceptible as listening conditions deteriorate.The number of substitution errors also increased as S:N ratio decreased. These increasesoccured at a steeper rate then they did for addition errors, especially from -6 to -12dB. Theincrease in substitution errors is possibly a reflection of the fact that as listening conditionsdeteriorate the speech signal becomes degraded and information is lost. It was observed thatindividuals compensate for this missing information in a variety of ways. Some of theseresponse patterns are described below:(1) the gist (i.e., semantic content) is repeated. For example:Target: 	 the apples should come just to the rim of the dish and no higherResponse: the apples should come just to the height of the dish and no higherAuditory: total mismatchSyntactic: total matchSemantic: total match(2) a syntactic substitution is given. For example:Target: 	 her ceasar salad is loaded with bacon bitsResponse: her ceasar salad is loaded with calories Auditory: total mismatchSyntactic: total matchSemantic: total mismatch68(3) an auditory substitution is given. For example:Target: 	 dense and moist and not too sticky sweetResponse: dance in the mist with not too sticky feetAuditory: Partial match 	 partial match 	 partial matchSyntactic: total mismatch 	 total mismatch 	 total mismatchSemantic: total mismatch 	 total mismatch 	 total mismatchThese responses are not particularly surprising. The listeners were trying to hear andunderstand what was being said. At the same time they are attempting to "Repeat back as muchand as best" they could. When all else failed they repeated phonetic information, even whenthe response was not particularly coherent with respect to the text topic. It is also notsurprising that the number of substitution response errors increased dramatically as S:Ndecreased from -6 to -12dB S:N. What these substitution errors are composed of is of interesthere. That is, what language element is being substituted for what? For example, does a wordsubstitue for a word, a word for a phrase, or a phrase for a sentence? How semantically ,syntactically, or auditorily similar is the response relative to the target? Although the formerquestion was not addressed in this study the latter was and will be elaborated upon in thefollowing sections.Preservation of Auditory Information: As S:N ratio decreased the overall number of errors increased, including responses that failedto match the target or matched it only partially. This is consistent with the fact that as listeningconditions become poorer, the clarity of the perceptual cues available in the speeech signal isreduced. As a consequence, the listener will hear and repeat back a greater number ofmisperceptions as the S:N ratio decreases.69When only substitution errors are considered, the number of partial auditory matches to thetarget are slightly greater than the number of total mismatches to the target. By definition,exchange errors always match the target auditorily, whereas Addition and deletion errors nevermatch the target auditorily, even partially. Therefore when all four types of errors areconsidered the number of total auditory mismatches exceeds the number of partial matches.Preservation of Syntactic Information: Degree of Syntactic Match & Partial Auditory Match Considering substitution errors that partially match the target auditorily (by definition additionand deletion  errors would not match the target even partially), in very few cases was syntacticinformation not preserved and in most cases it totally matched the target. The high percentageof responses in which syntactic information was preserved could be attributed to the limitedpossibilities dictated by the syntactic constraints of a language. That is, in English, althoughthe possible syntactic constructions are theoretically infinite, the actual occurrence of structuralforms is rather predictable. This can be likened to the observation that the vocabulary peopletend to use is fairly restricted when compared to their receptive vocabulary or the total numberof dictionary entries in a language. Furthermore, when a signal is perceptually degraded it islikely that listeners will attempt to reconstruct it as a syntactically correct utterance. Finally, itis reasonable that, while auditory misperceptions may lead to word level errors, it is less likelythat such misperceptions would result in errors in word class selection or other gross syntacticerrors. For example a listener might hear 'pot' instead of 'pond' but both are appropriateendings for the sentence "The little girl put her feet in theDegree of Syntactic Match & Total Auditory MismatchConsidering substitution errors alone when there is a total failure to match the target auditorily,70a similar pattern of results is seen and a similar explanation may be made as was the case forerrors that partially matched the target auditorily. When addition errors, which by definitiontotally mismatch the target auditorily, syntactically, and semantically, are also considered thenthe preservation of syntactic information, of course, looks much worse.A more detailed investigation of the patterns of syntactic breakdown that occur as S:N ratiodecreases is needed to gain a better understanding of the processes that are engaged in poorlistening conditions to effect comprehension. Such a study would involve the sub-categorization of syntactic errors and a documentation of their changes as S:N ratios changes.Preservation of Semantic Information: Degree of Semantic Match: Partial Auditory Match & Total Auditory Mismatch Considering substitution errors that partially match or completely mismatch the targetauditorily, total semantic mismatches significantly out-numbered partial and total semanticmatches. Based on these results alone, however, it is not possible to comment on the potentialloss of semantic information. To fully interpret the data, it is necessary to further examine inwhat way these responses are semantic mismatches. It may be the case that although one wordmay have been substituted for another, the substituted word does not detract from the overallmeaning of the text. For example:(1) Target: then dress himself in Arab robesResponse: and dress himself in Arab robesor for example:(2) Target: try the sauteed shrimp and crab on noodlesResponse: try the satay shrimp and crab on noodles.71In the first example the substitution of 'and' for 'then' does not detract from the idea that themain character of the story is dressing in a particular style of clothing, and in the secondexample although the substitution of 'satay' for 'sauteed' changes how the shrimp is preparedit does not change the overall notion that an entree containing seafood served over noodles is agood choice. These examples illustrate that a coding of 'semantic mismatch' is not necessarilya reflection of a loss of all relevant meaning. That is, the gist may be preserved andunderstanding of the text as a whole may not be significantly impaired.Degree of Semantic Match & Partial Auditory Match Considering substitution errors that partially match the auditory target, total semantic matchesout-numbered partial semantic matches at 0dB S:N and this relationship reversed as S:N ratiodecreased. The following explanantion has some intuitive appeal: As listening conditions aredegraded and perceptual errors increase, the number of total semantic matches is likely todecrease. These events occur because it becomes increaingly difficult to retrieve the exactwords, through the help of surounding context, however, it is possible to guess which wordbelongs.When substitution errors that partially match the target are examined it is found that thepercentage of responses in which semantic information is preserved decreases mostdramatically as S:N ratio decreases from -6 to -12dB S:N. Based on these findings it could besaid that it is not until -12dB S:N that the signal becomes so degraded that context can nolonger offer a means by which to preserve semantic information.Degree of Semantic Match & Total Auditory MismatchWhen error responses were complete auditory mismatches, the number of total and partial72semantic matches were similar in all three S:N conditions. This finding is also intuitivelyappealing for the following reasons: Since the accompanying perceptual information isincorrect, the probability of giving a response that contains the right target word orsemantically appropriate word are equally probable, because in order to make an appropriateguess, the