UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Modelling speech sound errors in aphasia : a case study Gravel, Marie Julie Sophie 1991

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

Item Metadata


831-UBC_1991_A6_7 G73.pdf [ 5.42MB ]
JSON: 831-1.0098464.json
JSON-LD: 831-1.0098464-ld.json
RDF/XML (Pretty): 831-1.0098464-rdf.xml
RDF/JSON: 831-1.0098464-rdf.json
Turtle: 831-1.0098464-turtle.txt
N-Triples: 831-1.0098464-rdf-ntriples.txt
Original Record: 831-1.0098464-source.json
Full Text

Full Text

MODELLING SPEECH SOUND ERRORS IN APHASIA: A CASE STUDY by MARIE JULIE SOPHIE GRAVEL B.Sc, University de Montreal, 1989  A THESIS SUBMITTED IN PARTIAL FULILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE  in THE FACULTY OF GRADUATE STUDIES (School of Audiology and Speech Sciences)  We accept this thesis as conforming to the required standard  THE UNIVERSITY OF BRITISH COLUMBIA August 1991 © Marie Julie Sophie Gravel, 1991  In  presenting  degree freely  at  this  the  thesis  in  University of  partial  fulfilment  of  of  department  this or  publication of  thesis for by  his  or  requirements  British Columbia. I agree that the  available for reference and study. I further  copying  the  representatives.  an advanced  Library shall make  It  this thesis for financial gain shall not  is  granted  by the  understood  head of  that  of^ A<±^&&4  (3j/  OuU^CjL&fojyy Oc^oL.  The University of British Columbia Vancouver, Canada  Date  DE-6  (2/88)  QJLK-^JUCQ)/  /</,  '/°t°ll  copying  my or  be allowed without my written  permission.  Department  it  agree that permission for extensive  scholarly purposes may be her  for  y  ^^<S^yA  ABSTRACT  The purpose of this study was to examine speech sound production errors in an aphasic subject in an attempt to determine the level of speech production processing from which such errors arose.  Speech production tasks included: 1) the Boston Naming Test; 2) picture  descriptions from the Boston Diagnostic Aphasia Examination, the Western Aphasia Battery and the Minnesota Test for Differential Diagnosis of Aphasia; 3) a story completion task; and 4) spontaneous conversation. Reading aloud was used to establish a comparison with the subject's silent reading abilities. It was hypothesized that the subject's deficit was at a phonological level; results supported this hypothesis.  The subject produced errors  characteristic of a phonological deficit at the level of what Shattuck-Hufnagel (1979) has termed the scan-copier.  iii  TABLE OF CONTENTS  ii  ABSTRACT ACKNOWLEDGEMENTS  iv  CHAPTER 1. LITERATURE REVIEW  1  The sound system  1  Phonetic versus phonemic error  3  Broca's, Wernicke's and conduction aphasia  4  Speech sound disturbance  4  Perception deficit in aphasia  23  CHAPTER 2. SPEECH PRODUCTION MODELS  32  CHAPTER 3. METHOD  46  Subject  46  Method  48  CHAPTER 4. RESULTS AND DISCUSSION  50  Relationship to previous studies on aphasic speech error data  50  Relationship to speech production models  56  CONCLUSION  69  BIBLIOGRAPHY  71  APPENDIX A  :  75  APPENDIX B  81  APPENDIX C  83  iv  ACKNOWLEDGEMENTS  Many people helped me throughout the various stages of this thesis. I would like to thank: G.F. Strong Centre for allowing A.K. to participate in my study; A.K. for her time, cooperation, and sense of humour; the aphasia group for their enlightening questions and answers; Charles for his unfailing reassurance and support; Chantal, for her sound (and clinical) advice; and John, for the tremendous support and source of confidence throughout this thesis and my years as a graduate.  1  CHAPTER 1 LITERATURE REVIEW  Aphasia is a "communication disorder ... characterized by complete or partial impairment of language comprehension, formulation and use" (Nicolosi et al., 1983, p.11) caused by lack of blood supply to the brain, an internal or external trauma to the head, or an infection. In some of the studies presented, the effects of age have been ruled out as a possible cause for the deficits. Although the site of lesion may reside in either hemisphere, I have limited the scope of my investigation to language disturbances due to lesions in the left hemisphere, since most often this side of the brain is dominant for language. As lesions are seldom confined to one area of the brain, the reality of a dissociated damage to language systems/subsystems, or a distinction between different types of aphasias, is not clear. This chapter will examine some of the disorders manifested in the speech production and perception systems of aphasics to determine whether different aphasias showing similar disturbances are indeed due to different impaired mechanisms.  I will focus on Broca's, Wernicke's and  conduction aphasics since the majority of studies included only these three types of aphasics, and because they suffice for the purpose of my research. The following definitions do not constitute facts: they serve as tools to clarify the presentation of the data.  The sound system The English sound system can be devided in various ways. One major distinction yields two classes of sounds: consonants and vowels. The production of consonants consists of obstructing the airflow coming from the lungs. This obstruction is relatively brief therefore  2 making consonants dynamic by nature. Four features further characterize consonants: the manner of articulation, in other words the degree to which the airflow is obstructed; the place of constriction in the vocal tract; the position of the velum (raised or lowered) during articulation; and the presence or absence of vocal fold vibration". 1  In the production of vowels, the passage of the airstream remains relatively unobstructed (Ladefoged, 1982). Consequently, the articulatory position of vowels can be maintained somewhat longer than that of the consonants: hence vowels are considered as static sounds. Furthermore, each vowel is distinguished by the position of the highest point of the tongue, the rounding/unrounding of the lips, and the elevation/lowering of the velum. Because of their contiguity in the speech stream, the features of sounds overlap: as a result, the articulatory features of each segment are rarely fully realized and thus they are not produced as separate entities. In other words, "by its very nature, speech production is a dynamic process and, in this sense, must depend upon maintaining appropriate timing and integration of articulatory movements" (Shinn &. Blumstein, 1983, p.91). This signifies that phonemes are simply abstract representations of the sound segments of a language, realized as articulatory gestures. If substituting a segment (consonant or vowel), for another segment that differs only by a single feature changes the meaning of a word, then these two segments are said to be phonologically, or linguistically, contrastive. 01: pat-pad > IV and 161 are contrastive pat-pet > la/ and lei are contrastive If by the same process the meaning of the word is preserved, then the segments are considered allophones.  1  02: pat-p^at > /p/ and /p*V are allophones. ^Note that the time elapsed between the release of a total airflow obstruction and the onset of vocal fold vibration is called voice-onset time.  .3 Phonetic versus phonemic error  When substitutions like those in examples 1 and 2 inadvertently occur in speech, they are often refered to as errors. According to Nicolosi et al. (1983), a phonetic or articulation error is one "in which the individual fails to plan or execute the proper motor gestures involved in the production of a speech sound" (p.16). This results in a distortion or substitution of the target phoneme for another phoneme or an allophone. The individual is incapable of correctly producing the target phoneme, or one of its features, in any context. 03: goat > coat or goat > g oat x  Contrastively, a phonological or phonemic error is one "in which the individual ... does not use the sound correctly in some linguistic productions ... yet may be able to produce (it) correctly in other syllables or words" (Nicolosi et al., 1983, p.16). 04: goat > coat, but other times goat> goat There is the difficulty of knowing at what level the substitution occurs in the speech production process; hence, the nature of articulatory and phonemic errors is a subject of dispute. If phonemic confusions occur in the presence of good agility of production and alterations in both phonemic and semantic components of language, then they are considered to be more a disorder of language than articulation, i.e., a disorganization of the linguistic system. Thus, a phonemic paraphasia would be a disturbance at a phonemic level involving only problems in the choice of number of phonemes and/or their serial integration in words (Burns and Canter, 1977), as opposed to a phonetic error, which would be a deficit at a level of articulatory planning. Apraxic errors, on the other hand, would be "characterized primarily by disturbances in word initiation, phoneme selection, and sound and syllable transitionalization or blending" (Burns and Canter, 1977, p.493), a disturbance in motor programming. MacKenzie (1982) viewed this difficulty with speech sounds not as a linguistic problem but rather as a motor speech disorder quite distinct from aphasia. Many aphasiologists associate apraxia of speech  4 to Broca's aphasia. Do different mechanisms produce the observed pattern of breakdown characteristic of each type of aphasia? Can aphasics be separated into two main groups on the basis of clinical and anatomical criteria, i.e., is there a phonetic-phonemic dichotomy?  "If this general  dichotomy is valid, (then) some evidence should be sought through the study of errors made by aphasics in phoneme production" (Guyard et al., 1981, p.19) and drawing relations with the potentially damaged systems.  Broca's. Wernicke's and conduction aphasia For the purpose of this research, I shall consider Broca's aphasia as synonymous to nonfluent and anterior aphasia. Generally, the lesion precipitating this type of aphasia lies in the third frontal convolution of the left hemisphere, and gives rise to difficulties with the sound production aspects of speech. Contrastingly, a lesion in the posterior first temporal gyrus usually results in auditory comprehension deficits, and also often phonologically and/or semantically disordered speech, two characteristics of Wernicke's aphasia, also referred to as fluent or posterior aphasia.  Conduction aphasics are also fluent although to a lesser degree  than Wernicke's. Conduction aphasia is charaterized by an impaired repetition comparatively to a good comprehension. The lesion associated with this aphasia lies in the arcuate fasciculus and/or deep in the supramarginal gyrus (Nicolosi et al., 1983). There is much debate about whether the fluent and nonfluent aphasias are in fact different, that is, whether their respective deficits lie at different levels in the speech production process.  Speech sound disturbance Many researchers have attempted to distinguish these aphasics on the basis of the disturbances found in their speech. If each type of aphasic pervasively produces a different kind of error, then this could indicate that the respective underlying impaired mechanisms  5 are different, or lie at different levels. On the other hand, if both types of aphasias show the same patterns of errors, then a common underlying deficit can be suspected. The simplest way  to examine speech sound errors is by  phonetic/phonemic  transcription. When an aphasic is attempting to say a word, the observer perceives a certain target but this target is not necessarily the end product the aphasic is attempting to produce. Thus, when trying to understand the failure, researchers need to restrict possible targets in order  to be  able to describe phonological/phonetic errors  and  explain the  closeness/differences between the actual production(s) and the target. Burns and Canter (1977) transcribed and compared the phonemic paraphasias of Wernicke's aphasics to those of conduction aphasics, in a repetition task and a naming task, to verify the conclusion common to many studies that "paraphasic speech may be manifested somewhat differently in the posterior aphasic syndrome of conduction and Wernicke's aphasia" (Burns and Canter, 1977, p.493).  They searched for factors such as type of production task,  phoneme position in word, and type of phonemic segment, that might influence the frequency of paraphasic errors. They found that the number of errors produced by their subjects was greater in spontaneous speech than repetition, increased with motor complexity of the phonemic segment type, and occurred in greater numbers with respect to place of articulation than to either manner, voicing or nasality. However, no distortions were observed in the speech errors of any of the aphasics. Whereas Shinn and Blumstein (1983; see below) found mostly single subphonemic-feature errors, Burns and Canter noticed a prevalence of two- (or more) -component subphonemic errors: the different nature of the tasks, repetition and continous speech, most likely accounts for the conflicting results of these two studies. Burns and Canter's analysis also revealed that the paraphasias of the posterior aphasics, Wernicke's and conduction alike, were compounded morphemic/semantic/phonemic errors and mostly substitutions. This could be explained by the fact that since spontaneous speech requires a  greater lexical search, errors produced in.this task can deviate more than in a repetition task. Similarly, Wernicke's aphasics, who  usually produce more semantic confusion errors,  showed greater deviations than conduction aphasics. The repetition task did not induce as many paraphasias as was  expected from conduction aphasics but more than expected from  Wernicke's, mainly on polysyllabic words. As most errors occured in final position, Burns and Canter hypothesized that "phoneme position differences reflect the manifestation of inadequate auditory projection to the motor system" (p.499), that is, a speaker would receive better auditory feedback about initial than final segments to Broca's area. They also stated that paraphasic errors cannot be solely explained as an attempt to simplify the articulatory production of words since compound substitutions (i.e., a single consonant substituted for two) were the most common type of error, as well as clusters replacing singletons. As with the lexical search pattern observed in normals, aphasics use available phonemic information about the target word, usually the initial consonant. Subjects first perform an audible search within words related in meaning, sound or both. Consequently, sound errors due to word-finding difficulty are most often word final.  If a (real) word  resembles the configuration of the target word by meaning/initial sound, it will dominate the response and may  be fused with or substituted for the target. This phenomena occurs  especially in Wernicke's aphasics who produce more phonemically/subphonemically deviated answers, that is, their productions  somewhat resemble the phonemic configuration of the  target word but has a different sound sequence, especially when the target is polysyllabic. MacKenzie (1982) tested the hypothesis "that aphasic subjects whose speech contains articulatory-phonemic errors are not a homogeneous group phonologically" (p.28-29). She postulated that two types of speech can be distinguished: apraxic and paraphasic. MacKenzie obtained very similar results to those of Burns and Canter (1977). Her naming, reading, imitation and discrimination tasks (containing distractors) revealed that  7 nonfluent aphasics showed primarily a phonetic impairment (disorder in speech sound production) with a mild phonological deficit. The main difficulty for these aphasics appeared to reside in achieving the correct positioning of the speech musculature: they seemed to grope in search of the articulatory position (or practice as MacKenzie called it) and this groping often resulted in distorted single consonants. This impediment increased with the articulatory complexity of the phonemes, i.e., it was greater in fricatives and affricates. This may be because these phonemes require a high degree of precision to be perceived as normal and even a small deviation in production interferes with correct perception.  