Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The effects of noise on identification of topic changes in discourse Tidball, Glynnis Anne 1995

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

Item Metadata

Download

Media
831-ubc_1995-0248.pdf [ 6.64MB ]
Metadata
JSON: 831-1.0086923.json
JSON-LD: 831-1.0086923-ld.json
RDF/XML (Pretty): 831-1.0086923-rdf.xml
RDF/JSON: 831-1.0086923-rdf.json
Turtle: 831-1.0086923-turtle.txt
N-Triples: 831-1.0086923-rdf-ntriples.txt
Original Record: 831-1.0086923-source.json
Full Text
831-1.0086923-fulltext.txt
Citation
831-1.0086923.ris

Full Text

THE EFFECTS OF NOISE ON IDENTIFICATION OF TOPIC CHANGES IN DISCOURSE by GLYNNIS ANNE TIDBALL B.A., Universite Laval, 1987 Dip.of Applied Linguistics, University of British Columbia, 1991 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE in THE FACULTY OF MEDICINE School of Audiology and Speech Sciences We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA April 1995 © Glynnis Anne Tidball 1995 r In presenting t h i s thesis i n p a r t i a l f u l f i l l m e n t of the requirements for an advanced degree at the University of B r i t i s h Columbia, I agree that the Library s h a l l make i t f r e e l y a v a i l a b l e for reference and study. I further agree that permission f o r extensive copying of t h i s thesis for scholarly purposes may be granted by the head of my department or by h i s or her representatives. I t i s understood that copying or pub l i c a t i o n of t h i s thesis for f i n a n c i a l gain s h a l l not be allowed without my written permission. iCf\cxs The University of B r i t i s h Columbia Vancouver, Canada Date A?r{\ ZG t l«Hr ABSTRACT The purpose of the present study was to investigate how adverse listening conditions affect the ability of normal-hearing listeners to identify the boundaries between discourse topics, or when "what is being talked about" has changed. Twelve subjects (21 to 35 years) listened to digitized recordings of a single speaker's monologues presented in three background noise conditions (+5, 0 and -5 dB S:N). Subjects were asked to push a button when they thought that a change of topic was about to occur in the monologue. Subject responses were analyzed for the latency of topic boundary identification and the number and location of responses. The role of prosodic cues in the identification of topic boundaries was also evaluated. It was found that as the listening condition became less favourable, listeners were slower to identify topic boundaries, were less certain as to where topic boundaries occurred, and relied more heavily on cues to topic initiation than on cues to topic termination for identification of topic boundaries. It was also shown that as the signal-to-noise ratio decreased, listeners were less able to utilize cues to topic boundary that are present in low amplitude utterances such as pitch range and contour, laryngealization and pre-boundary syllable lengthening, but that listeners relied on the prosodic cue of pause duration to identify topic boundaries equally in all three listening conditions. iii TABLE OF CONTENTS ABSTRACT ii LIST OF TABLES viii LIST OF FIGURES ix ACKNOWLEDGMENTS x Chapter 1. LITERATURE REVIEW 1 1.1 Introduction 1 1.2 Chapter Outline 2 1.3 Models of Speech Perception and Comprehension 3 1.3.1 A Model of Speech Perception: Aislin and Smith (1988) 3 1.3.2 A Model of Speech Comprehension: Cairns (1984) 4 1.3.2.1 The Lexical Processor 5 1.3.2.2 The Structural Processor 6 1.3.2.3 The Interpretative Processor 7 1.4 Cohesion and Its Role in Comprehension 8 1.5 The Role of Knowledge of Topic in Comprehension 10 1.5.1 A Model of Discourse Comprehension: Kintsch and van Dijk (1978) 11 1.5.2 How Knowledge of Topic Aids Comprehension and Recall 14 1.5.2.1 Increased Processing Ability and Memory Storage 14 1.5.2.2 Predictive Value of Knowledge of Topic 16 iv 1.5.2.3 Memory Capacity and Processing Efficiency 16 1.5.3 Consequences for the Hearing-impaired 19 1.6 Topic 22 1.6.1 Definition of Topic by Content 23 1.6.1.1 Sub-topics 25 1.6.2 Definition of Topic by Boundary Features 25 1.6.2.1 Description of Prosodic Cues 26 1.6.2.2 Lexical Cues 32 1.6.2.3 Listeners' Perception of Topic Boundary Cues 35 1.7 Hypotheses 38 Chapter 2. METHODS 41 2.1 Design 41 2.2 Subjects 41 2.3 Materials ...42 2.3.1 Elicitation of Materials 43 2.3.2 Recording of the Materials 45 2.3.3 Calibration of the Sound Level of the Monologues 45 2.3.4 Characteristics of the Competing Noise 48 2.4 Conditions of Presentation 49 2.5 Ordering of the Conditions 51 2.6 Experimental Task 52 2.7 Method of Measuring Responses 52 V 2.7.1 Definition of Topic Boundary 53 2.7.2 Defining a Subject Response 56 2.7.3 Defining Categories of Subject Response 56 2.8 Subject Observations 57 Chapter 3. RESULTS 60 3.1 Categorization of Responses: Best and Additiona Responses .' 60 3.2 Accuracy of Topic Boundary Identification : 61 3.2.1.1 Pre-Terminal Responses 61 3.2.1.1.1 Mid-topic Responses 61 3.2.1.1.2 Topic-Penultimate Responses 63 3.2.1.1.3 Topic-Final Responses 63 3.2.1.2 Topic Post-Terminal Responses 64 3.2.1.2.1 Inter-Topic Pause Responses 64 3.2.1.2.2 Post-Initiation Responses 65 3.2.1.2.3 Absent Responses 65 3.2.2 Summary of Findings About Accuracy of Topic Boundary Identification 66 3.3 Certainty of Topic Boundary Identification 66 3.3.1 Total Number of Additional Responses 67 3.3.1.1 Mid-topic Responses 68 3.3.1.2 Topic-Penultimate Responses 69 3.3.1.3 Topic-Final Responses 69 3.3.2 Topics Containing Multiple Responses 70 VI 3.3.3 Sumrnary of Findings About Certainty of Topic Boundary Identification 72 3.3.4 Latency of Topic Boundary Identification 72 3.4 Summary of Findings About the Latency of Topic Boundary Identification 75 3.5 Measurement of Prosodic Characteristics 75 3.5.1 Distribution of Pauses in the Monologues 75 3.5.2 Distribution of Utterance Amplitudes in the Monologues 76 3.6 Subjects' Observations 81 Chapter 4. DISCUSSION .'. 84 4.1 Review of hypotheses 84 4.2 Summary of Results 84 4.3 Accuracy of Topic Boundary Identification 85 4.4 Certainty of Topic Boundary Identification 86 4.5 Latency of Topic Boundary Identification 88 4.6 Prosodic Cues to Topic Boundary 89 4.6.1 Pauses and Additional Responses..... 89 4.6.2 Pause Durations: Present and Previous Findings 91 4.6.3 Variations in Utterance Amplitude and Changes in Response Latency and Certainty 93 4.6.4 Utterance Amplitudes: Present and Previous Findings 93 4.6.5 Additional Cues to Topic Boundary 94 4.7 Implications of Findings for Comprehension 95 4.8 Future Directions 97 vii REFERENCES 100 APPENDIX A Subjects' Audiometric Thresholds (dB HL) 104 APPENDIX B Transcripts of Monologues 105 APPENDIX C Instructions to Subjects 175 APPENDIX D Individual Subjects'Results 176 Vll l LIST OF TABLES Table 2-1. Output of White Noise on Madsen OB-802 Audiometer, Measured in One-Third Octave Bands 49 Table 2-2. Presentation Order for Monologues 1 to 6 51 Table 3-1. Distribution of "Best" Responses 62 Table 3-2. Distribution of "Additional" Responses 67 Table 3-3. Mean of the Median Latency of Subjects' Responses (in msec) 73 Table 3-4. Topic Boundary and Non-Topic Boundary Pause Durations (in msec) 76 Table 3-5. RMS Scale Values for Utterances in Topic-Initial, Topic-Final and Other Topic Positions 81 Table D - l . "Best" Responses Falling in Each Zone for +5 dB S:N Condition 176 Table D-2. "Best" Responses Falling in Each Zone for 0 dB S:N Condition 177 Table D-3. "Best" Responses Falling in Each Zone for -5 dB S:N Condition 178 Table D-4. "Additional" Responses Falling in Each Zone for +5 dB S:N Condition 179 Table D-5. "Additional" Responses Falling in Each Zone for 0 dB S:N Condition 180 Table D-6. "Additional" Responses Falling in Each Zone for -5 dB S:N Condition 181 Table D-7. Mean of the Median Latency of "Best" Responses (in msec) 182 Table D-8. Number of Topics Containing Multiple Responses 183 IX LIST OF FIGURES Figure 1-1. Cairns' Model of an Autonomous Comprehension System (Cairns, 1984) 5 Figure 1-2. Coherence Graph for a Set of Propositions (Kintsch & van Dijk, 1978) 13 Figure 2-1. Diagram of Experiment Set-Up 53 Figure 2-2. View of Speech Signal and Subject Response 53 Figure 2-3. View of Topic End-Point 55 Figure 2-4. View of Subject Response 58 Figure 2-5. Diagram of Response Types 59 Figure 3-1. Topics Containing Multiple Responses 71 Figure 3-2. Mean of the Median of Response Latency 74 Figure 3-3. Mean Duration of Inter-Topic and Within-Topic Pauses 77 Figure 3-4. Distribution of Within-Topic Pauses 78 Figure 3-5. Distribution of Inter-Topic Pauses 79 Figure 3-6. Mean RMS Scale Values of Topic-Initial and Topic-Final Utterances 80 Figure 3-7. Topics for which Topic-Final Utterances were Greater in Amplitude than Topic-Final Utterances 83 ACKNOWLEDGMENTS I would extend my sincere thanks to Kathy Pichora-Fuller for her constant support, patience and unceasing enthusiasm throughout this endeavour; to John Gilbert for his fresh ideas and editorial expertise; and to Noelle Lamb for her careful and insightful comments, from all corners of the globe. Thanks also to Anita Stel for her assistance with the analysis of the data. Finally, I would like to thank my husband, my friends and my family, for their encouragement and support throughout this project. 1 1. LITERATURE REVIEW 1.1 Introduction Perhaps the most significant disability experienced by a hard-of-hearing individual is a decreased ability to perceive speech. Diminished speech perception ability typically affects the ease and effectiveness of comprehension in some, if not most, listening situations; however, decreased speech perception does not affect comprehension equally for all individuals in all situations. Comprehension will also depend on characteristics of the speaker, the listener, the message and the listening environment (e.g., Erber, 1988). An important characteristic of the listener is his knowledge of the topic of conversation, or "what is being talked about." Many hard-of-hearing individuals report that when they know the topic of discussion, they experience little difficulty understanding what is said. Conversely, when the topic of discourse changes, and hard-of-hearing listeners no longer know "what is being talked about", they find it difficult to understand. In spoken language, identification of the topic of conversation requires being able to understand the content of the message. Identification of the topic of conversation also depends on the listener's ability to identify the structural cues to topic boundaries; that is, where one topic ends and another begins. In written text, the topic boundaries are generally represented by paragraph boundaries; in spoken discourse, speakers indicate topic boundaries by certain prosodic, lexical and syntactic cues. 2 To identify topics in conversation in ideal listening situations, normal-hearing listeners should benefit from both content cues and structural cues to topic boundaries. However, hard-of-hearing individuals, who experience diminished perception of speech, may be less able to exploit available content and structural topic boundary cues to identify when a topic changes. When a hard-of-hearing individual experiences difficulty understanding a conversation because the topic in unknown to him, it is possible that his difficulty resides partly in his diminished ability to identify when topics change in conversation. To investigate this possibility, it is first necessary to understand how well normal-hearing listeners identify topic changes in discourse in favourable and unfavourable listening conditions. To aid in this understanding, a study was undertaken to investigate the ways in which normal-hearing individuals identify topic changes in discourse when speech is masked by white noise. It is suggested that the results of such a study could ultimately be used to guide subsequent investigations, the purpose of which would be to determine where the breakdowns in identifying shifts in topic occur for hearing-impaired individuals, and whether or not they differ from normal-hearing listeners in this task. 1.2 Chapter Outline In this chapter we will review the concept of topic: how topic might be generated and represented, how topic supports a coherent interpretation of text and discourse, and how knowledge of topic provides contextual information which aids language comprehension for normal- and hard-of-hearing individuals. We will also examine proposed definitions of the 3 notion of topic, and how topic boundaries are signaled and interpreted in spoken discourse. To establish a framework within which to discuss these points, we will first discuss a model of language comprehension. 1.3 Models of Speech Perception and Comprehension The models of language used in the present discussion will cover language processing from the level of sensation of the speech signal to the level of conceptual representation of the message encoded in the signal. The models will include a general model of perception proposed by Aislin and Smith (1988), applied to speech perception, and the language comprehension models outlined by Cairns (1984) and Kintsch and van Dijk (1978). 1.3.1 A Model of Speech Perception: Aislin and Smith (1988) Aislin and Smith (1988) propose that perception involves three levels of processing: 1) the level of sensory primitives or elementary perceptual units such as pitch and loudness, 2) the level of perceptual representation, and 3) the level of higher-order representations, for example language and cognition. Frequency, intensity and temporal properties of the speech signal are mapped from the level of sensory primitives in the peripheral auditory system to the level of perceptual representation. Later input is categorized into those perceptual units' higher-order representations, and meaning is ascribed to the perceptual representations. An example of such a representation of perceptual units are the phonemic contrasts in one's native language. 4 According to the above perceptual model, an acoustic signal is presumably encoded as phonetic input at a level of perceptual representation. For the purposes of the present discussion, a more detailed model for the level of higher-order representations is required in order to explore the manner in which perceptual representations might interact with language and cognition. One such model of language comprehension has been presented by Cairns (1984). 1.3.2 A Model of Speech Comprehension: Cairns (1984) Cairns proposes that speech comprehension requires two types of processors which operate autonomously; the first performs purely linguistic functions, the second performs functions related to both linguistics and cognition. The first type of processor is composed of the Lexical and Structural Processors (see Figure 1-1). These processors analyze the purely linguistic properties of a sentence or an utterance1 and generate a representation of the structural organization of its component lexical items. During its initial operations, the Lexical Processor retrieves items from an internalized lexicon based on phonological information ; the Structural Processor creates a, syntactic representation of the utterance using grammatical information. 1 In this review, "utterance" refers to a spoken unit of language; in contrast, "sentence" refers to a written string of orthographic symbols, represented via visual processing (Lyons, 1977; cited in Brown & Yule, 1989). 2 While not explicitly stated in Cairns' article, we will assume for our purposes that phonetic information is encoded as phonological information at the level of the Lexical Processor. 5 1.3.2.1 The Lexical Processor The Lexical Processor involves two stages of operation during sentence comprehension. The first is a retrieval stage during which all possible lexical entries are Lexicon Conceptual Representation Interpretative Processor Structural Processor Lexical Processor Phonetic Input Real World Knowledge Figure 1-1. Cairns' Model of an Autonomous Comprehension System (Cairns, 1984) retrieved from the lexicon during the first pass of processing. The retrieval stage is not affected by semantic or real world information. The second stage of operation is a post-access decision stage during which the content of the lexical item selected is compared against the input signal to determine the accuracy of the retrieval process. In the case of a lexically ambiguous word, e.g. "watch", which may be a noun or verb, or a homophonous word, e.g. "bear" and "bare", all possible meanings and class forms are accessed during the 6 retrieval stage; selection of the correct interpretation of these words will depend on their context. It is at the post-access decision stage that context influences lexical selection. Thus, retrieval during the first-pass of processing is autonomous, i.e. distinct from higher-level process, whereas higher-level processes do influence selection during the post-access decision stage. During comprehension of speech in unfavourable listening conditions, the phonological information which is available for use in the retrieval and selection of lexical items is less complete than in favourable listening conditions. In the presence of incomplete phonological information, the interpretation of a retrieved item may be ambiguous until information available from higher-level processing acts on lexical selection at the post-access decision stage. Therefore the processors would have to rely more than usual on higher-level operations to determine which of the retrieved items is correct. 1.3.2.2 The Structural Processor The Structural Processor produces a syntactic representation of the items retrieved by the Lexical Processor. Included in the lexical entry is the form class of a word (i.e. noun, verb, etc.), and the sentence frame in which a word may appear (e.g., the verb "hit" requires that a direct object noun phrase follow the verb). Grammatical knowledge, lexical information and surface cues, such as cues to clause boundaries, are integrated at the level of the Structural Processor. As with the Lexical Processor, the Structural Processor functions independently of higher-level processes during initial or first-pass processing. Real-world 7 knowledge is available to the Structural Processor during second-pass processing. For example, real-world knowledge would be required to process "garden-path" sentences, e.g., 1. The horse raced past the barn fell. (Just & Carpenter, 1987). Here, the Structural Processor would initially interpret "the horse" as the noun phrase and "raced past the barn" as the verb phrase. The last word in the sentence, "fell", forces the Structural Processor to revise the initial syntactic analysis in order to produce a workable interpretation of the sentence. 1.3.2.3 The Interpretative Processor The product of the first-pass operations of the Lexical and Structural Processors is a representation comparable to the literal interpretation of an utterance, which is then processed by the second type of processor, the Interpretative Processor. The Interpretative Processor integrates information from the Lexical and Structural Processors with real-world knowledge and generates inferences accordingly. By applying real-world knowledge and inference operations to representations from the linguistic processors, the Interpretative Processor generates a conceptual representation of a message. Cairns proposes that the linguistic processors function autonomously during the first pass of processing; that is, they are distinct from the more general cognitive activity associated with the Interpretative Processor. Inference operations and real-world knowledge cannot influence lexical selection nor the representation of syntactic organization of an utterance or sentence during the first pass. In the case where a word's meaning is ambiguous, as in the case of homophonous words, the lexical processor will retrieve all 8 stored meanings of items based on phonological information alone. From this set of meanings, one meaning will be selected at the level of the Interpretative Processor where real-world knowledge will be applied to the selection process. Consider the following sentences from Just & Carpenter (1987): 2. Our store sells alligator shoes. 3. Our store sells horse shoes. A reader will interpret these sentences by applying his knowledge of the real world, i.e. alligators do not wear shoes, but horses do; shoes may be made from alligator skin, but are not generally made from horse hide. The linguistic processors will generate both meanings of each sentence. By integrating real-world knowledge into the comprehension process, the Interpretative Processor will estimate the plausibility of relationships within each sentence and then select the correct interpretation. The model of language comprehension proposed by Cairns implicates inferences and real-world knowledge in comprehension. Incorporating inferences and real-world knowledge allows a listener to generate more than a literal interpretation of text or discourse. In the next section we will explore the role of inference operations and knowledge types in comprehension. 1.4 Cohesion and Its Role in Comprehension Inference operations and real-world knowledge allow language users to make connections between units of discourse or text. It is by virtue of these processes that language users relate units in text or discourse to one another, and by which a discourse or 9 text achieves cohesion. The term cohesion describes the principle of semantic connectivity in language wherein the interpretation of some element in the text or discourse is dependent on that of another (Halliday & Hasan, 1976). This is illustrated by anaphoric reference, in which the interpretation of a pronoun is dependent on a referent, as in the following sentences (Halliday & Hasan, 1976): 4. Wash and core six cooking apples. Put them into a fireproof dish. Anaphoric reference is an example of cohesion or semantic connectivity which is explicitly expressed in the surface structure of the text. Cohesion may also be implicitly expressed, as in the following sequence: 5. A: There's the phone. B: I'm in the bath. In this example, interpretation of the second sentence depends on information presented in the first. As such, the two sentences are coherent, although the relation is not marked explicitly in the surface structure. A reader will interpret this sequence and similar sequences as a cohesive unit inferred on the basis of his real-world knowledge which, in this example, might include his knowledge of the situation to which the sequence refers. Real-world knowledge which contributes to a listener's or reader's evaluation of coherence includes: knowledge of semantic and syntactic relations, as in anaphoric reference; knowledge of the referential situation, or the situation being referred to, as in knowledge that the above sequence probably refers to an exchange between a couple in their home; knowledge of the causal and logical relations of the events described, as in knowledge that a person who is bathing cannot easily answer the phone (Just & Carpenter, 1987). 10 1.5 The Role of Knowledge of Topic in Comprehension Although language users can establish cohesive relations or ties between items in a text or discourse and use cohesion to expand their interpretation, cohesion is not always sufficient to produce a meaningful overall interpretation of a text or discourse. A reader or listener may be able to discern relations across a sequence of sentences or utterances but may not be able to achieve a coherent sense of what the text or discourse is about. This is illustrated by a segment of text from the beginning of a novel: 6. Through the fence, between the curling flower spaces, I could see them hitting. They were coming toward where the flag was and I went along the fence. Luster was hunting in the grass by the flower tree. They took the flag out, and they were hitting. Then they put the flag back and they went to the table, and he hit and the other hit. They went on, and I went along the fence. Luster came away from the flower tree and we went along the fence and they stopped and we stopped and I looked through the fence while Luster was hunting in the grass. This paragraph from William Faulkner's The Sound and the Fury contains examples of cohesive ties, and yet it is difficult for the reader to arrive at a coherent interpretation of the text without some additional information. Once the reader knows that the paragraph refers to a game of golf, then he can perform the necessary operations to produce a coherent interpretation of the text. This example illustrates how knowledge of the topic of a text or discourse segment, i.e., "what is being talked about", can enhance a reader's or listener's comprehension. This has been shown empirically by several studies. Bransford and Johnson (1973) found that readers' comprehension and recall of a text improved when they had prior knowledge of the text's topic, information that was not evident from the lexical information of the passage. 