listener is operating on the basis of context. Sometimes such guesses yield a correcttarget word; and other times such guesses yield a semantically similar word.The Preservation of Syntax in Relation to Semantics: When the response only partially matched or failed to match the target auditorily, semanticinformation was preserved wherever syntactic information was preserved. Only in a very fewinstances was semantic information preserved without preservation of syntax. These resultssupport the idea that more perceptual precision is required to achieve a semantic match than isrequired to achieve a syntactic match. That is, to hear a specific word in the presence ofbackground noise is a more difficult task than 'hearing ' the word class to which a particularlexical item belongs. (Where 'hearing' refers to an individual perceiving enough of theacoustic signal to surmise, at minimum, the appropriate word class if not the appropriate lexicalitem.) For these reasons, it could be said that in the presence of background noise, semanticconstructions are weaker and more difficult to preserve than syntax.When the response partially matched the target auditorily it was observed that the greatestchanges occured as S:N ratio decreased from -6 to -12dB. This is likely a consequence of aloss of both syntactic and semantic contextual cues. At higher noise levels, errors are nolonger restricted to lexical items or short phrases, but interfere with the perception of significantamounts of the speech signal to such a degree that the construction of a context is not possible.Overall, it has become evident that if only the number of errors based on auditory mismatches73were counted and used as a means by which to assess the listening abilities of an individual innoise, we would fail to appreciate the robustness of semantic and syntactic information.Future Directions The current investigation has shown that highly organized macrostructures facilitate listening totext in noise better than less organized macrostructures. In the process of studying this, it hasbecome evident that there is a need to examine the relationship between working memory andan individual's ability to understand speech in noise. If a relationship can be found, thenfurther insight into the relationship of bottom-up and top-down processing will be gained.The current study has also shown that auditory errors alone are not an accurate reflection of anindividual's ability to understand text in noise. In the process of uncovering this, otherquestions have surfaced: What is the exact nature of the syntactic and semantic information thatis being deleted and substituted?, To what extent does the listener paraphrase or rephrase theincoming information?, and To what degree is this reformulation syntactically and semanticallyaccurate?, And also, what is the relationship between the semantic information that is lost andthe extent to which comprehension is affected as listening conditions become poorer?It is the hope of this researcher that this study serves asoa starting point for further investigationinto the understanding of speech in noise and for the development of materials and procedureswhich accurately assess, and therefore reflect, an individual's ability to understand speech ineveryday listening situations.74REFERENCESAnderson, R. C., & Pearson, P. D. (1984). A schema-theoretic view of basic processesin reading comprehension. In P. D. Pearson (Ed.), Handbook of Reading Research(pp. 255-291). New York: Longman.Anderson, R. C., Spiro, R. J., & Anderson, M. C. (1978). Schemata as scaffolding forthe representation of information in connected discourse. American EducationalResearch Journal, 15.(3), 433-440.Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), ThePsychology of Learning and Motivation,Vol.  8,. New York: Academic Press.Bartlett, F. C. (1932). Remembering. Cambridge: Cambridge University Press.Bilger, R. C., Neutzel, J. M., Rabinowitz, W. M., & Rzeczowski, C. (1984).Standardization of a test of speech perception in noise. Journal of Speech and HearingResearch, 21, 32-48Brewer, C. C., & Resnick, D. M. (1983). A review of tests of speech discrimination.Seminars in 	, 4(3), 205-218.Brown, G., & Yule, G. (1989). Discourse Analysis. Cambridge: Cambridge UniversityPress.Cairns, H. S. (1984). Research in language comprehension. In R. Naremore (Ed.),Language Sciences: Recent Advances. San Diego: College Hill Press.75Carhart, R., Johnson, C., & Goodman, J. (1975, November). Perceptual Masking ofSpondees by Combination of Talkers. Paper presented to the 90th meeting of theAcoustical Society of America, San Francisco.Carhart, R., Tillman, T. W., & Greetis, E. S. (1969). Perceptual masking in multiplesound backgrounds. Journal of the Acoustical Society of America, 45, 694-703.Carpenter, P. A., & Just, M. A. (1989). The role of working memory in languagecomprehension. In D. Klahr and K. Kotovsky (Eds.), Complex InformationProcessing: The Impact of Herbert A. Simon (pp. 31-68). Hillsdale, New Jersey:Lawrence Erlbaum.Cohen, R. L., & Keith, R. W. (1976). Use of low-pass noise in word recognition.Journal of Speech and Hearing Research, 19, 48-54.Committee on Hearing, Bioacoustics, and Biomechanics (CHABA). (1988). Journal ofthe Acoustical Society of America, 83(3), 859-893.Cox, R. M., Alexander, & G. C., Gilmore, C. (1987). Development of the connectedspeech test (cst). Ear and Hearing, 8(suppl), 119S-126S.Cox, R. M., Alexander, G. C., Gilmore, C., & Pusakulich, K. M. (1989). The connectespeech test version 3: Audiovisual administration. Ear and Hearing, 10(1), 29-32.Craik, F. I. M. (1983). . On the transfer of information from temporary to permanentmemory. Philosophical Transactions of the Royal Society of London, B302, 341-359.Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framewrok formemory research. Journal of Verbal Learning and Verbal Behavior, 11, 671-684.76Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memoryand reading. Journal of Verbal Learning and Verbal Behavior, 12, 450-466.Daneman, M., & Green, I. (1986). Individual differences in comprehending andproducing words in context. Journal of Memory and Language, 25, 1-18.Davies, P. (1982). Scaredy Cat. In J. F. Parker (Ed.), The Writer's Workshop (p.45).Don Mills, Ontario: Addison Wesley Publishersde Beaugrande, R. (1980). Text, Discourse, and Process. London: Longman.DeFilippo, C. L., & Scott, B. L. (1978). A method for training and evaluating thereception of ongoing speech. Journal of the Acoustical Society of America, 63, 1186-1192.Den Uyl, M., & Van Oostendorp, H. (1980). The use of scripts in text comprehension.Poetics, 9, 275-294.Duquesnoy, A. J. (1983). Effect of a single interfering noise or speech source upon thebinaural sentence intelligibnility of aged persons. Journal of the Acoustical Society ofAmerica, 74(3), 739-743.Elliot, L. L. (1963). Prediction of speech discrimincation scores from other testinformation. Journal of Auditory Research, 3, 34-45.Erber, N. P. (1988). Communication Therapy for Hearing-Impaired Adults. Abbotsford,Australia: Clavis Publishing.77Fayol, M. (1991). Text typologies: A cognitive approach. In G. Denhiere and J. P.Rossi (Eds.), Advances in Psychology: Text and Text Processing,  Vol. 79 (pp. 61-76). Amsterdam: North Holland.Flesch, R. (1948). A New Readability Yardstick. The Journal of Applied Psychology,32, 221-233.Fodor, J. A. (1983). The Modularity of the Mind. Cambridge: MIT Press.Fodor, J. A., Bever, T. G., & Garrett, M. f. (1974). The Psychology of Language.New York: McGraw-Hill.Forster, K. (1979). Levels of processing and the structure of the language processor. InW. E. Cooper and E. C. T. Walker (Eds.), Sentence Processing: Psycholinguisticstudies presented to Merrill Garrett Hillsdale, New Jersey.Graesser, A. C. (1981). Prose Comprehension Beyond the Word. New York: Springer-Verlager.Graesser, A. C., Hauft-Smith, K., Cohen, A. D., & Pyles, L. P. (1980). Advancedoutlines, familiarity, and text genre on retention of prose. Journal of Experimentaleducation, 48 , 281-290.Grosjean, F. (1980). Spoken word recognition processes and the gating paradigm.Perception and Psychophysics, 2a, 267-283.Hiebert, E. H., Englert, C. S., & Brennan, S. (1983). Awareness of text structure inrecognition and production of expository discourse. Journal of Reading Behavior,2&4), 63-79.78Hirsh, I. J., Davis, H., Silverman, S. R., Reynolds, E. G., Eldert, E., & Benson, R. W.