Nonfluent subjects  usually deleted the most difficult consonant in a cluster: this would suggest, according to MacKenzie, that nonfluent aphasics were somehow aware of which phonemes involve greater articulatory demands. This awareness was also indicated by the observation that nonfluent aphasics deleted the most difficult consonant of a cluster, which in turn resulted in fewer distortions being found in clusters. Contrary to Burns and Canter (1977), MacKenzie found single component substitutions. The term "articulatory-phonemic" which she used to describe Broca's aphasics' errors is appropriate since the variation in one subphonemic component makes it difficult to distinguish whether the deficit is articulatory or phonemic. On the other hand, fluent aphasics showed no signs of articulatory difficulty nor struggle with any phoneme, regardless of articulatory complexity; in fact, only for distortions was the number of errors significantly different between the two groups. Instead, fluent aphasics produced more stuttering behaviours, they often repeated a correct response, suggesting a failure to inhibit or a lack of awareness of when a response is appropriate, possibly related to an auditory disturbance. The less severe nonfluent aphasics were aided by auditory presentation of the stimulus which suggests that the method of presentation influences their articulatory-phonemic accuracy.  MacKenzie proposed, however, that since  the less severe fluent aphasics were not aided by auditory presentation in imitation tasks, their "auditory image may be contaminated by other internal phonemic and possibly semantic  8 influences" (p.41). In other words, fluent aphasics had more difficulty in discriminating in picture choice tasks with foils one parameter away, thus demonstrating poorer semantic discrimination than nonfluent subjects. Like Bums and Canter (1977), MacKenzie observed that substitution errors occurred predominantly in final position in the fluent group, whereas for the nonfluent aphasics, difficulties arose across all word positions. In MacKenzie's study, and in Blumstein et al. (1980; see below), fluent aphasics made more errors on voiced consonants, and severely (my emphasis) disordered nonfluent aphasics on voiceless ones. The latter finding is probably due to the greater muscular effort required for voiceless segments; however, no explanation has been found for the former. MacKenzie and Blumstein et al. concluded from their respective data that the disturbed mechanism in each of these two types of aphasia are of a different order. Their inference is supported by the observation that fluent aphasics do not produce errors consistently across repetition of a task whereas nonfluent aphasics do, since their dificulty relates to articulatory complexity of segments, and that overall, they performed differently on the various tasks. Thus, there might appear to be two different underlying deficits: one articulatory, the other linguistic. The phonological disorder would usually accompany the articulatory one, as in Broca's aphasia, but the reverse is not necessarily true. One may suggest that aphasia is a disorder which affects linguistic processes at various levels: "The results of this study point to two distinct impairment patterns -one articulatory and the other linguistic- but it cannot be claimed that the non-fluent and type A (less severe non-fluent) subjects produce only articulatory errors, with no indication of phonological disruption" (p.44). Hence, it seems that fluent subjects evidenced a linguistic (phonological) impairment in the speech (and auditory) modality, and nonfluent aphasics have a phonetic deficit. Monoi et al. (1983) examined the errors of Broca's and conduction aphasics in naming and word repetition tasks. Their study yielded four findings about Broca's aphasics: first,  9 they were more likely to make errors on consonants than vowels, because of the greater articulatory complexity of the former class of sounds; second, their errors were mostly substitutions; third, the majority of errors deviated from the target phoneme by a single feature (see also Blumstein, 1973; Shinn & Blumstein, 1983); and last, all errors occurred in the same proportion on both tasks, confirming the "consistency nature" of their deficit. These data suggest the possibility of three different production disturbances located at various stages in the production system: either "true" substitution errors occur at the phoneme selection level resulting in an allophone of a neighbouring phoneme; or articulatory programming errors create phonetic distortions one feature away from the target phoneme; or, both levels may be affected thus giving rise to multiple feature errors. The latter type of these possible errors markedly predominated in the Monoi et al. experiment: in other words, complexity of articulatory gestures was directly proportionate to the number of errors in Broca's aphasics, further supporting the hypothesis that there is a speech production disturbance at the phonetic level (possibly at the stage of programming articulatory movements) accompanied by a mild phonological disorder (phonological processing). Conduction aphasics, on the other hand, made as many errors on consonants as on vowels on both tasks, suggesting that the two types of segments had the same susceptibility to be produced erroneously. This may be due, however, to the fact that the phonological structure of Japanese is different from English: in Japanese, the most common syllabic structure is CV, hence, the distribution and chance of making a production error on either type of segment is equal. These subjects made mostly substitution and transposition errors, more transpositions than substitutions especially with vowels. Monoi et al. affirm that these findings are "compatible with our contention that the disturbance of speech sound production exhibited by our conduction aphasic patients is located at the level of phonological processing" (p.189) since most of the errors they produced were involved with phoneme sequencing. In the naming task, conduction subjects made an equal number of single and two  1 0 subphonemic feature errors; in repetition, errors were mostly one feature away from the target (see also Burns and Canter, 1977). Hence, these result support the hypothesis that anterior and posterior aphasics differ not only with respect to their loci of brain damage, but also in their underlying impaired speech production mechanism. Guyard et al. (1981) submitted Broca's and Wernicke's aphasics to a delayed imitation and a naming task. Even though subjects in this study no longer exhibited the pre-therapy characteristics of their respective type of aphasia, they showed patterns that still allowed them to be distinguished. Although the aphasics' speech seemed unimpaired, Guyard et al. believed that their deficit could be revealed if a task were specially designed to amplify their respective difficulties. Thus they developed three series of triads of words: first, one in which the words had a semantic, but no phonetic, similarity; a second in which the initial phoneme of the CV word differed by one or two distincitve features; and a third containing trisyllabic words in which the order of syllabic initial consonants alternated amongst each other. With one subphonemic feature being the distance between two summits, Guyard et al. depicted a triangle: a normal speaker produces phonemes as if the features (sides) are equidistant, and respects the differences between distinctive features. Comparatively, Broca's aphasics have long and short distances, and often neglect fragile (short) features at the expence of another (long side) feature. In the Guyard et al. study, Broca's aphasics had trouble with the shorter words of the test, while Wernicke's aphasics performed worse on pollysyllables.  In short, these  researchers described Broca's aphasics as having a disturbance in motor control, being especially susceptible to coarticulatory effects in which some contrasts dominate over others. In contrast, Wernicke's aphasics would have a deficit at the pre-motor stage, i.e., a difficulty selecting one unit preferably to another thus suppressing the distinctive value of phonemic features.  11 From this study, it appears that different types of aphasics have difficulty with different types of phonological tasks: errors manifested by aphasics follow a trend, with these trends being different in each of the two types of aphasia. Mateer and Kimura (1977; in Guyard et al., 1981) stated that the difference between fluent and nonfluent aphasics is apparent in verbal and nonverbal tasks and that it reveals: at least two systems operating in the motor control of speech, one which is involved in the production of relatively discrete oral movements ... and the other operating to effect the transition from one discrete movement to another ... Presumably this latter system could also be involved in the selection or programming of the movements into longer sequences. (Guyard et al., 1981, p.28) Hence, there seems to be a dichotomy between the nature of the deficits in fluent and nonfluent aphasics, and normal phonological processing most likely is (at least) a dual system in which the two components cooperate in normal activity. In an  attempt to resolve this phonetic-phonemic error dichotomy, other  experimenters have used instrumental analyses, which are more precise than transcription (Ryalls, 1981), since the human ear may be unable to differentiate some linguistic subtleties present in the acoustic signal.  Moreover, instrumental analyses avoid the definitional  problem that is encountered when speech errors are classified. For example, what Monoi et al. (1983) consider to be a phonemic simplification, Blumstein et al. (1973; mentioned in Monoi et al., 1983) interpret as a phoneme omission. Blumstein et al. (1980) studied voice-onset time (VOT) in the speech of Broca's and Wernicke's aphasics, to see whether these subjects made mostly phonemic or phonetic errors. Essentially, their results reduplicated the ones they had obtained previously (see Blumstein et al., 1977). They noted that although the Wernicke's aphasics made few errors, they made an equal number of each type. It appeared that Wernicke's "phonetic" errors reflected an attempt to prevoice voiced consonants which, according to Blumstein et al., could represent either an attempt to produce a more distinct voiced-voiceless contrast or an experimental artefact, rather than an articulatory difficulty, since none of the prevoiced segments were  12 phonetically distorted and the distribution of VOT was non-overlapping.  If the Wernicke's  aphasics were trying to produce a more distinct voiced-voiceless contrast, then we would suspect a deficit in the selection of the correct phoneme of underlying phonological form, instead of a difficulty in articulatory programming. In spite of their ability to maintain the normal voiced-voiceless distinction, these aphasics most often produced phonemes having a VOT characteristic of the opposite category to that of the target phoneme; hence it must be concluded that their errors were of a phonological variety. Conversely, Broca's aphasics made significantly more phonetic than phonemic errors: their VOT did not result in two distinct categories, rather, the values sometimes fell either between or outside the normal boundaries for voiced and voiceless segments (see also Shinn & Blumstein, 1983). Since both types of errors made by these subjects occurred inconsistently across place of articulation, it is unlikely that a single mechanism is involved in speech production (Blumstein et al. 1980). Blumstein et al. noticed that the phonemically correct productions of the Broca's aphasics covered a wide phonetic continuum: these subjects produced few prevoiced consonants and long VOTs for voiceless segments which suggests a difficulty in initiating vocal fold vibration. Together with the frequent phonetic errors, this finding leads to the conclusion that Broca's aphasics demonstrate a pervasive phonetic disorder (see also Shinn & Blumstein, 1983), and not a phonological one. lt:does not seem likely that the Broca's aphasics' mechanism is the same as Wernicke's since their respective phonological errors were on different phonemic categories. Altogether, it would appear that if Broca's aphasics produce both phonemic and phonetic errors, then their deficit encompasses phonemic selection/organization and articulatory implementation, and it would be phonetic and phonological in nature, unless their phonological errors are extreme phonetic distortions. Broca's aphasics' phonemic errors can be mistaken for phonetic ones because their entire speech is "colored" by their phonetic deficit. If their phonemic errors are in fact extreme distortions, then do they originate at the  13 same level as Wernicke's aphasics errors , that is, do they result from a deficit at the same stage of the speech production process? Most important in the study of Blumstein et al. (1980), all types of errors were produced by all types of aphasics, i.e., all groups of aphasics evidenced some deviations in timing of articulatory movements and phonemic errors. Three explanations may be valid: different types of aphasias simply have a different degree of accuracy in their speech production system; the two different types of errors are due to two different impaired mechanisms; or hypothesis one applies to some aphasias and hypothesis two applies to others. If aphasics make errors on the basis of whether a consonant is voiced or voiceless, then the errors cannot reflect a common underlying deficit. If only one mechanism was involved, then errors would occur consistently across all types of phonemes: since a dissociation was also observed  in the distribution of phonemic/phonetic errors with respect to place of  articulation, then more than one mechanism must be involved in speech production. Voicing requires timing between two articulatory movements: the release/passing of the airstream and the initiation of vocal fold vibration. Some experiments on the speech production mechanism looked at articulatory controls that do not involve timing relations. Ryalls (1981) examined phonetic errors in vowel production of motor aphasics. Spectrographic analyses revealed that the values of the formants of certain vowels in the aphasics' speech varied more than for normal speakers: Ryalls interpreted this as "evidence (of) a problem in phonetic outputting, i.e., the patient is unable to maintain careful control of his articulatory mechanism to produce a stable phonetic output" (p.368). Moreover, Ryalls observed differences in the mean of the formant frequency of certain vowels, between aphasics and normals: in his opinion, this would be due to a change in the internal representation of the phonemic target, i.e., a phonetic disintegration of the target the aphasic is setting himself, thus leading to a change in the mean formant value actually produced. Unfortunately, Ryalls did not expand on this notion or its significance.  14 In short, Ryalls described the output deficit in motor aphasia as possibly arising at one of two levels: a higher order one that involves linguistic processes (phonemic target), or a lower order one in which articulatory realization takes place (phonetic outputting). In other words, a motor aphasic can have a language deficit, i.e., a difficulty in the abstract encoding at either of the various linguistic levels (semantic, syntactic and/or phonological), or a segment production deficit, i.e., a difficulty in implementing motor commands (motor outputting), both being different from a phonemic selection deficit. In the second part of his study, Ryalls acoustically analysized duration of words and sentences in the speech of the motor aphasics, and found a consistently significant difference in comparison to normal speakers. He remarked that aphasics slowed their speech not simply by increasing pauses but also by modifying factors at all other linguistic levels. This finding lends support to the classical description of the motor aphasics' speech as being slow and laborious. Shinn and Blumstein (1983) investigated whether the speech deficit of Broca's aphasics was reflected in spectral or in temporal cues.  "If anterior aphasics evidence  primarily a timing deficit, while maintaining the ability to reach the target configuration for the appropriate place of articulation, then the spectral characteristics of the production at the moment of release should be similar to normals." (Shinn and Blumstein, 1983, p.93). If the aphasics' templates were different from those of normal speakers, then the analysis of these differences might shed some light on the nature of the speech deficit in Broca's aphasics. To verify this hypothesis, Shinn and Blumstein measured the moment of consonantal release in the speech of Broca's and Wernicke's aphasics. They observed that Broca's aphasics could reach the appropriate articulatory configuration for place of articulation, although the subjects had some difficulties achieving target values for alveolar consonants.  