11 Lorch and Lorch (1985) found in their study of topic structure representation and text recall that subjects recalled more information about more topics if a "topic sentence" was included in the text. Garrett and Saint-Pierre (1980) demonstrated that relevant cues to the topic of a speech sample improved their subjects' estimates of intelligibility of speech in noise. These studies show that knowledge of topic provides a reader/listener with information which aids comprehension. Kintsch and van Dijk's (1978) model of discourse comprehension and production provides a framework in which to examine the processing and storage of cohesive units of text or discourse. Their model will serve as a basis for discussing the role of the reader's or listener's knowledge of topic in comprehension. 1.5.1 A Model of Discourse Comprehension: Kintsch and van Dijk (1978) Kintsch and van Dijk (1978) propose that a sequence of sentences or utterances, at the level of Conceptual Representation in Cairns' model, is represented by a set of propositions. For each sentence or phrase, propositions are generated and stored in a short-term memory buffer of limited size which forms part of the listener's or reader's working memory. Propositions are processed several at a time. When one chunk of propositions has been processed and stored, a new incoming chunk is compared to the chunk stored in the short-term memory buffer. There are three means by which cohesion between propositions is established. First, if there is an overlap between new propositions and those retained in the buffer, the new input is considered to be coherent with the previous text or discourse segment. Second, if there is no overlap, the processor will search for a coherent referent in previously processed 12 propositions held in long-term memory. Third, if the long-term memory search is not successful, the processor will perform inference operations. Propositions generated by inference operations and which connect the incoming proposition to already processed propositions are then added to the short-term memory buffer. These three processes by which a coherent relation is established are considered to represent three degrees of processing ease. Kintsch and van Dijk consider that when the referent for a new proposition is in the buffer, as in the case of anaphoric reference given in example 3, the operation of determining cohesive ties places the fewest demands on processing resources. Long-term memory searches for propositions previously processed place considerably more demands on processing resources, and inference operations are the most resource-consuming operations of all. Once propositions are processed, they are integrated into a structure representing the semantic relations between propositions. Kintsch and van Dijk propose that propositions in discourse are structured at two levels: the local and the global. At the local level is the microstructure, the structure of individual propositions and their relations. At the global level is the macrostructure, or the "meaningful whole" which Kintsch and van Dijk refer to as the topic of discourse. As described by Kintsch and van Dijk, discourse topic is a semantic structure or a sequence of propositions arranged linearly and/or hierarchically. This structure is obtained through a set of semantic mapping rules or macrorules, with micro structural information as input and macrostructural information as output. The three semantic mapping rules or macrorules are as follow: 13 (i) deletion: delete only those propositions that do not form an interpretation condition of a following proposition (ii) generalization: substitute each sequence of propositions by a general proposition denoting an immediate superset. (iii) construction: substitute a proposition denoting a global fact for a sequence of propositions which represent those facts which are normal conditions, components, or consequences, (p. 366) Topic of Discourse 1 5 8 2 3 4 6 7 9 10 11 Figure 1-2. Coherence Graph for a Set of Propositions (Kintsch & van Dijk, 1978) By cycling microstructure-level propositions through these macrorules, the processor generates a network of coherent propositions. Kintsch and van Dijk represent this network with a coherence graph, the nodes of which are propositions with the connecting lines joining propositions with shared referents (see Figure 1-2). The top-most proposition is the discourse topic, the second level is composed of propositions (1,5 and 8) related to the 14 discourse topic, the third level contains propositions (2 to 4, 6, 7, and 9 to 11) connected to propositions in the second but not the first level. 1.5.2 How Knowledge of Topic Aids Comprehension and Recall The model of discourse comprehension proposed by Kintsch and van Dijk (1978) provides a framework in which to examine ways in which knowledge of topic might aid comprehension and recall of language as has been shown in the studies cited earlier (Bransford & Johnson, 1973; Garrett & Saint-Pierre, 1980; Lorch & Lorch, 1985). In the following section, we will discuss specific applications of Kintsch and van Dijk's model to language processing. 1.5.2.1 Increased Processing Ability and Memory Storage At any given time, those discourse propositions most recently processed or retrieved from long-term memory are retained in the reader's or listener's short-term memory buffer. As propositions are cycled through the set of macrorules, they are compared against the propositions residing in the short-term memory buffer and semantic connections between the propositions are established. These most recently processed propositions may then be held in long-term memory, along with other previously processed propositions. If no connections are found in the short-term memory buffer, then the processor initiates a search for related propositions stored in long-term memory; if the search is unsuccessful, inference operations are required. The related propositions derived from these searches and inference operations are moved to the short-term memory buffer. Such searches and inference operations are thought to make heavy demands on the reader's or listener's comprehension resources. 15 Propositions are more likely to be stored in long-term memory if they are frequently cycled through the short-term memory buffer; that is, if they are related semantically to other propositions in the text. Thus a proposition which is related to a considerable number of other propositions, such as a discourse topic, is more likely to be remembered. This has been shown empirically: Kintsch and van Dijk.cite studies which demonstrate that propositions belonging to high levels of the hierarchy of the macrostructure are better recalled than propositions belonging to low levels in the hierarchy (see Kintsch & van Dijk, 1978, for a review). The authors consider that better recall of high-level propositions is related to the frequency with which they are stored in the short-term memory buffer, since high-level propositions will be connected to more of the total set of propositions contained in a coherent text or discourse segment. Kintsch and van Dijk propose that a topic of discourse is generated by cycling propositions through a series of macrorules. They consider it possible that knowledge of topic will reduce the amount of work the Interpretative Processor will need to carry out for three reasons. First, the proposition or concept representing the top level of the semantic structure will already have been established. With fewer resources needed for generating a discourse topic, more resources will be available for language comprehension and memory tasks. Second, the discourse topic proposition should be relevant to a high proportion of incoming propositions, and will frequently be stored in the short-term memory buffer as new propositions are being processed. The Interpretative Processor will need to perform fewer long-term searches and inference operations to locate propositions which link new propositions to the set of text or discourse propositions at hand. Third, knowledge of topic 16 may constrain the number and types of propositions which will be expressed in discourse, and may therefore reduce processing demands involved in inference operations during comprehension. The means by which knowledge of topic might reduce processing demands will be explored further in the following section. 1.5.2.2 Predictive Value of Knowledge of Topic It is reasonable to assume that a reader's or listener's knowledge of the discourse topic, or of the theme of the discourse segment, might enable him to predict to some extent the content and possibly the structure of language in advance of hearing an utterance. As stated earlier, the language processor may need to carry out long-term memory searches or inference operations in order to interpret a proposition. A reader's or listener's knowledge of what is being written or talked about might enable the processor to limit the number and type of inferences and propositions searched during comprehension operations. Thus the processor would not need to search the reader's or listener's real-world knowledge in its entirety, nor all the propositions related to the discussion or text at large, but rather could search through a sub-set of semantically relevant information contained in long-term memory to establish a coherent relation between propositions. This, in turn, would reduce the amount of processing required to interpret a proposition. 1.5.2.3 Memory Capacity and Processing Efficiency The aforementioned factors, which may influence processing efficiency, are related to properties of the text or discourse. Factors related to the characteristics of the reader or listener, such as memory capacity, will also influence processing efficiency. If working 17 memory capacity is great, then the greater the number of propositions that could be held in the buffer, or the greater the resources that could be available to be allocated to other processing tasks. Conversely, the size of the short-term memory buffer will depend to some degree on the amount of resources that must be devoted to other aspects of processing, such as perceptual decoding, lexical accessing, syntactic analyses, inference generation and integration of information. The more automatic these processes are, the greater the capacity of the storage buffer will be. This trade-off between processing and storage functions was the focus of Daneman and Carpenter's (1980) study on individual differences in working memory and reading comprehension. They examined how processing efficiency during reading might influence the amount of information stored in working memory, when working memory is conceptualized as encompassing both the processes required during comprehension and the storage of information,in short-term memory. Specifically, it was proposed that if there is a trade-off between processing and storage functions in working memory, then an efficient reader would have more capacity for storage than would an inefficient reader for whom the processing demands of reading would be relatively heavy. Processing by better readers would not consume all of the available capacity in working memory, thus leaving more capacity for storage of information and possible integration of particular facts into a more general representation. To measure the trade-off between these two functions, subjects were given a reading span test which required both language processing and storage of information. The reading span test consisted of a series of sentences presented individually. Subjects read the 18 sentences in sets of increasing size. At the end of each set, subjects were asked to repeat back the final word in each sentence of the set. The maximum number of sentence-final words that each subject could recall served as an index of their working memory capacity. Results from the study demonstrated that readers who achieved high comprehension scores recalled more items than did poor readers. The more efficient an individual was at processing information during reading, the more residual capacity he had to store facts in short-term memory. The trade-off between efficiency of processing and storage capacity was not specific to reading comprehension; Daneman and Carpenter (1980) also found that listening span, like reading span, correlated well with comprehension abilities. To measure listening span capacity, subjects listened to a set of sentences and were asked to judge whether each sentence was true or false, and remember the last word in each sentence. As with the reading span test, the number of items recalled served as an indication of each subject's listening span. Each subject's listening comprehension was assessed by measuring his ability to (i) recall facts from a passage read to them, (ii) identify a pronominal referent from this passage, and (iii) generate a title for the passage. Analysis of the error types showed that subjects with larger spans made errors that were nonetheless consistent with a good understanding of the passage read to them; in contrast, subjects with smaller spans made errors reflecting a fundamental misunderstanding of the passage. Subjects with larger spans were also better at identifying a theme from the passage than were subjects with smaller spans. Results from this study demonstrate that there is a trade-off between ease of processing and the storage capacity of the short-term memory buffer. Heavy processing 19 demands do not simply decrease an individual's capacity to retain items in memory, they also appear to affect ability to generate a discourse topic (as described by Kintsch & van Dijk, 1978), and to fully understand the content of a segment of discourse or text. 1.5.3 Consequences for the Hearing-impaired The previous section discussed how topic knowledge may improve language retention and comprehension by reducing the amount of cognitive resources allocated to language processing. With this in mind, we will now explore some possible explanations as to why hard-of-hearing listeners would experience difficulty understanding speech when they are unaware of the topic of conversation. First, hard-of-hearing listeners are impaired in their perception and, in turn, in their comprehension of speech. For a hard-of-hearing individual, a speech signal can be either totally below the level of hearing sensitivity, or it can be partially inaudible and is perceived as distorted. In addition to distortion arising form the partial inaudibility of the speech signal, distortion may also be due to loudness recruitment (e.g., Brunt, 1985), abnormal difference thresholds for frequency or intensity (e.g., Ross, Huntington, Newby, & Dixon, 1965), and wider than normal critical bands (e.g., Ross et al., 1965). In a noisy environment, the difficulties experienced by a hard-of-hearing listener in both the perception of speech and the comprehension of language are compounded (e.g., Findlay & Denenburg, 1977), since competing noise masks sound, thereby reducing the amount of speech information available to the listener (e.g., Miller & Nicely, 1955). Listening to speech in quiet may present few difficulties for some hard-of-hearing listeners. However, even when speech is 20 amplified to a level at which it is as audible as possible in quiet, in noise, when compared to normal-hearing listeners, hard-of-hearing listeners generally require a greater signal-to-noise ratio to reach a specified level of performance (e.g., Keith & Talis, 1972; Stelmachowicz, Jesteadt, Gorga, & Mott, 1985). Second, it has been shown that as listening conditions become less favourable (when signal-to-noise ratio decreases), listening becomes increasingly effortful for individuals with normal hearing (Broadbent, 1958; Rabbit, 1966). That is, when noise masks parts of the speech signal, listening places considerable demands on the cognitive resources required for listening. For a person who is hard-of-hearing, the effort of listening to speech in noise must be relatively greater than it is for normal-hearing listeners. Results from several studies (e.g., Carhart, Johnson & Goodman, 1975; Cooper & Cutts, 1971; Pichora-Fuller, Schnieder & Daneman, 1995) show that the effects of masking noise on the perception of speech are greater for hard-of-hearing than for normal-hearing listeners. Even when fitted with amplification, hard-of-hearing individuals experience considerable difficulty understanding speech in adverse listening situations (e.g., Olsen, Jabaley & Pappas, 1966, as cited in Cooper & Cutts, 1971). We may conclude from these and other results that hard-of-hearing listeners probably allocate more mental resources to lower level processes when perceiving speech than do normal-hearing listeners (Pichora-Fuller et al., 1995). Third, Kintsch and van Dijk (1978) propose that the more resources are diverted to tasks such as perceptual decoding, the smaller the short-term memory buffer becomes. This is consistent with Daneman and Carpenter's (1980) finding of a relationship between reading and listening comprehension ability and working memory capacity. Presumably, as 21 the storage capacity of the buffer decreases, fewer propositions will be available for establishing semantic connections. This in turn will adversely affect the comprehension abilities of the listener. Fourth, if hard-of-hearing listeners do not know the topic of conversation, they do i not have a topic of discourse against which input propositions can be checked for coherent references. According to Kintsch and van Dijk's (1978) model of discourse comprehension, once an utterance has generated a proposition, it is checked against propositions held in the short-term memory buffer for semantic connections and then, if no match is found in the buffer, the long-term memory is searched. If a hard-of-hearing individual does not know the topic of discourse or conversation, then presumably the processor will have to search for propositions held in long-term memory or will have to perform inference operations, both of which are considered by Kintsch and van Dijk to be resource-consuming operations. Furthermore, not knowing the discourse topic will mean that the listener's Interpretative Processor will have a broader range of information and proposition types to search than if the topic is known and the processor can limit the scope of the search. For a hard-of-hearing individual, familiarity with the topic of conversation would improve ability to utilize segmental and suprasegmental information, especially in the presence of competing background noise. Ability to identify when shifts in topic occur should also be important for a hard-of-hearing listener. Failure to identify topic changes in conversation would lead a listener to attempt to relate input propositions to an unrelated set of discourse propositions, and perhaps to make false predictions regarding the content and structure of the discourse. Identification of topic boundaries might not explicitly inform 22 listeners of the discourse content, but it might help them to know when the set of relevant propositions is about to shift, and when some or all of the set of propositions stored in short-term memory might be discarded in anticipation of propositions related to a new topic. Kintsch and van Dijk suggest that in discourse, speakers might signal an appropriate chunk size to the listener by employing certain surface cues. Thus, a good speaker will clearly mark sentence boundaries so that a listener can use these cues for chunking purposes, thereby making the processing of chunks easier. It is of value, therefore, to review linguistic features which cue listeners to topic shifts in discourse, and how a listeners' ability to make use of such cues is affected by adverse listening conditions. 1.6 Topic We have thus far discussed how a listener trying to comprehend discourse might benefit from knowing the topic of the discourse. In their model of language comprehension, Kintsch and van Dijk (1978) assume that for a coherent discourse segment, the topic of discourse can be represented by a theme to which all propositions contained in the discourse relate. Kintsch and van Dijk propose that the topic of discourse can be represented by a proposition. Discourse analysts have not, however, reached consensus on how the notion of topic in discourse is to be defined. The matter of identifying and isolating the topic of a segment of discourse continues to be controversial. Brown and Yule (1989) define topic by establishing internal and external criteria; that is, by defining parameters to specify the content of the topic, and by isolating the topic in discourse by its structural features. The following consideration of the definition of topic 23 will follow the same outline as Brown and Yule; we will first define topic in terms of its content, and second in terms of its structural or boundary characteristics as indicated by prosodic and lexical cues. 1.6.1 Definition of Topic by Content A general meaning of the term topic is "what is being talked about" (Brown & Yule, 1989), a definition which may be applied at the sentence or discourse levels. At the sentence level, "topic" refers to the constituent which, in English, is usually the subject of the sentence. In the following example, "John" is considered to be the topic and the predicate is considered to be the comment or what is said about the topic: 7. John / is off to visit his family, topic comment about topic At the discourse level, formal identification of "topic" is less straight forward than at the sentence level. We may still consider it "a matter of a what is being talked about", yet it is often difficult to establish in a simple phrase or sentence what constitutes the topic of a text or conversation. Keenan and Schieffelin (1978) consider the term "discourse topic" to refer to a proposition or set of propositions which may be represented by a single primary supposition. It is likely that topic boundaries also correspond to syntactic clause boundaries, and that syntactic clause completion could provide a third category of topic boundary cues. This said, the effectiveness of syntactic clause completion as a cue to topic boundary is probably minimal, given that clause completion is likely to occur many times within a topic. Therefore, syntactic cues to topic boundaries will not be discussed further in the present work. 24 This definition conforms with Kintsch and van Dijk's (1978) definition of discourse topic. For example, in the following conversational dyad, Keenan and Schieffelin state that both speakers are contributing information about the proposition "we need something for the diaper": 8. Mother: (trying to put too large diaper on doll, holding diaper on) Well we can't hold it on like that. What do we need? Hmm? What do we need for the diaper? Allison (approximately. 20 mos. old): Pin. However, not all topics may be so easily reduced to a single primary supposition. Brown and Yule (1989) point out that Keenan and Schieffelin's definition of topic may be too limited to properly illustrate all that the term "topic" encompasses. Any fragment of conversational discourse may be judged to have different topics by the participants in the conversation, and these judgments may change at different points in conversation. Consider the following example (Brown & Yule, 1989, p. 76; "+" represents pauses 1000-1900 milliseconds in length): 9. R: in those days + when we were young + there was no local fire engine here + it was just a two-wheeled trolley which was kept in the borough - in the borough eh store down on James Street + and whenever a fire broke out it was just a question of whoever saw the fire first yelling "Fire" + and the nearest people ran for the trolley and how they got on goodness knows + nobody was trained in its use + anyway everybody knew to go for the trolley + well + when we were children + we used to use this taw + and it smouldered furiously + black and thick smoke came from it and we used to get it burning and then go to a letter box and just keep blowing + open the letter box + and just keep blowing the smoke in + you see + till you'd fill up the lower part of the house with nothing but smoke + just to put the breeze up + just as a joke + and then of course + when somebody would open a window or a door the smoke would come pouring out + and then + everybody was away then for the trolley + we just stood and watched all of them ++ S: so that's what "smoke the house" is? R: probably + probably + we called it "the taw" 25 Speaker S might describe the topic of this fragment of discourse as "smoking the houses", speaker R "the taw". Another possible topic might be "the prank", though, as Brown and Yule point out, this does not fully describe "what is being talked about" in this segment. The speaker is also talking about a certain time and place, and about a specific person. Thus the topic of any one fragment of conversation can include several elements. 1.6.1.1 Sub-topics As is evident in the above example, a topic may also be broken down into smaller units or sub-topics. Sub-topics may be defined as the different foci or aspects within a topic Brown, Currie & Kentworthy, 1980). Thus we could break down the above discourse fragment into sub-topics such as "how the people of Stornoway reacted to a fire in the town", "how the speaker and his friends would make the townspeople believe a house was on fire", and "speaker and his friends' reaction". For the purposes of this study, sub-topics are relevant in that the cues that signal sub-topic boundaries may be mis-perceived in adverse listening situations by listeners who may consequently mis-interpret a sub-topic as a topic rather than a segment thereof. 