(1952). Development of materials for speech auiometry. Journal of Speech andHEaring Disorders, 17, 321-337.Hodgson, W. R. (1980). Basic Audiologic Evaluation. Williams & Wilkins: Baltimore.Horowitz, R. (1985). Text patterns: part I. Journal of Reading, 28(5), 448-454.Just, M. A., & Carpenter, P. A. (1987). The Psychology of Reading and LanguageComprehension. Boston: Allyn and Bacon.Kalikow, D. N., Stevens, K. N., & Elliott, L. L. (1977). Development of a test ofspeech intelligibility in noise using sentence materials with controlled wordpredictability. Journal of Acoustical Society of America, 61(5), 1337-1351.Kintsch, W. (1977). On comprehending stories. In P. Carepenter and M. Just (Eds.),Cognitive Processes in Comprehension. Hillsdale, New Jersey: Erlbaum.Kintsch, W. & van Dijk, T. A. (1978). Toward a model of text comprehension andproduction. Psychological Review,  $5, 363-394.Kintsch, W. & Yarbrough, J. C. (1982). Role of rhetorical structure in textcomprehension. Journal of Educational Psychology, 74(6), 828-834.Kintsch, W. & Young, S. R. (1984). Selective recall of decision relevant informationfrom texts. Memory and Cognition, 12(2), 112-117.Kirk, R. E. (1982). Experimental Design: Procedures for the Behavioral Sciences.Belmont, California: Brooks/Cole.79Kozminsky, E. (1977). Altering comprehension: the effect of biasing titles on textcomprehension. Memory and Cognition, 5, 482-490.Kozminsky, E., Kintsch, W., & Bourne, L. E. (1981). Decision making with texts:Information analysis and schema acquisition. Journal of Experimental Psychology:General, 110(3), 363-380.Mandler, J. M. & Johnson, N. S. (1977). Remembrance of things parsed: Story structureand recall. Cognitive Psychology, 9, 111-151.Marshall, L., & Bacon, S. P. (1981). Prediction of speech discriminationscores fromaudiometric data. Ear and Hearing, 2, 148-155.Marslen-Wilson, W. D., (1973). Linguistic Structure and speech shadowing at very shortlatencies. Nature, 244, 522-523.Marslen-Wilson, W. D., (1975). Sentence perception as an interactive parallel process.Science, 189, 226-228.Marslen-Wilson, W. D., (1976). Linguistic descriptions and psyhological assumptions inthe study of sentence perception. In R. J. Wales & E. C. T. Walker (Eds.), NewApproaches to the Study of Language. Amsterdam: North Holland.Marslen-Wilson, W. D., & Tyler, L. K., (1975). Processing structure of sentenceperception. Nature, 257, 784-786.Marslen-Wilson, W. D., & Tyler, L. K., (1980). The temporal structure of spokenlanguage understanding. Cognition,l, 1-71.80Marslen-Wilson, W. D., & Tyler, L. K. (1981). Central processes in speechunderstanding. Philosophical Transactions of the Royal Society of London, B 295,317-332.Martin, F. H. (1986). Introduction to Audiology. Englewood New Jersey: PrenticeHall.Meyer, B. J. F. (1975). The organization of prose and its effects on memory.Amsterdam: North-Holland.Meyer, B. J.F. (1977). The structure of prose: Effects on learning and memory andimplications foreducational practice. In R. C. Anderson, R. Spiro, & W. Montague(Eds.), Schooling and the Acquisition of Knowledge  (pp. 179-200). Hillsdale, NJ:Lawrence Erlbaum.Meyer, B. J. F. & Freedle, R. 0. (1984) The effects of different discourse types on recall.American Educational Research Journal,  21, 121-143.Meyer. B. J. F., & McConkie, G. W. (1973). What is recalled after hearing a passage?Journal of Educational Psychology, 65(1), 109-117.Miller, G. A. (1956). The magical number seven. plus or minus two: Some limits on ourcapacity for processing information. Psycological Review, 63, 81-97.Miller, G. A., & Isaard, S. (1963). Some perceptual consequences of linguistic rules.Journal of Verbal Learning and Verbal Behavior,  2, 217-228.Miller, G. A. , & Nicely, P. E. (1955). An analysis of perceptual confusions amongsome English consonants. Journal of the Acoustical Society of America,  27(2), 301-315.81Minsky, M. (1975). A framework for representing knowledge. In P. H. Winston (Ed.),The Psychology of Computer Vision. New York: McGraw-HillNoble, W. (1983). Hearing, hearing impairment, and the audible world: A theoreticalessay. Audiology, 22, 325-338.Olsen, W. 0., & Matkin, N. D. (1991). Speech Audiometry. In W. R. Rintelmann(Ed.), Hearing Assessment, second edition. Austin, Texas: Pro-Ed.Penrod, J. P. (1985). Speech Discrimination Testing. In J. Katz (Ed.), Handbook ofClinical Audiology. (pp. 235-255). Baltimore: Williams and Wilkins.Pollack, I. & Pickett, J. M. (1963). Intelligibility of excerpts from conversation.Language and Speech, 28, 97-102.Rumelhart, D. E. (1977). Understanding and summarizing brief stories. In D. La Berge& S. J. Samuels (Eds.), Basic Processes in Reading: Perception and Comprehension(pp. 265-303). Hillsdale NJ: Lawrence Erlbaum.Rumelhart, D. E., & Ortony, A. (1977). The representation of knowledge in memory. InR. C. Anderson, R. J. Spiro, & W. E. Montague (Eds.), Schooling and theAcquisition of Knowledge. Hillsdale, New Jersey: Erlbaum.Schank, R. C., & Abelson, R. P. (1977). Scripts, Plans, Goals and Understanding : AnInquiry Into Human Knowledge Structures. Hillsdale, New Jersey: Erlbaum.Schultz, J. (1985). Dining Out in Edmonton. Edmonton, Alberta: Tree Frog Press82Schwartz, A. (1985). The Case of the Arabian Jewels. In A. Schwartz (Ed.),  Tales ofTrickery: From the Land of Spoof (p.23). New York: Farras, Straus & Giroux.Silverman, S. R., & Hirsh, I. J. (1955). Problems related to the use of speech in clinicalaudiometry. Annals of Otology. Rhinology. and Laryngology, 64, 1234-1244.Simon, H. A. (1974). How big is a chunk? Science, 	 482-488.Speaks, C., & Jerger, J. (1965). Method for measurement of speech identification.Journal of Speech and Hearing Research, 8, 185-194.Stein, N., & Glenn, C. (1979). An analysis of story comprehension in elementary schoolchildren. In R. 0. Freedle (Ed.), New Directions in Discourse Processing.Norwood, New Jersey: Ablex.Stine, E. L., Wingfield, A., & Poon, L. W. (1989). Speech comprehension and memorythrough adulthood: the roles of time and strategy. In L. W. Poon, D. C. Rubin, &B. AA. Wilson (Eds.),  Everyday Cognition in Adulthood and Late Life (pp.195-221).New York: Cambridge University Press.Tannen, D. (1979). What's in a frame? Surface evidence for underlying expectations. InR. 0. Freedle (Ed.), New Directions in Discourse Processing. Norwood, New Jersey:Ablex.Thorndyke, P.W. (1977). Cognitive Structures in comprehension and memory ofnarrative discourse. Cognitive Psychology, 9, 77-110.83Tillman, T. W., & Carhart, R. (1966). An Expanded Test for Speech DiscriminationUtilizing CNC Monosyllabic Words: Northwestern University Auditory Test No. 6.Technical Report no. SAM-TR-66-55. USAF School of Aerospace Medicine, BrooksAir Force Base, Texas.Tun, P. A. (1989). Age differences in processing expository and narrative text. Journalof Gerontology: Psychological Sciences, 44(1), 9-15.Tye-Murray, N, & Tyler, R. S. (1988). A critique of continuous discourse tracking as atest procedure. Journal of Speech and Hearing Disorders, 53, 226-231.Van Dijk, T. A., & Kintsch, W. (1983). Strategies of Discourse Comprehension. NewYork: Academic Press.Wadden, P. (1990, July). The loyal dispatcher: Rendezvous in a gale. The WashingtonPost, p.159.Warren, R. M. (1970). Perceptual restoration of missing speech sounds. Science, 167,392-393.Wingfield, A. Lahar, C. J., & Stine, E. A. L. (1989). Age and decision strategies inrunning memory for speech: Effects of prosody and linguistic structure. Journal ofGerontology, 44(4), 106-103.Wingfield, A., & Nolan, K. A. (1980). Spontaneous segmentation in normal and in timecompressed speech. Perception and Psychophysics, a(2), 97-102.Wingfield, A., Poon, L. W., Lombardi, L., & Lowe, D. (1985). Speed of processing innormal aging: Effects of speech rate, linguistic structure, and processing time. Journalof Gerontology, 40, 579-585.,84Young, M. A. & Gibbons, E. W. (1962). Speech Discrimination scores and thresholdmeasurements in a non-normal hearing population. Journal of Auditory Research,  2,21-33.Young, L. L.., Parker, C., & Carhart, R. (1975, November). Effectiveness of Speechand Noise Maskers on Numbers Embedded in Continuous Discourse.  