Since labial  and velar productions were within normal limits, Shinn and Blumstein attributed the difficulty with alveolars to the greater demand of motor control involved in alveolars. These  15 observations suggest that the overall static aspects of speech production are relatively well preserved in Broca's aphasics. The phonetic distortions found in alveolar consonants were evidenced by a loss of high-frequency energy. Shinn and Blumstein imputed this distortion to the laryngeal deficit that is often present in Broca's aphasics, and maybe a lack of articulatory control that would impede total articulatory closure. If a good closure is not achieved, then the burst of the release will contain less high energy since the two cavities will not be separated by the closure; thus, the spectrum of the burst in alveolars will be changed, in voiced and voiceless segments alike. This loss of high-frequency energy is consistent with the breathy/lowfrequency energy characteristic of Broca's aphasics' voice. If one can explain the errors in terms of vocal tract configuration, then it is more likely that the deficit is phonetic than phonemic. The selective occurence of this deficit on alveolars shows that this is not a pervasive across-the-board disorder. A deficiency in laryngeal control, i.e., a timing/dynamic deficit of speech production, in Broca's aphasics was further supported by the greater number of voicing errors than place errors, (in contrast to the Wernicke's aphasics; see also Blumstein et al., 1977a), and the loss of the voiced-voiceless distinction. On the whole, whereas dynamic aspects of the speech signal evidence an impairment, the static aspects of speech production seem to be mostly preserved  in Broca's aphasics.  "the deviations which occurred are explainable by  modifications of the vocal tract configuration, problems with place of articulation seem to reflect more nearly phonetic, i.e., articulatory implementation, rather than phonological, i.e., phonological planning errors" (p.112). In Shinn and Blumstein's study, no subject showed two-feature errors; moreover, all phonological substitutions were phonetically correct, more voicing than place errors were found in Broca subjects' speech, and vice versa for Wernicke's aphasics. On the basis of these results, we would hypothesize the presence of a phonological planning disorder rather than an  16 articulatory implementation deficit. Both types of aphasics had the same percentage of phonemic paraphasias, thus contradicting other findings (Blumstein et al., 1977a), although Shinn and Blumstein had a small number of subjects. Also, the nature of the reading task probably left less opportunities for the Wernicke's aphasics to produce phonemic paraphasias than they might have in a spontaneous speech task. In the second part of their study, Shinn and Blumstein (1983) investigated whether "the acoustic characteristics of a perceived [t] whose target was a [p] to be different from a perceived [t] whose target was in fact a [t]?" (p.100).  They submitted the aphasics'  utterances to a template fitting/rejection (expected configuration of burst).  When an  utterance was perceived correctly, the aphasics had produced appropriate cues for perception of place of articulation, except for alveolar consonants produced by Broca's aphasics. An utterance which was perceived as the correct target but did not match the template accurately was most likely to represent a phonetic distortion, otherwise it would not have been perceived as the correct target phoneme. A planning/selection error could be detected by a template analysis of the actual phoneme produced: the phoneme produced would fit the template of the phoneme perceived, but not that of the target phoneme. To verify the phonetic-phonemic dichotomy hypothesis, Katz (1988) examined another timing factor, anticipatory coarticulation, which reflects some measure of the preplanning of upcoming segments in the speech stream. "Coarticulation may be defined as a dynamic process in which the integrity of presumed segments is obscured at the articulatory level.  In anticipatory, or "right-to-left" coarticulation, a speaker may initiate the  production of selected features of subsequent speech sounds in advance of their target attainment" (p.341). Previous studies had obtained results suggesting that nonfluent and verbal apraxic aphasics had a deficit in anticipatory coarticulation. Katz investigated whether anterior aphasics showed sound-transitional deficits in addition to evidencing timing problems at the phoneme production level. If anterior aphasics evidence such a deficit, then  17 this would tell us much about the role of Broca's area in speech production. If these aphasics do not show this impairment, then this would suggest that some timing capacities are retained regardless of the cortical damage to the pre-Rolandic area, contrarily to results obtained in other studies. To resolve the dichotomy, Katz also investigated whether Wernicke's aphasics demonstrated a deficit in anticipatory coarticulation, since some phonetic deficits had been found in the speech of these fluent aphasics (see also Blumstein et al., 1977a). The subjects, anterior and posterior aphasics, read or repeated monosyllabic (C)CV sequences of which only phonemically correct tokens were examined in terms of the distribution and intensity of their spectral peaks. In analyses of late and early regions of the aperiodic waveform of the consonants, there was evidence of coarticulation, indicated by acoustic correlates which were identical to those of normal subjects.  Hence, this suggests that aphasics, anterior and  posterior alike, have retained coarticulatory planning processes for CV- and CCV-size utterances, but contradicts the finding that anterior aphasics have a timing deficit in production. gestures  Rather, it would appear that the implementation of planned coarticulatory  is only partially affected in anterior aphasics.  Conversely, anticipatory  coarticulation is intact in posterior aphasics and hence does not accompany other phonetic deficits found in this population by Blumstein et al. (1977a).  Katz postulated that the  discrepancy between his results and that of other studies may be due to individual differences or subject classification difficulties, that is, subjects could have divergent underlying bases for their speech deficits which would be reflected in differing degrees of coarticulatory ability. In the second part of Katz's experiment, normal subjects had to identify the vowels excised form the (C)CV sequences produced by normal and aphasic speakers in the first part of the experiment. The listeners were able to accurately and consistently identify all tokens. This result suggests that planning and implementation of anticipatory coarticulation are intact  1 8 in anterior and posterior aphasics, as was also demonstrated by the acoustic data. However, identification scores of anterior aphasics' utterances were lower, agreeing with the general finding that the speech of these aphasics is nonfluent and poorly articulated, causing some of their productions to lack in coarticulatory information. Conversely, anterior aphasics were able to reach the appropriate place of articulation for stop consonants, as in Shinn & Blumstein (1983).  In accord with their difficulty in producing voiced consonants, these  subjects experienced difficulty coordinating two different articulators: for example, in [tV], the sequence requires simultaneous programming of the tongue and lips. Moreover, the anterior aphasics had greater difficulty with [t] because of the fine tongue blade movement needed and also because stops demand a relatively more rapid transition to the vowel. Thus, in describing the articulartory deficits of anterior aphasics, one must consider timing and motor information. On the other hand, the identification of the posterior aphasics' utterances was similar to that of normal speakers, and not that of the anterior aphasics.  Hence, the  phonetic-phonemic speech production distinction is supported by the findings of this study. Assuming that the anticipation of the features of an upcoming vowel is a process distinct from phoneme selection and ordering, the finding of reduced perceptual coarticulatory information in the pre-vocalic consonants of anterior aphasics' productions suggests that these subjects have a "phonetic" outputting deficiency. Similarly, the fact that posterior aphasics pattern very closely to normal subjects suggests that both "phonemic" and "phonetic" levels of processing are intact in these subjects. This finding also suggests that the "subtle phonetic deficit" noted to exist in posterior subjects' speech (e.g., Blumstein et al., 1980) does not extend to anticipatory coarticulation processes. (Katz, 1988, p.361-362) Anticipatory coarticulation has been considered an index of the size and extent of speech planning "units". It is measured in terms of the amount of acoustic change in an early part of a target utterance as a function of the rounding feature of a subsequent vowel. In an analysis of acoustic cues for coarticulation in the speech of anterior aphasics compared to that of normal speakers, Katz et al. (1991) used electromagnetic articulatography to analyze trisyllabic words contrasting minimally in the rounding feature  1 9  of the vowel of the middle syllable. Words were imbedded in a carrier phrase, and speech errors were eliminated from the analysis. To avoid any controversy, Katz et al. obtained not only "indirect" (acoustic and perceptual) data, but also "direct" (kinematic or EMG)  data in  another study. Moreover, they correlated these two forms of evidence to investigate whether a given acoustic attribute reflects the involvement of more than one articulatory gesture. The analysis consisted of looking at the temporal and spectral properties of productions, as well as segment duration; the latter factor was examined since degree of coarticulation overlap varies with speaking rate. The absolute duration of the anterior aphasics' consonants, syllables and mostly their vowels was greater than that of normal speakers. Unfortunately, few of the acoustic effects reached statistical significance, and the number of subjects was very small. Katz et al. were not able to determine the time course of coarticulation as precisely as in their kinematic study, but nevertheless, the analysis of the spectra preceding the target vowel allowed them to determine that the anterior aphasics showed similar acoustic cues and patterns for anticipatory lip rounding and in the same spectral regions as do normal speakers, even though at slower rate of speech. According to the kinematic study by Katz et al. (in press, mentioned in Katz et al., 1991), aphasic subjects began lip rounding as early in the word as did the normal speakers although it appeared that one of the aphasic subjects did not show the same degree of acoustic cues for anticipatory labialization in the acoustic part of the study. This may be due to the fact that this aphasic's productions were highly variable in terms of target formant frequencies and distribution of spectral peaks; hence, this imprecision added "noise" to the acoustic measures and made it difficult to acoustically analyze anticipatory coarticulation. Coarticulatory shift was noted in the acoustic analysis of the aphasic's productions, though not consistently in all target vowels, again indicating that an acoustic analysis may "miss" certain effects which are detected through a kinematic analysis. For example, the  20 consonant [I] preceded the vowel in some of the stimuli: it has a prominence in the low frequency regions thus possibly masking the (low frequency) acoustic cues for the anticipatory coarticulation of certain vowels. Since Katz et al. noted as many acoustic cues for anticipatory coarticulation in the segment preceding the vowel as in the segment preceding the segment preceding the vowel, they hypothesized that there may be more acoustic cues to anticipatory labialization than the ones they examined (e.g. other formants). The Katz et al. kinematic study showed that anterior aphasics' coarticulatory patterns are more highly variable than those of normal speakers, especially with respect to spatial positioning of the articulators, but that their timing of coarticulation was comparable to that of normal speakers. This constitutes tentative counterevidence to the claim that uniform delays in coarticulation are a likely property of anterior aphasic speech. If it is the case that highly automatized behavior such as anticipatory labialization remains intact in adults presenting with massive damage to anterior brain structures, then neurolinguistic models implicating specific anterior regions as general "speech programming" centers will certainly require revision. It may instead be more realistic to consider these cortical regions as being critically involved in the discrete timing propeties of individual phonemes but not necessarily in "programming" coarticualtion of a phonological string. (Katz et al., to appear, quoted in Katz, 1991) Based on the observation that many studies suggest most aphasics have an impairment in production of temporal parameters, Baum et al. (1990) devised an experiment to investigate whether anterior and posterior aphasics exhibit deficits in temporal coordination, and if they do, whether these deficits are the result of the same impaired underlying mechanisms. Their study included a detailed analysis of lesion data (to study more closely any possible neuroanatomical correlation with speech production deficits), measures of VOT, intrinsic and contrastive fricative duration, and intrinsic and contrastive vowel duration in Broca's  aphasics with anterior lesions (group  A), nonfluent aphasics with an  anterior-posterior lesion (AP), and fluent aphasics with a posterior lesion site (P). In most studies examining phonetic characteristics of aphasics speech, it has been found that nonfluent aphasics show an overlap in VOT values, but fluent aphasics typically  21 show little or no overlap; at times, posterior aphasics even "exaggerate" the spread of the category boundary. Both types of aphasics, however, have been shown to frequently produce the phoneme of the opposite category to that of the target. The findings of Baum et al. replicated those of earlier studies: the posterior aphasics' productions showed two well-separated VOT  categories, the nonfluent subjects had significant overlap in their  distributions, and two of three anterior (Broca's) subjects also had overlap. Interestingly, one of the anterior aphasics studied showed two distinct categories with a very clear-cut category boundary region containing no productions at all. In their investigation on intrinsic duration of fricatives, Baum et al. found that the aphasic subjects showed normal proportions in terms of which voiceless fricative is shorter/longer, but the absolute duration of these segments proved to be respectively greater/shorter than those of normal speakers. Conversely, the C/V duration proportions of aphasics were similar to those of normal speakers, i.e., there was not a significant difference between the two groups. When comparing duration between voiced/voiceless segments produced by aphasics, Baum et al. did not find the expected pattern of voiced fricatives being shorter than the voiceless, but then again, normal speakers did not show that trend either. Vowel duration in aphasic productions followed the same trend as for normal speakers: tense vowels were longer than lax ones, hence they were of normal intrinsic duration. In some syllables, the anterior aphasics produced vowels longer than those of both posterior and anterior-posterior aphasics, as was found by Ryalls (1981; see above). Baum et al. did not know why AP subjects did not also show longer vowels since they share some a common lesion site with the A subjects. A close examination of the extent of neurological damage revealed that subjects who demonstrated deficits on the production of VOT all had lesions in Broca's area, the anterior limb of the internal capsule and the lowest motor cortex areas for tongue and larynx. Although Baum et al. did not claim that VOT is "located" in any of those areas, they postulated that those  22 sites must be intact if a speaker is to accurately produce that parameter, and possibly other features which require the integration of articulatory movements. This hypothesis is further supported by the observation that subjects which had no lesion in anterior regions displayed normal VOT values. The P aphasics did show some subtle speech errors, thus suggesting that areas other than Broca's, the anterior limb of the internal capsule and the lowest motor cortex areas also contribute to normal speech production. This in turn suggests that the different aphasias are most likely to result from different mechanisms of impairment. Baum et al. did not, however, make any proposals as to the role of these other neuroanatomical areas in speech production. "The constellation of impairments for the anterior aphasics including both the A and AP patients suggests that their disorder primarily reflects an inability to implement particular types of articulatory gestures or articulatory parameters rather than an inability to implement particular phonetic features" (Baum et al., 1990, p.52). If the disorder had been due to a deficit in implementing a phonetic feature, then it would have been observed more pervasively across all acoustic parameters that contribute to the marking of that feature.  