1.6.2 Definition of Topic by Boundary Features Thus far we have discussed some criteria for identifying topics and sub-topics in discourse by examining the content of a topic, and we have defined very generally topic content as "what is being talked about". A second means of defining a topic in a text or discourse sample is to identify where one topic ends, and another begins, i.e. the boundaries of a topic. By characterizing the properties of the boundaries of a topic, a structural basis for 26 defining topic can be developed. Using structural criteria to isolate a topic strengthens the definition of topic in formal theory; identification of a topic on the basis of its structural properties eliminates the need for an a priori definition of the topic on the basis of content. In text, paragraph divisions are one way in which topics are structurally defined. In spoken language, speakers may signal a topic boundary by use of linguistic devices such as lexical markers (Shiffrin, 1987), by supra-segmental markers such as variations in fundamental frequency (Brazil, 1978; Brown, Currie, & Kentworthy, 1980; Menn & Boyce, 1982), or by body movements (Duncan & Fiske, 1977). While lexical and supra-segmental markers may function inter-dependently to cue topic changes or topic boundaries in discourse, in much of the literature they are treated as independent cues. In the following sections, we will review the prosodic and lexical cues that are characteristic of topic boundaries in discourse, and how listeners use specific cues to identify topic boundaries. 1.6.2.1 Description of Prosodic Cues The prosody of utterances provides strong cues to topic change in conversation. The prosody of speech, as perceived by the listener, consists of four primary features: length, loudness, pitch and pausing. The following discussion will examine these four primary features as they occur within intonation groups and within larger discourse units, namely topics.4 A fifth prosodic feature, laryngealization, is significant when describing the prosody 4 An intonation group is defined as a segment of connected speech containing one word bearing a pitch accent or nucleus, i.e. an obvious obtrusion of pitch from the pitch of surrounding syllables, and having a pitch pattern or a direction in which the pitch changes, beginning at the nucleus (e.g., rising-falling, falling-rising) (Cruttenden, 1986). The term 27 of large discourse units, and will be also be described in this section. The prosodic characteristic rate of speech, as it pertains to shifts in topic, will also be discussed in this section. The first primary prosodic feature, length, refers to the length of a syllable. Syllable length is influenced by a) the "innate" length of the vowel within a syllable (e.g., as in "peat" vs. "pit"), b) the stress that the syllable carries, c) the position of the syllable within an intonation group, and d) the position of the syllable within a larger discourse unit.5 As a prosodic feature, length can serve to give a syllable prominence, as in the following: 10. John didn't do it versus 11. John didn't do it. A crucial cue to intonation group boundaries is the phenomenon of syllable-final lengthening or pre-pausal lengthening. It has been observed that in English, as well as in other languages, the last syllable before a pause is lengthened and thus provides listeners with a cue to the structure of the intonation phrase. Cruttenden (1986) suggests that pre-pausal lengthening marks the boundaries of intonation groups. Syllable lengthening may also mark a speaker's hesitation, and as such may serve as a sort of filled pause or pause substitute, as in the following: 12. He's [z:] in the middle of doing it now. "intonation group" corresponds to Crystal's (1969) "tone-unit" and Brown, Currie and Kentworthy's (1980) "tone group". 5 Cruttenden (1986) considers the vowel's "innate" length to be a property of the vowel and therefore unrelated to the prosodic feature of length. 28 Lengthening may also serve to mark larger units of discourse. An example of this is phenomenon of lengthening where the last syllable in a sentence preceding a topic boundary is longer than the when it occurs within a topic (Lehiste & Wang, 1977). Lehiste and Wang (1977) examined listeners' ability to identify within-paragraph sentence boundaries from between-paragraph sentence boundaries on the basis of pre-boundary lengthening. Results from their study indicated that subjects could not consistently distinguish within-paragraph sentence boundaries from between-paragraph sentence boundaries on the basis of different degrees of pre-boundary lengthening alone. They found, however, that lengthening interacts with other cues such as pause length and laryngealization, described below, to cue listeners to paragraph boundaries. The second prosodic feature, loudness, refers to the perceptual correlate of intensity. The loudness of a syllable is partly determined by the intensity intrinsic to the vowel. For example, open vowels have a greater intensity than closed vowels (see Cruttenden, 1986, for a review). For linguistic purposes, an accented syllable may be uttered with greater intensity in order to distinguish it from unaccented syllables. At the supra-sentential level, the overall amplitude of an utterance, in relation to other utterances in a discourse segment, may serve to indicate topic boundaries. Brown, Currie and Kentworthy (1980) found that speech amplitude appeared to rise at the start of a new topic and fall at the end. We might therefore expect variations in intensity to be another prosodic feature which signals topic boundaries. The third prosodic feature, voice pitch, is the perceptual correlate of the rate of vibration of the vocal folds or the fundamental frequency of speech. Changes in fundamental frequency over time correspond to changes in pitch. The degree and direction 29 of change over time in amplitude fundamental frequency form the basis of intonation contours in speech. As with the prosodic features of length and loudness, some variations in voice pitch are related to the "innate" fundamental frequency of vowels and should be considered distinct from the prosodic cue of voice pitch. For example, Peterson and Barney 1952) demonstrated that the average fundamental frequency of segments of the vowel [u] is typically 14% higher than that of the vowel [a] when the sample of segments are produced by an adult male, and 9% higher when they produced by an adult female. The prosodic feature of pitch depends on the transition of a speaker's fundamental frequency over time. Specifically, the range and contour of vocal excursion as well as the register or range of vocal fold vibration typical c>f a particular speaker contribute to a listener's perception of intonation. Typically, the first utterance in a topic carries a high pitch relative to the average pitch of the speaker's voice; the beginning of a topic is also marked with a large pitch excursion. In contrast, the end of a topic is generally marked by a very low pitch with a declining pitch contour (Brazil, 1978; Brown et al., 1980; Geluykens & Swerts, 1991; Lehiste, 1980; Lehiste & Wang, 1977; Menn & Boyce, 1982; Swerts et al., 1992). The prosodic features described above generally do not occur independently. For example, voice pitch, loudness and syllable length all contribute to relative prominence in speech, thereby distinguishing between words of different lexical meaning or grammatical class: 13. insult (noun) versus 14. insult (verb) 30 or emphasizing a word in a sentence: 15. Pat didn't do it. versus 16. Pat didn't do it. Lieberman (1967) proposed that speakers can use prosody to disambiguate syntactically ambiguous sentences that have more than one possible surface structure. Boundaries of intonation phrases often correspond to syntactic phrase boundaries and thus appear to aid listeners to distinguish between structurally ambiguous sentence pairs, such as the following (Price, Ostendorf, Shattuck-Huffnagel, & Fong, 1991): 17. I read NP[a review PP[of NP [nasality in German]]] versus 18. I read NP[a review PP[of nasality] PP[in German]] This last example illustrates the fourth prosodic feature, pausing. Cruttenden (1986) proposes that pauses (indicated by "...") typically occur at three places in an utterance: (i) at major constituent boundaries, i.e., between clauses or between subject and predicate, generally indicating an intonation group boundary (indicated by "/"): 19. The Prince of Wales /...is visiting Cardiff tomorrow; (ii) before words of high semantic content or low predictability, as might indicate word-finding difficulties: 20. The Minister talked at length about the ... redeployment of labour; (iii) after the first word in an intonation group, which seems to serve as a holding device while the speaker plans the rest of the sentence: 31 21. I do like Elgar's violin concerto./ It's ... quite the most perfect work of its kind. Only type (i) pauses indicate intonation-group boundaries, primarily because the portions of utterance which pause types (ii) and (iii) separate do not contain a word which carries a pitch accent or nucleus; therefore, these utterance portions fail to qualify as an intonation-group. Categorization of pause types would thus appear to be directly related to the syntactic and semantic content of the utterance, as well as to prosodic features such as the pitch contour of the speaker's voice. Pause type may be categorized by length. For example, pauses of relatively short duration tend to occur at sentence boundaries within a topic or paragraph, whereas longer pauses tend to occur at topic boundaries. Brown et al. (1980) found that long pauses (600 to 1800 msec) correlated with shifts in topic, whereas shorter pauses separated shorter units of speech within topics, a finding replicated by Lehiste (1980) who found that longer pauses were more closely associated with perceived paragraph boundaries than with sentence boundaries. Lehiste also found that pauses between sentences within a paragraph averaged 737 msec, and that those between paragraphs within a conversational turn averaged 1659 msec. Pauses between turns averaged 5045 msec. The fifth prosodic feature relevant to topic boundaries is laryngealization of the speaker's voice. Laryngealization or creaky-voice is a vocal characteristic produced by tightly adducting the arytenoid cartilages located in the posterior region of the larynx. This manner of vocal fold adduction prevents the posterior portion of the vocal folds from vibrating during voicing, and produces a low-pitched sound which occurs at the end of a falling intonation contour for some speakers of English (Ladefoged, 1982). Lehiste (1980) 32 has noted that speaker's often laryngealize speech as they approach the end of a paragraph, and has hypothesized that laryngealization may cue listeners to a forthcoming topic boundary. In addition to the five features of syllable length, loudness, pitch, pausing and laryngealization, an additional characteristic, rate of speech, is considered to change as a function of the location within a topic. Lehiste (1980) demonstrated that identical sentences are of longer duration when uttered in paragraph-final position than when uttered in paragraph-initial position; that is, rate of speech is slower for paragraph-final sentences than for paragraph-initial sentences. As rate of speech is a function of syllable length as well as pause length, changes in rate of speech may be a by-product of changes in syllable and pause length. 1.6.2.2 Lexical Cues The discussion of cues to topic shift in discourse has thus far been limited to the supra-segmental features of speech. Some lexical items may also cue listeners to topic changes and these lexical items can be considered as markers of topic boundaries rather than as part of the content of the topic. Such lexical items mark the hierarchy of propositions in the discourse and do not contribute to the propositional content. Two types of lexical cues that will be reviewed are lexical markers and lexical phrases. Lexical markers are the lexical cues to topic structure in discourse. Schiffrin (1987) defines lexical markers (words such as "oh", "well", "and", "but", "or", "so", "because", "now", "then", "I mean", "y'know") as "sequentially dependent elements which bracket 33 units of talk" (p. 31). Schiffrin defines "a unit of talk" as a syntactic unit, a propositional and/or a tone unit. The term "sequentially dependent" is used to indicate that these markers are devices that function at the level of discourse. Markers such as "well", "now", "right", "you know", do not provide a listener with information about the syntactic structure of a forthcoming utterance, but might provide a listener with information on how its discourse content relates to previous discourse items or propositions. For example, the marker "well" might preface a response that is an insufficient answer to a question: 22. Well, I'm not sure I know what you mean. The marker "now" may precede a comparison: 23. Now that's what I call fantastic. Another function of "now" is to mark the speaker's focus on the forthcoming topic or sub-topic in the discourse. "Then", in contrast,, marks succession in discourse from one topic to another by pointing to preceding information in the discourse. Schiffrin cites the following to illustrate how the marker "now" can be used to point to forthcoming discourse units but how "then" can only be used when it introduces a sub-topic related to previous units in the discourse. 24. a. It depends on where you live. b. Nowl*then our street is not that nice. c. Because there's so much traffic. d. Now/then their street is nice. In descriptive lists, "now" signals the addition of topics which accumulate as specific cases of a more general topic: 34 25. Now another one, eh - what's the other one that eh - they made a lot of money with. And everybody jumped out of their seats. Whatever the marker "now" introduces, it must be interpreted as a subordinate unit of a larger cumulative structure, and hence marks development of the conceptual representation. "Now" also functions as a temporal adverb. Listeners can use intontation to distinguish between "now" as an adverb and "now" as a discourse marker. As a time adverb, "now" receives tonic stress (the pitch accent that is most prominent in an intonation group) and high pitch, but it receives neither prosodic feature when it is a discourse marker. If "now" receives both tonic stress and high pitch, however, and is followed by a pause, Schiffrin considers "now" to be part of a separate intonation group and as such it can therefore be considered to be a discourse marker: 26. Now, for us to do that... Two other markers related to topic structure in discourse structure are "and" and "so". "And" can be used to link events within a discourse topic; "so" can distinguish them from one another. By switching from "and" to "so", events are globally differentiated from each other: Topic 1 EVENT and EVENT and EVENT so Topic 2 EVENT and EVENT and EVENT 35 "And" can also mark topic segments, thereby differentiating between discourse topics at a local level. Topic 1 EVENT EVENT and Topic 2 EVENT EVENT Therefore, the lexical marker "and" can be used at both local and global levels of discourse. The second type of lexical cues to topic boundaries are lexical phrases, that is, distinct lexical phrases which speakers often use to mark topic boundaries (Brown & Yule, 1989). For example, speakers may mark the beginning of a new topic with phrases that relate to the here and now, as in, "I've often thought,", "I suppose," or "Was it you who was telling me...?". To signal the end of a topic, speakers trail off lexically, analogous to speakers decreasing the pitch range and loudness of their voice at the end of a topic. Speakers indicate the end of a topic by not adding "matter" to the topic but rather adding lexical items previously introduced, or by inserting lexical phrases such as "and so on," "and things like that," "that's how I see it." 1.6.2.3 Listeners' Perception of Topic Boundary Cues It is evident from the foregoing discussion that speakers tend to mark topic boundaries by variations in fundamental frequency, syllable length and voice quality (i.e., laryngealization), voice amplitude, pause duration and/or by certain lexical markers. It therefore follows that listeners should be able to use these cues to identify topic boundaries in everyday listening situations. When identifying discourse boundaries in favourable 36 listening situations, normal-hearing persons should be able to base their judgments of where topic boundaries occur by using prosodic and lexical cues, in addition to using the semantic content of the topic. Nevertheless, in unfavourable listening situations, where segmental information may be masked by noise, listeners may have less information on which to base judgments regarding the occurrence of topic boundaries. Several studies have examined listeners' perception of topic boundary cues, especially their perceptions of prosodic cues such as voice pitch and pause length. These studies demonstrate that even in the absence of segmental information that relates the content of a topic, listeners perceive topic boundaries in speech. In a study by Swerts and Geluykens (1993), subjects listened to a monologue consisting of five successive instructions plus the beginning of a sixth utterance. "Topic" was defined as an instruction by a speaker to a listener on how to manipulate a single item in a set of items. Speech was filtered through a 310-Hz low-pass filter and a 260-Hz high-pass filter, thereby minimizing speech intelligibility while preserving much of the prosodic structure of the speech. Subjects listened to three versions. Version 1 had the original prosody of the 50-Hz band-pass signal. Version 2 had a constant pause duration of 920 milliseconds with the original fundamental frequency. In Version 3, the fundamental frequency was made monotonous at 200 Hz and the pause durations were unaltered. The task of each subject was to respond verbally whenever he thought an instruction had ended. Results from the study showed that even in absence of segmental information, subjects were able to identify topic boundaries in the stream of speech well above chance when the prosody of the signal was unchanged. Results also showed that, based on changes in subject perception, variations in fundamental 37 frequency were a stronger cue to topic change than was pause duration. Swerts and Geluykens (1993) also suggest that topic-final cues are communicatively more important than topic-initial cues that serve as confirmations that a new unit has started. In a study of listeners' perception of sentence boundaries and topic or paragraph boundaries, Lehiste (1980) found that although some prosodic features indicating topic boundaries may be perceptually more significant than others, the features that distinguish sentence boundaries from paragraph boundaries appear to be inter-dependent. She measured voice pitch, syllable lengthening at sentence and topic boundaries, pausing and laryngealization, and then compared listeners' identification of topic boundaries, with and without segmental information. To minimize segmental information while preserving prosodic information, the speech signal was spectrally inverted. Lehiste found that the three phonetic cues (lengthening, pausing and laryngealization) may counteract each other so that a sentence followed by a short pause may be interpreted as a paragraph boundary if sufficient pre-boundary lengthening and laryngealization are present. She also found that subjects identified considerably more paragraph boundaries when listening to an inverted speech signal, for which they could only base their judgments on prosodic information, than when they listened to the uninverted speech signal. Presumably this was because listeners could not verify the occurrence of topic boundaries by comparing prosodic information against the content of the paragraphs. Furthermore, despite what might be expected, subjects' reaction times were reportedly slower for normal speech than for inverted speech. Slower reaction time may reflect time needed to perform additional processing tasks related 38 to language comprehension, as the uninverted materials could be comprehended whereas the inverted materials could not be comprehended. 1.7 Hypotheses In the present chapter the following were reviewed: the notion of topic and how knowledge of topic may help language comprehension, how topic boundaries are identified, and how a listener's perception of topic boundaries contributes to their identification of topic. We have also discussed how deterioration of the speech signal, by either a hearing impairment or competing noise, might influence a listener's identification of topic and of topic boundaries in discourse and, consequently, language comprehension. Several studies discussed above have demonstrated that prosodic cues are instrumental in the identification of topic boundaries. Based on these findings, it is hypothesized that listeners for whom speech is distorted by hearing loss and/or obscured by noise might identify a greater number of topic boundaries in conversation than would normal-hearing individuals in a quiet environment. It is also hypothesized that under difficult listening conditions, listeners might be slower to identify topic boundaries than they would be in favourable listening conditions because difficult listening conditions increase processing requirements and decrease prosodic information. In the present study, we examine ways in which the ability of normal-hearing listeners to identify topic changes in discourse is affected by the presence of competing noise. 39 The following hypotheses were tested: Hypothesis 1: As the signal-to-noise ratio decreases and the listening condition becomes less favourable, the accuracy with which a normal-hearing listener identifies topic boundaries will not differ significantly from the accuracy with which he identifies topic boundaries in a more favourable listening condition. Accompanying research hypothesis: As the signal-to-noise ratio decreases, normal-hearing listeners will identify topic boundaries less accurately. Hypothesis 2: As the signal-to-noise ratio decreases and the listening condition becomes less favourable, the certainty with which a normal-hearing listener identifies topic boundaries will not differ significantly from the certainty with which he identifies topic boundaries in a more favourable listening condition. Accompanying research hypothesis: As the signal-to-noise ratio decreases, normal-hearing listeners will incorrectly identify a greater number of sub-topic or non-topic boundaries as topic boundaries. Hypothesis 3: As the signal-to-noise ratio decreases and the listening condition becomes less favourable, the speed with which a normal-hearing listener identifies topic boundaries will not differ significantly from the speed with which he identifies topic boundaries in a more favourable listening condition. 40 Accompanying research hypothesis: As the signal-to-noise ratio decreases, normal-hearing listeners will be slower to identify boundaries. , 41 2. METHODS 2.1 Design The purpose of this experiment was to determine how competing noise affects an individual's accuracy, certainty and speed in identifying topic changes in discourse. Each subject attended one session during which he listened to a total of four pre-recorded monologues, one presented in quiet to familiarize him with the voice in the monologue, and three presented in experimental conditions in differing levels of competing noise. In the three experimental conditions, the monologues and competing noise were presented monaurally at each of three signal-to-noise ratios: +5, 0 and -5 dB. Two groups of subjects served as listeners. Group I listened to the monologues in the following order: +5 dB signal-to-noise ratio (S:N) in the first experimental condition, 0 dB S:N in the second condition, and -5 dB S:N in the third. The order of presentation was reversed for Group II. 2.2 Subjects Two groups of six volunteer subjects participated in the study. Al l were native Canadian English speakers between the ages of 21 and 35 years of age. Group I, subjects Sl to S6, was comprised of three males and three females; the mean age of this group was 27.3 years. Group II, subjects S7 to S12, was comprised of five males and one female; the mean age of this group was 26.2 years. The mean number of years of post-secondary school education was 6.3 and 6.4 for Groups I and II respectively. None of the subjects had received any instruction in linguistics; therefore, it was assumed that they were relatively 42 naive about the lexical, syntactic and prosodic features associated with topic changes in discourse. The hearing of each subject was tested prior to their participation in the study to ensure that it was within normal limits for at least the best ear (i.e., pure-tone air-conduction thresholds obtained at a level of less than or equal to 25 dB HL for 250, 500, 1000, 2000 and 4000 Hz). Threshold asymmetries of 15 dB or greater were noted in 2 subjects. Details of subjects' hearing thresholds are presented in Appendix A. 2.3 Materials The experimental materials consisted of six pre-recorded monologues, each containing eleven topics and therefore ten topic changes. The rationale for using monologues rather than interactional speech samples such as conversational dyads was to avoid the possibility that changes in topic would co-occur with changes in the turns of talkers. Conversational partners tend to negotiate topics of conversation (Brown and Yule, 1983), and the point at which one talker's turn begins often coincides with a change in topic. Avoidance of turn-taking in the materials also reduces the need for the talker to indicate, prosodically and lexically, when he is ready to concede his turn to his conversational partner (Cutler & Pearson, 1986). Using monologues permitted the evaluation of topic change cues in the absence of possibly confounding cues to turn-taking. An important consideration in preparing the materials was that topic changes needed to be clearly identifiable. To elicit materials that would contain distinct topics, descriptions of pictures from a family photo album were used. The manner in which monologues were elicited is described below. 