Paper presentedto the 90th Meeting of the Acoustical Society of America, San Francisco.85APPENDIX APure-Tone Thresholds, Babble Thresholds (BT), and Age of Individual SubjectsSubject Right Ear Left Age.25 .5 1 2 3 4 BT .25 .5 1 2 3 4 BTMM 10 10 15 0 5 10 0 5 0 5 10 0 5 0 25FH 0 0 0 5 0 0 0 5 0 0 0 0 10 0 26DD 0 0 5 0 0 0 0 10 10 10 5 0 10 0 29SK 0 0 0 0 5 5 0 0 0 0 0 5 5 0 26TE 0 0 0 0 5 0 0 0 0 0 0 5 0 0 0 23SM 0 0 5 0 0 0 0 5 5 0 0 0 0 0 29JG 5 5 0 5 5 20 0 5 10 5 5 15 15 5 27SW 0 0 5 0 0 0 0 10 5 0 0 0 0 0 29SP 0 0 0 0 0 0 0 0 5 0 5 10 5 0 29CK 5 10 10 0 0 5 0 5 5 5 5 5 5 0 28MS 10 5 5 0 5 0 0 10 0 0 0 0 0 0 27NT 0 0 0 0 0 0 0 0 0 0 0 0 0 0 22NB: All threshold values are expressed in dBHLAPPENDIX BText for the narrative "Scaredy Cat" 1: when I was small2: the most frightening place in the world for me was the basement in my ownhome3: there were so many places in the basement4: where a bogey man could hide5: I believed him to be waiting and watching from the cool dark corners of the room6: just longing to catch me alone7: typical of an old-fashioned cellar, the basement had a low ceiling8: a bulb hanging at the end of the room was the only lighting9: shadows from haunted old furniture played games with my mind10: making my imagination run wild11: a door to the right of the main room led to a small dim laundry room and workshop12: one day my eldest sister asked me to run downstairs and fetch some clean dish towels13: as she was busy cleaning the kitchen14: "well I'm kinda busy right now", I replied15: as I sat at the table, twiddling my thumbs16: "oh come on, don't be so lazy"17: "unless of course you're afraid of the bogey man," she said gleefully18: to avoid weeks of harassment from my younger sisters19: I hurriedly assured her20: that I would be happy to get the towels21: and bounced down the stairs to the basement22: the gloomy quiet atmosphere seemed to close in on me23: as I cautiously stepped into the room24: my footsteps echoed loudly along with my heart25: as I walked slowly across the floor26: the laundry door room whined protestingly27: a warning of my presence to all who lurked in the shadows28: the shelves of the clean laundry remained hidden by a black curtain29: taking a deep breath30: I drew the curtain to one side31: and a pair of evil green eyes stared malevolently32: from within the deep dark recesses of the cupboard33: a coronary arrest threatened to overtake me34: as the cat shot over my shoulder and through the open window like a bullet35: I crawled upstairs, a nervous wreck36: and told my sister of my fright86("Scaredy Cat" continued)37: swearing emphatically that I would never ever go downstairs alone again38: she was extremely sympathetic39: "oh," she said40: "but you've forgotten the tea towels"41: Hmmph, sympathetic indeed87Text for the narrative "Taxi" 	 881: it's seven a.m. in Denville2: and the taxi company has just called a second time3: to say they can't find my house4: once again I spell out directions even a blind cabbie could follow5: only two hours remain until my flight leaves6: and it's an hour-and-a-half trip to the airport7: outside the torrential rains are threatening to sweep my little house8: off the mountain slope on which it teeters9: a place so far north in denville10: that city buses lurch past only three times a day11: the telephone rings again12: terribly sorry, begins the dispatcher13: then I realize what's happened14: flooded with calls the company's maximizing profits by handling only in-city runs15: I'd heard this happens when the weather gets bad16: desperately I shout into the phone17: that I have a plane to catch18:and I'll meet the taxi a few hundred metres away on a bridge over the Clifton River19: standing over the roaring gale-swelled river20: horizontal-driven rain drenching my overcoat21: I gaze up and down the street22: No taxi23: finally, struggling with my umbrella and suitcase24: I begin to hitchhike25: a pickup truck goes by26: driver and passenger staring at the well-dressed business person27: walking backward in the downpour28: from the other direction a white car approaches, passes29: and then jams on its brakes30: a man throws the door open31: gesturing for me to get in32: shaking with cold and anger I get inside33: in the most humble manner the man identifies himself as the dispatcher34: with whom I've spoken three times this morning35: to get me to my plane36: he's abandoned his post37: and raced from the company in his personal car38: he apologizes profusely39: but does not explain why a taxi couldn't pick me up("Taxi" continued)40: except to say they are very, very busy this morning41: delivering me to the airport bus42: he refuses the money I press into his hand43: and with more apologies implores me to patronize his company in the future44: later settling back into my seat as the plane takes off45: I open the newspaper46: On the second page my eyes wander to a headline47: Taxi strike begins this morning in Denville89Text for the narrative "Jewel Thief'	 901: when a trickster named Jim Moran2: learned that the crown prince of Saudi Arabia was in hollywood for a long visit3: he decided to pose as the prince one night4: and see what it was like to be treated as royalty5: he would wait until the prince was out of town6: then dress himself in Arab robes7: and have dinner in a fancy restaurant8: that was only part of his plan9: he hired three actors to help him10: one to pose as his Arab dinner companion11: the other two as his servants12: he arranged with a costume company to rent the robes and other clothing they would need13: then he bought a big batch of glass jewels and a large amethyst14: an inexpensive gemstone that looks more valuable than it is15: he stored his so-called precious jewels in a large leather pouch16: when Moran heard that the prince had left Hollywood for a few days17: he called an expensive restaurant18:and reserved a table for that night19: he also rented a chauffeur driven limosine20: when they arrived at the restaurant21: the two servants went inside to make certain22: that the table was suitable for the prince and his friend23: after they seated them24: they stood nearby25: their table was at the end of the dance floor26: whenever the dance band played27: the dancers jostled one another28: to get a good look at the royal party29: at one point Moran sent one of his servants30: to ask the band leader to play this is the life31: a song that was one of his favorites32: after the band played the song33: the prince smiled at the band leader34: and nodded his approval35: then he pulled out his pouch of jewels and emptied it onto the tablecloth in front of him36: as everyone in the restaurant watched37: he selected the amethyst38: and then sent it to the band leader as a sign of his appreciation39: soon after dinner the prince decided to leavef"Jewel Thief' continued)40: his party started across the dance floor toward the door41: suddenly the prince's jewel pouch opened42: and the glass jewels fell to the floor spilling in all directions43: the servants started to pick them up44: but the prince said, "leave them"45: and swept from the room without looking back46: as the Arabs were seating themselves in the limosine47: everyone in the restaurant was on hands and knees48: scrambling for the jewels the prince had left behind9192Text for the restaurant review "Siggy's Place" 1: the glass trophy case in the entranceway of Siggy's place2: is a reflection of one man's work3: chef-owner Siggy Biewald is a familiar face4: at international culinary salons across europe5: and the hardware he's collected from them is impressive6: the case bulges with medals, plaques7: and various certificates proclaiming his excellence in competition8: a guest book overflows with rave reviews from grateful customers9: although the daily fare at Siggy's is not the stuff of international salons10: this is good homecooking for a reasonable dollar11: there are three reasons to keep Siggy's in mind12: one, the sunday brunch; the best in town for a sunday meal13: two, his special menu featuring german dishes14: and three, in the buffalo business Siggy is a groundbreaker15: during the summer season when your house overflows with out of towners16: and you're frantic to find a novel dining experience for them17: buffalo might be the answer18: the appetizer of choice is smoked goose breast19: which is lightly smoked so the strong goosy flavor still comes through20: the amount of choice in this restaurant is staggering21: apart from buffalo and german dishes the regular menu includes22: pork tenderloin, rib-eye beef23: and skewered beef with papaya and pineapple24: lamb chops and even a pheasant dish25: if you're a red meat fan26: you'll enjoy Siggy's special buffalo menu27: buffalo chili, buffalo steak with mushrooms, buffalo stew28: buffalo short ribs, or buffalo bratwurst with sauerkraut29: there's an item you won't find on many menus30: the buffalo cheeseburgers are okay31: but I'd be hard pressed to tell them apart from beef32: the buffalo stroganoff33: which Siggy makes with less sour cream and more dill pickle than some cooks would like34: is a comfortable dish for a big appetite35: with all of this the menu promises fresh vegetables36: and sometimes they are37: other times you get squishy frozen carrots38: avoid if you can the shrimp crepes39: which are a soggy effort involving some canned shrimp and some strange gray bits(Siggy's Place continued)40: that are in fact oysters in less than good eating condition41: for dessert ask about the black forest torte42: if it's fresh it's wonderful43: the wine list is short44: the service is good45: dinner for two is about thirty-five dollars, complete93Text for the restaurant review "Ninth Street Cafe"1: this is a comfortable hole in the wall near the university2: dim, warm, and almost always full3: it's become a favorite spot for students4: with its reasonable prices, pretty good cooking5: and a menu that's as varied as the choice of music6: classical is interspersed with jazz and rock-n-roll7: the regular menu runs to three pages8: and daily specials are chalked on an overhead blackboard9: if you're looking for a light meal perhaps a top notch sandwich10: you're in the right place11: give this kitchen a couple slices of bread or a burger bun12: and they're off and running13: The burgers are made of coarse juicy beef with sprouts, cheese, avocado slices14: and other good extras dribbling out the side15: this cafe makes an upscale version of a clubhouse16: a sandwich which is dipped in egg and fried17: it's definitely a knife and fork affair18: as well there is a house special sandwich19: ham, cheese, tomatoes, mushrooms, onions served open faced20: this one is a plate licker21: the soups, which ought to be first rate in a place like this, are ordinary22: one special had recently been a can of creamed corn23: which was made smooth24: and served not quite hot25: the french onion soup26: a dark consomme with cubes of bread27: and lots of fresh onions under a thick layer of baked cheese28: is a much better deal29: there's a ceasar salad30: there's always a ceasar salad31: but this one involves big juicy ribs of romaine lettuce32: a rich dressing33: and a healthy sprinkling of freshly grated parmesan on top34: it's a better deal than the strips of darkening avocado35: on either side of defrosted soggy shrimps36: bathed in a bland cream and celery dressing37: pasta is available in great variety38: and you're invited to match any pasta with any six or seven sauces94(Ninth Street Cafe continued) 	 9539: the lasagna is made with fresh spinach and carrots40: which sounds bizarre but tastes just fine41: this bistro also offers a selection of heavier meals42: including peppered steak43: beef stroganoff44: veal in cream45: and various species of fish and fowl46: for dessert have the homemade ice cream47: the cheesecake 's dry48: the cafe has an adequate wine list49: and champagne is available by the glass50: if you're more thirsty than hungry51: there's a lounge upstairsText for the restaurant review "Mrs. T's" 1: There's a small restaurant on seventh avenue2: with a large rock in the middle of the room3: and a free form fireplace that burns real wood on cold evenings4: Welcome to Mrs. T's5: a cafe where well fed customers dawdle happily over their coffee6: while the refills keep coming7: the chef in this kitchen actually cooks8: as opposed to those who merely defrost9: the food is plain here but good10: for openers there're garlic laced chicken wings11: smoked salmon with capers12: and jumbo garlic shrimp13: with the smoked salmon leading the pack14: and the garlic wings running a close second15: the shrimp are just ordinary16: among the main courses the chicken kiev is large,golden, a little dry outside17: some would call it extra crispy18: its a question of semantics19: but full of parsley and melted butter20: which spurts nicely when you slice into it21: the vegetables are likely to be peas with a scoop of mashed potatoes on the side22: if you like seafood23: try the sauteed shrimp and crab on noodles24: its a well put together plate25: with lots of mushrooms and a gentle white wine sauce that goes well with the seafood26: the big plus in this restaurant is the homemade corned beef27: which is rosy and lean and just spicy enough28: Mrs T's pride and joy29: made weekly in discrete quantities in this very kitchen30: and it's good31: following in the time honored tradition of this city's finest corned beef sources32: take note of the house special sandwich33: it starts with rye bread34: and builds into a great wad of thin-sliced wonderful beef35: it's a terrific mouthful36: if you like your beef uncorned37: there's beef stroganoff38: or the roast beef dinner9697("Mrs. T's continued)39: the corned beef also shows up in one of several omelettes40: but the sandwich is its big moment41: other big moments here include42: the honestly homemade pumpkin pie with whipped cream43: and the classic european cheese cake44: dense and moist and not too sticky sweet45: the chocolate cheesecake is a certified diet buster46: but Mrs. T ain't perfect47: her ceasar salad is loaded with bacon bits48: and the dressing doesn't quite make it49: if you're lusting after the definitive ceasar50: this won't be it51: lunch and sunday brunch are both on Mrs. T's schedule52: and the restaurant is closed on monday evening53: there's a short wine list54: service is friendly and efficient55: in spite of the fact that one waitress seems to handle it allText for the recipe "Apple Tart" 1: This recipe makes six to eight servings2: and requires the following nine ingredients3: fourteen tablespoons of sweet butter4: a two third cup of granulatd sugar5: seven tablespoons of dark brown sugar6: six large red delicious apples7: one short crust pastry8: a quarter teaspoon of nutmeg9: a quarter teaspoon of cinnamon10: one tablespoon of water11: sweetened whipped cream12: and now for the method13: preheat the oven to four-hundred-and-fifty degrees14: the ideal dish for making this tart is a metal dish15: ten inches in diameter and about two inches deep16: in the dish put twelve tablespoons of the butter17: one-third cup of the white sugar18: and half the dark brown sugar19: blend thoroughly with your fingers20: and then pat this mixture around the sides and bottom of the dish21: next peel and core the apples22: slice them an eigth to a quarter inch in thickness23: arrange them in an overlapping symmetrical petal-like fashion24: over the butter and sugar base25: when you're finished26: the apples should come just to the rim of the dish and no higher27: melt the remaining two tablespoons of butter28: and add the nutmeg, cinnamon, and remaining brown sugar29: stir this mixture30: then sprinkle it over the apples31: then roll out the pastry into a circle about one-eighth inch in thickness32: cut it to fit as precisely as possible over the apples33: fit the pastry over the apples34: making sure it doesn't overlap the rim of the dish35: also cut a small slit in the center of the pastry to allow steam to escape36: bake the tart for thirty minutes37: or until the pastry is lightly browned38: afterthirty minutes remove the dish from the oven98("Apple Tart" continued) 	 9939: and increase the oven heat to five-hundred-and-fifty degrees40: make a round of foil to fit over the dish41: to prevent the pastry from burning42: bake, covered, for forty-five minutes to an hour43: until the liquid that forms around the apples has changed from a runny yellow44: to a dark, oozing sticky amber45: this can be noted by carefully tilting the pan46: and looking under the crust47: it's important to note that oven thermostats in homes vary48: and it may be necesssary to adjust the oven heat49: if the tart starts to burn50: on the other hand the oven must be hot enough to thicken the filling51: when the tart's done52: place a serving plate over the top of the dish53: then quickly invert the tart54: The apples should appear dark in color55: finally melt the remaining sugar with the water in a thick bottomed saucepan56: when the sugar is melted and is dark amber57: remove it from the heat58: using a pastry brush59: and working as quickly as you can60: paint a thin layer of caramel over the surface of the apples61: let the dish cool62: serve warm with sweetened whipped creamText for the recipe "Chocolate Cake" 1: This recipe makes twelve to sixteen