Two  observations evidence a deficit involving only temporal coordination: first, A and AP subjects had difficulty temporally coordinating the release of closure/constriction with the onset of vocal fold vibration and other segments involving laryngeal control. Second, the anterior aphasics were able to produce contrastive differences, that is, they were able to distinguish final voiced and voiceless stop consonants on the basis of the preceding vowel duration (which does not involve temporal coordination). Hence, the anterior aphasics' impairment lies in an inability to implement some of the parameters that indicate a contrast and not in marking the functional nature of the contrast itself. Thus, the disorder of these A and AP subjects is phonetic and not phonological in nature. The results of Itoh et al. (1986; see below) also demonstrated a difficulty in temporally integrating two articulatory movements. Since posterior subjects exhibited impairments in the production of intrinsic and  23 contrastive fricative duration, Baum et al. concluded that these aphasics have a subtle phonetic impairment. However, the nature of the impairment in these aphasics would be different from that of A or AP subjects, since P' aphasics did not show any abnormal patterns of laryngeal control. Moreover, P subjects evidenced increased variability in their productions. Baum et al. suggested that one of four factors might be precipitating the P aphasics' deficit: 1) impaired control of temporal parameters; 2) impaired structures involving global durational patterns of speech; 3) projections from posterior brain areas to motor/articulatory systems being damaged; and 4) the auditory feedback system for the control of normal articulatory parameters of speech being damaged. Regardless of which of these factors is correct, these posterior aphasics had no lesions in the areas of the brain typically implicated in speech production, yet still they evidenced a phonetic deficit. The findings of Baum et al. complicate the dichotomy between A and P. It appears from their study that the neural implementation of speech involves more than the perisylvian area previously thought to be exclusively involved in normal speech production: their observations suggest that posterior areas are also possibly critical involved in normal speech production.  Perception deficit in aphasia In normals, there are assumed to be at least two levels of processing in speech perception: the first is based on functioning of a set of property detectors, a second is seen as using these properties for linguistic processing. As with speech production, researchers have also attempted to determine whether different perceptual mechanisms are impaired in different types of aphasics showing perceptual deficits. Basso et al. (1977) were among the first researchers to investigate the extent to which "the ability to identify the acoustic cues that underlie the phonemic distinctions of spoken language is impaired in aphasia" (p.85). They also assessed whether auditory comprehension is affected by a phonemic identification defect in fluent and nonflueni aphasics.  24 Phoneme identification of synthetic speech was normal in only a third of the aphasics they studied; for the remainder of their subjects, the boundary zone could not be determined. Almost all of the nonfluent aphasics had a phoneme identification defect, although this was observed in only two thirds of the fluent aphasics. In conjunction with results of other studies which have examined the major deficits in Broca's and Wernicke's aphasics, this latter finding suggests that poor comprehension is not due to poor phonemic discrimination. On the other hand, the observation indicates that most of the subjects with phoneme identification defect also have a disorder of phonemic output, i.e., they produced phonemic paraphasias. Blumstein et al. (1977a) had Broca's and Wernicke's aphasics (the same who participated in the production study) discriminate and label synthetic speech sounds with various VOTs. The only consistent pattern found was in Wernicke's aphasics: they could discriminate but not label stimuli consistently, a perceptual difficulty which Blumstein et al. attributed to an inability to use sounds contrastively. Some aphasics retained the ability to discriminate between phonemic categories but had lost category labels, i.e., they could not maintain a stable configuration of phonemes and use phonological information in a linguistically relevant way to then comprehend the message. Some Broca's aphasics could both label and discriminate, others could only discriminate, and some Broca subjects could do neither. In this experiment, it was clear that the perceptual ability of anterior aphasics did not relate to their production ability, i.e., they were able to distinguish voiced consonants from voiceless ones, but still made phonemic and phonetic errors when producing either category. Blumstein et al. also observed that language comprehension abilities do not correlate with the ability to perform VOT perception tasks, since some aphasics who showed very good comprehension were unable to label and/or discriminate phonemes. The performance of aphasics with poor comprehension on the labelling and discrimination tasks showed great individual variability.  This result may be attributable to the synthetic nature of the stimuli  25 and its lack of coarticulatory and situational cues. On the basis of their findings, Blumstein et al. suggested that perception is the integration of two levels of processing: a first basic prelinguistic level in which selective differences between auditory stimuli may  be  discriminated; and a second linguistic level in which the acoustic categories from the first level are used linguistically, i.e., this is the level at which labels are assigned. By integrating these two levels, a listener establishes the relation between sound and meaning, which then allows understanding of linguistic labels and ultimately the linguistic system of a language. In another experiment, Blumstein et al. (1977b) investigated the notion that difficulties in phonemic discrimination might account for comprehension deficits in Wernicke's aphasics.  "If the phonological form of a word or utterance is incorrectly  perceived, its contribution to the semantic or syntactic content of the message may  be  erroneous" (p.19). Their experiment comprised three tasks, which required discrimination between pairs of real words, and between pairs of nonsense words, with similar phonological relations. The first phoneme discrimination task comprised one and two syllable words in which the initial consonant of the first or second syllable differed with respect to voicing, place of articulation, or voicing and place. In the second task, syllable discrimination, pairs of bisyllabic words were constructed so that they varied with respect to the unstressed syllable, but having the same stressed syllable. The phoneme order discrimination task was composed of mono- and bisyllabic words differentiated by the order of the consonants (e.g. "name-main", "tax-taks"). As had been found in other studies (Basso et al., 1977; Blumstein et. al., 1977a), there did not appear to be a correlation between comprehension and discrimination since the Wernicke's aphasics, whose comprehension was  poorest did not perform worst on  discrimination tasks. Broca subjects made fewer errors with nonsense words than with real words, whereas Wernicke's aphasics made twice as many errors on nonsense words. Broca's aphasics performed quite well in discrimination tasks involving place and voicing contrasts;  26 Wernicke's aphasics, however, had more difficulty with the place feature, and did not perform as well as Broca subjects on either task. The Wernicke's aphasics' greater difficulty with place may be due to the rapidly changing nature of the cues that indicate this feature; VOT, in contrast, depends on a timing relation.  From this finding, Blumstein et al. (1977b)  hypothesized "that the decoding of formant transition patterns underlying place contrasts imposes a particular demand on the left cerebral auditory association area" (p.28), and not on the pre-Rolandic area.  Although it appears that there is a link between temporal lobe damage  and phonemic discrimination, all types of aphasics had a phonemic perception disturbance, varying in degree of severity. Nonetheless, this study suggests that the the perception deficit observed in Broca's aphasia is of a different nature than that observed in Wernicke's aphasia. In a later study, Blumstein et al. (1984) asked if lengthening the transition (in synthetic stimuli) between a consonant and a vowel would help perception of place of articulation of stop consonants, and whether labelling depends on discrimination abilities for place of articulation, in the same way that it does for VOT. In the perception tasks, lengthening the transition helped to minimally improve the performance of Broca's, Wernicke's and conduction aphasics. On the other hand, the additional cues, which enhanced the saliency of the transition, helped all the subjects significantly: if subjects were already able to perform the task, the added cues helped, but if the subjects had been unable to perform the task, they remained unable to perfrom the task regardless of added cues. As in the previous experiment of Blumstein et al. (1977a) in which stimuli were used which varied with respect to VOT,  aphasics could label stimuli differring in place of  articulation only if they could also discriminate them. This finding supports the view that discrimination underlies the ability to use information in a linguistic way (Blumstein et al., 1977a; 1984) and that a subject's ability to discriminate tells us more about his percetual ability than labelling does.  Again, the perception of place of articulation requires the  integration of acoustic events over a relatively short period of time as the articulators move  extremely rapidly from the consonantal release to the following vowel; hence, it requires more rapid processing than the perception of cues like VOT.  Many aphasics had great  difficulty perceiving the cues for place of articulation in stops consonants. Contrary to results obtained in experiments with VOT stimuli, Wernicke subjects were able to label and discriminate at least some of the stimuli. Also, in contrast to results with VOT stimuli, Broca's aphasics showed a dissociation between labelling and discrimination for place of articulation. The difference in performance between these two types of aphasics provides more evidence for the observation that their respective deficits are different in some subtle ways. From these results it may again be seen that there is no relation between auditory comprehension and the ability to discriminate/label (computer simulated) segments: comprehension of language relies not on an analysis of the acoustic waveform but rather on the parsing of both phonological and syntactic grammars.  Speech perception tasks using  monosyllables contrasting in only one phonetic dimension are not as sensitive a measure of auditory comprehension as tasks using acoustic information greater than the syllable. This was demonstrated by the observation that the Wernicke's aphasics who had the poorest auditory comprehension could label and discriminate the stimuli, whereas the Wernicke's aphasic with the highest auditory comprehension score could neither discriminate nor label the stimuli. The fact that no perceptual shifts were obtained for the discrimination or labelling functions compared to normals, even in those patients who could not label a place of articulation continuum, underscores the stability of the categorical perception phenomenon in the face of brain damage and supports the view that the ability to discriminate categories of speech underlies the ability to use them linguisticallu for labelling. (Blumstein et al., 1977b cited in Blumsteinet al., 1984) In a discrimination and labelling task of computer simulated velar stop cognates, Itoh et al. (1986) found that Broca's and Wernicke's aphasics in general (most other subjects being amnesic aphasics), had difficulty perceiving/identifying VOT,  and in using other  28 information about the acoustic properties and phonetic features of speech stimuli. Although no relation was found between the type of aphasia and errors in VOT perception, Wernicke's aphasics performed less well than Broca subjects. Since, in the study of Itoh et al., Wernicke's aphasics were able to discriminate but unable to label, thus evidencing an impairment at the linguistic/labelling level, Itoh et al. questioned the way in which Blumstein et al. (1977a) interpreted their results. First, they described the criteria used by Blumstein et al. as ambiguous and quantitatively imprecise, and the data as inappropriately interpreted. Blumstein et al. had based their interpretation of results on peaks of discrimination and phoneme boundaries of aphasics compared to those of normal speakers, without having defined the values for normal controls. Itoh et al. suggested that the values obtained by Blumstein et al. might indicate categorical perception by Wernicke's aphasics, whereas Blumstein et al. had concluded otherwise. Itoh et al., however, did not verify this suggestion. Second, Itoh et al. disputed the statement of Blumstein et al. that labelling is selectively impaired, since some of the subjects described by Blumstein et al. failed to both discriminate and label. Blumstein et al. meant this assertion to apply only to Wernicke's aphasics who only failed in the labelling task. To understand the level of the comprehension/processing deficit, i.e., whether it lies at the phonetic or phonemic level, or "higher", Itoh et al. analyzed their data following a model of speech perception developed by Tatsumi et al. (1978; 1980). This model proposes that speech first undergoes an acoustic analysis, during which the acoustic properties of speech sounds are analyzed; this first stage constitutes a prelinguistic stage of analysis. Temporal and spectral patterns, such as the timing relationship of which characterizes VOT in stop consonants, are analyzed, but the system can only perceive differences equal to or greater than 25 msec. This temporal limitation creates the perception of two distinct categories, each with a reversed order of successiveness between voicing onset and burst. At the second stage, which is linguistic, "the phonetic features of speech stimuli are extracted with reference to the  29 information in long term memory" (Itoh et al., 1986, p.76), and phonemes are identified by the assignment of a category label based on the phonetic features that have been extracted. Finally, the identified phoneme is expressed through a response mode whereas stages 1 and 2 process continous information, stage 3 and 4 process discrete/categorized information which is more robust in noise, especially in normal subjects. Both types of information are retained in long term memory. Any noise present at stages 1 or 2 will affect the accuracy of identification, especially if a stimulus is near the category boundary. "Because the acoustic properties and phonetic features of a given stimulus may not necessarily be firmly connected or associated with a category label in these subjects, their responses will fluctuate even when they deal with discrete information" (p.77). If no noise is present at stages 1 and 2, then an incorrect assignment of label (stages 3 and 4) may be due to: 1) a failure to read all possible foils, 2) a literal paraphasia, 3) perseveration from the previous response, or 4) the presence of noise at stage 3. The aphasics showed decreased ability to analyze timing relationships and to utilize other information about acoustic properties, thus suggesting an impairment in the acoustic and/or phonetic stages of auditory processing (stages 1 and 2). Since their confusion rate was also greater, this was interpretated as indicating a confusion at the phoneme identification stage, that is, an inability to assign a category label to a set of phonetic features for a given stimulus. Using a number of different tasks, Baker et al. (1981) "attempted to understand more fully the auditory processing abilities of Wernicke's aphasics in relation to phonemic discrimination, on the one hand, and semantic decoding, on the other" (p.2-3). Their first task involved judgeing whether two auditorily presented words were same/different, thus requiring no need for semantic processing from the subjects. Their second task involved a judgement of same-different of auditorily and pictorially presented words, and required only minimal semantic processing. In a third task, the aphasics judged similarities of auditorily presented words in a four-foil picture choice paradigm.  