43 2.3.1 Elicitation of Materials The talker who produced the stimulus materials was an English-speaking male, eighteen years old and a native of Vancouver. He was requested to bring in a large selection of family photos from which seventy-two photos were chosen by the experimenter. Each photograph depicted a unique event and was taken to constitute a distinct topic. To ensure that the talker was familiar with all of the photographs that were chosen, the experimenter and the talker briefly discussed the content of each photograph before beginning the recording. The 72 photos were divided into 6 groups of 11 pictures which were used in the experimental conditions; the remaining 6 photos were reserved for use as practice materials for the talker. The purpose of having the talker practice the task of describing photographs was to allow him to become comfortable with the task. This was important because pilot recordings had indicated that talkers tended to settle into a pattern of descriptive style after a trial run. Following the practice session, the talker was asked to describe the photos in the first set of eleven pictures, and to pause and wait for the experimenter's signal before continuing on to the next set. (See Appendix B for transcriptions of the monologues). The talker described a total of eleven photographs in each set, thereby producing ten "topic" changes in each monologue. No attempt was made to control the length of the descriptions of individual photographs or the length of individual monologues. Any constraints on the length of the descriptions would, it was felt, have detracted from the naturalness of the talker's descriptions. The. length of the monologues ranged from 4 minutes 45 seconds to 6 minutes 22 seconds. 44 The talker was instructed to describe the pictures in each set so that mture listeners might be able to imagine the picture for themselves. It was suggested that he include information such as who was in the photograph, where and when it was taken, and any other details he believed to be relevant. He was also asked to describe each photograph individually, without referring to any information mentioned in the descriptions of previous photos. This was done for two reasons: first, to obtain sets of descriptions that could be ordered randomly, and second, to limit the possibility that a listener might rely on information presented previously. In addition to descriptions of the photographs, the talker also recorded materials unrelated to the materials that were used in the experimental conditions. This recording was used to familiarize listeners with the talker's voice prior to the experimental task. The familiarization monologue was made by asking the talker to instruct a listener to arrange a set of felt pieces to make a house. The listener was a student volunteer. The talker, seated in the sound-attenuated booth, had in front of him a set of felt pieces and a picture of the house that was to be constructed from the felt pieces. The listener, seated outside the booth and out of the view of the talker, listened to the talker through headphones. The listener had in front of her a set of felt pieces identical to that of the talker, but did not have the picture of the house to be assembled from the felt pieces. The listener could hear the talker but could not speak to him; the talker could speak to the listener but could not hear her. 45 2.3.2 Recording of the Materials The experimental monologues and the familiarization monologue were recorded by the talker while seated in a sound-attenuating, double-walled IAC booth. Both the photograph descriptions and the familiarization monologue were recorded using a Sennheiser model K3U microphone positioned approximately six inches from the talker's mouth. The monologues were recorded in mono, via a Proport model 656 stereo-audio DSP port interface, onto a NeXT computer system sound-recording programme, Sound Works 3.0 Version 2. The sound files were recorded at a sampling rate of 16,000 Hz and stored on the hard disk of the computer. These monologues were also archived onto a 512 megabyte optical disk. 2.3.3 Calibration of the Sound Level of the Monologues Each monologue was calibrated using an in-house calibration programme (Liang, 1994). The programme calculates the root mean square (RMS) of the sound pressure level of a speech signal contained in a sound file. Once the programme has calculated the RMS scale value of the speech signal, the programme creates a second sound file containing a calibration tone of a user-specified frequency and for a user-specified duration such that the RMS scale value of the calibration tone equels the RMS scale value calculated for the speeh signal. A text file containing the computed RMS scale values and durations of both the utterances and the pauses in the sound file is also output. The purpose of generating a calibration signal was to permit each separate recording to be presented at approximately 46 equal levels. The calibration signals created for the monologues used in this study were ten-second 1000 Hz tones. Before the RMS programme can process a sound file, the user must prepare the sound file by inserting a one-second silent segment at the beginning of the sound file, followed by a two-second sample of typical background noise. In the present study, the one-second silent interval was generated using Sound Works 3.0 Version 2; the two-second sample of typical background noise was copied into position from the end of each sound file. Once the RMS programme is launched, the prepared sound file is opened and the RMS programme prompts the user to specify first an amplitude threshold, and second a duration threshold. The RMS programme will eliminate any segment from the calculation of the RMS scale value for the osund file if the segment has an RMS scale value below the amplitude threshold value for a duration greater than the duration threshold; the programme will effectively count these segments as pauses. This ensures that the programme includes only the amplitude of the speech signal in the calculation of the RMS scale value and excludes from the calculation the amplitude of the background noise that is present in pauses. The elimination of pauses from the average is especially important when discourse materials are used. Such a correction would not be required if single words or even if single intact sentences were recorded. To specify the amplitude threshold level, the user chooses a value between 1 and 10 as the criterion for the amplitude threshold. The programme then calculates the RMS scale value for the background noise of the sound file based on the two-second sample of typical background noise that has been inserted at the beginning of the recording. The resulting 47 amplitude threshold is equal to the RMS scale value for the two-second sample of the recording's background noise, multiplied by a factor of 1 to 10 as set by the user. An amplitude threshold of 3 was used in the present study. To specify the duration threshold, the user sets a duration threshold in milliseconds. A duration threshold of 200 msec was used in the present study. Thus the programme identified as pauses all portions of the sound file in which the amplitude value never exceeded 3 times the RMS scale value of the background noise for an interval of at least 200 msec. Conversely, the programmed identified all other portions of the signal as utterances, with each utterance being bounded by a pause. The pauses and utterances as defined by the programme were later verified by a listening check to ensure that all segments that had been identified as utterances contained speech and not extraneous noises. Where the programme had incorrectly identified an extraneous noise (such as the sound of the speaker moving in his chair) as an utterance, the experimenter added the duration of the noise to the duration of adjacent pauses. The only non-speech sounds that were retained as utterances were coughing and laughing. This was done to reflect what a listener might have perceived in terms of pauses in the monologues. That is, a listener who heard a cough or a laugh was not likely to consider the sound as part of a pause, but rather as an interruption in a pause. Transcriptions of the utterances, utterance amplitudes and pause durations are presented in Appendix B. Before presenting a monologue to a subject, the experimenter opened and played the sound file containing the calibration tone that had been generated by the RMS programme for that specific monologue. The intensity of the calibration tone was used to adjust the level of the output of the audiometer. Each recording was presented to the listener at 70 dB 48 SPL (50 dB HL on the audiometer dial), the average level of conversational speech (Davis, 1947). 2.3.4 Characteristics of the Competing Noise White noise generated by a Madsen OB-802 audiometer was used as competing noise. To determine the sound pressure level of the broad band noise used in this study, the output of the noise, set at 65 and 75 dB SPL on the audiometer, was measured in the sound booth using a Quest Model OB-300 filter and Model 1800 sound level meter, coupled to Madsen TDH 39P 10W earphones using a 6-cc coupler. Two types of measures were taken. First, the overall SPL of the noise through the right and left earphones was measured using the wide-band filter setting with an A-weighting. The wide-band output of the noise with an intensity setting of 65 dB SPL and then of 75 dB SPL on the audiometer, were measured as 68.2 dB A and 74.8 dB A respectively. Second, the outputs of the right and left earphones were measured using the one-third octave band filters with centre frequencies ranging from 100 to 10,000 Hz on the automatic mode of the sound pressure level meter using a linear-weighting. In the automatic mode, the sound level meter averages the sound level for each frequency band sampled over a period of approximately 25 seconds. The average sound level for each frequency band is then recorded automatically by the sound level meter. The one-third octave band output of the noise was recorded manually; this information is presented in Table 2-1. 49 Table 2-1. Output of White Noise on Madsen OB-802 Audiometer, Measured in One-Third Octave Bands Output of White Noise (dB SPL) Centre Frequency (Hz) of 1/3 Octave Band Noise set at 65 dB SPL Noise set at 75 dB SPL 100 42.7 50.6 125 42.7 52.1 160 41.6 49.8 200 40.1 49.1 250 39.7 49.1 315 41.9 49.4 400 41.2 49.8 500 42.3 51.3 630 43.4 52.4 800 43.4 52.4 1000 44.6 53.2 1250 44.9 53.9 1600 44.9 54.7 2000 46.1 55.4 2500 48.7 58.4 3150 51.7 61.1 4000 50.6 58.4 5000 52.4 61.4 6300 53.6 63.3 8000 51.7 54.7 10000 49.4 52.1 2.4 Conditions of Presentation ^ Al l monologues were presented monaurally to the subject's better ear, defined as the ear with the lower average of the pure-tone thresholds obtained at 500, 1000 and 2000 Hz. The monologues were presented at 70 dB SPL (50 dB HL on the audiometer dial), and the 50 noise was presented at one of three levels: 65 dB SPL, 70 dB SPL or 75 dB SPL; i.e. +5, 0 or -5 dB S:N. These levels were chosen to represent easy to difficult listening conditions. In a pilot trial in which individuals listened to the monologues at various signal-to-noise ratios, listeners reported that they had no difficulty understanding the speech presented in a +5 dB S:N condition, a little difficulty in the 0 dB S:N condition, and that they could understand very little of the monologues when they were presented at a -5 dB S:N. Experimental sessions took place with the subject seated in a sound-attenuating, double-walled IAC booth and the experimenter outside the chamber, seated at the controls. The digital recordings of the talker's voice were played from the NeXT Sound Works 3.0 (Version 2) programme and relayed to the listener via a Madsen OB-802 audiometer. Subjects listened through Madsen TDH 39P 10W headphones. The initial task of a subject was to listen to the familiarization monologue described earlier. The purpose of this initial listening task was to familiarize the subject with the talker's voice. Following the familiarization task, the subject listened to one of the six recorded monologues presented in competing noise. Each subject listened to a total of three monologues, each one presented in a different signal-to-noise ratio condition, for a total of three experimental conditions per subject. The three conditions were -5, 0 and +5 dB S:N. Each subject heard three of the six recorded monologues such that for each group, each monologue was presented once in each condition (see Table 2-2). Therefore, differences due to the monologues should not be confounded with differences due to the signal-to-noise ratio. 51 Table 2-2. Presentation Order for Monologues 1 to 6 Signal-to-noise ratio +5dB OdB -5 dB Group I Sl 1 2 3 S2 3 6 5 S3 6 3 2 S4 2 1 4 S5 4 5 6 S6 5 4 1 Group II Sl 3 2 1 S2 4 1 2 S3 1 4 5 S4 6 5 4 S5 5 6 3 S6 2 3 6 2.5 Ordering of the Conditions The order of presentation of the three conditions was +5, 0 and -5 dB S:N for Group I, and -5, 0 and +5 dB S:N for Group II. The order of presentation was reversed for the second group to determine if there was any learning effect or fatigue effect which might contribute to a subject's increased accuracy, certainty, and/or speed in identifying topic changes. That is, if response latency decreased and the number of identifications of topic changes increased as a function of increased signal-to-noise ratio, then the order of 52 presentation of the signal-to-noise ratio conditions should not affect the results. If, however, subjects fatigued or adapted to the experimental task, then response latency and the number of topic changes identified would vary as a function of the order of presentation of conditions and not simply as a function of the signal-to-noise ratio that was employed. 2.6 Experimental Task Subjects were asked to press a response button placed on a table in front of them when they thought that the talker was about to switch to a new topic. To reduce inter-subject variability concerning the definition of topic, the term "topic" was pre-defined by the experimenter; each photograph being described by the talker was considered to constitute a topic, and thus subjects were asked to judge "when the talker was about to describe a new photograph." Subjects were encouraged to guess if they were uncertain about where this point in the monologue occurred. Subjects read the instructions from a series of 10 x 15 cm cards placed on a table in front of them (see Appendix C). When a subject pushed the response button, a broad-band FM signal was generated and recorded onto one track of an audiotape. The monologue to which the subject listened was recorded simultaneously onto the second track of the same audiotape (see Figure 2-1). 2.7 Method of Measuring Responses The audiotape recordings of the FM signals marking the subject responses and the simultaneously recorded monologues were later re-recorded digitally in stereo using the NeXT Sound Works programme. The subject responses and monologues were recorded onto separate tracks of the sound files; the time waveforms of the monologue and subject 53 responses could be viewed simultaneously (see Figure 2.2). Measurement of the latency of each response was obtained by locating the boundaries of topics, utterances, pauses and subject responses along the time waveform. Details of the criteria used to determine the location of these boundaries is discussed below. ^ Response But ton F M Signal Generator Sound R o o m O O Tape Recorder • • N e X T Computer Figure 2-1. Diagram of Experiment Set-Up 2.7.1 Definition of Topic Boundary The end of each photo description (topic) for each monologue was determined by visually locating the end of the last periodic sound. A periodic sound is a complex or 54 55 56 sinusoidal sound wave which repeats itself at regular and equal intervals in time (Durrant & Lovrinic, 1984). The last periodic sound was defined as the last wave in a series of at least four sound waves of similar frequency and amplitude. To ensure consistency across the measures obtained for different subjects, a print-out of the waveform of the end of each topic, including the final periodic wave of each photo description, was used as a template by which to locate the same point in subsequent measures involving the same description for other subjects (see Figure 2-3). An intra-tester test-retest reliability check of 10% of the data indicated that measurements of the last periodic wave in an utterance were the same within 0.9 msec 95% of the time. An inter-tester reliability check on 10% of the data indicated that the measurements of the last periodic wave in an utterance were consistent to 9 msec 95% of the time. 2.7.2 Defining a Subject Response A subject response was defined as the first zero-crossing of the broad-band noise generated when the subject hit the response button to indicate that he thought that a new topic was about to begin (see Figure 2-4). 2.7.3 Defining Categories of Subject Response Al l responses were categorized into mutually exclusive categories in terms of where they occurred in relation to the end of a topic. The two major categories were pre-terminal and post-terminal responses (see Figure 2-5). Pre-terminal responses were defined as those responses which occurred before a topic had ended (end of topic is defined in section 2.7.1). Post-terminal responses were defined as those responses which occurred after a topic had 57 ended. Pre-terminal responses were then further categorized into three sub-categories: mid-topic responses, topic-penultimate responses, and topic-final responses. Mid-topic responses were all responses occurring following the first 2000 msec of the topic (a, in Figure 2-5) and the beginning of the topic-penultimate utterance (b{ in Figure 2-5). Topic-penultimate responses were responses occurring between the beginning of the penultimate utterance (b] in Figure 2-5) and the beginning of the topic-final utterance (c, in Figure 2-5). Topic-final responses were all responses occurring between the beginning of the final utterance (cl in Figure 2-5) and the end of the topic (d, in Figure 2-5). The post-terminal responses were further divided into two sub-categories: responses in inter-topic pauses and post-topic initiation responses. Responses in inter-topic pauses were all responses occurring between the end of a topic (dj in Figure 2-5) and the beginning of the next topic (e[ in Figure 2-5). Post-topic initiation responses were all responses occurring between the beginning of the next topic (e! in Figure 2-5) and the end of the first 2000 msec (a2 in Figure 2-5) of the first utterance in the next topic. 2.8 Subject Observations At the end of the experiment, the experimenter interviewed each subject regarding what strategies they believed they used to identify topic changes when they could hear the talker but could not hear "what was being talked about". Subjects' comments were audiotaped and summarized. 59 60 3. RESULTS 3.1 Categorization of Responses: Best and Additional Responses Because for each topic, subjects could fail to respond or could respond more than once, it is not possible to differentiate "correct" from "incorrect" responses. Nevertheless, it seems that responses falling in the inter-topic pauses should be definitely considered as correct. Responses occurring in the post-initiation period also seem likely to be delayed but correct responses. Furthermore, responses occurring in the topic-penultimate utterance or topic-final utterance seem likely to be correct anticipations of an imminent topic change. Therefore, the only responses which are unlikely to be correct identifications of topic boundaries are those categorized as mid-topic responses. When a subject responded more than once for a given topic boundary, the response most likely to correspond to the correct identification of the topic boundary was taken as the subject's "best" response. Specifically, for the present purposes it was decided that the hierarchy of responses would be as follows, from "best" to "worst": 1. inter-topic responses 2. post-initiation responses 3. topic-final responses 4. topic-penultimate responses 5. mid-topic responses. 61 The justification for this hierarchy will be discussed below. All responses given in addition to the "best" response for each topic boundary were counted as "additional" responses. If a subject failed to respond for a given topic, then this was counted as an absent response. 3.2 Accuracy of Topic Boundary Identification Accuracy was evaluated by counting the number of "best" responses occurring in each of the above-defined pre- and post-terminal sub-categories respectively: mid-topic responses, topic-penultimate responses, topic-final responses, inter-topic responses, and post-initiation responses. An increase in the number of "best" responses in the mid-topic category of response would be the strongest indicator of a change for the worse in subject accuracy of topic boundary identification. The number of "best" responses in each sub-category is presented in Table 3-1. Results for individual subjects are presented in Appendix D. 3.2.1.1 Pre-Terminal Responses Pre-terminal responses were those responses occurring after the first 2000 msec of a topic and prior to the end of the same topic. The pre-terminal responses were divided into three sub-categories: mid-topic responses, topic-penultimate responses, and topic-final responses. The effect of signal-to-noise ratio on the number of responses in each of these sub-categories is reported below. 62 3.2.1.1.1 Mid-topic Responses The number of responses occurring after the first 2000 msec of a topic and before the penultimate utterance in a topic was small and varied only slightly as the signal-to-noise ratio decreased. The number of mid-topic responses in the +5, 0 and - 5 dB S:N conditions was 1, 6 and 5 respectively. An analysis of variance was conducted with order of presentation (increasing versus decreasing change in the signal-to-noise condition) as a between-subjects factor with Table 3-1. Distribution of "Best" Responses Signal-to-noise ratio +5 dB 0 dB -5 dB Pre-terminal responses Mid-topic responses Topic-penultimate responses Topic-final responses Post-terminal responses Inter-topic responses Post-initiation responses Total "best" responses Absent responses 1 6 5 4 7 6 4 3 3 97 84 75 12 16 28 118 116 117 2 4 3 63 two levels, and signal-to-noise condition (+5, 0 and -5 dB S:N) as a within-subjects factor with three levels. The analysis of variance confirmed that there was no significant effect of signal-to-noise ratio on the number of mid-topic responses (F(2,20) = 2.966, p > .05). No other effects were significant. 3.2.1.1.2 Topic-Penultimate Responses The number of responses occurring within the penultimate utterance in a topic and in the pause following this utterance varied only slightly as the signal-to-noise ratio decreased. The number of topic-penultimate responses in the +5, 0 and -5 dB S:N conditions was 4, 7 and 6 respectively. An analysis of variance was conducted with order of presentation (increasing versus decreasing change in the signal-to-noise condition) as a between-subjects factor with two levels, and signal-to-noise condition (+5, 0 and -5 dB S:N) as a within-subjects factor with three levels. The analysis of variance confirmed that there was no significant effect of signal-to-noise ratio on the number of topic-penultimate responses (F(2,20) = .273, p > .05). No other effects were significant. 3.2.1.1.3 Topic-Final Responses The number of responses occurring in the last utterance of a topic was small and varied only slightly as the signal-to-noise ratio decreased. The number of topic-final responses in the +5, 0 and -5 dB S:N conditions was 4, 3, and 3 respectively. An analysis of variance was conducted with order of presentation (increasing versus decreasing change in the signal-to-noise condition) as a between-subjects factor with two levels, and signal-to-noise condition (+5, 0 and -5 dB S:N) as a within-subjects factor with three levels. The 64 analysis of variance confirmed that there was no significant effect of signal-to-noise ratio on the number of topic-final responses (F(2,20) = .033, p > .05). No other effects were significant. 