servings2: and requires the following nine ingredients3: Four cups of sifted all purpose flour4: One and a half cups of butter5: Half pound of semi-sweet chocolate6: One third cup of warm water7: Half a teaspoon of salt8: Two cups of heavy cream9: One-pound of the finest quality sweet chocolate10: Confectioner's sugar11: Chocolate curls12: To make the cake layers13: Place the flour in the bowl14: Using a pastry blender, two knives, or your fingertips15: Work the butter into the flour as though making pastry16: Until the mixture resembles coarse oatmeal17: Don't allow the butter to become oil18: Melt the semi-sweet chocolate very slowly over hot water in a double boiler19: Then beat in the water and salt until the mixture is smooth20: Fold the chocolate mixture into the flour and butter mixture21: Divide the dough into three parts22: Wrap each part in waxed paper23: and chill twenty minutes or until firm in the refrigerator24: Between sheets of waxed paper, roll each third of the dough into a rectangle25: about eight by twelve inches26: Peel off the top paper27: and fold the rectangle of dough into threes28: peeling the bottom waxed pager off as you go29: Now wrap in fresh waxed paper and chill about twenty-minutes30: Repeat this procedure of rolling between waxed paper100("Chocolate Cake" continued)31: folding and chilling twice more32: for a total of three times33: Wrap each third of the dough in waxed paper and chill until very firm34: This takes about two hours35: Preheat the oven to three-hundred and twenty-five degrees36: After removing the dough from the fridge37: Between sheets of waxed paper, roll out each third of the dough into a circle38: and remove the top sheet of waxed paper39: Cut a nine-to-ten inch circle with a sharp knife40: Leaving a round of waxed paper beneath the layer41: Place the circle, waxed paper side down, on an ungreased baking sheet42: Bake thirty minutes or until done43: Cool on the sheet44: Repeat with the other two thirds of the dough45: When the layers are completely cold46: Place the cream in a chilled bowl buried in ice47: Melt the fine quality chocolate very slowly over hot water in a double boiler48: and when its melted, fold it into the cream49: which has been whipped until stiff50: Spread the chocolate cream filling between the round layers51: It will be quite thick52: Sprinkle the top with confectioner's sugar53: and garnish with the chocolate curls54: Chill well55: The cake should be removed from the refrigerator56: at least one hour before it is to be served57: Use a sharp serrated knife for cutting101Text for the recipe "Poached Trout" 1:This recipe makes about eight servings2: and requires the following nine ingredients3: Two, three pound trout4: Two eggs, separated5: Half a cup of all purpose flour6: Quarter teaspoon of nutmeg7: Salt8: Three tablespoons of melted butter9: Half a cup of milk10: One cup of heavy cream11: Four cups of boiling salted water12: In preparing this dish, it's important that all ingredients be quite cold13:before they're put together14:When purchasing the fish15:have the trout deboned and filleted by someone in the fish department16:The bones and skin will be needed for a stock, so be sure to get them17: Keep the fish, bones, and skin refrigerated until ready to use.18:There should be about one-and-a-half pounds of fish fillets19:Combine the egg yolks, flour, half of the nutmeg20: and the salt in a saucepan, and stir rapidly with a whisk21: while stirring, add the melted butter22: Bring the milk just to a boil23: and beat it into the egg and flour mixture24: Continue stirring and cooking rapidly25: until the mixture pulls away from the sides of the saucepan26: Let the mixture cool27: This mixture is called a panade28: Cover and refrigerate until quite cold29: or put the mixture in the freezer30: but don't allow it to freeze31: Next place each fish fillet on a flat surface102("Poached Trout" continued)32: and proceed to trim away and discard the tiny fishline down the center of each33: Cut the remainder of the fillets into two-inch lengths34: and add a few at a time to the container of an electric blender, and blend35: Continue blending stirring down with a rubber spatula36: Until all the fish is blended well37: As the fish is blended, transfer it to the bowl of an electric mixer.38: Season the fish with salt and the remaining nutmeg39: and beat well with the electric mixer on medium speed40: Continue beating while adding the panade a little at a time41: When this is well blended, add the egg whites a little at a time42: Beat the cream in gradually43: Butter a large deep pan44: about seventeen by twelve by two and a quarter inches45: Using two large dessert spoons dipped into hot water46: Shape the fish morsels into oval rounded shapes47: They should resemble large duck eggs.48: As they are shaped arrange them neatly on the buttered pan49: Dip the spoons into hot water between shapings50: Butter a large length of wax paper51: and place it buttered side down over the oval shaped fish morsels52: Pour the hot salted water on top of the wax paper53: Letting the water flow outward into the pan54: Bring to a boil on top of the stove and simmer five to ten minutes55: Drain and serve with a hot mushroom sauce103APPENDIX CReadability, word, and syllable counts for each text NarrativesReadability ) #words #syllablesScaredy Cat 70 351 502Jewel Thief 72 401 553Getting a Taxi 67 371 545Restaurant ReviewsSiggy's Place 63 377 573Mrs. T.'s 70 416 586Ninth Street Cafe 69 373 533RecipesApple Tart 71 398 569Chocolate Cake 72 375 529Poached Trout 76 416 5701 A Flesch readability score of 60 to 70 is interpreted as a standard reading score for seventh to eighth104graders.APPENDIX D105Examples of boundary decisions: line 1: The prince smiled at the band leaderline 2: and nodded his approvalline 1: The prince smiled at the band leader and nodded his approvaluse two lines as per the experimenterline 1: Stir this mixtureline 2: then sprinkle it over the applesline 1: Stir this mixture then sprinkle it over the applesuse two lines as per the experimenterline 1: This is a comfortable hole in the wall near the university dim, warm,and almost always fullline 1: This is a comfortable hole in the wall near the universityline 2: dim, warm, and almost always fulluse two lines as per the pilot subjectline 1: There's a small restaurant on seventh avenue with a large rock in themiddle of the roomlinel: There's a small restaurant on seventh avenueline2: with a large rock in the middle of the roomuse two lines as per the pilot subjectExample 1: Experimenter:Pilot subject:Decision:Example 2: Experimenter:Pilot subject:Decision:Example 3: Experimenter:Pilot subject:Decision:Example 4: Experimenter:Pilot subject:Decision:Example 5: Experimenter:Pilot subject:DecisionLine 1: A place so far north of Denville that city buses lurch past only threetimes a dayline 1: A place so far north of Denvilleline 2: that city buses lurch past only three times a dayuse two lines as per the pilot subject106APPENDIX EA Distribution of the Words (from 1 to 19) per Line Across Texts.I 2 2 4 5.6 2 a 2 la 12 11 14 11 16 1.2 la 12TEXTNarrativeTaxi 1 	 1 4 5 7 7 4 	 4 5 4 2 2 	 - 1Scaredy - 	 2 4 1 5 7 4 	 5 2 1 4 1 	 - 	 1 3	 1Jewel - 	 1 3 3 8 8 5 	 6 4 3 2 2 	 1 	 1 	 1 1Restaurant ReviewSiggy - 	 1 2 5 4 6 9 	 5 4 3 3 1 	 1 	 1Mrs. T - 	 2 6 10 4 11 8 	 2 2 4 1 2 	 1 	 2Ninth 1 	 4 4 5 4 11 4 	 6 4 2 3 2- 	 1RecipeTart - 	 3 4 9 12 7- 9 6 3 1 2 	 4- 	 1 -Cake 1 	 2 1 8 13 6 5 	 6 6 2 2- 4 	 1	 - -Trout 1- 1 5 6 8 5 5 	 5 3 5 5- 2 	 2 	 2-107APPENDIX FListening InstructionsDuring this task you will be listening to a passage in background noise. The noise issomewhat like that encountered at a cafeteria or cocktail party.The passage will be presented in a line by line fashion.Both the noise and the passage will be presented to your right ear.After each line is presented please repeat back aloud as accurately as possible what you havejust heard. It is important to listen closely to the words in each of the lines of the passage,so that you can repeat them back to me as accurately as possible Each line will bepresented only once.If AT ANY TIME you feel you repeated something back incorrectly or feel that you forgotto include some information in a previous statement, I want you to tell me what informationyou want to change or add.Also should you be unsure of what you heard, feel free to say whatever possible answerscome to mind . For example if the sentence "Suddenly the