30 Wernicke's aphasics made significantly more errors than Broca's, no subject performed randomly, and few errors were made on "same" responses.  Rather, both groups  made significantly more errors on "different", although Wernicke subjects made more than Broca's, especially when the errors were two features away. Broca's aphasics had shorter response times than Wernicke's aphasics in all conditions, and all subjects' response times were quicker when the correct response was "same". Broca's aphasics had similar latencies for both conditions whereas Wernicke's aphasics took significantly longer when both words were auditorily presented and different. The two groups did not show a significant difference in latency for the "same" conditions.  Wernicke's aphasics  latency varied with  feature/number of feature, and with respect to same/different condition; this was not so for any Broca subjects. In the second task, Wernicke's aphasics took significantly longer than Broca's to make "same" judgements, thus showing increased processing time for auditory decoding. In the third task, "If poor auditory comprehension in Wernicke's aphasia were caused by a disturbed auditory analyzer it would have been predicted that the preponderance of errors should have been to the phonemic foil"  (p.10),  but in fact, Wernicke's aphasics made more semantic  errors indicating that they were operating in the appropriate domain. The Wernicke subjects' deficit cannot be purely phonological since the introduction of semantic mediation required longer processing time and they made more errors on semantic than phonemic contrasts. Thus, the Wernicke's aphasics impairment implicates meaning (or semantic processing) as well as phonological processing. Their deficit seems to lie in the relation between the phonological structure of a word and its semantic representation, one suffering at the expense of the other when demands or confusion possibilities increase. In other words, Wernicke's aphasics seem to evoke the phonological structure of a word before they make a semantic judgement. Hence, their deficit is not a low-level one: it seems to be at a semantic level and influenced by phonological factors.  31 Overall, Wernicke's aphasics appeared to be inferior to Broca's in same-different judgements of phonemic contrasts. Broca's aphasics made the same number of errors on each type of contrast and responded faster than the Wernicke's.  The results of this study thus  suggest that these two types of aphasias have a different underlying impairment.  The aims of the research reported in this thesis were: first, examine speech production errors of an aphasic; second, determine the level at which these errors arise; and third, attempt to explain why this aphasic uses certain strategies in attempting to overcome difficulties in speech production.  32  CHAPTER 2 SPEECH PRODUCTION MODELS  The mental grammar is the internal knowledge that enables an individual to be a competent speaker/listener. An adequate speech production model must explain the processes that take place when a speaker, using that knowledge, produces a sentence in a form that another speaker of the language can understand.  Like any theoretical scientific model, one  that tries to explain speech production must account for any observed phenomena, and must predict any phenomena that might occur, disordered or normal. It must describe all the steps of normal speech production, the outcome of each step, the errors that may arise and why and/or how these errors might occur. The model needs to include rules and constraints of sentence formation to enable a speaker/lisener to judge the well-formedness of a sentence, and/or monitor her own speech to prevent speech errors; and if there are various constraints to be applied, then it must be postulated that there is more than one step in the process of speech planning and production. A speaker never expresses an idea in the exact same form twice, but on the other hand, s/he cannot store an infinite set of utterances. It must then be concluded that speaking is done by combining together, in an infinite number of ways, a finite set if units. One of the difficulties in describing speech production is in determining and defining the size of these units, their nature, and the order in which these units will occur in the speech stream. Fromkin (1971; cited in Fromkin, 1991) based her model on the observation that "speech errors, hesitation phenomena, false starts, etc. have provided indirect evidence for the units, stages, and cognitive computations that are involved in speech production" (p.3).  33  This procedure, of studying of speech errors and constraints on the kind of errors that may occur to obtain insight on what needs to be in a speech production model, is not typical: usually, when devising a theory, one postulates a hypothesis and then tests it. Rather, this method is ad hoc, i.e., first, the normal error-generation mechanisms (as opposed to damaged mechanisms as in aphasia) are analysed, and then a hypothesis of the normal speech production mechanism is formed. Fromkin (1971; described in Fromkin, 1991) developed the Utterance Generator model, which attempted to account for speech error phenomena. Her top-down model of speech production describes six stages of utterance representation, between which are processes which translate the representation from one level to the next. At the FIRST stage, the speaker focusses her/his thoughts on a specific fact s/he wants to express and generates the 'meaning' to be conveyed (Fromkin, 1991) in a form yet unspecified by the model. At the SECOND stage, the message is mapped onto a syntactic structure (syntactic tree), which specifies the form (e.g. number, tense) and grammatical category (e.g. noun, verb) of the semantic features (major elements of the message; e.g. ideas of "chase", "cat", "dog").  In other words, the semantic and syntactic information are "slotted" at the lexical  nodes of the phrase (marker). Evidence for the existence of this stage and for the fact that phrases are a production unit has been found in that phrases can be exchanged or reversed as a whole (see example 1), and that movement of parts of a phrase, which would result in an ungrammatical sentence, have not been observed. 01: a hummingbird was attracted by the red color of the feeder --> the red color was attracted by a hummingbird of the feeder  1  1  All examples are taken directly from Fromkin, 1991, or Garrett, 1984.  Similarly to phonemes, which are composed of a hierarchy of features, syntactic phrases have  34  a hierarchical structure: sentences/clauses are composed of smaller clauses, which in turn are composed of various types of phrases. When a speaker attempts to correct her/his error, s/he is more likely to start the repair at the beginning of the syntactic constituent break (phrase), showing that syntactic boundaries, specified at the second stage, serve a role and are immutable. 02: The doctor looked up Joe's nose--that is, up Joe's left nostril The phrase unit provides another important piece of information: anticipation errors and exchange errors often occur over many phrases, suggesting that speech is not simply produced/selected in succession, one word at a time, but possibly as much as two phrases at at time. At the THIRD stage, the intonation contours, phrasal and sentence stress "are generated on the basis of the syntactic representations" (Fromkin, 1991, p.28) formed at the previous level. The prosodic information is assigned to the sentence before the lexical selection is performed because prosodic information is based on the hierarchical tiers of the phrasal and clause construction, whereas lexical stress is independent of the syntactic structure but dependent on the lexical items and thus is determined later in the process. Fromkin justified the inclusion of this third level by hypothesizing that, if stress can be disordered like the other units of a language, then it should be considered as an independent production feature/unit. She argued that, in a word, the primary stress remains on the same syllable, even if the word is involved in a movement error. She also proposed that sentence stress is immuable, regardless of the presence of any speech errors. Stress could be distorted by shifts, not by stranding or exchange errors. Sudies of stress errors in tone languages have yet to discern, however, whether stress errors are the result of stress-related speech errors or lexical selection errors. At the FOURTH level, semantic features and syntactic categories, which have been  35 specified at the second stage, regulate word selection, thus setting the stage for possible word mis-selection errors.  Garrett (1984) explained that, in Fromkin's model, this lexical  retrieval comprises two aspects: the lexical items are chosen on a semantic basis, and they are also retrieved in the form of stems, of which the phonemic segments and serial syllable positions are specified, and already marked in terms of the morphemes needed. The lexicon "box", as defined in Fromkin's model, does not explain word substitutions -phonological or semantic- and hence needs to be revised. Fromkin (1991) has suggested that some word substitutions occur because the word search activates/evokes words that are related to the target paradigmatically or syntagmatically: during the search, a cohort is formed, and in the case of an error, an entry of the cohort is mis-selected in place of the target. 03: too many irons in the fire --> too many irons in the smoke (semantic substitution) "in word substitution and blends, the words involved are semantically and/or phonologically similar" (Fromkin, 1991, p.15). These errors, however, arise only after the semantic features have been assigned their grammatical/syntactic role since they never result in ungrammatical sequences. As shown in examples 3 and 4, word substitutions and blends can result in non-occuring but possible word/morpheme combinations, similarly to stranding and morpheme exchange errors. 04: grab (V)/reach (V) --> greech (V)(word blend) preserves the grammatical category so that the resulting clause is not ungrammatical Word substitutions and blends, as well as word-exchange errors, must also occur before lexical items are inserted in the syntactic frame as the result of such errors is always morphologically correct. This suggests that morphological rules are active throughout the various stages of transformation of the sentence. Once lexical items have been retrieved, they are assigned to their position in the syntactic frame, and morphophonemic rules are applied to  36  specify the phonetic shape of the morphemes. All types of morphemes can be involved in speech errors: e.g., stem and affixes, and of the latter, derivational and inflectional prefixes and suffixes. Example 5 shows how,"in certain errors, morphemes are combined into possible, but non-occurring words. 05: a New Yorker --> a New.Yorkan (and not a New York): the lexical search has specified that a morpheme indicating "nationality" is required here, but an incorrect one has been selected in the application of the morphemic rule. When contrasted with word blends and substitutions, words and word stems involved in exchange errors do not necessarily share semantic or phonological features. Morphemes, however, need to be adjusted to accomodate their environment when an exchange in word root or a free morpheme occurs. Errors involving syllables have also been observed, thus suggesting they are also units of speech, although morphemes are more commonly involved in errors than non-meaningful syllabic strings: "very little evidence from spontaneous error corpora favors the syllable as a moveable unit" (Shattuck-Hufnagel, 1979,  quoted in Shattuck-Hufnagel, 1987, p.18).  Rather, many errors of movement or replacement of subunits of the syllable are found (onsets, rhymes, nuclei, and codas), influenced by syllable structure; in other words, segmental errors follow a structural law with respect to syllable place.  Garrett (1988)  concluded from this observation that "the evidence indicates that syllabic structure provides a processing framework rather than units of retrieval in a processing vocabulary" (quoted in Fromkin, 1991, p.8).  This hypothesis has been supported and in turn it supports a  syllable-based structural representation.  Shattuck-Hufnagel (1987) also supported this  contention, and further hypothesized that word-onset consonants would form a special processing class, that is, the "early" portion of words (still not well defined) would play an important role in lexical search and access.  37  At the FIFTH stage of Fromkin's model, phonological pronunciations rules specify the phonetic segments, and can cause phonological exchange errors, in which the segments exchanged share phonetic features. That is, as segments are being mapped onto lexical items, they are still abstract representations of phonemes which need to be adjusted with respect to their phonemic environment (Shattuck-Hufnagel, 1987).  Evidence to support this  hypothesis is found in errors in which a phoneme is realized differently in its "erroneous" context compared to how it would have been produced in its intended context. 06: "skop the tar" for "stop the car", the "t" is aspirated in the error position, whereas it would not have been aspirated in the word "stop". In other words, a segment is (mis)-placed before it takes its phonetic surface shape. This phenomena also applies to the duration of vowel nuclei when two consonants exchange. The smallest sized-errors are "phonetic" ones, in which a feature-sized unit can be anticipated, persevered, deleted, added or exchanged. In phonological errors, the same processes can occur but between phonemic-sized units. There are some constraints, however, on the type of errors that can be made, (e.g., a speaker rarely stops in the middle of a segment, phonetic errors occur amongst vowels and consonants but never between the two). Moreover, the misordering and blending of segments usually occurs between segments that have a similar structure and position in a bigger unit, and arise at the step at which segments are to be assigned to their slot: "the error constraints support the claim that such dimensions as suprasegmental structure, lexical stress, and distinctive features are reflected in the processing representation" (Shattuck-Hufnagel, 1987, p.22). In other words, the segmentto-slot association process can fail, resulting in misordering or blending of segments, since slots are defined with the same parameters the segments thus allowing "confusions" to arise. the model that accounts for sublexical errors as malfunctions in a serial ordering process involving separate representations of the segments and their organizing framework can account for five major sublexical error types: exchanges, substitutions, additions, omissions, and shifts ... (and) word blends. (Shattuck-Hufnagel, 1987, p.23)  38  Shattuck-Hufnagel (1987) hypothesized that since the information about the order of segments of a word is already available as part of the definition of the word in the lexicon, errors in the serial ordering of phonemes occur during the translation of the sentence from one level to the next. Moreover, she found that most studies support the hypothesis that mis-selections arise because the target and intrusion segments have a very similar form, or a similar position in the phrase, not because the intruding segment is "stronger" or more frequent. "When two target segments of an utterance interact in an error, the similarity of their target positions in the larger structure of the utterance provides information about the representation of that structure" (Shattuck-Hufnagel, 1987, p.40). In the penultimate stage of Fromkin's model, the already determined phonetic features of the segments are mapped and sent in the form of motor commands to be executed, at the last stage, by the oral motor system. Fromkin's (1971) model has been found to contain flaws as it does not predict all possible speech production errors that have been observed in various studies: it only accounts for the types of errors mentioned above. It does not explain why major class items do not shift, but that minor class items do, nor does it account for errors in which syntactic rules are misapplied, as in (a grammatically and phonologically correct) blend of two sentences. Garrett (1984) developed a model of (the normal process of) speech production by observing, analyzing and comparing speech errors of normal speakers with those of aphasics, drawing a parallel between sound errors and phonemic paraphasias, word substitutions and verbal paraphasias, etc. His working model of normal language production is primarily based on real-time language-processing data, i.e., it is based on what really happens in speech production, and aphasics' production errors are correlated with the normal speech process, deducing at what level these errors arise.  