3.2.1.2 Topic Post-Terminal Responses Post-terminal responses are those responses occurring after the end-point of a topic; that is, in the inter-topic pauses or in the first 2000 msec following the initiation of the subsequent topic. Post-terminal responses were divided into two types of responses: inter-topic pause responses and post-initiation responses. The effect of signal-to-noise ratio on the number of responses in each of these sub-categories is reported below. 3.2.1.2.1 Inter-Topic Pause Responses The number of responses occurring in the pauses between topics was large and decreased slightly as the signal-to-noise ratio decreased. A total of 97 inter-topic pause responses was given in the +5 dB S:N condition; the number decreased to 84 in the 0 dB S:N condition, and to 75 in the -5 dB S:N condition. An analysis of variance was conducted with order of presentation (increasing versus decreasing change in the signal-to-noise condition) as a between-subjects factor with two levels, and signal-to-noise condition (+5, 0 and -5 dB S:N) as a within-subjects factor with three levels. The analysis of variance confirmed that there was no significant effect of signal-to-noise ratio on the number of inter-topic responses (F(2,20) = 3.46, p_ > .05). No other effects were significant. 65 3.2.1.2.2 Post-Initiation Responses The number of responses occurring in the first 2000 msec after the beginning of the new topic increased as the signal-to-noise ratio decreased. A total of 12 post-initiation responses was given in the +5 dB S:N condition; the number increased to 16 in the 0 dB S:N condition, and to 28 in the -5 dB S:N condition. An analysis of variance was conducted with order of presentation (increasing versus decreasing change in the signal-to-noise condition) as a between-subjects factor with two levels, and signal-to-noise condition (+5, 0 and -5 dB S:N) as a within-subjects factor with three levels. The analysis of variance confirmed that there was a significant effect of signal-to-noise ratio on the number of post-initiation responses (F(2,20) = 4.485, p < .05). No other effects were significant. A Student-Newman-Keuls test of multiple comparisons showed that a significantly (p < .05) higher number (M = 2.3) of post-initiation responses was given in the -5 dB S:N condition than in either the 0 dB (M = 1.3) or +5 dB S:N conditions (M=l-0). There was no significant difference between the number of post-initiation responses observed in the 0 and the +5 dB S:N conditions. 3.2.1.2.3 Absent Responses The number of topics for which subjects gave no response was small and varied only slightly across the three experimental conditions. The number of absent responses in the +5 dB, 0 dB and -5 dB S:N conditions was 2, 4 and 3 respectively. An analysis of variance was conducted with order of presentation (increasing versus decreasing change in the signal-66 to-noise condition) as a between-subjects factor with two levels, and signal-to-noise condition (+5, 0 and -5 dB S:N) as a within-subjects factor with three levels. The analysis of variance confirmed that there was no significant effect of signal-to-noise ratio on the number of absent responses (F(2,20) = 1.385, p_ > .05). No other effects were significant. 3.2.2 Summary of Finding About Accuracy of Topic Boundary Identification The above results reveal that only the number of "best" responses occurring in the post-initiation zone changed significantly as a function of signal-to-noise ratio conditions. Subjects gave significantly more post-initiation responses in the -5 dB S:N condition than either the 0 or +5 dB S:N condition. These results suggest that subjects were only slightly less accurate at identifying topic boundaries as the signal-to-noise ratio decreased, according to the hierarchy of "best" to "worst" response types outlined above. 3.3 Certainty of Topic Boundary Identification Three types of analyses were performed to determine the certainty with which subjects identified topic boundaries. First, the total number of "additional" responses observed in each of the three signal-to-noise conditions was analyzed; second, the number of "additional" responses in each sub-category was analyzed; and third, the number of topic boundaries for which subjects gave multiple responses was analyzed. The less certain a subject was of where a topic boundary occurred, the more likely he would be to respond frequently. Thus an extremely uncertain subject would give a high number of "additional" responses for many topic boundaries. 67 3.3.1 Total Number of Additional Responses The total number of "additional" topic boundary identifications increased as a function of signal-to-noise ratio. The total number of "additional" responses was 44 in the +5 dB S:N condition; this number decreased to 41 in the 0 dB S:N condition, and then increased to 75 in the -5 dB S:N condition. The results are shown in Table 3-2. Results for individual subjects are presented in Appendix D. An analysis of variance was conducted with order of presentation (increasing versus decreasing change in the signal-to-noise condition) as a between-subjects factor with two levels, and signal-to-noise condition (+5, 0 and -5 dB S:N) as a within-subjects factor with three levels. The analysis of variance Table 3-2. Distribution of "Additional" Responses Signal-to-noise ratio +5 dB OdB -5 dB Pre-terminal responses Mid-topic responses 37 Topic-penultimate responses 5 Topic-final responses 2 32 7 2 55 15 5 Post-terminal responses Inter-topic responses Post-initiation responses 0 0 0 0 0 0 Total "additional" responses 44 41 75 68 confirmed that there was a significant effect of signal-to-noise ratio on the number of "additional" topic boundary identifications (F(2,20) = 4.31, p_ < .05). A Student-Newman-Keuls test of multiple comparisons showed that the number of "additional" topic boundary identifications was significantly greater in the -5 dB S:N condition (M=6.3) than in either the 0 dB (M=3.4) or +5 dB (M=3.7) conditions (p < .05) with no significant difference being observed for the latter two conditions. Importantly, no "additional" responses were given in the inter-topic pause, nor were any given in the first 2000 msec of the beginning of a topic. Therefore, all "additional" responses were pre-terminal responses, i.e. mid-topic, topic-penultimate or topic-final responses. 3.3.1.1 Mid-topic Responses The number of "additional" identifications of topic boundaries occurring after the first 2000, msec of a topic and before the penultimate utterance in a topic increased slightly as the signal-to-noise ratio decreased. The number of "additional" mid-topic responses was 37 in the +5 dB S:N condition; this number decreased to 32 in the 0 dB S:N condition, and then increased to 55 in the -5 dB S:N condition. An analysis of variance was conducted with order of presentation (increasing versus decreasing change in the signal-to-noise condition) as a between-subjects factor with two levels, and signal-to-noise condition (+5, 0 and -5 dB S:N) as a within-subjects factor with three levels. The analysis of variance did not confirm that there was a significant effect of signal-to-noise ratio on the number of "additional" mid-topic responses (F(2,20) = 2.657, p > .05). No other effects were significant. 69 3.3.1.2 Topic-Penultimate Responses The number of "additional" identifications of topic boundaries occurring within the penultimate utterance of a topic increased as the signal-to-noise ratio decreased. The total number of "additional" topic-penultimate responses was 5 in the +5 dB S:N condition; this number increased to 7 in the 0 dB S:N condition, and increased further to 15 in the 0 dB S:N condition. An analysis of variance was conducted with order of presentation (increasing versus decreasing change in the signal-to-noise condition) as a between-subjects factor with two levels, and signal-to-noise condition (+5, 0 and -5 dB S:N) as a within-subjects factor with three levels. The analysis of variance confirmed that there was a significant effect of signal-to-noise ratio on the number of "additional" penultimate responses (F(2,20) = 3.50, p < 0.5). No other effects were significant. A Student-Newman-Keuls test of multiple comparisons showed that the number of "additional" topic-penultimate responses was significantly (p < 0.1) greater in the -5 dB S:N condition (M^l.3) than in the +5 dB S:N condition (M=0.4). The number of "additional" penultimate responses in the 0 dB S:N condition (M=0.6) was not significantly greater than the number in the +5 dB condition, nor was the number significantly greater in the -5 dB S:N condition than in the 0 dB S:N condition (p_ > 0.1). 3.3.1.3 Topic-Final Responses The number of "additional" identifications of topic boundaries occurring in the final utterance of a topic increased slightly as the signal-to-noise ratio decreased. The number of "additional" topic-final responses in the +5 dB, 0 dB and - 5 dB S:N conditions was 2, 2 and 70 5 respectively. An analysis of variance was conducted with order of presentation (increasing versus decreasing change in the signal-to-noise condition) as a between-subjects factor with two levels, and signal-to-noise condition (+5, 0 and -5 dB S:N) as a within-subjects factor with three levels. The analysis of variance did not confirm that there was a significant effect of signal-to-noise ratio on the number of topic-final responses (F(2,20) = .091, p > .05). No other effects were significant. 3.3.2 Topics Containing Multiple Responses The total number of topic boundaries for which at least one subject gave more than one response was 29 (24% of all topic boundaries presented) in the +5 dB S:N condition; this number increased negligibly to 30 (25%) in the 0 dB S:N condition, and then more than doubled to 52 (43%) in the -5 dB S:N condition. Results are shown in Figure 3-1. Results for individual subjects are presented in Appendix D . An analysis of variance was conducted with order of presentation (increasing versus decreasing change in the signal-to-noise condition) as a between-subjects factor with two levels, and signal-to-noise condition (+5, 0 and -5 dB S:N) as a within-subjects factor with three levels. The analysis of variance confirmed that there was a significant effect of signal-to-noise ratio on the number of topics containing multiple responses (F(2,20) = 5.13, p_ < .05). A Student-Newman-Keuls test of multiple comparisons confirmed that the number of topics triggering multiple responses was significantly greater in the -5 dB S:N condition (M = 4.3) than in either the 0 dB (M=2.7) or +5 dB (M=2.4) conditions (p < .05) with there being no significant difference observed between the number of topics triggering multiple responses in the latter two conditions. 71 T3 O O S - I on o 00 -a in + o o in o o o 72 3.3.3 Summary of Findings About Certainty of Topic Boundary Identification The above results indicate that subjects gave significantly more "additional" responses in the -5 dB S:N condition than in either the 0 or -5 dB S:N conditions. More specifically, they gave significantly more "additional" topic-penultimate responses in the -5 dB S:N condition than in the +5 dB S:N condition. Finally, an analysis of the total number of topic boundaries for which at least one subject gave more than one response indicates that the number of topics triggering multiple responses was significantly greater in the -5 dB S:N condition than in either the 0 or -5 dB S:N condition. These results suggest that subjects were least certain of where a topic boundary occurred in the -5 dB S:N condition. 3.3.4 Latency of Topic Boundary Identification For each topic boundary for each subject in each condition, one latency measure was obtained unless the subject failed to respond. In the event that a subject responded more than once for a given topic boundary, the latency of the subject's "best" response was measured and included in subsequent analyses. Therefore, the median latency for a maximum of ten topic boundaries for each subject in each condition was determined and the median value was employed in the subsequent analyses. The mean of the median latencies increased as the signal-to-noise ratio decreased. Specifically, the mean of the median latencies was 1538 msec in the +5 S:N dB condition; it increased to 1877 msec in the 0 dB S:N condition and it increased further to 2172 msec in the -5 dB S:N condition. The results are shown in Figure 3-2 and Table 3-3. Results for 73 individual subjects are presented in Appendix D. An analysis of variance was conducted with order of presentation (increasing versus decreasing change in signal-to-noise condition) as a between-subjects factor with two levels, and signal-to-noise condition (+5, 0 and -5 dB S:N) as a within-subjects factor with three levels. The analysis of variance confirmed that there was a significant effect of signal-to-noise ratio on the mean of the median latencies of subjects' responses (F(2,20) = 2.99, p < .01). No other effects were significant. A Student-Newman-Keuls test of multiple comparisons showed that the mean of the median latencies was significantly greater in the -5 dB S:N condition than in either the +5 dB condition (p < .01) or the 0 dB S:N condition (p < .05). The mean of the median latencies was not significantly greater in the 0 dB S:N condition than it was in the +5 dB S:N condition (p >.05). Table 3-3. Mean of the Median Latency of Subjects' Responses (in msec) Signal-to-noise ratio +5 dB OdB -5 dB Median Response Latency M SD 1625 491 1964 642 2194 754 74 _CU 'o fl cu •s k - J cu cn fl O DH OO CU c 2 T3 cu cu o c c« CU C N m cu bp I T3 m O O '"ci CU ' S fl 03 fl 'ex! CQ xt m + o o o cn o o m C N o o o C N o o i n o o o o o m (ossui) Axnrarer ireipaTAT 75 3.4 Summary of Findings About the Latency of Topic Boundary Identification The above results indicate that subjects were significantly slower to identify topic boundaries in the -5 dB S:N condition than in either the 0 or the -5 dB S:N conditions. 3.5 Measurement of Prosodic Characteristics Recall that prosodic characteristics possibly related to the identification of topic boundaries include: vowel length, relative amplitude of topic-initial and topic-final utterances, variations in fundamental frequency, duration of inter-topic pauses, rate of speaking and degree of laryngealization of the voice. Of these characteristics, only pause duration and utterance amplitude were examined to determine whether or not they were related to changes in the accuracy and latency of topic boundary identification in the present study. While analysis of all the prosodic characteristics of the monologues used in the present study would have been ideal, such detailed analysis was beyond the scope of this study. 3.5.1 Distribution of Pauses in the Monologues The pauses between topics were longer than the within-topic pause durations. The mean duration was 3028 msec for inter-topic pauses and 703 msec for within-topic pauses (see Table 3-4, Figure 3-3, Figure 3-4 and Figure 3-5). The mean duration of inter-topic pauses was significantly higher than the mean duration of within-topic pauses. This description was confirmed by a t-test for independent samples (1(713) = 32.513, p < .001). 76 Table 3-4. Topic Boundary and Non-Topic Boundary Pause Durations (in msec) M SD Range Topic Boundary Pauses (n=60) 3028 1128 272 to 5139 Within-Topic Pauses (n=655) 703 437 201 to 3086 3.5.2 Distribution of Utterance Amplitudes in the Monologues Amplitude of topic-initial utterances was generally greater than that of topic-final utterances. The median RMS scale value was 1537 for topic-initial utterances and 995 for topic-final utterances. Because the mean and median values for topic-initial and topic-final utterances were similar, the mean values were used in the present analyses. The mean RMS scale value for topic-initial utterances (M = 1627) was significantly higher than the mean RMS scale value for the topic-final utterances (M = 1151). This description was confirmed by a t-test for independent samples (t(130) = 5.12, p < .001). While the mean RMS scale values for topic-initial utterances are generally greater than those for topic-final utterances, note however that the maximum RMS scale value for topic-final utterances is 5731, and as such is greater than the maximum RMS scale value for topic-initial utterances. There are, then, some exceptional cases in which the final utterance of a topic may be greater than the topic's initial utterance. Examination of topic boundary 77 cn li-en P H P H O H - > cn CD cn P H O ' P H o cu (oasra) U O U R I T L Q 78 o m i n o m C N i n o i n o i n C N cn cu cn u o 0 f—1 1 A • i-H -fl o fl o ' f i cn m cu j—< fl E o >n t - -m o i n C N m o i n r - -C N o m C N o i n o i n C N ssstved j o J3quinj\[ o o i n i n o o i n o o o CD PL, O o H CD O c o m m CD o o o o j j l p l VO C N ssstred jo J3qum^[ 80 o o CN o o o CN o o o o o o o snreA areas SWS. 81 Table 3-5. RMS Scale Values for Utterances in Topic-Initial, Topic-Final and Other Topic Positions M SD Range Topic-initial Utterances (n=66) 1627 424 893 to 2908 Topic-final Utterances (n=66) 1151 737 253 to 4233 Other Utterances (n=491) 1332 682 325 to 6649 utterances for each of the topics revealed that eight (12%) had RMS scale values for the topic-final utterance that exceeded the RMS scale value of the corresponding topic-initial utterance. The difference between topic-initial and topic-final utterance RMS scale values for these eight topics is shown in Figure 3-7. In seven of the eight topics, the RMS scale value for the topic-final utterance also exceeded the RMS scale value of the second utterance in the corresponding topic. 3.6 Subjects' Observations Ten of the twelve subjects in this study were asked how they identified topic boundaries when they could not understand "what was being talked about". Eight of the subjects interviewed observed that when they experienced difficulty understanding what the talker in the monologues was saying, they listened for the talker to pause so that they could better identify when the topic was about to change. Three of these subjects reported that using pauses to identify topic changes was sometimes misleading if the talkerpaused and 82 then continued talking about the photograph. Five subjects noted that the talker would begin a topic in a "strong" or "loud" voice, and that his voice would "fade" as he neared the end of the photograph description. 83 o o o o o o o o i n o m o •^ f "d* cn cn o o o o o o o o i n o i n o C N C N r—1 1—1 anreA areos siATH 84 4. DISCUSSION 4.1 Review of hypotheses The present study examined how the accuracy, certainty and latency of normal-hearing listeners' identification of topic changes in discourse is affected by the presence of competing noise. The following null hypotheses were tested: As the signal-to-noise ratio decreases and the listening condition becomes less favourable, 1) the accuracy with which normal-hearing listeners identify topic boundaries will not differ significantly from the accuracy with which they identify topic boundaries in a more favourable listening condition; 2) the certainty with which normal-hearing listeners identify topic boundaries will not differ significantly from the certainty with which they identify topic boundaries in a more favourable listening condition; and 3) the speed with which normal-hearing listeners identify topic boundaries will not differ significantly from the speed with which they identify topic boundaries in a more favourable listening condition. 4.2 Summary of Results The present study demonstrates that even in a favourable listening condition, normal-hearing listeners are not perfect at identifying topic shifts. As the listening condition worsens, normal-hearing listeners are not less accurate in their ability to identify topic 85 boundaries; however, they do judge topic shifts to occur more frequently and they take longer to identify topic boundaries. Under favourable listening conditions (+5 and 0 dB S:N), topic-terminal cues are generally sufficient to allow listeners to identify topic boundaries; however, when the listening condition is unfavourable (-5 dB S:N), normal-hearing listeners rely more heavily on topic-initiation cues than on topic-final cues to identify topic changes. The cues that seem to be the most useful for listeners in adverse listening conditions are inter-topic pause duration and amplitude changes at topic initiation. In terms of the prosodic characteristics of the materials used in the present study, pause durations are generally greater at topic boundaries than within topics, and utterance amplitudes are generally greater at the beginning of a topic than at the end of a topic. 4.3 Accuracy of Topic Boundary Identification It was hypothesized that listeners would be less accurate when identifying topic boundaries under unfavourable listening conditions than under favourable listening conditions. Rather than categorizing responses as "correct" or "incorrect", when there was more than one response for a topic, responses were ranked according to response types which might represent a subject's "best" identification of topic boundaries. It was assumed that the most accurate responses were those occurring after a topic had ended, either in the pause between topics (inter-topic responses) or within the first two seconds of a new topic (post-initiation responses). Responses occurring within the last two sentences of the topic (topic-final and topic-penultimate responses) might represent less accurate identifications of topic boundaries. Alternatively, these responses might represent a subject's correct 86 anticipation of a topic boundary. Swerts et al. (in press) have demonstrated that listeners can estimate how far a given utterance is from the end of a topic, though the scope of this prediction appears to be limited to the last two utterances of a topic. Mid-topic responses were the only response types which were considered to truly represent inaccurate topic boundary identifications. Subjects were not less accurate at identifying topic boundaries as the signal-to-noise ratio decreased. Of the five response types, only the number of post-initiation responses increased significantly as the signal-to-noise ratio decreased. If post-initiation responses represent correct but delayed topic boundary identifications, then this result suggests that listeners are not necessarily less accurate at identifying topic boundaries in unfavourable listening conditions but that they are merely slower. 4.4 Certainty of Topic Boundary Identification It was hypothesized that listeners would be less certain when identifying topic boundaries under unfavourable listening condition than when doing so under favourable listening conditions. The certainty of topic boundary identification was gauged by the number of times that subjects identified topic boundaries more than once for a single topic. Subjects gave "additional" responses more frequently and for more topics in the least favourable listening condition (-5 dB S:N) than in either of the more favourable listening conditions (0 and +5 dB S:N), indicating that they were less certain of where a topic boundary occurred as the listening condition worsened. In particular, the number of "additional" responses occurring within the penultimate utterance of a topic and the pause 87 following this utterance increased significantly as the signal-to-noise ratio decreased. It is not clear why the number of "additional" responses in the topic-penultimate interval increased in the least favourable listening condition. One possible explanation is that in the -5 dB S:N condition, subjects were able to perceive enough cues in or before the topic-penultimate interval to roughly identify that a topic was about to end shortly, and leading them to respond during the topic-penultimate interval. Note that following this "early" response, subjects gave another response in the topic-final, intef-topic or post-initiation interval; this later response was considered to be the "best" response. In the least favourable listening condition, then, subjects may have been early to identify topic boundaries, but were able to recognize that they had responded prematurely as the topic continued or a new topic started. This may indicate that although listeners can distinguish topic-penultimate from topic-final utterances under favourable listening conditions (Swerts et al., in press), they may be less able to do so under unfavourable listening conditions. One surprising result from this analysis is that subjects were not perfect at identifying topic boundaries even in the most favourable listening condition (+5 dB S:N). Even in the most favourable signal-to-noise condition, one subject failed to identify one topic boundary in monologue 1, and another subject failed to identify one topic boundary in monologue 5. Furthermore, subjects gave a considerable number of "additional" responses in the +5 dB S:N condition (M=3.7). It would appear that even under favourable listening conditions, for which subjects reported little difficulty understanding the monologues, listeners were not perfect at identifying topic boundaries. Subjects may have missed topic boundaries entirely because they experienced lapses of attention. In the cases where subjects 88 gave "additional" responses, subjects may have responded when they identified a sub-topic structure within a topic. As discussed in the first chapter, there can be several sub-topics within a topic (photograph description). It is probable that within some of the photograph descriptions there are two or more distinct sub-topics. Each of these sub-topics is likely to possess some of the prosodic characteristics of a full topic change such as lowering of voice pitch and intensity, pre-boundary syllable lengthening, laryngealized voice quality, decreased rate of speech, and increased pause length. Though the monologues were not analyzed for sub-topic boundaries, it is possible that subjects may have identified "additional" topic boundaries at sub-topic boundaries. This possibility could be further investigated in future research. 