fans rushed the stage" is presentedand you repeat back "The fans ran up to the stage" and then immediately think that what youreally heard was "The fans rushed the stage", and then realize that what you really heard was"Suddenly the fans rushed the stage", then give me all three responses, or more if they occurto you.This listening and repeating task is not under any time limits. Take the time you need torepeat back what you hear. And take the ti me to consider any changes you may want tomake to the answers you have given: Additions, corrections, whatever..108When you have finished repeating back what you have heard and have made the changes youwant to what you have just said, or changes to what you have previously said press the buttonto indicate that you are ready to go onto the next item.Finally,it is important to listen to and understand the passage itself as you will be askedto answer some questions after the entire passage has been presented.The responses you make are being tape-recorded for analysis.APPENDIX GQuestions for the Narrative: Taxi1. The narrator's housea. has been swept off a mountain slopeb. teeters on a mountain slopec. has good bus serviced. is close to the airport2. The taxi company in this story isa. trying to maximize profitsb. servicing its customers as best it couldc. poorly rund. doing its best to cope with the rainy weather3. The taxi drivera. is a regular full time driverb. is also the owner of the taxi companyc. is also a dispatcherd. is a casual part-time driver4. The narrator decides to hitchhike becausea. there's lots of trafficb. she's given up on the taxi companyc. its legald. she's getting wet and cold5. At the start of the story the narrator isa. anxiousb. sleepingc. in a good moodd. crying6. Which of the following vehicles drives by the narrator?a. a dump truckb. a vanc. a motorcycled. a pickup truck1097. The taxi drivera. makes a lot of money from this driveb. charges only half the farec. refuses money for his serviced. receives a large tip for his service8. The taxi drivera. wants the narrator to make her planeb. has had a hard time finding the narrator's housec. has risked his life to get the narrator to the airportd. is just learning the city9. When the story begins it'sa. eleven in the morningb. seven in the morningc. noond. seven in the evening10. The narrator waits for a taxi outside on a bridge becausea. she thinks its the only way she will get a taxib. she is tired of waiting insidec. she thinks her house is too hard to findd. she can wait for a bus at the same time110Questions for the Narrative: Jewel Prank1. The jewels were stored ina. a safeb. a pouchc. a pursed. a case2. At the end of the story the restaurant customersa. were excitedb. were upsetc. were tiredd. were satisfied3. When the band played "This is the Life"a. Jim Moran was feeling like he was in a dreamb. Jim Moran was feeling like a drinkc. Jim Moran was feeling like dancingd. Jim Moran was feeling like royalty4. Jim Moran showed his approval to the band leadera. by nodding his headb. by clapping his handsc. by shaking his handd. by raising his eyebrows5. In this story the main charactera. waited until the timing was right to go to a fancy restaurantb. wasted no time in making reservations at a fancy restaurantc. bought expensive jewelsd. enjoyed eating at fancy restaurants6. Upon arriving at the restaurant who entered the restaurant first?a. Jim Moranb. The band leaderc. one actord. two servants1117. Jim Moran left the restaurant aftera. finding a fly in his soupb. eating his dinnerc. the real prince entered the restaurantd. spilling his drink8. In this story, Jim Morana. liked to be generousb. liked to hurt peoplec. liked to lied. liked to have fun9. Jim Moran wasa. a mobsterb. an actorc. a trickstere. a prince10. The band leadera. could not become rich by selling the gemstone he was givenb. could make lots of money with the gemstone he was givenc. could become famous by selling the gemstone he was givend. was a friend of Jim Moran's112Questions for the Narrative: Scaredy Cat1. The basement in this story is typically old fashioned becausea. it has a high ceilingb. it has a low ceilingc. it has poor sealantd. it has no lighting2. The sister of the narratora. is understandingb. is sympatheticc. is also scared of the bogey mand. is a typical sister3. The narrator goes down into the basementa. because she cares what other people thinkb. because she knows she is being sillyc. because she hates her sistersd. because nobody else will4. When the narrator goes into the laundry room what happens?a. the bogey man jumps out at hertb. a monster with green eyes attacks herc. a cat jumps out at herd. her sister jumps out at her5. Our imaginationsa. are strange thingsb. are very creativec. are what make us humand. are fun to have6. The narrator goes into the basement to geta. bath towelsb. flourc. dish soapd. dish towels1131147. When returning upstairs the narratora. runsb. tripsc. skipsd. crawls8. At the end of the storya. the narrator has overcome her fearsb. still believes in the bogey manc. realizes there's no such thing as a bogey mand. is glad her sister forced her to confront her fears9. According to the narrator, her basement is a frightening place becausea. the bogey man hides thereb. its hard to see in the darkc. nobody likes itd. scarey creatures hide there10. The narrator doesn't want to go down to the basement when her sister asks becausea. she is too busyb. she is lazyc. she has other things to dod. she is afraid of the bogey manQuestions for the Restaurant Review: Ninth Street Cafe1. This restaurant is the place to eata. nachosb. light mealsc. dessertd. french cuisine2. The dessert menu at this bistro isa. extensiveb. limitedc. for chocolate loversd. for cheesecake lovers3. The ceasar salad at this restauranta. is like any other ceasar saladb. is worth ordering if you like saladsc. is terribled. is not a traditional ceasar salad4. The lasagna is made witha. fresh cottage cheeseb. fresh ground beefc. fresh carrots and spinachd. fresh pasta5. This restaurant hasa. an extensive menub. a special kid's menuc. a different menu every night of the weekd. all of the above6. Which of the following was a soup special at this restaurant?a. canned cream of mushroomb. canned cream of cornc. canned cream of asparagusd. canned cream of celery1157. The best choice for dessert isa. the cheese cakeb. the chocolate cakec. the berries and creamd. the homemade ice cream8. If you're craving pastaa. order the spaghetti with garlicb. avoid the pasta menu herec. there's a limited choice hered. you're likely to find something to satisfy your tastebuds9. This restaurant is neara. a universityb. a subway stationc. a malld. a college10. The burgers in this restauranta. are tastyb. are ordinaryc. are just like those you get at McDonald'sd. are made with tofu116Questions for the Restaurant Review of Mrs. T's Cafe1. Which of the following is an "Opener" at Mrs. T's?a. chicken kievb. shrimp and crab on noodlesc. smoked salmon with capersd. potato skins2. The efficiency of the restaurant's one waitress isa. revolutionaryb. comfortingc. terribled. amazing3.The best corned beef dish on the menu isa. the omeletteb. the sandwichc. the pied. the salad4. The house special sandwich is described asa. typical deli foodb. too much for one personc. larger than lifed. a big moment5. Mrs T's is the kind of place wherea. you get a good meal for your dollarb. you never feel rushed to leavec. you eat and rund. everything is microwaved6. The pride and joy of Mrs. T'sa. are her childrenb. is the roast beefc. is the shrimp and crab on noodlesd. is the corned beef1177. The restaurant is closed ona. Sundaysb. Sunday nightsc. Monday nightsd. Mondays8. The pumpkin pie served at Mrs. T's is madea. from scratchb. from a box mixc. at a local bakeryd. only in October9. In the middle of Mrs. T's there's aa. coal stoveb. rockc. tabled. crock pot10. The vegetable side dish at Mrs. T's isa. predictableb. always a surprisec. never the samed. not something typically made at home118Questions for the Restaurant Review of Siggy's Place1.The Sunday brunch at Siggy'sa. features a buffetb. is a good meal to take your mother toc. is the best deal in townd. starts at eleven in the morning2. When is the best time to order the black forest torte?a. Soon after it's been bakedb. When it's listed as the dessert specialc. In the eveningsd. When it's served with fresh whipped cream3. This restaurant will appeal toa. fish eatersb. meat