In other words, from these errors, Garrett  inferred what levels and processes need to exist to produce normal speech. Garrett believed  39  this method of investigation could be more revealing than infering from normal speech itself: "If speech error patterns are taken to reflect normal processing structure, the properties of error types and their interactions should tell us what structures are being computed by the system at given points in the elaboration of a sentence" (Garrett, 1984, p.181). Garrett's and Fromkin's models have in common three main stages/levels, which are subdivided  differently in each model: conceptual,  language-specific, articulatory.  Additionally, Garrett's model made explicit some of the phenomena which was left unexplained in Fromkin's (1971) model. For example, Garrett specified that semantically related words exchanges occur at the functional level. Garrett also explained that the segmental structure of lexical items is determined at the positional level when they are assigned a position in the sentence frame. As in most language production models proposed, Garrett's has a general conceptual level which he calls the message level, a functional level, a positional level, a phonetic level for sentence processes, and an articulatory level of motor control processes. At the MESSAGE LEVEL, the speaker elaborates a conceptual construct (idea) on the basis of her/his conceptual and affective state (what s/he is thinking and feeling), as well as on the basis of her/his knowledge of the world, including linguistic and nonlinguistic facts. Once the message representation has been established, the steps between the message and the functional level are logical and syntactic processes. At the functional level, the first specifically linguistic structures (as opposed to general inferential) are formed, because, at the end of the process, sentences need to be communicatively appropriate for appropriate interpretation by the listener.  The subject  must perform a lexical search and selection based on the meaning relations contained in her/his message, determine the functional structures of the sentence (e.g. who does what to whom), and assign the elements to their proper structural role (what semantic category plays  40  what role in the message). These steps of sentence planning have been inferred from observing errors such as meaning-based word substitutions and whole-word exchanges between phrases. As seen in example 8, two words exchanging positions are usually in different phrases but occupy the same grammatical role in their respective phrase. 07: "He rode his bike to school tomorrow." (yesterday) 08: "They JeJi it and forgot it behind." In errors from normal speakers, word substitutions are meaning px form dependent, i.e., the "confusion" in lexical selection is of one type or the other. Parameters of form such as initial phonetic element, syllable length and stress locus can "cause" form-based errors (see example 9); meaning-based errors are induced by a paradigmatic relation (see example  09: "a slip which considered " (consisted) At this stage, the syntactic structure of the sentence is planned and constructed and hence, grammatical/syntactic errors such as sentence blends, word/phrase shift errors, or a words/phrase movement needed and not applied, or not needed but applied may occur. Minor category elements (bound and free morphemes and inflectional elements) are introduced as features of the planning frames, but are not subject to exchange processes, like the major category items. Instead, minor category elements are usually stranded or misplaced "in the process of interpreting features of the phrasal planning frame and siting them in the terminal string" (Garrett, 1984, p.180), because free and bound morphemes set the constraints of which type and how many lexical items can be assigned to a specific phrasal position. Meaning-related word substitutions are very similar, if not parallel with, anomic aphasics' errors in which the speaker demonstrates access to the meaning of the target word by producing circumlocutions, definitional phrase, gestures, etc. Form-related errors in normal speakers parallel errors in conduction aphasics: these subjects often make many  41  attempts at pronouncing a word which are very close to the target but contain sound (form-related) errors. In jargon aphasics, it would seem that meaning-based word selection and sound interpretation processes are both impaired. Some researchers have proposed that, in these subjects, a word-formation process forms "gap-fillers" to keep the speech output fluent. Some of the gap-fillers are composed of meaningful morphemes, but, assembled together, they form meaningless words/phrases, similarly to word blends, stranding and shift errors in normal language production which are combinations of meaningful morphemes. These errors indicate that there is a separation between morphological structure and meaning relations. Jargon aphasics do not appear to differentiate inflectional from derivational morphemes, but the model assumes that these two types of morphemes to be specified at different levels; hence, the model has to be revised to explain the jargon aphasics' behaviour. Minor and major class words can also be referred to as closed- and open-class words, respectively. Words substitutions, sound exchanges and word exchanges take place between open class items; shift errors are movement of closed class elements. "Closed-class and open-class vocabularies play distinct computational roles in sentence construction. That difference implicates the processes of phrasal construction, which in turn implicates the contrast on meaning-based and form-based processes in production" (Garrett, 1984, p.185). The two classes of words are also distinguished in that a specific type of aphasic will have a disturbance with only one of the two classes: posterior aphasics usually demonstrate preserved local phrasal organization and use of closed-class vocabulary, but their jargon evidences a difficulty with open-class items (nouns, verbs, some lexical adverbs). Contrastingly, anterior aphasics fail in tasks tapping the processing of grammatical morphemes (closed-class items) but produce and comprehend open-class items, although in a reduced amount. Although anterior aphasics are able to retreive major lexical items at the  42  functional level, they have a deficit in mapping these elements from the functional level to the positional level of representation, i.e., they often have difficulty in assigning the correct lexical item to its proper thematic role. Some instances are found, however, in which prepositions, which are considered to be closed-class items, are exchanged. This contradicts the hypothesis that only major category words appear in word-exchanges at the functional level of representation. On the other hand, prepositions cannot be considered as major class elements since, like minor class elements, they are never involved in sound exchange errors. Garrett (1984) suggested that, at the functional level, prepositions be considered as lexical items, but not at the positional level. This would make sense, considering that the functional level should be considered as logical and/or syntactic and the positional level as phonemic/phonological: prepositions are not subject to any phonological processes since no derivational/morphological rules apply to them, but they do play a major role in defining the syntactic frame of a sentence. Fromkin (1991) agrees with Garrett that free and bound morphemes (inflectional and derivational morphemes) do not participate in exchange or revearsal errors. Morphemes are stranded in errors and are subject to phonological rules of the new word they are attached to. These morphological rules are applied after the morpheme is stranded or shifted since the phonemic sequence that results from a stranding error is often permissible and the stranded element is still considered as an acceptable grammatical morpheme.  Like Garrett, she  mentioned that morphemes and minor category elements are involved in shifts, not exchanges, but that this does not affect the syntactic stress pattern. The processes that transform the representation previously and at the functional level are logic-oriented, that is they ensure that the order and meaning of the message constituents are appropriate for interpretation by the listener. The processes at and those subsequent to the positional level are mostly pronunciation-oriented.  Between the functional and the  43  positional levels, retrieval of the segmental structure of lexical items (word form) takes place; thus, form-based word substitution may occur at this stage. Also, the surface phrasal geometry is determined, i.e. planning frames are formed in terms of what elements are needed (e.g. agent, verb, object), depending on the features of the verb. The assignment of lexical formatives to phrasal positions (i.e. assignment of segmental and prosodic structure of words), and interpretation and siting of grammatical formatives in the surface sequence of sentence elements (node information) also takes place at the positional level. Sound exchanges, stranding (grammatical) and shifts of word and morphemes take place at that level. Garrett (1988) hypothesized that morpheme shift errors occur after all exchanges "as a consequence of operations that interpret features of phrasal frames for insertion into the terminal string of phrase markers" (quoted  in Fromkin, 1991,  p.21).  Whereas  morphological and phonological rules apply to stranded morphemes, only phonological rules apply to shifts. Fromkin assumed that both kinds of rules apply to both types of errors. Shattuck-Hufnagel (1987) postulated that in the case where two (target) segments are mistakingly evoked/retrieved instead of only one, both have an equal chance of displacing the other in an error, regardless of any features of the segment. Additions, omissions, and shifts can be accounted for by a breakdown in a "slot-to-segment" association mechanism, breakdown in the check-off monitor which eliminates a unit once it has been used, or a breakdown in both (Shattuck-Hufnagel, 1987). In structural role assignment, the cohort of lexical items is kept in a short-term processing store, and the set of phonemic segments of each item is available. The syllabic structure and main lexical stress is constructed, and non-onset portions of words are assigned to the phrasal structure, word-onset portions of the structure remaining empty at this stage. Eventually, word-onset portions are associated with their proper locations. This process applies only to open-class words, and often to words which have the same vowel in their first syllable. This stage can also explain word-onset  44  consonant shifts, even with vowel-initial words, since they have an empty slot for consonants. Lastly, this phenomena can explain word blends, since the model does not specify the number of segments until later in the process. Shattuck-Hufnagel also argued that word-onset consonants are the most susceptible segments to be involved in exchange/interaction errors, since so many errors occur in words in which the main stress is on the first syllable, and by the same token, on the word-onset consonants. She observed, in a specifically designed experiment controlling word-initial consonants and location of stress in words, that word-onset consonants are more error-prone that lexical stress, and subject to particularly interaction errors: "word-onset position similarity has a more powerful effect on sublexical errors than does one based on stress" (p.44). Garrett (1984) found that phrasal stress is preserved under exchange situations but is "violated" in shift errors as it is a feature of surface structure of the sentence, i.e., is determined only late in the process. Word and sound exchanges are not subject to the same constraints: the former occurs with respect to vocabulary type, syntactic and phonological roles of exchanged words, and takes place between phrases since lexical items have not been assigned to their position yet. The latter relies on morphological structure, segmental and prosodic similarity of the words containing the exchanged sounds, which must be within the same phrase, as sound exchanges occur during the positional level (assignment segment-to-slot) which is a single-phrase planning level Subsequently, at the positional level, the detailed phonetic character of the segments is once more modified, at a phonetic level of representation: at this level, the phonological constraints of the language are applied according to the phonetic/phonemic environment of each lexical item. At the articulatory level, phonetic and prosodic information about the sentence is translated from sentence level structures into distinctives features matrices  45 (articulatory  structures)  so that  appropriate motor commands.  the  respiratory  and  articulatory  systems have  the  46  CHAPTER 3 METHOD  Subject A.K. is a 67-year-old, right-handed, grade 12 educated white female. In August, 1990, she suffered a thrombosis in the distribution of the left middle cerebral artery, resulting in a right hemiplegia and an expressive aphasia. She had a history of TIAs, the first of which occurred several years ago, the second about three years ago, and the third occurred less than 24 hours prior to her CVA.  All four episodes resulted in speech production  difficulties (speech sound errors, slurred speech, and/or semantic paraphasias), which all resolved spontaneously except for the last two episodes. At admission to hospital, A.K. was diagnosed as having profound Broca's aphasia: she had become confused, could understand others but was effectively mute except for occasional single words, and had a right facial weakness and hemiplegia. A.K.'s native language is English: she was an office worker prior to her marriage, returned to work as her children grew up, and did not retire until three years ago, around which time she was widowed. Before her stroke, she enjoyed needle-craft, sewing, knitting, reading, and was actively involved in her church. Since her stroke, A.K. has not resumed all of her activities since she is limited by her right-sided weakness: she now enjoys swimming, group exercizes, classical music, television, and is an active member of stroke club and a church. In November, 1990, 3 months after her stroke, A.K. was  transferred to a  rehabilitation centre: her communication was then formally assessed for the first time. The Boston Diagnostic Aphasia Battery (Goodglass and Kaplan, 1972) and the Boston Naming Test  47 (Kaplan et al., 1976) were administered between November 27 and December 4. Results revealed that auditory comprehension was reliable and adequate for daily needs. A.K. had good conversation skills, she could perform single word to picture matching, and respond appropriately to sequential command, but demonstrated some mild deficits on higher level tasks. A brief neuropsychological examination revealed no memory problems; threshold (dB SPL) hearing was within normal limits. A.K.'s verbal output was moderately to severely limited in quantity, her answers to yes-no questions were felt to be reliable, and her word-finding was adequate for simple familiar words, but impaired with increasing complexity and unfamiliarity of words.  