4.5 Latency of Topic Boundary Identification It was hypothesized that as the signal-to-noise ratio decreased and the listening condition became less favourable, subjects would require more processing time to identify topic changes. Analysis of subject response latencies confirms that subjects took significantly longer to identify topic changes in discourse under unfavourable listening condition (-5 dB S:N) than under favourable listening conditions (0 dB and +5 dB S:N). The analysis of subject accuracy in identification of topic boundaries discussed above revealed that subjects gave more post-initiation responses in the least favourable listening condition (-5 dB S:N) than in either of the more favourable listening conditions (0 and +5 dB S:N). Recall that a post-initiation response is an identification of topic shift which occurs after one topic has ended and the next topic has begun. Therefore, it appears 89 that in the most adverse listening condition, responses not only became slower but they became so slow that many occurred in the post-initiation interval rather than the inter-topic interval. This finding suggests that normal-hearing listeners rely more on topic-initiation cues to identify topic boundaries as the listening condition becomes less favourable. This finding is consistent with the finding of increased latency reported above; as the signal-to-noise ratio decreases, listeners were evidently less able to anticipate that a topic was ending and needed to wait until the next topic began before they were able to identify a topic boundary. 4.6 Prosodic Cues to Topic Boundary Measurements of pause durations and the relative amplitudes of topic-initial and. topic-final utterances were completed to explore how prosodic characteristics might relate to the ability of listeners to identify topic boundaries, and to determine if the materials used in the present study had prosodic features like those reported by other researchers to characteristic of topic boundaries. 4.6.1 Pauses and Additional Responses The analysis of the relative durations of inter- and within-topic pauses in the materials used in this study demonstrate that pauses were typically longer when they were between two topics than when they occurred within a topic. Lehiste & Wang (1977) have demonstrated that listeners use pause length to identify paragraph boundaries. In the present study, pause length may have been an important cue to identifying topic boundaries, particularly in the least favourable listening condition (-5 dB S:N) where the identification, 90 discrimination or even the detection of segmental cues and of other prosodic cues would be difficult because of the presence of competing noise. This explanation was supported by subjects' observations that they used pause duration to identify topic boundaries when they could not understand "what was being talked about". One reason why pause duration could be an especially strong cue, or mis-cue, to topic boundaries under adverse listening conditions is that for listeners to perceive pauses it is necessary for them to detect wheteher or not a signal is present; they do not need to be able to identify or discriminate segments of the speech signal. Detection might occur even if the speech signal is very degraded, whereas identification or discrimination of speech segments would require a more intact speech signal. If subjects used pause length to decide that the topic changed, then we might expect that the within-topic pauses in which "additional" responses occurred would be of greater duration than the duration of within-topic pauses in general. A post hoc analysis of within-topic pause durations revealed that the mean duration of within-topic pauses in which "additional" responses were given was 1285 msec (n = 160). This value is considerably greater than the mean duration (703 msec) of all the within-topic pauses combined (n = 655). Specifically, a pause duration of 1285 msec falls in the 90th percentile of the distribution of all within-topic pause durations. This result suggests that pause duration may indeed have led subjects to make additional identifications of topic boundaries when they encountered some of the longest within-topic pause durations. To determine if subjects were miscued by pause duration more often in the least favourable condition (-5 dB S:N), the mean duration of within-topic pauses in which "additional" pauses were given was calculated for each 91 condition. The mean durations were 1339 ms, 1406 msec and 1285 msec for the +5 dB, 0 dB and -5 dB S:N conditions respectively. An analysis of variance was conducted with order of presentation (increasing versus decreasing change in signal-to-noise condition) as a between-subjects factor with two levels, and signal-to-noise condition (+5, 0 and -5 dB S:N) as a within-subjects factor with three levels. The analysis of variance confirmed that there was no effect of signal-to-noise ratio on the mean duration of within-topic pauses containing "additional" responses (E(2,75) = 0.46, p > .05). This suggests that pause length may have been a fairly strong cue, or miscue, to topic boundaries in all three conditions. This finding is also consistent with the idea that the detection of pauses is relatively immune to noise. Interestingly, this finding does not appear to support the observation made by several subjects that they had used pause duration to identify topic boundaries more often in the least favourable listening condition (-5 dB S:N). It may be that in the most difficult listening condition, the very low amplitude portions of the speech signal were effectively masked by the noise so that subjects could not detect the presence of portions of the speech signal. If so, then subjects may have perceived low amplitude speech segments as an extension of the inter-topic pause. This may have contributed to their impression that they identified topic boundaries by pause length more often in the noisiest condition. 4.6.2 Pause Durations: Present and Previous Findings It was expected that inter-topic pauses in the materials used in this study would be similar in duration to the inter-topic pauses described by Brown et al. (1980) and Lehiste (1980), and that the pauses occurring at topic boundaries would be amongst the longest 92 pauses in the monologues. Brown et al. (1980) reported that inter-topic pause durations in the conversations that they sampled ranged from 600 to 1800 msec. Lehiste (1980) reported that inter-paragraph pause durations within a conversational turn averaged 1659 ms, that those between turns averaged 5045 ms, and that the mean duration of within-topic pauses was 737 msec. An analysis of pause durations in the present materials revealed that the mean duration of inter-topic pauses is 3086 ms, and that the mean duration of within-topic pauses is 703 msec. The mean duration of within-topic pauses in the present materials is within 50 msec of the mean within-topic pause duration reported by Lehiste (1980). The mean duration of inter-topic pauses in the present study is almost twice as long as the maximum duration of inter-topic pauses reported by Brown et al. (1980), and it is mid-way between the inter-paragraph and the inter-turn pause durations reported by Lehiste (1980). The duration of inter-topic pauses in the monologues employed in the present study seems most like the duration of inter-turn pauses in conversational dyads. Of course, in a monologue situation pauses serve only to separate topic or sub-topics; they do not function as turn-taking cues. Although the durations of inter-topic pauses in the present monologues are greater than those observed by others for conversational materials, the inter-topic pauses in the present monologues are nevertheless typically longer than within-topic pauses. The fact that the duration of the inter-topic pauses in the present monologues are more like the between-turn pauses than the inter-topic pause durations observed in conversations suggests that inter-topic pauses should have been a strong cue to topic boundaries in the present study. 93 4.6.3 Variations in Utterance Amplitude and Changes in Response Latency and Certainty Topic-initial utterances were generally of higher amplitude than topic-final utterances in all six of the monologues used in the present study. It follows, then, that topic-initial utterances should be more easily perceived than topic-final utterances as the signal-to-noise ratio decreases. This could account for the increase in the number of post-initiation responses given as the signal-to-noise ratio decreased. Under favourable listening conditions, listeners are able to identify topic-penultimate and topic-final utterances based on their prosodic characteristics (Swerts et al., in press). Under unfavourable listening conditions, it appears that higher amplitude topic-initiation cues are more important for cueing listeners to topic boundaries than are lower amplitude and therefore less audible topic-final cues. Therefore, subjects may have relied more on topic-initial cues than on topic-final cues to identify topic boundaries in the -5 dB S:N condition because topic-initial utterances were higher in amplitude and therefore more easily perceived than topic-final and topic-penultimate utterances. 4.6.4 Utterance Amplitudes: Present and Previous Findings Analysis of the amplitudes of topic-boundary utterances revealed that the patterns of variation in utterance amplitudes in the materials used in the present study were consistent with the patterns reported in previous studies (Lehiste, 1980; Brown et al., 1980). There were exceptions to this pattern, however; for approximately 12% of the photograph descriptions, the amplitude at the end of the topic exceeded the amplitude at the beginning of the topic. One possible explanation for these exceptions is that because the RMS scale 94 values represent only the mean amplitude of an utterance over its duration, the RMS scale values do not reflect variations in amplitude within an utterance. Therefore, although a topic-final utterance may have ended with relatively low amplitude, this might not be accurately represented by the RMS scale value of the utterance if the onset of the utterance is high in amplitude. Conversely, the RMS scale value for a topic-initial utterance would under-estimate the peak value for the utterance.- Nonetheless, the majority of topics did conform to the expected pattern of amplitude changes. 4.6.5 Additional Cues to Topic Boundary The cues to topic boundaries in discourse that we have examined in this study have been limited to the prosodic cues of pause duration and utterance amplitude. These analyses demonstrate that these features might contribute to the correct and possibly incorrect identification of topic boundaries in discourse, and that the materials used in this study generally conform to the prosodic descriptions of topic changes in discourse that have been reported by other researchers. This said, it is difficult to wholly appreciate how utterance amplitude and pause duration might account for changes in accuracy, certainty and latency of topic boundary identification without examining other aspects of the monologues such as the range and contour of fundamental frequency excursion, both within and across utterances; pre-boundary lengthening; laryngealization; rate of speaking; amplitude contour; and the content of the utterances themselves. Cues other than pause duration and utterance amplitude have been shown to be instrumental to the identification of topic boundaries (e.g., Swerts & Geluykens, 1993; Swerts et al., 1994; Lehiste, 1980). Furthermore, these cues 95 appear to operate in a somewhat complex manner. Swerts, Bouwhuis and Collier (1994) have shown that voice pitch, and range of pitch excursion as well as pitch contour are cues which not only independently influence a listener's perception of topic boundaries but that these cues also combine additively in producing cues to topic finality. Conversely, the presence of one topic boundary cue could compensate for the lack of another topic boundary cue (Lehiste, 1980). For example, a topic-final utterance possessing sufficient laryngealization and pre-boundary lengthening may cue a listener to identify a topic boundary, even though the pause following the utterance is atypically short for an inter-topic pause. The present analyses of pause durations and utterance amplitudes may not, therefore, fully account for the patterns of responses reported in this study; other prosodic features may have contributed to listener perception of topic boundaries. 4.7 Implications of Findings for Comprehension The motivation for the present study was to determine how normal-hearing individuals identify topic changes in favourable and unfavourable listening conditions, and ultimately to better understand how and where breakdowns in the identification of topic boundaries and consequent difficulties in comprehension might occur for hard-of-hearing listeners. It is impossible to know exactly how normal-hearing listeners' performance in the present experimental task might be similar to the ability of hard-of-hearing individuals to identify topic boundaries in everyday listening situations. It is possible, however, to speculate about how the changes in a listener's ability to identify topic boundaries might { 96 affect comprehension not at the perceptual level of processing, but at higher processing levels for normal- and hard-of-hearing listeners alike. Knowledge of topic aids language comprehension. Listeners identify topics by identifying the content as well as the structure of discourse. Surely, then, if a listener is less certain of where a topic boundary occurs and is slower to identify when a topic changes, then this will have an impact on his ability to use knowledge of topic to aid comprehension. First, it is possible that decreased certainty as well as increased latency might indicate that listeners are having to assign more processing resources to the task of judging where topic boundaries are located. Any increase in processing effort can be expected to adversely affect the efficiency of language comprehension secondary to increased demands on working memory (Daneman and Carpenter, 1980). Second, if a listener must wait for topic-initiation cues to identify a topic boundary, rather than using topic-terminal cues, then the language processor will delay storage and integration of information related to the most recently completed topic in working memory and therefore have less time in which to process new information as the next topic begins. According to Kintsch and van Dijk's (1978) model of discourse comprehension, the language processor generates and stores in short-term memory a proposition for each sentence or phrase. These propositions are processed in chunks of several propositions at a time. When one chunk of propositions has been processed and stored, the new incoming chunk of propositions is compared to the chunk stored in short-term memory. If speakers signal an appropriate chunk size by prosodic and lexical cues, as suggested by Kintsch and van Dijk, then a listener's delay in identification of these cues could result in less efficient processing of propositions. The language processor would have 97 less time to prepare for a new chunk of propositions. If so, then this delay would certainly reduce the time available to process new propositions, straining further the processing system which is attempting to process degraded speech information. 4.8 Future Directions This investigation has examined how normal-hearing individuals identify topic boundaries in favourable and unfavourable listening conditions. As stated earlier, the findings presented here could ultimately be used to guide a subsequent investigation to determine where the breakdown in identification of topic shifts occurs for hard-of-hearing individuals. Also, replications of the present study using a larger subject group might yield more robust results. Future research could also examine how binaural perception of prosodic features such as intonation aids identification of topic boundaries in discourse. It has been shown that for normal-hearing individuals, there is an advantage of binaural over monaural listening for speech reception in noise when the interaural differences are not the same for the speech as for the noise (see Zurek, 1993, for a review). This binaural advantage is due to the ability of the binaural auditory system to separate the target or signal from the noise by making use of the intra-aural differences in the sounds received at the two ears. Enhanced detectability of a signal presented in a dichotic listening condition is greatest for signal frequencies between 100 and 500 Hz (see Humes, 1994, for a review). Prosodic information such as intonation is carried primarily in those frequencies for which listeners derive a binaural advantage. Given that one of the functions of intonation is to signal topic 98 boundaries in speech, further investigation could address whether or not binaural processing of speech in noise could enhance the intelligibility of speech for discourse-level materials. The analysis of the prosodic characteristics of the materials used in the present study was limited to the prosodic features of utterance amplitude and pause duration. Further analysis could examine the occurrence and contribution of variations in fundamental frequency, laryngealization, pre-boundary syllable lengthening, rate of speech and lexical cues to topic boundaries and how these cues might be more or less important for topic boundary identification as the signal-to-noise ratio decreases. This might be especially valuable for those points in the monologues where there was high agreement across subjects regarding topic boundary identification. Monologue 3, for example, contains a pause in which all six subjects identified what is likely to be a sub-topic boundary, where the pause was located in the middle of a photograph description, more than 10 seconds before the end of the photograph. Analysis of such points in the monologues could provide further insight into what might miscue listeners to think that a new topic was about to begin, and what makes for an easily identifiable topic boundary under adverse listening conditions. The experiment carried out in the present study also has potential clinical application. Current audiologic practise typically measures a hard-of-hearing individual's hearing disability by administering tests of speech discrimination at the word or sentence level. Results from such tests correspond poorly with reports by hard-of-hearing individuals regarding the difficulties they experience understanding speech in everyday listening situations, perhaps because audiologic testing of speech discrimination does not accurately reflect the demands posed by listening comprehension at the level of discourse. Future 99 directions in audiologic evaluation could aim to incorporate comprehension measures at the level of discourse, and include measures like those taken in the present study that indicate how hard-of-hearing listeners use suprasegmental information to aid comprehension. Such evaluations might provide a clearer picture of the impact of an individual's hearing impairment on communication. Finally, a hard-of-hearing individual's ability to understand speech depends on multiple factors, including variables related to the speaker, the listener, the environment and the message (Erber, 1988). One of the current trends in aural rehabilitation is to examine how these variables can be manipulated to achieve maximum effectiveness in communication. Certainly, then, one of the areas which merits further investigation is the role of the perception of discourse structure cues in comprehension, and how such cues can be maximized by communication partners to allow for better communication of various kinds of messages in various realistic situations. 100 REFERENCES Aislin, R.N., & Smith, L.B. (1988). Perceptual development. Ann. Rev. Psychol., 39, 435-473. Bransford, J.D., & Johnson, M.K. (1973). Considerations of some problems of comprehension. In W.G. Chase (Ed.), Visual Information Processing. New York: Academic Press. Brazil, D. (1978). Discourse Intonation II. University of Birmingham, UK: Department of English. Broadbent, D.E. (1958). Perception and Communication. London: Pergamon. Brown, G., & Yule, G. (1983). Discourse Analysis. New York: Cambridge University Press. Brown, G., Currie, K.L, & Kentworthy, J. (1980). Questions of Intonation. London: Croom Helm. Brunt, M.A. (1985). Bekesy audiometry and loudness balance testing. In J. Katz (Ed.), Handbook of Clinical Audiology (pp. 273-291). Baltimore: Williams & Wilkins. Cairns, H.S. (1984). Research in language comprehension. In R. Naremore (Ed.), Language Sciences: Recent Advances. San Diego: College Hill Press. Cutler, A., & Pearson, M . (1986). On the analysis of prosodic turn-taking cues. In C. Johns-Lewis (Ed.), Intonation in Discourse (pp. 139-155). San Diego, CA: College-Hill Press. Cooper, J.C., & Cutts, J.C. (1971). Speech discrimination in noise. Journal of Speech and Hearing Research, 74,332-337'. Cruttenden, A. (1986). Intonation. New York: Cambridge University Press. Crystal, D. (1969). Prosodic Systems and Intonation in English. Cambridge: Cambridge University Press. Daneman, M. , & Carpenter, P.A. (1980) Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behaviour, 19, 450-466. Davis, H. (1947). Acoustics and psychoacoustics. In H. Davis and S.R. Silverman (Eds.), Hearing and Deafness. NY: Holt, Rinehart and Winston. 101 Duncan, S., & Fiske, D. (1977). Face-to-Face Interactions: Research, Methods and Theory. Hillsdale: Lawrence Erlbaum Associates. Durrant, J.D., & Lovrinic, J.H. (1984). Bases of Hearing Science. Baltimore, MD: Williams and Wilkins. Erber, N.P. (1988). Communication Therapy for Hearing-Impaired Adults. Abbotsford, Australia: Clavis Publishing. Faulkner, W. (1984). The Sound and the Fury. New York: Random House. Findlay, R.C, & Denenberg, L.J. (1977). Effects of subtle mid-frequency auditory dysfunction upon speech discrimination in noise. Audiology, 16, 252-259. Garrett, R.L., & Saint-Pierre, L. (1980). Message comprehension as a function of intelligibility and advance knowledge of topic. Journal of Auditory Research, 20,71-80. Geluykens, R., & Swerts, M. (1991). Local and global prosodic cues to discourse organization in dialogues. Proceedings of the ESCA Workshop of Prosody, Lund, Sweden, September, 1993, pp. 96-99. Halliday, M.A.K., & Hasan, R. (1976). Cohesion in English. New York: Longman. Humes, L. (1994). Psychoacoustic considerations in clinical audiology. In J. Katz (Ed.), Handbook of Clinical Audiology. Baltimore: Williams & Wilkins. Just, M.A. & Carpenter, P.A. (1987). The Psychology of Reading and Language Comprehension. Boston: Allyn and Bacon, Inc. Keenan, E.O., & Schiefflelin, B.B. (1976). Topic as a discourse notion: A study of topic in the conversation of children and adults. In C.N. L i (Ed.), Subject and Topic (pp. 337-383). New York: Free Press. Keith, R.W., & Talis, H.P. (1972). The effects of white noise on PB scores of normal-hearing and hearing-impaired listeners. Audiology, 11, 177-186. Kintsch, W., & Keenan, J. (1973). Reading rate and retention as a function of the number of propositions in the base structure of sentences. Cognitive Psychology, 5, 257-274. Kintsch, W., & van Dijk, T.A. (1978). Toward a model of text comprehension and production. Psychological Review, 85, 363-394. Ladefoged, P. (1982). A Course in Phonetics. New York: Harcourt, Brace, Jovanovich, Inc. 102 Lehiste, I. (1980). Phonetic characteristics of discourse. Paper presented at the Meeting of the Committee on Speech Research, Acoustical Society of Japan. Lehiste, I., & Wang, W. S-Y. (1977). Perception of sentence and paragraph boundaries with and without semantic information. In Dressier & Pfeiffer (Eds.), Phonologica 1976, pp. 277-283. Innsbruck: Institut fur Sprachwissenschaft der Universitat Innsbruck. Liang, B. (1994). RMS Value: Calibration Tone (version 1). In-house computer programme, School of Audiology and Speech Sciences, University of British Columbia. Lieberman, P. (1967). Intonation, Perception, and Language. Cambridge, MA: MIT Press. Lorch, R.F., & Lorch, E.P. (1985). Topic structure representation and text recall. Journal of Educational Psychology, 77, 137-148. Menn, L., & Boyce, S. (1982). Fundamental frequency and discourse structure. Language and Speech, 25, 341-384. Miller, G.A., & Nicely, P.E. (1955). An analysis of perceptual confusions among some English consonants. Journal of the Acoustical Society of America, 27, 338-352. Olsen, W.O., Jabaley, T., & Pappas, N.J. (1966). Speech discrimination in quiet and in competition with good fidelity system and with three hearing aids. Paper read at the Annual Convention, American Speech and Hearing Association, Washington, D.C. Peterson, G.E., & Barney, H.L. (1952). Control methods used in a study of the vowels. Journal of the Acoustical Society of America, 24, 175-184. Pichora-Fuller, M.K., Schneider, B.A., & Daneman, M . (1995). How young and old adults listen to and remember speech in noise. Journal of the Acousical Society of America, 97, 593-608. Price, P.J., Ostendorf, M. , Shattuck-Hufnagel, S., & Fong, C. (1991). The use of prosody in syntactic disambiguation. Journal of the Acoustical Society of America, 90, 2956-2970. Rabbit, P. (1966). Recognition memory for words correctly heard in noise. Psychonomic Science, 6, 383-384. Ross, M. , Huntington, D.A., Newby, H.A., & Dixon, R.F. (1965). Speech discrimination of hearing-impaired individuals in noise. Journal of Auditory Research, 5, 47-72. Schiffrin, D. (1987). Discourse Markers. Cambridge: Cambridge University Press. 