eatersc. vegetariansd. everyone4. At Siggy's restauranta. the desserts are the best in townb. the vegetables are always freshc. the cheeseburgers are outstandingd. the choice of foods is staggering5. Many of Siggy's customersa. have experienced indigestionb. have come from Europec. think highly of his cookingd. are chefs6. This restaurant specializes ina. beef dishesb. buffalo dishesc. seafood dishesd. pork dishes1197. The food at Siggy's isa. goodb. terriblec. honestd. excellent8. Siggy's shrimp crepes area. made with the freshest of ingredientsb. a demonstration of his expertisec. a poor reflection of his abilitiesd. a typical example of Siggy's fine cooking9. At the front of Siggy's restaurant is a glass case which displaysa. the menub. a guest bookc. medalsd. pictures of Siggy with famous chefs10. According to this restaurant review, Siggy's restauranta. is reasonably pricedb. will put a dent in your walletc. is known for its blue plate specialsd. is typical of eateries120Questions for the Recipe: Apple Tart1. Which of the following does not happen to the apples?a. they are peeled and coredb. they are arranged in a petal-like fashionc. they are slicedd. they are boiled2. The apple tart is meant to be eatena. immediately after it comes out of the ovenb. with ice creamc. five minutes after it comes out of the ovend. when it's not too hot and not too cold3. In this recipe, the instructions provided area. typical of european recipesb. described with a fair degree of precisionc. nonsensicald. outlined in a very loose fashion4. A round of foil is fitted over the dish toa. speed up the cooking timeb. prevent the pastry from burningc. prevent the apples from spilling outd. prevent the apples from bubbling5. This recipe describes aa. main courseb. soupc. dessertd. appetizer6. After the pastry is rolled out it is puta. on the bottom of the metal dishb. over the apples in the metal dishc. on the bottom of the dish and on top of the applesd. in the freezer1217. Which of the following is painted in a thin layer over the apples?a. caramelb. whipped creamc. butterd. cinnamon8. According to the instructions, the tart may burna. because thermostats frequently breakdownb. because thermostats are poorly designedc. because ovens are often too hotd. because not all ovens have an accurate thermostat9. Nutmeg, cinnamon, & brown sugar are sprinkled over thea. pastry to prevent it from burningb. pastry to prevent it from turning brownc. apples to provide them with flavord. apples to provide them with calories10. This recipe calls fora. one appleb. three applesc. six applesd. twelve apples122Questions for the Recipe: Chocolate Cake1. The chocolate mixture is folded intoa. the warm waterb. the cream and salt mixturec. the flour and butter mixtured. the bowl2. A sharp serrated knife is recommended for cuttinga. because the cake has several layersb. because the cake is jelledc. because cream filling is thick and stiffd. because the cake is frozen3. The number of layers this cake has isa. twob. threec. fourd. five4. Each of the cake layers is bakeda. with a round of waxed paper underneath themb. with a round of tinfoil underneath themc. with a round of tinfoil on top of themd. with a round of waxed paper on top of them5. At the beginning of this recipe the butter is combined with the floura. to make a doughb. to keep the butter from turning to oilc. to make oatmeald. to make a batter6. The dough isa. rolled between two sheets of wax paperb. kneadedc. left to rised. placed in the freezer four times1231247. The top of the cake is decorated witha. almond sliversb. chocolate sliversc. chocolate curlsd. almond sauce8. The cake is baked on an ungreased cookie sheet becausea. the cake is made with cooking oilb. the bottom of the cake is meant to be crunchyc. the the recipe calls for a teflon coated cake pand. the cake is baked with a round of waxed paper beneath it9. This recipea. uses one dozen eggsb. requires nine ingredientsc. serves two peopled. requires two ovens10. Which of the following techniques is used in this recipe to make dough rolling easier?a. Flouring the rolling pin prior to rolling out the doughb. Flouring the dough prior to rolling it outc. Chilling the rolling pin prior to rolling out the doughd. Rolling the dough between sheets of waxed paperQuestions for the recipe: Poached Trout1. In the preparationof this dish all ingredients need to bea. bonedb. coldc. at room temperatured. fresh2. A hot mushroom saucea. turns this dish from a cold appetizer to a hot entreeb. goes well with this dishc. allows the subtle flavors of the fish to fully developd. none of the above3. T o make this recipe, you would requirea. A vast array of herbsb. A well equipped kitchenc. A good knowledge of fish habitsd. A wok4. The fish balls are shaped usinga. an ice cream scoopb. a melon bailerc. two large spoonsd. your hands5. This recipe recommends that youa. catch the fish yourselfb. have an expert prepare the fishc. wash the fish thoroughlyd. soak the fish in lemon juice6. The egg, flour, butter mixture is called aa. panadeb. doughc. canardd. banger1257. This recipe involvesa. gravyb. salted waterc. cranberry sauced. lemon juice8. Which of the following statements best describes this recipe?a. "This is your basic fish recipe"b. "This dish is a favorite among French Canadians"c. "This dish is always served in French restaurants"d. "This recipe requires time"9. This recipe calls for the following amount of nutmega. one tablespoonb. quarter teaspoonc. half cupd. a pinch10. When the panade pulls away from the side of the saucepana. It's time to move onto the next stepb. Something has gone wrong with the recipec. The cooking heat is too highd. It's ready for freezing126APPENDIX HFamiliarity QuestionaireSubject code: 	 Session: 	 Date:Text Type: 	Text Exemplar: 	S/N: 	Familiarity Rating1. I never encounter { Stories Recipes	 RestReviews }2. I occasionally encounter {Stories Recipes	 RestReviews}3. I frequently encounter { Stories Recipes 	 RestReviews }127128APPENDIX IExamples of target messages, responses and coding1. Examples of DeletionTarget:	 Hooded with calls the company's maximizing profits by handling only in city runsResponse: Flooded with calls the company's maximizing profitsCoding: DeletionAuditory:	 no matchSemantic:	 no matchSyntactic:	 no match2. Examples of AdditionTarget:	 "terriby sorry", begins the dispatcherResponse: "I'm terribly sorry", begins the dispatcherCoding: AdditionAuditory:	 no matchSemantic:	 no matchSyntactic:	 no match3. Examples of SubstitutionTarget:	 and a freeform fireplace that burns real wood on cold eveningsResponse: and a free form fireplace that makes it real hotCoding: SubstitutionAuditory:	 no matchSemantic:	 no matchSyntactic:	 no match129Target: 	 to say they can't find my houseResponse: to say they can't find my place Coding: SubstitutionAuditory: 	 no matchSemantic: 	 total matchSyntactic: 	 total matchTarget: 	 unless of course you're afraid of the bogey man she said gleefullyResponse: unless of course you're afraid of the bogeyman she said gruffilyCoding: SubstitutionAuditory: 	 partial matchSemantic: 	 no matchSyntactic: 	 total matchTarget: 	 the gloomy quiet atmosphere seemed to close in on meResponse: the gloomy dark air started closing in on meCoding: SubstitutionAuditory: 	 partial matchSemantic: 	 partial matchSyntactic: 	 total match4. Examples of ExchangeTarget:	 and various certificates proclaiming his excellence in competitionResponse: and certificates proclaiming his excellence in various competitionsCoding: ExchangeAuditory:	 no matchSemantic:	 no matchSyntacti:	 no matchTarget:	 then he bought a big batch of glass jewels and a large amathstResponse: then he bought a big batch of large glass jewels and an amethystCoding: ExchangeAuditory:	 no matchSemantic:	 no matchSyntactic:	 no match5. Examples of Multiple Codings Target: Terribly sorry begins the dispatcherResponse: I'm terribly sorry rings the dispatcherCoding: Addition	Auditory:	 no match	Semantic:	 no match	Syntactic:	 no matchCoding: Substitution	Auditory:	 partial match	Semantic:	 no match	Syntactic:	 total match130131	Target:	 well I'm	 kinda busy right now I repliedResponse: well I kinda visitedCoding: Deletion	Auditory: 	 no match	Semantic: 	 no match	Syntactic: 	 no matchCoding: Substitution	Auditory: 	 partial match	Semantic: 	 no match	Syntactic: 	 no matchCoding: Substitution	Auditory: 	 no match	Semantic: 	 no match	Syntactic: 	 no match


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