Her  output was also characterized by a phonological searching/groping that sometimes made her responses difficult to interpret; it was noted that a first letter/sound cue helped her attain the target. The subject's oral motor functions were within normal limits, and no facial asymetry, reduced range or speed of oral movement were noted. Reading comprehension was good in simple tasks, but showed moderate difficulty with higher level material. Similarly, writing (using the non-preferred hand) of simple, single words was adequate, but moderately impaired for more complex productions. In short, A.K.'s initial speech and language evaluation indicated a moderate nonfluent aphasia, with slow and laborious output, functional auditory skills, and some preserved reading and writing abilities. A.K. received individual speech therapy four times per week, and group therapy twice per week, from early December, 1990 to late March, 1991, at which time she was discharged from the centre. Treatment focussed primarily on verbal expression: improving clarity of speech by pacing and breaking up long words; silent practice/rehearsal of phonologically complex and long words, and articulatory placement, so as to reduce the number of approximations to reach a target. Other goals included increasing functional reading comprehension, and improving the writing of single words and simple sentences.  48 By late January 1991, A.K.'s speech was still slow and laborious: her sentence production consisted of up to 8-10 words (correctly using content and function words), the clarity of her speech was still greater on short phonologically simple words (i.e., two syllables or less), she was using silent rehearsal and was easily cued by articulatory position. She could read two or three consecutive paragraphs, answer comprehension questions, and write single words. Therapy continued until discharge with the same frequency of sessions. At discharge, A.K.'s spontaneous speech was still slow with some phonological alterations. If she did not succeed in producing the target, her attempts were usually close enough to the target to be recognized. In language production, her use of sentence structures was within normal limits; she had increased her reading up to paragraph length material equivalent to a sixth grade level, and her writing was functional for daily needs (e.g. signature, writing notes and lists). Although she seemed to have reached a plateau, A.K. was refered to a private practice speech-language pathologist at her own request. Testing for this research was conducted 10 months after A.K.'s stroke: she was then 68-years old, and had residual speech difficulties as was indicated on her discharge summary. At the time of testing, she was being seen by a private speech-language pathologist, but since the testing sessions were close in time, therapy was not thought to have made significant effects on test results.  Method Testing was done in two sessions of approximately one hour each, separated by seven days, and one session of approximately one hour, five weeks later. In the first two sessions, the subject was asked to carry out a total of six (6) tasks which all aimed at examining speech production errors. First, A.K.'s naming abilities were assessed with the Boston Naming Test (Kaplan et al., 1976) in order to determine the ease with which her targets were recognizeable. Second, she was asked to describe the pictures from the Boston Diagnostic  49 Aphasia Examination (Goodglass & Kaplan, 1972), the Western Aphasia Battery (Kertez, 19??) and the Minnesota Test for Differential Diagnosis of Aphasia (Schuell, 1965). In the second session, she performed a story completion task from Goodglass et al. (1972) and she was asked to engage in a spontaneous conversation. In the third session, the subject read aloud a word list containing real words which she had had difficulty producing in the first two sessions, and nonwords chosen from Shattuck-Hufnagel (1991; see Appendix B). In the Boston Naming Test and the picture descriptions tasks, the standard instructions were employed in the administration of each test. In the story completion, the subject was told: "I am going to read you some short stories. They are very straightforward. After each story, I will ask you a simple question. Let's try one." For the spontaneous conversation, she looked at some family pictures and talked about herself and her family; questions were used only as a means of eliciting as much speech as possible. In the reading task, A.K. was asked to read aloud largely printed words and nonwords presented to her on cards. All tasks were recorded onto a TDK  IEIC l/type I Normal position, AR-X  90 super  precision anti-resonance cassette mechanism normal bias 120 us EQ, with a Sony cassette corder TC-150 (catalogue number 3-545-686-01; serial number 22222) at a recording level of -10 to -5, and a SONY Dynamic microphone, The distance from the subject's mouth to the microphone was approximately 18 inches. All tasks were then transcribed manually using the international phonetic alphabet and analyzed for phonemic and multiphonemic errors.  50  CHAPTER 4 RESULTS AND DISCUSSION  In the first chapter, I discussed some speech production errors in aphasia and the possibility that different errors may arise from different levels of impairment in the speech production mechanism.  I presented definitions of various production errors as well as  standard descriptions of main aphasia types. I then reported studies which tried to establish relationships between types of aphasia and types of speech sound error. The main thrust of these studies hasbeen twofold: first, they have argued that the speech production mechanism contains many levels, and second, they have postulated that different types of aphasia may reflect an impairment at a different level of the mechanism. In other words, by comparing types of speech errors of Broca's, Wernicke's and conduction aphasics, researchers have been able to deduce that the normal speech production mechanism is composed of more than one stage, and that each of these stages can be independently impaired. I ended the chapter by presenting studies which argue that there may also be a differentiation between speech perception and comprehension, such that (and analogously to production) different types of aphasia may reflect disturbances at different levels of perception or comprehension. In the second chapter, I presented two models of speech production which described the levels in more detail, that is, the transformation of a sentence, from message to utterance. These models (developed by Garrett (1984) and Fromkin (1971)) were based on the study of normal speech errors, and possible errors and impairments.  Relationship to previous studies on aphasic speech error data The tasks administered to A.K. were chosen because they elicit oral speech.  No  51  comprehension tasks were included since A.K. was only mildly impaired in language comprehension. This investigation focussed solely on production.  Pre-articulatory  programming errors/deficits are particularly apparent in "target-bound" tasks, i.e., when a subject has to identify a stimulus which the listener already knows, i.e. is fixed; these tasks tend to induce a larger number of pre-articulatory errors because a subject cannot change the target word. The Boston Naming Test (Kaplan, Goodglass & Weintraub, 1976) was administered first to verify the recognizability of A.K.'s targets, provided she was unable to produce the target correctly at a first attempt. Her word retrieval appeared to be moderately impaired, but phonemic cueing proved helpful. It appeared that, in most of instances when A.K. knew the target word, she was able to retreive the first consonant/vowel/syllable of the word, but then had difficulty executing the remaining segments. Thus, it was easy to infer what her intented targets were, even when it proved to be other than the target-picture. 01: (globe): "a ... a ... atl-as ... at ... a-... atlas" In many of her attempts in all tasks (except reading aloud), it appeared that A.K. was gradually "accessing" the phonological word, as if she was gradually "seeing and reading" the letters of the target word in her mind's eye.  1  With simple mono- and bisyllabic words, it  appeared that she would simply "read" the words regularly, until she had accessed all segments; in longer and more complex words (e.g. polysyllabic words, words with consonant clusters) A.K. was also able to access the first consonant/vowel/syllable, but produced many phonemic errors: displacements, additions, omissions, and various substitutions. In Guyard et al. (1981), Broca's aphasics had difficulty with shorter words, whereas A.K.'s errors were 1  Note:  We  are  only looking at those  instances where A.K.  experienced difficulty  articulating  segmental strings, i.e., her errors can be explained within the speech production model presented in the  second  chapter.  52  independent of word length. We might therefore ask: Did A.K. know what the segments were, but not where they went in the lexeme? Or, did she have difficulty planning how to assign the non-onset segments of the lexeme to their respective slots? Perhaps when attempting to retrieve a correct phoneme she failed to do so: this failure then resulted in the emergence of some other segment, which then occurs as an error. experiencing  difficulty was  It was  apparent that A.K.  was  at a level lower than what Garrett (1984) has termed the  "Functional Level": "sound exchanges are presumed to arise in the process of segmental interpretation of the lexical items that are assigned to positions in the phrasal frame of this (the positional) level" (Garrett, 1984, p.179). In spontaneous speech tasks (i.e., all tasks excluding reading aloud) and on the Boston Naming Test, A.K. made many phonological errors: 36% were substitutions, 23% were phoneme substitutions which could be accounted for by the influence of their surrounding phonemes (assimilation within a word, assimilation across word boundaries, metatheses, all of which fall under the term  "displacement"), 25% were additions of phonemes/syllables,  and 16% were simplifications (loss of phoneme/syllable) (Nespoulos et al., 1987; Joanette et al., 1980; Blumstein, 1973). Of the substitution errors, 64% involved a single feature. Of these, 30% involved voicing, 41% place of articulation, 19% manner of articulation, 11% nasality. The other 36% of substitution errors involved two or more features. In reading aloud, 45% environment, 18%  of A.K.'s phonological  additions, 25%  simplifications. 61%  errors were substitutions, 11% of all her phonological errors  involved only one feature and 39% were multifeatured errors. Of the substitutions involving only one feature, 64% were with respect to voice, 0% manner, 27% place of articulation, and 9% nasality. Overall, the subject made more phonological substitutions involving only a single feature.  The analysis presented in Appendix C was  performed assuming that  spontaneous speech of aphasics does not vary from that of normal subjects in terms of  53  phoneme frequency of occurrence. The subject mentioned that, at times, she saw the word written in her mind, and that this helped her to say a word although sometimes the word still did not "come out" the way she saw it. All these comments indicate that A.K. knew the word she was trying to say, i.e., she had already retrieved its meaning. If it is assumed that A.K. had retrieved the meaning of the target word, then her problem may be viewed not as a disturbance in semantic organization, but rather as disturbance at a lower level of execution. This matter will be discussed below. A seventh task was administered five weeks later to verify whether A.K. was indeed "internally" reading the target. It was hypothesized that reading aloud would be an easier task than silent reading, this hypothesis was confirmed: that is, A.K. succeeded in producing the target at the first attempt, roughly 60% of the time (with some phonological errors and effort due to her verbal apraxia). A.K. would appear to be a good example of an aphasic who is difficult to classify as only one type of aphasia strictly according to site of lesion and linguistic behaviour. Speaking is obviously arduous for A.K., her articulation is laborious and she does much oral groping, as attested by many aborted attempts at articulating a word. Lecours and Rouillon (1976) stated that Broca's aphasics evidence an arthric disorder: their speech output is slowed, laborious and even often syllabic, possibly aprosodic. A.K. showed all of these characteristics. During hesitations, her voice often creeked, that is, she mistakingly initiated voicing, suggesting a lack of coordination between vocal fold activity and supralaryngeal musculature. Moreover, she did not respect stress agreement, i.e., she did not apply correct stress on words, nor in phrases. Since her speech was so laborious, most vowels were distorted, and she had almost no melodic line. MacKenzie (1982) had observed articulatory groping in her aphasic subjects which increased with articulatory complexity of phonemes a behaviour she termed "articulatory- phonemic". A.K.'s single feature errors could well be exaggerated distortions  54  that fall in the cognate category. As she tired, A.K. demonstrated a mild qualitative reduction, and a quantitative reduction in rate of speech. She was often slowed by a seemingly arduous lexical/phonological search. As for most Broca subjects, it was difficult to determine whether in some instances her phonetic deficit concealed a phonemic deficit. Her comprehension was spared, as in conduction and Broca's aphasics, which thus can not be used as a basis for classification. If A.K. is a Broca's aphasic, then we expect her to show all features characteristic of a Broca. On the other hand, some of A.K.'s verbal behaviours lead to the hypothesis that she might be termed a conduction aphasic. Regardless of her articulatory deficit, she made many attempts at reaching target words, attempts which contained multiphonemic transformations, in addition to phonetic distortions. Her repetition (although not formally assessed in this research) was poor and often lead to the production of many phonemic transformations - the same was true, but to a lesser degree in reading aloud: this is partially characteristic of conduction aphasia.  Lecours & Rouillon (1976) have also made these observations of  conduction aphasia. As regards types of errors in Broca's aphasics, Nespoulos et al. (1987) found more substitutions of a voiceless for a voiced cognate and observed unvoiced segments to be replaced by segments of a different place. A.K. showed the same number of substitutions between voicing cognates (voiced to voiceless and vice versa) as she had voiceless segments substituted by segments of a different place of articulation. Conduction aphasics reported by Nespoulos et al. made significantly more serial ordering errors than Broca's aphasics. In Burns and Canter (1977), subjects produced more place errors than any other type, and these errors were chiefly substitutions (as in A.K.'s case), most of which were in final position for fluent aphasics and across word positions for nonfluent aphasics. Shinn and Blumstein (1983) found more voicing than place errors in Broca's aphasics speech.  