103 Stelmachowicz, P.G., Jesteadt, W., Gorga, M.P., & Mott, J. (1985). Speech-perception ability and psychophysical tuning curves in hearing-impaired listeners. Journal of the Acoustical Society of America, 77, 620-627. Swerts, M. , & Geluykens, R. (1993). The prosody of information units in spontaneous monologue. Phonetica, 50, 189-196. Swerts, M. , Collier, R., & Terken, J. (1992). Prosodic correlates of discourse units in spontaneous speech. Proceedings of the International Conference on Spoken Language Processing, Banff, Alberta, September, 1992. Swerts, M. , Bouwhuis, D.G., & Collier, R. (1994). Melodic cues to the perceived "finality" of discourse. Journal of the Acoustical Society of America, 96, 2064-2075. Swerts, M. , Collier, R., & Terken, J. (in press). Prosodic predictors of discourse finality in spontaneous monologues. 104 APPENDIX A Subjects' Audiometric Thresholds (dB HL) Test frequency 250 Hz 500 Hz 1000 Hz 2000 Hz 4000 Hz Ear to which materials were R L R L R L R L R L presented Subject Sl 20 10 5 0 5 5 5 20 5 10 Right S2 0 5 0 0 0 5 0 5 0 5 Right S3 15 15 10 10 5 5 0 0 -5 0 Right S4 10 15 5 5 -10 0 0 15 -5 -5 Right S5 5 10 5 5 -5 -5 0 0 -5 5 Right S6 0 \ 5 -5 0 -5 -10 -5 -5 0 -5 -5 Left S7 10 0 5 0 5 -5 0 0 0 Right S8 5 5 10 10 0 5 0 0 10 -10 Right S9 5 5 -5 0 10 -5 -5 0 10 0 Right S10 5 10 5 5 10 5 5 10 10 15 Right S l l 10 15 0 0 0 -5 0 0 -5 5 Left S12 5 5 0 5 -5 -5 -5 5 5 15 Right APPENDIX B Transcripts of Monologues Monologue 1 This is a photograph of my mom and my sister and my cousin. RMS: 2567 Pause: 863 ms Um it's at my mom's birthday. She's holding up her uh -RMS: 2794 Pause: 586 ms it's one of her presents. It's a record. I RMS: 3906 Pause: 313 ms can't see which one it is. RMS: 1029 Pause: 1363 ms And uh RMS: 1532 Pause: 423 ms I guess this is about RMS: 2358 Pause: 1438 ms hmm RMS: 1513 Pause: 500 ms ten years ago RMS: 2048 Pause: 1425 ms and uh i(t's) -RMS: 1330 Pause: 289 ms it's at my grandparents' apartment in their living room. RMS: 4233 106 Pause: 3141 ms END OF TOPIC 1 This photograph is of my aunt and myself. I'm sitting on my aunt's lap. RMS: 1766 Pause: 881 ms This is RMS: 1299 Pause: 647 ms a pretty old photograph. My uh -RMS: 1740 Pause: 700 ms Hook RMS: 1323 Pause: 758 ms like a little kid here. I guess I'm about eight. RMS: 1453 Pause: 764 ms And we're sitting in the back yard of my grandparents' uh RMS: 1476 ' Pause: 797 ms or on the -RMS: 1294 Pause: 418 ms on the porch of my grandparents' apartment. RMS: 1870 Pause: 253 ms looking out over the uh RMS: 910 Pause: 1101 ms courtyard. RMS: 1973 Pause: 445 ms [cough] 107 RMS: 716 Pause: 2228 ms END OF TOPIC 2 This photograph is of RMS: 2908 Pause: 391 ms Princess Anne RMS: 2378 Pause: 246 ms talking to uh -RMS: 1839 Pause: 991 ms talking to this RMS: 1512 Pause: 243 ms this woman uh Mrs. McDonald that we know. She's a hundr(ed) -RMS: 1391 Pause: 280 ms a hundred years old. RMS: 1723 Pause: 870 ms And uh standing with her is um RMS: 1542 Pause: 1040 ms my great-great-aunt RMS: 5731 Pause: 628 ms um RMS: 764 Pause: 558 ms Ella Lockhart. RMS: 2426 Pause: 252 ms And uh she's ninety-nine in the photograph. RMS: 1089 Pause: 1306 ms.; And uh RMS: 1111 Pause: 760 ms don't know where it is RMS: 1922 Pause: 435 ms but uh RMS: 783 Pause: 645 ms looks pretty busy. It's out on the street RMS: 1594 Pause: 309 ms and there's people all over the place. RMS: 1176 Pause: 618 ms It's so uh... RMS: 691 Pause: 4387 ms END OF TOPIC 3 This photograph was taken in Barkerville, B.C. RMS: 1855 Pause: 543 ms My family made a trek up there one RMS: 1252 Pause: 466 ms one summer where we sort of did the tour of the interior. RMS: 1246 Pause: 794 ms And uh we stopped at Barkerville. RMS: 1622 Pause: 266 ms 109 And uh this photograph is my mum, myself and my sister standing in front of this church and we look like little tiny peanuts in front of it RMS: 1233 ms Pause: 239 (be)cause it was an enormous church and my dad RMS: 1165 Pause: 646 ms took the photograph from uh RMS: 1324 Pause: 1126 ms a couple blocks down RMS: 1501 Pause: 331 ms so it's kind of a RMS: 959 Pause: 418 ms funny photograph. This was taken RMS: 1020 Pause: 821 ms a couple - no three or four years ago I guess it was now. RMS: 1061 Pause: 3169 ms END OF TOPIC 4 This photograph is of my dad standing next to a RMS: 1744 Pause: 690 ms saguaro cactus RMS: 1068 Pause: 322 ms in uh - in Arizona. RMS: 1470 Pause: 1563 ms They went about 110 RMS: 3363 Pause: 907 ms nine years ago - eight years ago to uh -RMS: 1851 Pause: 381 ms to visit RMS: 1723 Pause: 382 ms Tucson RMS: 1842 Pause: 270 ms and Phoenix. RMS: 1201 Pause: 859 ms And they had such a great time there that they uh -RMS: 1249 Pause: 936 ms they decided they'd bring the whole family down next time. RMS: 1748 Pause: 2946 ms END OF TOPIC 5 This is a photograph of me at one of my birthdays. RMS: 1886 Pause: 769 ms I got a -1 got a T-shirt that had all of the major muscle groups printed on it. RMS: 1813 Pause: 372 ms And so when you wore it, all of the muscle groups on the T-shirt lined up with the muscles RMS: 1286 Pause: 742 ms on the person wearing it. RMS: 1521 Pause: 833 ms I'm about RMS: 1457 Pause: 379 ms [cough] I'm about RMS: 2331 Pause: 1617 ms seven. RMS: 2412 Pause: 952 ms And I'm displaying my T-shirt. RMS: 1655 Pause: 3219 ms END OF TOPIC 6 This is a picture of my dad and my sister and I standing in front of RMS: 1741 Pause: 881 ms some uh cholla cactuses RMS: 1280 Pause: 601 ms um in -RMS: 1444 Pause: 570 ms in California. We went to RMS: 1434 Pause: 731 ms Disneyland and Universal Studios whenJ was in about grade five. RMS: 1488 Pause: 4472 ms END OF TOPIC 7 My uh -RMS: 1825 Pause: 912 ms this photograph is of my RMS: 3008 Pause: 320 ms dad's best friend and my mom's best friend who actually got married RMS: 2172 Pause: 818 ms um Ken and Susan Skinner. RMS: 1759 Pause: 331 ms And uh RMS: 1124 Pause: 1208 ms I'm just uh eating dinner RMS: 1961 Pause: 302 ms at the cabin. RMS: 1330 Pause: 203 ms Um. RMS: 706 Pause: 407 ms They have a cabin at -RMS: 1630 Pause: 308 ms at uh Grand Beach RMS: 1232 Pause: 205 ms in Manitoba RMS: 833 Pause: 972 ms they go to every year. RMS: 1730 Pause: 1043 ms And they took us up one summer. RMS: 1572 113 Pause: 260 ms And we're just having dinner there. RMS: 1572 Pause: 260 ms Looks like they're - looks like we're having a great time too. Everyone's laughing pretty hard. RMS: 1624 Pause: 1255 ms END OF TOPIC 8 This photograph was taken just last year. RMS: 2051 Pause: 939 ms Um. RMS: 934 Pause: 2158 ms It's of my RMS: 1212 Pause: 297 ms dad and RMS: 964 Pause: 594 ms myself and my sister and my grandmother and my mom's friend and her three boys. RMS: 1110 Pause: 499 ms And we're all sitting around the dining room table. It's - it's Christmas dinner. RMS: 1537 Pause: 1195 ms And uh I guess I'm about eighteen years old here. RMS: 1107 Pause: 4021 ms END OF TOPIC 9 This photograph was taken when I was in RMS: 1836 114 Pause: 241 ms grade five. RMS: 1495 Pause: 862 ms I'm standing - it's up at Whistler. And I'm standing on the beach. RMS: 1995 Pause: 367 ms It's the summer time. It's hot as could be (be)cause no - everyone's wearing their bathing suits and there's people windsurfing on the lake behind me. RMS: 1431 Pause: 1273 ms And I RMS: 1212 Pause: 954 ms look like I'm having a pretty good time. Got a big grin on. RMS: 1416 Pause: 1256 ms END OF TOPIC 10 This is a great photograph. It's of my aunt and my uncle, my mum's sister Joanne and her husband Len. RMS: 1407 Pause: 463 ms They went to some party RMS: 1939 Pause: 306 ms [cough] RMS: 1967 Pause: 890 ms and they met Mr. T. there. RMS: 2218 Pause: 742 ms So this is a photograph of Mr. T. with my aunt and my uncle. RMS: 2059 END OF TOPIC 11 115 Monologue 2 This is a photograph of my grandfather and my sister RMS: 2081 Pause: 813 ms sitting in a - sitting in one of the armchairs in my old house RMS: 1754 Pause: 642 ms in the living room. RMS: 1717 Pause: 368 ms And my sister's holding her favorite stuffed animal, this frog that was about RMS: 1412 Pause: 1276 ms five feet tall RMS: 1162 Pause: 436 ms and my sister RMS: 1021 Pause: 488 ms at the time was about three years old and certainly nowhere near five feet tall so she looks pretty ridiculous holding this giant frog. RMS: 1317 Pause: 453 ms And she's sitting on my grandfather's lap. RMS: 1077 Pause: 1871 ms END OF TOPIC 1 This photograph was taken um RMS: 2087 Pause: 1196 ms at Hallowe'en RMS: 2799 Pause: 480 ms 116 um RMS: 1030 Pause: 251 ms I guess RMS: 809 Pause: 2024 ms nineteen eighty-RMS: 1141 Pause: 1798 ms four, RMS: 1330 Pause: 627 ms maybe nineteen eighty-five. RMS: 1073 Pause: 834 ms And uh RMS: 1136 Pause: 307 ms it's -RMS: 936 Pause: 236 ms it's myself and my sister RMS: 1274 Pause: 849 ms and uh three - my three friends that lived in the neighbourhood RMS: 1532 Pause: 1037 ms in the hallway of my RMS: 1909 Pause: 588 ms old house just sort of RMS: 1257 Pause: 345 ms between the kitchen and the living room. And I'm dressed as a wizard 117 RMS: 1459 Pause: 647 ms and my friend David is dressed as Robin Hood RMS: 1521 Pause: 570 ms and his brother's dressed as a pirate RMS: 1287 Pause: 977 ms and the third friend is dressed as a ghost and my little sister looks like Strawberry Shortcake. RMS: 1236 Pause: 5139 ms END OF TOPIC 2 This is a photograph of my dad and I RMS: 1986 Pause: 531 ms up at Whistler. RMS: 1657 Pause: 413 ms Um I'm just getting off the orange chair RMS: 1722 Pause: 1272 ms which tells you something about the age of this place - this photograph. RMS: 998 Pause: 467 ms Um. RMS: 1160 Pause: 444 ms I look about RMS: 984 Pause: 1492 ms ten or eleven. RMS: 1273 Pause: 918 ms 118 It's probably one of the first -RMS: 857 Pause: 264 ms first times I went skiing too cause I didn't start till I was about nine or ten. RMS: 960 Pause: 4372 ms END OF TOPIC 3 This is a photograph of my sister standing uh RMS: 2155 Pause: 1447 ms at the railing looking over at Niagara Falls. RMS: 1774 Pause: 746 ms We made a trip to Toronto when I was in grade six RMS: 1298 Pause: 335 ms to visit my aunt and my uncle RMS: 1061 Pause: 732 ms and cousin. RMS: 1186 Pause: 260 ms And uh RMS: 1122 Pause: 248 ms we RMS: 540 Pause: 670 ms made a stop off at uh RMS: 1618 Pause: 678 ms Niagara Falls RMS: 1486 Pause: 324 ms 119 and it's actually a really good photograph of my sister. RMS: 1019 Pause: 821 ms She looks like she's about -RMS: 623 Pause: 1137 ms I guess RMS: 577 Pause: 1251 ms at that time she'd be about RMS: 779 Pause: 658 ms seven or eight. RMS: 770 Pause: 3137 ms END OF TOPIC 4 This is a photograph from my tenth birthday party RMS: 1394 Pause: 760 ms which I had up at Bowen Island (be)cause we had a RMS: 1154 Pause: 256 ms a cabin there. RMS: 1184 Pause: 870 ms And the photograph is of myself sitting in a lawn chair with my grandfather kneeling next to me RMS: 1306 Pause: 626 ms on the porch of the house with my uh -RMS: 6649 Pause: 683 ms with my haul sitting next to me on the -120 RMS: 982 Pause: 585 ms on the RMS: 858 Pause: 285 ms deck. RMS: 1032 Pause: 773 ms It's a hot sunny day and we're all wearing our sunglasses. RMS: 794 Pause: 3561 ms END OF TOPIC 5 Um I was working at a RMS: 1547 Pause: 266 ms [cough] Y MCA summer camp. RMS: 2076 Pause: 797 ms And uh my family came up to visit me one time. It was the first summer I'd worked there. RMS: 1298 Pause: 1047 ms So this is a photograph of me and my sister and my cousin RMS: 1509 Pause: 925 ms s-standing uh RMS: 1773 Pause: 1246 ms in the chapel RMS: 2006 Pause: 1114 ms which was sort of a bunch of benches RMS: 1398 Pause: 423 ms ^ 121 in a row RMS: 1367 Pause: 638 ms facing this podium. RMS: 3607 Pause: 748 ms And it looks out onto the uh -RMS: 1246 Pause: 355 ms onto the ocean there. It's just on the sunshine coast. It's a beautiful background. RMS: 1018 Pause: 797 ms It's a sunny day RMS: 984 Pause: 1014 ms and I'm holding a plate of food there. I guess it was lunch time. RMS: 3915 Pause: 2832 ms END OF TOPIC 6 This photograph is RMS: 2078 Pause: 469 ms an old one from before I was born. RMS: 1704 Pause: 778 ms It's of my grandfather RMS: 1702 Pause: 882 ms and his dog. RMS: 1556 Pause: 769 ms Don't know the dog's name RMS: 1381 Pause: 322 ms 122 but I've seen photographs of this dog before. RMS: 994 Pause: 576 ms And uh they're at Terracotta which is where my grandfather and uh RMS: 1254 , Pause: 221 ms grandmother lived RMS: 1176 Pause: 1229 ms before they moved to Vancouver. RMS: 958 Pause: 2724 ms END OF TOPIC 7 This photograph was taken this year. RMS: 1834 Pause: 681 ms It's in North Hatley RMS: 1531 Pause: 302 ms in Quebec. RMS: 1271 Pause: 945 ms My parents came out to visit me while I was at school RMS: 2844 Pause: 920 ms and uh RMS: 1143 Pause: 1264 ms took a day and went and spent it with them in North Hatley which is where they were staying. RMS: 1183 Pause Duration: 849 ms My dad and I are leaning over the railing looking at the Massawipi 123 RMS: 1274 Pause: 450 ms Lake. RMS: 854 Pause: 1826 ms It looks -RMS: 381 Pause: 414 ms it's a beautiful day and it's fall. RMS: 1069 Pause: 364 ms I guess it's October - late October or RMS: 959 Pause: 445 ms it's about mid-October and all the trees are gorgeous colours. There's bright yellows and reds and oranges. RMS: 1354 Pause: 2913 ms END OF TOPIC 8 This is a photograph of a friend of um -RMS: 1733 Pause: 785 ms a friend of my mom RMS: 1605 Pause: 1046 ms from university RMS: 1319 Pause: 250 ms and her husband and three kids RMS: 1251 Pause: 866 ms who we still see a lot of. They just live RMS: 886 Pause: 790 ms 124 a few blocks away. It's uh RMS: 1150 Pause: 752 ms a Christmas card RMS: 1291 Pause: 1076 ms and the five of them are all RMS: 1281 Pause: 898 ms dressed up and looking all smiley and festive for Christmas. RMS: 2307 Pause: 2534 ms END OF TOPIC 9 Here's a photograph of my sister in her RMS: 1918 Pause: 870 ms ballet attire, RMS: 1610 Pause: 265 ms holding a flower um standing in front of the fireplace in our old house. RMS: 1234 Pause: 1726 ms And she looks like she's about RMS: 741 Pause: 1348 ms six years old. RMS: 989 Pause: 2681 ms END OF TOPIC 10 This is a photograph from RMS: 1552 Pause: 739 ms 125 one Hallowe'en. RMS: 1650 Pause: 1261 ms I guess that I was about RMS: 1178 Pause: 385 ms twelve in this photograph and I'm standing there with my sister RMS: 1431 Pause: 1400 ms and uh RMS: 784 Pause: 230 ms she'd be about eight RMS: 683 Pause: 921 ms and my two friends from down the block RMS: 1282 Pause: 491 ms Paul and Jamie and RMS: 1209 Pause: 514 ms I'm dressed as a hunter and uh RMS: 1305 Pause: 815 ms my sister's dressed as a witch I guess. She's got this crazy wig on and she's dressing -RMS: 926 Pause: 207 ms dressed in black. RMS: 980 Pause: 1401 ms Paul's dressed as a samurai and uh RMS: 1794 Pause: 227 ms Jamie's dressed as a hockey player. RMS: 1013 Pause: 286 ms We're all standing at the front door of my old house RMS: 1076 Pause: 501 ms with a big RMS: 751 Pause: 277 ms jack-o'-lantern at the - at our feet. RMS: 963 END OF TOPIC 11 127 Monologue 3 This is a picture of my dad and I RMS: 2874 Pause: 316 ms um RMS: 1073 Pause: 417 ms at the Bowen Island Festival. RMS: 2479 Pause: 950 ms They have this big festival every year where um RMS: 1311 Pause: 1246 ms there's a big parade and it all ends up at this RMS: 4864 Pause: 472 ms empty field where they set up RMS: 1445 Pause: 392 ms various stands to -RMS: 2480 Pause: 437 ms to play games and RMS: 1578 Pause: 748 ms have races. This particular one is my dad and I kneeling down at the slug race RMS: 1772 Pause: 374 ms with our slug Slimey. RMS: 1419 Pause: 1054 ms Um. RMS: 1187 Pause: 1136 ms 128 I've got my two sticks so I can manipulate Slimey uh RMS: 1304 Pause: 463 ms through the race track. RMS: 1587 Pause: 929 ms And uh RMS: 1525 Pause: 424 ms this year RMS: 2753 Pause: 345 ms it was really funny. This uh little girl sabotaged the whole race by sprinkling salt down the track of each slu(g) -RMS: 1455 Pause: 359 ms of each slug's raceway except her own RMS: 1547 Pause: 434 ms ensuring her victory. RMS: 1050 Pause: 836 ms But uh RMS: 1012 Pause: 258 ms she was caught when she was the only one who finished the race. RMS: 1580 Pause: 1952 ms And this would be RMS: 1547 Pause: 1856 ms I guess in nineteen RMS: 1618 Pause: 2659 ms eighty-RMS: 1066 Pause: 1597 ms four - nineteen eighty-three. RMS: 1259 Pause: 3494 ms END OF TOPIC 1 This photograph is RMS: 2307 Pause: 254 ms um RMS: 1139 Pause: 841 ms one taken in my old house in the bathroom. RMS: 1422 Pause: 717 ms We had sort of RMS: 1287 Pause: 291 ms twin sinks RMS: 5094 Pause: 265 ms one side by - uh side by each. RMS: 927 Pause: 738 ms Um. RMS: 1207 Pause: 352 ms One of them, my cat -RMS: 1410 Pause: 526 ms our old cat Pepper is drinking from RMS: 1129 130 Pause: 569 ms with her bum up in the air and her head right down in the sink and on the RMS: 1296 Pause: 732 ms at the other sink is my sister with her bum in the air and her head down in the sink also drinking water. RMS: 1376 Pause: 884 ms And uh RMS: 1000 Pause: 592 ms it's actually a pretty funny photograph. RMS: 793 Pause: 420 ms Um. RMS: 1045 Pause: 487 ms I guess I'd have been about RMS: 1442 Pause: 252 ms thirteen or fourteen in this photograph -RMS: 833 Pause: 424 ms taking this photograph. I took this photograph. RMS: 706 Pause:-2676 ms END OF TOPIC 2 This photograph is of my sister RMS: 2065 Pause: 355 ms icing her cake. RMS: 1418 Pause: 543 ms 131 Our elementary school had a RMS: 2022 Pause: 1000 ms a cake walk every year RMS: 1636 Pause: 383 ms and to be allowed to participate in the cake walk you had to produce a cake. RMS: 1272 Pause:. 1000 ms And uh RMS: 1323 Pause: 467 ms I'd have been in grade five here so my sister Martha would have been in about grade -RMS: 1023 Pause: 1344 ms in grade RMS: 618 Pause: 404 ms one or grade two I guess. RMS: 670 Pause: 828 ms And she's uh RMS: 991 Pause: 260 ms slapping the icing on this cake that's in the shape - the shape of a -RMS: 1222 Pause: 622 ms a weight-lifter. RMS: 1401 Pause: 424 ms And he's holding RMS: 924 Pause: 721 ms uh 132 RMS: 689 Pause: 737 ms weights above his head. RMS: 1208 Pause: 3227 ms END OF TOPIC 3 This is a photograph of me skiing RMS: 1721 Pause: 1024 ms at Mount Baker. And I look like I'm about nine. RMS: 1117 Pause: 876 ms Sort of one of the first -RMS: 1016 Pause: 230 ms first runs of my life. RMS: 1573 Pause: 753 ms And it's got great scenery. It's a -RMS: 1395 Pause: 593 ms it's a beautiful day and there's RMS: 1422 Pause: 205 ms trees and the snow's everywhere. RMS: 1198 Pause: 1408 ms And it looks warm (be)cause all I'm wearing is my ski pants and a - and a RMS: 1685 Pause: 375 ms long-sleeved T-shirt. RMS: 1021 Pause: 2018 ms And this one would have been. RMS: 1227 Pause: 1823 ms ) I guess -RMS: 399 Pause: 1231 ms yeah taken in sort of -RMS: 1058 Pause: 2417 ms well I don't remember what year it was. RMS: 1050 Pause: 885 ms I look about nine though. RMS: 1123 Pause: 1464 ms END OF TOPIC 4 This is of my dad's friend Ken Skinner. RMS: 1739 Pause: 1004 ms And uh RMS: 1710 Pause: 1886 ms it's taken in my grandmother's house in Winnipeg. RMS: 2179 Pause: 645 ms And uh RMS: 1103 Pause: 920 ms he's wearing -RMS: 2736 Pause: 466 ms he's wearing his son's hat. And his son's about RMS: 1612 Pause: 633 ms 134 a year old. So he looks pretty silly wearing this RMS: 1577 Pause: 1024 ms little baby's hat RMS: 1442 Pause: 493 ms on his head. RMS: 1052 Pause: 426 ms And he's smiling away. RMS: 1226 Pause: 1924 ms END OF TOPIC 5 This is a neat photograph. It's a photograph of a statue a bronze statue that we have. RMS: 1703 Pause: 577 ms Just a small - a small br(onze) - bronze statue of a RMS: 1226 Pause: 976 ms man riding a horse RMS: 1377 Pause: 774 ms And um RMS: 1372 Pause: 1791 ms he - this statue - this small statue is in the foreground RMS: 1529 Pause: 256 ms and in the background is my front yard. RMS: 1198 Pause: 865 ms So between the foreground and the background is the glass - the window. RMS: 1317 135 Pause: 754 ms And you can see the garden of the front yard RMS: 1382 Pause: 563 ms in the back RMS: 655 Pause: 778 ms and um RMS: 926 Pause: 1150 ms this statue in the front. It's - it's a really neat photograph. RMS: 1159 Pause: 2591 ms END OF TOPIC 6 This photograph was taken RMS: 2155 Pause: 987 ms at what was probably my RMS: 5658 Pause: 644 ms sixth or RMS: 1143 Pause: 619 ms fifth birthday party. RMS: 1963 Pause: 739 ms And uh RMS: 1465 Pause: 995 ms we're all wearing these crazy .glasses that have the eyes painted on them. And uh I look pretty silly. RMS: 1631 Pause: 686 ms 136 We're making great faces. RMS: 1415 Pause: 1400 ms I'm still friends with most of the people in this photograph too. RMS: 1204 Pause: 2062 ms END OF TOPIC 7 This is a really old photograph RMS: 1458 Pause: 229 ms [clears throat] RMS: 853 Pause: 585 ms of my sister and I. RMS: 1325 Pause: 764 ms And my sister's still a little infant. RMS: 1240 Pause: 708 ms So I'm probably about four or five. RMS: 966 Pause: 748 ms And um we're sitting in front of the Christmas tree with all the Christmas presents behind us, RMS: 990 Pause: 500 ms probably just RMS: 922 Pause: 424 ms dying to get this photograph over with to dive into the presents. RMS: 1105 Pause: 2763 ms END OF TOPIC 8 This is a photograph that was taken last summer. I -RMS: 1466 Pause: 478 ms I went on Outward Bound RMS: 1469 Pause: 442 ms um RMS: 829 Pause: 266 ms for three weeks RMS: 1352 Pause: 771 ms So here's -RMS: 1179 Pause: 554 ms here's the uh -RMS: 1107 Pause: 450 ms the eight of us sitting at the top of Mount Ayregorn RMS: 1215 Pause: 844 ms Um it's one of the Coastal Mountains here in B.C. RMS: 1444 Pause: 876 ms And um RMS: 1217 Pause: 1814 ms it's a fantastic day. It's just RMS: 1025 Pause: 204 ms beautiful. Hot and sunny. RMS: 1386 Pause: 692 ms The background is this amazing mountain range RMS: 1250 Pause: 510 ms and uh you can see for miles and miles. RMS: 886 Pause: 805 ms We all look RMS: 1704 Pause: 324 ms pretty happy to have made it to the top of this mountain. RMS: 3731 Pause: 3135 ms END OF TOPIC 9 This is a photograph that was also taken just last summer. RMS: 1121 Pause: 937 ms Um it's of my mom and my self and my sister and my dad. RMS: 1144 Pause: 690 ms We're standing in the driveway of our new house RMS: 1539 Pause: 360 ms uh next to the RMS: 602 Pause: 629 ms rhododendron bush. RMS: 1344 Pause: 973 ms And uh RMS: 1011 Pause: 1319 ms this would have been uh RMS: 1279 Pause: 909 ms 139 just before uh RMS: 1130 Pause: 950 ms going to my sister's confirmation RMS: 969 Pause: 449 ms actually. RMS: 576 Pause: 4214 ms END OF TOPIC 10 This photograph is taken in the courtyard of my RMS: 1497 Pause: 219 ms grandparents ... Vancouver -RMS: 1448 Pause: 292 ms um RMS: 802 Pause: 593 ms of their apartment. RMS: 1810 Pause: 509 ms We're - I'm RMS: 1146 Pause: 743 ms next to the pool RMS: 1965 Pause: 331 and it's a hot sunny day. RMS: 668 END OF TOPIC 11 Monologue 4 This is a photograph taken of my father and my sister RMS: 1491 Pause: 1195 ms standing on the RMS: 1744 Pause: 252 ms passenger deck of the Bowen Island ferry. RMS: 3450 Pause: 424 ms And they're looking out at all the cars parked in the -RMS: 1192 Pause: 310 ms waiting to get on to the other ferries. RMS: 1363 Pause: 1830 ms And it's a sunny day. RMS: 1232 Pause: 1098 ms It's funny: you can see all the rows of RMS: 962 Pause: 281 ms seats for people to sit down on. RMS: 747 Pause: 2272 ms And it's - this would have been at the RMS: 906 Pause: 498 ms Horseshoe Bay ter(minal) - terminal. RMS: 2476 Pause: 538 ms At that end of it as opposed to the RMS: 638 Pause: 314 ms 141 Bowen Island end. RMS: 1076 Pause: 2674 ms END OF TOPIC 1 This is a photograph RMS: 1373 Pause: 1340 ms of a whole bunch of friends RMS: 5769 Pause: 249 ms of um my family's. RMS: 864 Pause: 951 ms . ° Every year we get together. There's sort of five or six families. And we get together and we have this RMS: 1176 Pause: 598 ms big family softball game. RMS: 1566 Pause: 687 ms And we divide up into two teams: the green team and the blue team. RMS: 1079 Pause: 386 ms And this is a photograph of the green team. RMS: 1126 Pause: 647 ms Um. RMS: 569 Pause: 1353 ms We always take a photograph. Sort of the RMS: 1159 Pause: 461 ms annual 142 RMS: 1750 Pause: 1099 ms recording of what everyone looks like. RMS: 1168 Pause: 1223 ms And this - we always play at Quilchena Park, RMS: 1348 Pause: 853 ms 33rd and Arbutus there. RMS: 777 Pause: 2012 ms This one would have been RMS: 839 Pause: 2158 ms 5 grade eight -RMS: 834 Pause: 511 ms grade eight. So RMS: 553 Pause: 1067 ms five years ago, RMS: 532 Pause: 2024 ms six years ago. RMS: 253 Pause: 1206 ms END OF TOPIC 2 This is a photograph of my dad and my mum's friend Susan. RMS: 1564 Pause: 1245 ms And uh we had dinner RMS: 1824 Pause: 476 ms 143 at their house one evening. RMS: 1788 Pause: 749 ms And I guess this would have been grade seven RMS: 1145 Pause: 1212 ms or grade eight. RMS: 1041 Pause: 1148 ms And uh RMS: 1265 Pause: 2625 ms Susan's husband Ken is a really good friend of my dad's and he's sort of a crazy guy. He's really funny. And he started out RMS: 1744 Pause: 271 ms folding his napkin into a little gumby hat and put it onto his head. So this is a photograph of my dad and my aunt RMS: 1359 Pause: 715 ms and half of me -RMS: 1633 Pause: 692 ms we're all wearing gumby hats -sitting in the dining room of my aunt's - of my -RMS: 1570 Pause: 496 ms of Susan and Ken's living - house. RMS: 1233 Pause: 428 ms Uh it's sort of a RMS: 624 Pause: 219 ms funny photograph. We all look pretty ridiculous. RMS: 1089 Pause: 1845 ms END OF TOPIC 3 This is a photograph of myself RMS: 1630 Pause: 210 ms and uh RMS: 1281 Pause: 977 ms my - my RMS: 1169 Pause: 335 ms exchange billet RMS: 1514 Pause: 361 ms from uh -RMS: 1043 Pause: 591 ms from Quebec. When I was in grade seven RMS: 1297 Pause: 653 ms we did a Quebec exchange. RMS: 1421 Pause: 336 ms And uh Frederic Morency came to visit me. RMS: 1620 Pause: 761 ms And we're uh just -RMS: 1390 Pause: 228 ms [cough] sitting out - standing outside RMS: 1876 Pause: 845 ms my old house at the side. RMS: 1961 Pause: 644 ms And it's -RMS: 920 Pause: 269 ms and uh RMS: 1069 Pause: 410 ms it's still got the old fence up and all the rose bushes. RMS: 1332 Pause: 1330 ms And it's - it doesn't look like a very nice day. It's sort of gray and rainy. RMS: 1315 Pause: 1072 ms One of your typical Vancouver RMS: 1452 Pause: 219 ms spring days. RMS: 1390 Pause: 2343 ms END OF TOPIC 4 This photograph is of myself and my dad and my dad's friend and his daughter RMS: 1378 Pause: 430 ms and my sister and we're all -RMS: 1874 Pause: 417 ms we're all in this power boat -RMS: 2266 Pause: 204 ms speed boat. It's up in Kelowna. Every summer we used to go up to Kelowna RMS: 1682 Pause: 1070 ms 146 and spend it with these two other families - spend a couple of weeks with these two other families at this resort. RMS: 1287 Pause: 759 ms In this photograph I look like RMS: 1115 Pause: 492 ms I'm in maybe grade RMS: 1063 Pause: 922 ms six, RMS: 1579 Pause: 418 ms maybe grade five. RMS: 1208 Pause: 693 ms And uh RMS: 1193 Pause: 1038 ms it looks like we're just getting ready to tow somebody for water skiing but in this photograph you can't see who it is. RMS: 1200 Pause: 1849 ms END OF TOPIC 5 This is a photograph of my grandmother, RMS: 1508 Pause: 810 ms my mom's mom. It's taken out front of uh -RMS: 929 Pause: 821 ms of Susan's and Ken's house who uh RMS: 1333 Pause: 352 ms live a few blocks away from her 147 RMS: 962 Pause: 257 ms in Winnipeg. RMS: 614 Pause: 4630 ms END OF TOPIC 6 This is a photograph of my little sister when she was about RMS: 1194 Pause: 477 ms two. RMS: 1028 Pause: 1088 ms And uh when she was two years old she RMS: 1584 Pause: 979 ms never quite grasped the concept of picking the food up and putting it in her mouth. She sort of brought her face down to the food and RMS: 1927 Pause: 836 ms ate it that way. So it's a photograph of her RMS: 1489 Pause: 1355 ms licking her cake -. RMS: 1366 Pause: 263 ms licking the icing off of her cake. RMS: 1298 Pause: 203 ms Her face is RMS: 960 Pause: 233 ms practically right inside RMS: 1253 Pause: 443 ms the icing. RMS: 866 Pause: 983 ms It's a pretty crazy photograph. It's funny. RMS: 707 Pause: 1391 ms END OF TOPIC 7 This is a photograph of me standing next to a - a river in Arizona. RMS: 1363 Pause: 935 ms Um we went for a walk -RMS: 1236 Pause: 225 ms my mum and I went for a walk and looked at the RMS: 1097 Pause: 515 ms plants and the animals and uh RMS: 1442 Pause: 897 ms Mum took a photograph of it. It's uh really interesting. The RMS: 1268 Pause: 1690 ms dry RMS: 2575 Pause: 939 ms dirt has been RMS: 1990 Pause: 502 ms. sort of RMS: 1033 Pause: 231 ms carved 149 RMS: 1716 Pause: 758 ms to allow this -RMS: 1012 Pause: 1042 ms well it looks as though it's been carved to allow this river to flow. RMS: 1068 Pause: 755 ms It's a really neat photograph. This would have been taken RMS: 1232 Pause: 412 ms when I was in grade eleven. RMS: 867 Pause: 687 ms So RMS: 492 Pause: 1007 ms two years ago, three years ago. RMS: 611 Pause: 4193 ms END OF TOPIC 8 This is a photograph of me riding a - a horse RMS: 1470 Pause: 726 ms or a pony I guess it is. RMS: 1248 Pause: 825 ms Nah, it's a horse. That's a big horse. RMS: 569 Pause: 729 ms And um RMS: 1419 Pause: 1313 ms I think I went with this U.B.C. day camp I was doing. RMS: 1267 Pause: 900 ms Uh, it was a sports day camp and we used to go and do various activities and one day went RMS: 992 Pause: 809 ms and uh rode - rode ponies. RMS: 5094 Pause: 235 ms That was a lot of fun. RMS: 864 Pause: 2521 ms END OF TOPIC 9 This is a photo of my graduation. It's taken in the gardens of Queen Elizabeth Park. RMS: 1431 Pause: 886 ms And I'm standing with sort of my four closest friends from school: RMS: 1296 Pause: 365 ms Mike, Eric, Neil and Elan. RMS: 1082 Pause: 691 ms And uh RMS: 952 Pause: 653 ms we're all RMS: 2120 Pause: 365 ms dressed up. Eric's wearing his double-breasted RMS: 1138 Pause: 738 ms tuxedo and Mike's got a cape on RMS: 1156 Pause: 839 ms and Neil's wearing his RMS: 1468 Pause: 302 ms tuxedo RMS: 1273 Pause: 512 ms and uh RMS: 1208 Pause: 650 ms Elan's wearing a tuxedo with - with tails and I'm wearing my -RMS: 1390 Pause: 504 ms my kilt. RMS: 853 Pause: 1167 ms And uh it's a - a great photograph. RMS: 829 Pause: 3937 ms END OF TOPIC 10 This is a photograph of my little sister sitting on Santa Claus' RMS: 1534 Pause: 246 ms knee. RMS: 1152 Pause: 464 ms taken at Woodward's RMS: 1204 Pause: 445 ms and we used kids RMS: 1112 Pause: 309 and this about RMS: 1091 Pause: 859 ms four RMS: 743 Pause: 993 ms maybe 5 RMS: 655 Pause: 995 ms and she's sitting there telling Santa what she wants for Christmas. RMS: 711 153 Monologue 5 This is a photograph of my mum one Hallowe'en. Um my parents got all dressed up to go to this Hallowe'en party. RMS: 1295 Pause: 327 ms And uh RMS: 902 Pause: 372 ms my mum's dressed up as a witch. She's got sort of RMS: 1068 Pause: 220 ms white spray-on stuff in her hair and she's got RMS: 904 Pause: 294 ms make-up on that - black and white make-up around her eyes, black lipstick and her nails are painted black and she's - she's holding a pumpkin and she's got a spider ring and a snake around her neck and she's wearing a black RMS: 932 Pause: 207 ms witch's hat with a black RMS: 679 Pause: 308 ms cloak. It's a great photograph. RMS: 1034 Pause: 298 ms She's uh - she dressed -RMS: 551 Pause: 261 ms she - it's a great costume. RMS: 716 Pause: 1872 ms END OF TOPIC 1 This is a photograph of me. standing in my back yard RMS: 1525 154 Pause: 480 ms at my old house. RMS: 735 Pause: 320 ms And I'm holding my - my baseball glove and my Softball and I'm wearing my softball hat. Looks like I'm just getting ready to go play a softball game. RMS: 1215 Pause: 1241 ms This would have been taken RMS: 884 Pause: 215 ms when I was RMS: 664 Pause: 858 ms about RMS: 687 Pause: 915 ms eleven or twelve. RMS: 915 Pause: 2252 ms END OF TOPIC 2 This is a photograph of my little sister at the tab(le) - at the dinner table. She's RMS: 1261 Pause: 429 ms dead asleep. RMS: 1384 Pause: 574 ms And when she was a little kid she used to suck her two middle fingers. RMS: 1159 Pause: 570 ms And so she's sucking her two middle fingers and she's totally unconscious on the dining - on the dining room table. RMS: 1206 Pause: 424 ms 1 And she's sitting next to my grandfather who's holding her up from falling out of her chair. RMS: 1116 Pause: 4272 ms END OF TOPIC 3 This is an old photograph. I look about RMS: 1206 Pause: 674 ms eight. And uh I'm standing next to my sister RMS: 964 Pause: 255 ms and my dad and my sis(ter) - and my mum, RMS: 851 Pause: 205 ms my aunt and my uncle RMS: 969 Pause: 558 ms and uh -RMS: 644 Pause: 759 ms and my other uncle, my dad's brother; and my grandparents. And we're all standing in the living room of my house RMS: 1129 Pause: 1124 ms uh RMS: 624 Pause: 264 ms in front of the fireplace RMS: 1138 Pause: 380 ms at the old house. RMS: 523 Pause: 239 ms And it's Christmas time 156 RMS: 765 Pause: 217 ms (be)cause uh you can tell (be)cause the stockings are hung RMS: 1252 Pause: 220 ms next to the fireplace. RMS: 1064 Pause: 4466 ms END OF TOPIC 4 This is a photograph taken in RMS: 1468 Pause: 677 ms uh - in Kelowna. RMS: 955 Pause: 537 ms My dad and my sister and I went go-cart riding. RMS: 1107 Pause: 277 ms Um. RMS: 589 Pause: 717 ms It was great fun. We got little driver's licenses made and uh put on these helmets and drove these little RMS: 1204 Pause: 293 ms cars - they look like RMS: 1101 Pause: 220 ms miniature formula ones - around the race course. Boy did they go fast. RMS: 1279 Pause: 288 ms This one was taken RMS: 1024 Pause: 3086 ms 157 I guess about RMS: 822 Pause: 271 ms nine years ago, RMS: 1054 Pause: 220 ms eight years ago. Yeah, I look about ten or eleven. RMS: 972 Pause: 645 ms And you can see Fred Flinstone in the background from I guess Bedrock City or whatever it's called there RMS: 943 Pause: 334 ms in uh -RMS: 574 Pause: 1152 ms ' in the interior. RMS: 698 Pause: 272 ms END OF TOPIC 5 This photograph was taken RMS: 999 Pause: 223 ms from - in Ottawa. And it's a photograph of my friend David and I. We're standing on the bridge that connects Hull and Ottawa. RMS: 1722 Pause: 611 ms And I'm in Hull RMS: 1632 Pause: 411 ms and my friend's in Ottawa. We're standing right bet(ween) - right be(tween) - right on the -the border there. RMS: 1554 Pause: 231 ms And I went out to visit him when I was in grade six (be)cause he moved there for a year RMS: 1210 Pause: 2631 ms END OF TOPIC 6 This is a photograph of my little sister RMS: 1825 Pause: 546 ms and uh RMS: 1764 Pause: 274 ms Ken and Susan -RMS: 1614 Pause: 415 ms my parents' friends' RMS: 998 Pause: 277 ms dog named Gucci. RMS: 1371 Pause: 484 ms and uh we're -RMS: 1138 Pause: 203 ms she's on the front lawn of my -RMS: 1401 Pause: 755 ms of their house -RMS: 975 Pause: 560 ms of their - of RMS: 722 Pause: 874 ms their house there. She looks like she's maybe a couple of years old, maybe three. She's RMS: 1212 Pause: 371 ms tiny little girl. RMS: 878 Pause: 317 ms And the dog is this - this funny little scruffy thing. RMS: 1268 Pause: 275 ms Really cute dog. RMS: 955 Pause: 3383 ms END OF TOPIC 7 This is an old photograph. It's a black and white photograph of my mother. RMS: 1032 Pause: 440 ms And um RMS: 634 Pause: 492 ms this would have been taken in sort of nineteen RMS: 1113 Pause: 330 ms seventy-eight, RMS: 1041 Pause: 1309 ms nineteen seventy-six. RMS: 857 Pause: 396 ms Um. RMS: 764 Pause: 2041 ms It's sort of a non-topic background. It's a white wall RMS: 1080 Pause: 917 ms and she's just sitting there wearing a sweater and her blouse. 160 RMS: 781 Pause: 1890 ms She looks really pretty in this photograph. RMS: 619 Pause: 3904 ms END OF TOPIC 8 This is a photograph taken of my best friend and I taken at summer camp one year. And we're sitting on a log with a - around a fire. RMS: 1347 Pause: 268 ms Um we look like we're maybe twelve. RMS: 3770 Pause: 276 ms And we're both smiling away. RMS: 1000 Pause: 2581 ms END OF TOPIC 9 " This is a crazy photograph of myself sitting at my desk at university RMS: 1423 Pause: 394 ms in my uh -RMS: 1219 Pause: 588 ms in my dorm room at the computer. I'm a - I've got this goofy face on. The computer's not even on. RMS: 1232 Pause: 351 ms I um [laughs] RMS: 1502 Pause: 639 ms but it's a -RMS: 1104 Pause: 351 ms 161 in - it's a good photograph of my desk and my work area. RMS: 1595 Pause: 717 ms And uh this was taken when my parents came to visit me RMS: 1904 Pause: 552 ms at uh university this year. RMS: 1123 Pause: 1258 ms So I'm eighteen in this photograph. RMS: 702 Pause: 497 ms I'm not sure. Eighteen, yeah, eighteen in this photograph. RMS: 728 Pause: 660 ms END OF TOPIC #10 This is a photograph of myself. Looks like I'm all set to play a soccer game. I'm wearing a soccer uniform and I'm RMS: 1509 Pause: 566 ms standing next to my dad's car RMS: 1187 Pause: 913 ms and uh RMS: 736 Pause: 419 ms I look about RMS: 652 Pause: 1703 ms seven or eight in this photograph RMS: 1097 END OF TOPIC 11 Monologue 6 This is a photograph of uh RMS: 2391 Pause: 320 ms my dad and I. RMS: 2538 Pause: 691 ms And I'm probably RMS: 1637 Pause: 746 msec about a year old, maybe two years old. RMS: 1669 Pause: 1551 ms I'm sitting on my dad's lap and we're-RMS: 1238 Pause: 1111 ms we're fishing off the back of a boat. RMS: 1205 Pause: 625 ms I'm so small, I don't -1 don't remember where or when this photo was taken RMS: 892 Pause: 267 ms or who's boat we're on. RMS: 1047 Pause: 362 ms But it's a beautiful day RMS: 985 Pause: 216 ms and uh RMS: 633 Pause: 488 ms a really pity -RMS: 1017 Pause: 790 ms pretty background. It's RMS: 823 Pause: 359 ms the ocean and there's an island RMS: 1137 Pause: 3136 ms snd the sky is a RMS: 842 Pause: 903 ms clear blue. RMS: 804 Pause: 609 ms END OF TOPIC 1 This is a photograph of RMS: 1188 Pause: 2814 ms my mum and myself and my sister. RMS: 1135 Pause: 1499 ms And I look like I'm in maybe RMS: 748 Pause: 1142 ms grade RMS: 785 Pause: 825 ms four, RMS: 802 Pause: 758 ms maybe grade five. RMS: 821 Pause: 810 ms And um RMS: 895 Pause: 1156 ms I don't know where this is taken. The background is too dark to RMS: 723 Pause: 538 ms anything or anybody. RMS: 943 Pause: 3572 ms But uh [laughs] RMS: 710 Pause: 1789 ms my sister's got this huge smile on her face. RMS: 726 Pause: 343 ms END OF TOPIC 2 v This is a photograph taken last year. RMS: 893 Pause: 238 ms It's uh RMS: 882 Pause: 254 ms in the back yard at our new house. RMS: 1510 Pause: 521 ms It's of my RMS: 923 Pause: 485 ms uncle and my cousin and my dad. RMS: 859 Pause: 213 ms And their just sitting around in the backyard drinking their -, RMS: 708 Pause: 1045 ms drinking their coffees RMS: 646 Pause: 4765 ms after dinner. RMS: 628 Pause: 226 ms And it's a pretty evening. The sunset's in the background. RMS: 651 Pause: 879 ms END OF TOPIC 3 This is an old photograph RMS: 937 Pause: 238 ms from before I was born RMS: 779 Pause: 1355 ms of my - my dad and my RMS: 889 Pause: 956 ms aunt and my uncle RMS: 547 Pause: 1131 ms and my grandparents. It's RMS: 785 Pause: 1075 ms the family photo of my dad's family. RMS: 867 Pause: 242 ms It's taken -RMS: 883 Pause: 577 ms it's taken outside (be)cause RMS: 965 Pause: 1131 ms it looks pretty cold. Everyone's wearing sweaters and stuff. Probably the fall. RMS: 710 Pause: 651 ms And there's uh RMS: 645 Pause: 802 ms trees - big evergreen trees in the background. RMS: 744 Pause: 2506 ms And it's a nice photograph. RMS: 618 Pause: 892 ms It's a neat photo. RMS: 430 Pause: 446 ms END OF TOPIC 4 This is a photo taken at the RMS: 994 Pause: 288 ms Bowen Island Festival. RMS: 1859 Pause: 1123 ms And it's of my sister and myself and my mum sitting at a table RMS: 1072 Pause: 922 ms at the pancake breakfast. RMS: 993 Pause: 916 ms um to which the Beachcombers all came. RMS: 918 Pause: 380 ms And uh RMS: 999 Pause: 208 ms the festival all started with a big parade. RMS: 1906 Pause: 529 ms And uh went to the field where they had the uh pancake breakfast RMS: 896 Pause: 389 ms and they had all sorts of competitions and RMS: 899 Pause: 313 ms it was a lot of fun. RMS: 998 Pause: 1152 ms A n d l - I -RMS: 975 Pause: 1117 ms This was taken RMS: 626 Pause: 598 ms quite a while ago. My sister and I RMS: 1039 Pause: 235 ms both look RMS: 794 Pause: 793 ms really young. RMS: 980 Pause: 212 ms I'm probably RMS: 531 Pause: 4875 ms about six. My sister looks like she's about RMS: 648 Pause: 811 ms two. RMS: 817 Pause: 450 ms END OF TOPIC 5 This is a photograph of my mom's sister RMS: 1140 Pause: 766 ms and uh - Joy -RMS: 962 Pause: 926 ms and her granddaughter, her first grandchild. RMS: 878 Pause: 641 ms My cousin got married and -RMS: 873 Pause: 864 ms a long time ago. RMS: 766 Pause: 251 ms And uh RMS: 813 Pause: 560 ms this is his -RMS: 656 Pause: 209 ms his first -RMS: 738 Pause: 863 ms first kid RMS: 1001 Pause: 1051 ms sitting with my aunt. RMS: 948 Pause: 680 ms And I can't tell where it's taken (be)cause the background is too dark. RMS: 831 Pause: 241 ms Um, RMS: 407 Pause: 4524 ms but RMS: 325 Pause: 901 ms she's a pretty baby. RMS: 731 Pause: 1758 ms END OF TOPIC 6 This is a photograph taken RMS: 1445 Pause: 312 ms at um -. RMS: 1183 Pause: 1042 ms at my dad's RMS: 1503 Pause: 376 ms friend's fortieth birthday RMS: 1400 Pause: 805 ms to which RMS: 1380 Pause: 434 ms they invited a belly dancer. So it's a photograph of this belly dancer RMS: 1317 Pause: 780 ms dancing about in the living room RMS: 1763 Pause: 841 ms of uh Mr. Cook's house. RMS: 854 Pause: 3853 ms And Mr. Cook and my dad are RMS: 1133 Pause: 845 ms standing side by side, laughing pretty hard, it would seem. RMS: 937 Pause: 1698 ms END OF TOPIC 7 This is a photograph at um RMS: 1286 Pause: 915 ms the christening of my cousin, Jordan. RMS: 1566 Pause: 503 ms My mum is holding Jordan and my dad is standing with her. RMS: 1564 Pause: 4849 ms And uh it's just outside the church by the garden. RMS: 891 Pause: 211 ms Beautiful flowers behind them. RMS: 730 Pause: 719 ms END OF TOPIC 8 This is a photograph of one of my favorite days. RMS: 1544 Pause: 466 ms Big day for me. RMS: 1393 Pause: 852 ms We all went fishing, my dad and myself and RMS: 1502 Pause: 310 ms my dad's friend Ken, RMS: 994 Pause: 397 ms and we just RMS: 693 Pause: 784 ms went out on the boat RMS: 1797 Pause: 677 ms and went fishing for a day. RMS: 1439 Pause: 252 ms And nobody caught anything except for me. I caught two rainbow trout. RMS: 1879 Pause: 1321 ms So in this photograph, I'm RMS: 1622 Pause: 1521 ms standing on the dock next to the boat proudly displaying my catch of the day. RMS: 1722 Pause: 3363 ms I look like I'm about RMS: 1059 Pause: 266 ms seven or eight years old. RMS: 1247 Pause: 994 ms END OF TOPIC 9 This is a photograph of RMS: 1320 Pause: 407 ms my soccer team when I was in - when I was in grade four. RMS: 1538 Pause: 309 ms Um, RMS: 1220 Pause: 370 ms [laughs] RMS: 809 Pause: 748 ms uh we all look so little. RMS: 1168 Pause: 1044 ms Um, RMS: 899 Pause: 861 ms we played for the South-West Marine Dolphin's Club. RMS: 1413 Pause: 438 ms And there's a bunch of us all smiling with our three coaches outside. RMS: 1089 Pause: 248 ms That was a lot of fun, RMS: 960 Pause: 916 ms that soccer team. RMS: 597 Pause: 808 ms All those - all those boys lived RMS: 1275 Pause: 916 ms within the same two blocks RMS: 1393 Pause: 797 ms 173 as me. RMS: 1165 Pause: 201 ms So we were all friends before we joined the soccer team. RMS: 1529 Pause: 3545 ms It was fun. RMS: 864 Pause: 201 ms It was a lot of fun. RMS: 940 Pause: 3545 ms END OF TOPIC 10 When I was in grade seven I got into building rockets RMS: 1390 Pause: 757 ms out of cardboard and wood RMS: 1175 Pause: 353 ms that um -RMS: 997 Pause: 363 ms that would actually fly. You could shoot them up into the air. RMS: 1207 Pause: 871 ms And so this photograph is one of my rockets. I'm proudly displaying on the stand. RMS: 1231 Pause: 525 ms It's ready for -RMS: 1500 Pause: 220 ms for take off. RMS: 2036 174 Pause: 276 ms And it's taken in the RMS: 1318 Pause: 304 ms field of McKechnie School which is the elementary school that was near my house. RMS: 1482 Pause: 1214 ms Looks like we're ready to send it off into -RMS: 815 Pause: 335 ms into orbit. RMS: 1543 175 APPENDIX C Instructions to the Participant Card 1 Thank you for taking the time to participate in this study. As outlined in the consent form, we require that all participants undergo a hearing test to determine that hearing sensitivity is within normal limits. Following completion of the hearing test, you will be asked to listen through headphones to recordings of a speaker talking about various subjects. The first recording will be presented in quiet. You are asked only to listen so that you become familiar with his voice. YOU DO NOT NEED TO DO ANYTHING BUT LISTEN AT THIS POINT. Card 2 The next recording will be of the same speaker who will now be talking about some family photos he had in front of him at the time of recording. This recording will be presented along with some background noise. Try to ignore the noise and listen to speaker's voice. Card 3 Your task will be to press the red button on the table in front of you each time and AS SOON AS you think the speaker is about to start talking about a new picture. It is important that you do this as soon as you think the speaker is going to switch to a different photograph. You will be listening to three different recordings with background noise. Each successive trial will have a more background noise. Even if you are not sure but you think the speaker might have begun talking about a new photo, press the button. If you have any questions about the study, now or after the study, please direct them to the experimenter. Note: If you are going to use your right hand to press the red button, keep your right hand to the right of button while you are waiting to respond. If you are going to use your left hand, keep this hand to the left of the button. 176 APPENDIX D Individual Subjects' Results Table D - l . Number of "Best Responses" Falling in Each Zone for the +5 dB S:N Condition Zone Inter- Post- Topic- Topic- Mid- No topic initiation final penultimate topic response Subject Sl 9 0 0 1 0 0 S2 8 0 1 1 0 0 S3 6 3 0 1 0 0 S4 10 0 0 0 0 0 S5 6 3 1 0 0 0 S6 6 1 2 0 1 0 S7 9 1 0 0 0 0 S8 7 3 0 0 0 0 S9 9 0 0 0 0 1 S10 10 0 0 0 0 0 S l l . 8 0 0 1 0 1 S12 9 1 0 0 0 0 177 Table D-2. Number of "Best Responses" Falling in Each Zone for the 0 dB S:N Condition Zone Inter- Post- Topic- Topic- Mid- No topic initiation final penultimate topic response Subject Sl 10 0 0 0 0 0 S2 10 0 0 0 0 0 S3 1 6 0 0 1 2 S4 10 0 0 0 0 0 S5 5 2 0 3 0 0 S6 3 3 2 0 2 0 S7 10 0 0 0 0 0 S8 9 1 0 - 0 0 0 S9 6 1 1 1 1 0 S10 6 1 0 2 0 1 S l l 7 0 0 1 1 1 S12 7 2 0 0 1 0 178 Table D-3. Number of "Best Responses" Falling in Each Zone for the -5 dB S:N Condition Zone Inter- Post- Topic- Topic- Mid- No topic initiation final penultimate topic response Subject Sl 6 1 1 1 1 0 S2 9 0 0 1 0 0 S3 4 5 0 . 0 0 1 S4 7 2 1 0 0 0 S5 7 3 0 0 0 0 S6 6 1 2 0 1 0 S7 9 1 0 0 0 0 S8 7 2 0 0 1 0 S9 5 1 0 3 0 1 S10 6 2 0 1 0 1 Sl 1 8 0 1 " 0 1 0 S12 8 1 0 0 1 0 179 Table D-4. Number of "Additional" Responses Falling in Each Zone for the +5 dB S:N Condition Zone Inter- Post- Topic- Topic- Mid- Total topic initiation final penultimate topic "Additional" Responses Subject Sl 0 0 0 0 3 3 S2 0 0 0 0 7 7 S3 0 0 0 0 0 0 S4 0 0 0 1 6 7 S5 0 0 0 0 5 5 S6 0 0 1 0 1 2 S7 0 0 0 0 2 2 S8 0 0 0 0 7 7 S9 0 0 0 0 0 0 S10 0 0 0 1 0 1 S l l 0 0 1 3 6 10 S12 0 0 0 ' 0 1 1 180 Table D-5. Number of "Additional" Responses Falling in Each Zone for the 0 dB S:N Condition Zone Inter- Post- Topic- Topic- Mid- Total topic initiation final penultimate topic "Additional" Responses Subject Sl 0 0 1 0 2 3 S2 0 0 0 1 6 7 S3 0 0 0 0 2 2 S4 0 0 0 0 1 1 S5 0 0 0 0 0 0 S6 0 0 0 0 0 0 S7 0 0 0 1 5 6 S8 0 0 0 0 0 0 S9 0 0 0 0 2 2 S10 0 0 0 2 2 4 S l l 0 0 0 3 8 11 S12 0 0 1 0 4 5 181 Table D-6. Number of "Additional" Responses Falling in Each Zone for the -5 dB S:N Condition Zone Inter- Post- Topic- Topic- Mid- Total topic initiation final penultimate topic "Additional" Responses Subject Sl 0 0 0 1 2 S2 0 0 0 3 9 12 S3 0 0 0 0 2 2 S4 0 0 0 1 6 7 S5 0 0 0 0 2 2 S6 0 0 0 0 0 0 S7 0 0 0 2 9 11 S8 0 0 1 0 3 4 S9 0 0 0 1 2 3 S10 0 0 1 2 5 8 S l l 0 0 1 1 10 12 S12 0 0 1 2 2 5 Table D-7. Mean of the Median Latency of "Best" Responses (msec) Signal-to-noise ratio +5 dB 0 dB -5 dB Subject Sl 1277 1260 1670 S2 962 1223 1212 S3 1916 3461 3534 S4 1254 1796 1393 S5 1797 2387 3287 S6 2219 2376 2653 S7 1606 1691 2105 S8 1941 2158 2928 S9 1741 1956 1611 S10 2096 1680 1668 S l l 641 1230 2120 S12 2049 2314 2153 Table D-8. Number of Topics Containing Multiple Responses Signal-to-noise ratio +5 dB 0 dB -5 dB Subject Sl 2 2 3 S2 5 6 6 S3 0 2 2 S4 5 i 8 S5 3 0 2 S6 2 0 0 S7 2 5 8 S8 4 0 4 S9 0 3 3 S10 1 3 5 S l l 4 7 7 S12 1 3 4 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

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"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0086923/manifest

Comment

Related Items