55  Blumstein et al. (1980) observed that Wernicke's aphasics produced phonemes of the opposite VOT category to the target, and Broca subjects produced distortions (i.e. between-category allophones). Because no instrumental analysis was performed, we cannot determine what kind of distortions A.K. produced. Since A.K. produced mainly voicing and place errors, it can be stated that she performed like most conduction and Broca's aphasics. During testing, A.K. distorted more vowels than consonants; however, she made most phonological errors on consonants. In Monoi et al., conduction aphasics made as many errors on vowels than on consonants, however, they hypothesized that the structure of Japanese tends to precipitate more substitution errors. That motor aphasics produced vowels with formants varying more than in normal subjects led Ryalls (1981) to suggest that these aphasics experience a difficulty in phonemic encoding or a segment production deficit but that they do not have a phoneme selection problem. We will attempt to determine where the breakdown occurs for A.K.. In their study, Burns and Canter (1977) found more multicomponent, subphonemic errors than single component errors. They postulated that spontaneous speech would induce more errors because of the greater lexical search it requires. However, since A.K.'s errors do not appear to lie in an access problem (see further), the nature of the task in terms of lexical search did not appear to affect A.K.'s susceptibility to phonemic errors. Unlike the present data for A.K., Monoi et al. (1983) noted that Broca's aphasics made errors mainly on consonants, less on vowels. Furthermore, most of their errors were single feature errors as in the studies by Blumstein (1973), and Shinn and Blumstein (1983). On the other hand, in a naming task, conduction aphasics produced the same amount of multi- and single-feature errors. Blumstein (1973) stated that: "in the phonological dissolution of aphasic speech, it would be expected that the number of substitution errors would decrease as the phonological distance between phonemes increased." (p. 125). Monoi et al. suggested three levels at which  56  errors may arise: 1) at the stage of phoneme selection, which would result in substitutions, 2) at the level of articulatory programming, which would result in errors one feature away from the target phoneme, or 3) at both levels. In their study, the complexity of articulatory gestures created an increase in the number of errors and hence, they suggested that their subjects were demonstrating a low level phonetic deficit and a milder phonological disorder, possibly due to an impairment at the stage of articulatory programming of movements. On the basis of previous studies, more substitutions should occur between phonemes separated by one feature than between phonemes separated by two or more features, regardless of aphasia type or underlying deficit; this was the case for A.K..  However, in the  case of A.K. it does not appear that increasing complexity of articulatory gestures generated a larger number of errors.  Relationship to speech production models With respect to the kind of phonological analyses described in the previous section, Joanette et al. (1980) stated: "This 'static' approach to the analysis of phonemic paraphasias may not capture all the information usually taken into account in the clinical observation of patients who show such paraphasias in their symptomatology" (p.31). Thus, it is important to describe not only what a subject does in terms of speech errors, but also why these errors arise, and from which level they might originate in the speech production mechanism. Such information (together with recent phonological theory might) provide better direction for therapy programmes. As demonstrated, A.K. might be viewed as a combination of Broca's and conduction aphasia, since she evidenced the articulatory characteristics of the former and the phonological deficit of the latter. However, if A.K. were a Broca's aphasic, then we might expect her to omit function words (articles, pronouns, connectives, etc. (Blumstein, 1973)).  57  We might also expect her repetition to be better than demonstrated. If she were a conduction aphasic, then it would be reasonable to expect many selection and sequencing errors (Blumstein, 1973). It is difficult to classify A.K. as one type of aphasia since she clearly evidenced deficits characteristic of both conduction and Broca's aphasia. Kohn (1984) wrote that the label "conduction aphasia" has been applied.to a wide range of neurological and behavioural features.  Similarly, it has been postulated that it might not be a unitary  disorder. In the following, I will orient my efforts to explaining A.K.'s phonological disorder as it relates to models of speech production. Henaff Gonon et al. (1989) suggested that three features characterize an impairment at the phonological level: 1) good semantic knowledge, 2) poor effect of phonemic cue, and 3) tip of the tongue response. A.K. demonstrated two of these features which again justify why I considered her as demonstrating a deficit at the phonological level. Brown (1972; 1975, mentioned in Kohn, 1984) "argues that conduction aphasics suffer from a general deficit in phonemic encoding" (Kohn, 1984, p.98).  It has been suggested that the phonological  impairment can be at the level of the planning or execution of the phonological aspect of an utterance, or it can be an impairment of the internal phonological representation, and/or monitoring system for speech. As in Joanette et al. (1980), I only analyzed those sequences in which two or more attempts were made. This level of analysis was based on the premise that, if sequences of approximations become more and more similar to the target, then it could be assumed that A.K.'s internal phonological representation of the target is adequately preserved, and the target could be correctly inferred (after Joanette et al., 1980). If, on the other hand, A.K.'s sequences of phonemic approximations did not show a particular trend either toward or away from the target, then it would not be possible to draw any specific inferences concerning her phonological output mechanism" (following Joanette et al., 1980). A.K.'s sequences of  58  approximations were quite lengthy at the start of her speech-language therapy, and she also tended, at times, to diverge from the target. As part of her treatment, she was instructed to think of the first sound of the word she was attempting to say: this strategy in conjunction with spontaneous recovery seems to have helped her shorten her sequences of attempts, and also helped her produce attempts more similar to the target. Internal phonological representation can be differentially impaired in various tasks and types of aphasia. Joanette et al. hypothesized that a stonger internal representation would instigate a stronger (closer) progression toward the target, whereas fatigue and longer sequences of approximations could cause a decrease in the strength of the internal representation and a weaker progression of approximations toward the target, if it is reached at all. A speaker needs an intact internal representation and internal monitor to produce normal speech: "If it is the phonological form of their utterances that causes them to make the correction, (then) we must suppose that they have some internal phonological representation, which they can refer to during the monitoring process" (Joanette et al., 1980, p.41). In Dell's (1986) theory, the retrieval mechanism is based on spreading activation, and the lexicon is a network in which phonemes and phonemic features are nodes, amongst other nodes. The activation level of a node/unit is "j" at a time "t" and "an error occurs when a wrong item is more activated than the correct one and is selected and tagged" (Dell, 1986, p.291). If the subject has retrieved all phonological information, then the lexical search is terminated. If the subject has been able to only retrieve partial phonological information, the system will be less effective in terminating the search and more errors may  arise  (Butterworth, 1979). I will not pursue this theory further since it deals simply with why a specific segment is chosen instead of the target one: because it shares segmental and positional features with the target segment. Rather, I will attempt to explain how segments exchange  59  positions and are mistakingly inserted into an incorrect slot and why access to the phonological information may be impaired. Kohn (1984) suggested a single-word production model in which there would be a lexicon containing abstract word representation linked to their respective word sounds. To produce a word, the lexical information is accessed and retrieved from the lexicon and transmitted to the working memory.  Although the representation is converted to the  pre-articulatory programming level (and subsequent levels), the working memory would keep a trace of the phonological representations. The phonologically specified targets are formed at the pre-articulatory programming stage. An error at this level would result in a phonemic paraphasia.  Kohn postulated the  presence of two feedback loops, one between each the pre-articulatory and articulatory programming levels to the working memory. "In this manner, Working Memory has the capacity to operate as a verbal short-term memory buffer (cf Shattuck-Hufnagel, 1979)" (Kohn, 1984, p.107). Kohn (1984) believed that the breakdown observed in conduction aphasics mostly occurs at the pre-articulatory programming level and which can show itself as a difficulty in conjoining syllables, and either selecting, or sequencing phonemic targets, to produce a form with sufficient information for articulatory realization: such a difficulty is manifested by conduction aphasics in two types of phonologically- oriented sequences. In the first type of sequence,'words are spoken with halting syllable transitions so that they are clearly articulated, yet produced with equally stressed syllables ... to facilitate auditory self-monitoring. In the second type of sequence, difficulty at the syllable boundary halts word production in midstream, forming word fragments. (Kohn, 1984, p.109) A.K. exhibited both of these verbal behaviours. If it is assumed that her programming system is fragile and misfires and/or breakes down in midstream, then this could explain the halting quality of her spontaneous speech. Kohn hypothesized that, on the other hand, syllabification may be used to avoid a programming error, i.e., improve monitoring of speech output. Given  60  A.K.'s frequent use of syllabification, and that pacing was used as a treatment technique, it would appear that she was using this as a production strategy. The large number of attempts at a target observed mostly in conduction aphasics (although sometimes in Broca's aphasics), is evidence that conduction aphasics in some sense know the correct target (internal representation) and are thus able to monitor themselves through an intact feedback loop; if this was not the case, it is difficult to understand their frequent attempts to self-correct. If the conduction aphasics produce a long string of phonemic approximations, then we can also infer that their verbal working memory is unimpaired. If a subject does not recognize that s/he has produced the correct target, then we should not necessarily conclude that their monitoring system is impaired but rather that they may  be  subjected to an attentional overload due to their many consecutive attempts and that their system may have become saturated with conflicting phonemic information. We  might thus  infer that A.K. very likely possesses the correct internal representation and was aware of her errors, i.e., she had an intact monitor and feedback loop. This inference is based on the observation that she often made many attempts at each target and frequently commented on her awareness of the target. Regardless of these observations, it sometimes appeared that A.K. did not know the target word, whether in spontaneous conversation or in a target-bound task. It is important to note, however, that A.K. was always satisfied with the word she produced: I therefore assume that her pauses were usually resolved (i.e, her system did not have to resort to a sequence generator to form neologisms), and that the length of her pauses depended on the amount of phonological information she was able to retrieve (Butterworth, 1979). LeDorze & Nespoulos (1989) explained that what the subject produces when faced with a word-finding occurrence depends on the subject's "reaction", compensatory behaviours, and the information s/he possesses on the formal characteristics of the lexical item. It is not  61  surprising, however, that A.K. showed some word-finding difficulty since it is "typically observed in aphasia and (is) considered a central feature of this linguistic disturbance" (LeDorze & Nespoulos, 1989, p.382). Benson (1979) described word-finding disorders as being due to one of three different levels of breakdown in the naming process: 1) a selection anomia due to a difficulty in choosing the correct word from the lexicon; 2) a semantic anomia which impedes the use of words as symbols; or 3) a word-production anomia usually seen in nonfluent aphasics, like A.K.. In the latter case, the motor planning and/or production of words may be disordered and may also result in the paraphasic errors evidenced by conduction and nonfluent aphasics. The "Inability to express the desired word constitutes one of the more frequent sources of what appears to be anomia" (Benson, 1979, p. 300). It is debatable, however, whether this is a true type of anomia since when provided with even minimal cueing, subjects can then often produce the word suggesting that the word was in fact available to them but that they were only having a difficulty in initiating the articulation.  Benson explained that such subjects will often  comment that they knew the word but were not able to say it (see examples below) and that usually this behaviour is observed in nonfluent/frontal patients.  Nonetheless, some  occurrences of "true word-finding" problems, i.e., the "pure" inability to retrieve the word even with strong cueing, may occur in these (nonfluent) aphasics. Garrett (1984) emphasized the clinical observation that aphasics often have access to word meaning but not form, evidenced by the subjects' statement that they know a word they are unable to utter. The following examples from A.K. illustrate these preceding points: 02: "I know the word but I can't say it." 03: (target: the name of her grandson's girlfriend) (S.G.) "What's her name?" (A.K.)"I forgot. I do know but I can't remember."  62  04: (target: the name of the city in which her son lives) (A.K.)"I  [fogAt]...  uh, I can't remember. I know what it is but I can't remember."  (S.G.) "Prince George?" (A.K.) "No, it's a- farther... in mimi m- [mwen]-land ... on right the border... on [bi pitwid] ..." (S.G.)"between B ..." (A.K.) "yes, B.C. and uh ... the next ah, pro- province." 05: (target: volcano): A.K.: "The top blew out"; 06: (target: stethoscope): A.K.: "Listens to your heart." 07:  (target: tripod): "These [jar] for [meri  [ware] the [or]- road".  me3 mere] measuring the [rei]  •  08: (target: sphynx): "I don't know but I've seen one like it"; 09: (target: yoke): "Something for put on a their [n- ne]- neck"; 10: (target: abacus) "It is this ah ah- [tje] Chinese [ke] [kontirj] word."  Anomia may be a lexical retrieval problem at the later level of lexical selection of Garrett's (1984) model, where the lexical item is identified in terms of its phonological characteristics. If an aphasic can benefit from partial phonemic cueing, then her/his deficit is not in accessing the lexical semantics, but rather the difficulty is associated with accessing the complete lexical-phonological representation/information. Circumlocutions ... imply(ing) some preservation of processing at the semantic level. If aphasic patients are capable of such cognitive processing at the semantic level, a true semantic deficit is an unlikely cause for their anomia ... The alternative possibility is that their problem lies in accessing a lexical-phonological representation. (LeDorze & Nespoulos, 1989, p.395-396)  In their investigation of naming deficits and related compensatory strategies in aphasia, LeDorze & Nespoulos (1989) observed that Broca subjects differed with respect to the compensatory naming strategies they each used, and that most conduction aphasics produced few  63  modalizations, but many attempts (phonemic errors) at the targets, as well as a great number of neologisms. A.K.'s strategies appear to most closely parallel those of conduction aphasics in this study, although she did not produce any neologisms. Anomic aphasics produced an equal number of modalizations and attempts, and many semantic paraphasias, while Wernicke subjects produced many modalizations. A.K.'s impairment is not like that observed in anomics, since their deficit does not appear to be associated with a phonological processing deficit (and I am assuming A.K.'s deficit is, see below). A.K. is also unlike the fluent Wernicke's aphasics. In Garrett's (1984) model, the translation of the sentence from a logico-syntactically oriented repesentation to a pronunciation-oriented representation occurs between the functional and the positional levels. During this translation, the retrieval of the segmental structure of the lexical items takes place, governed not only by meaning, but also by word form: initial segments, syllable length and stress locus. Consequently, words/parts of words can be exchanged because of similarity in form Q_L meaning (Garrett, 1984).  More or less  simultaneously with lexical retrieval, positional level planning frames are formed, and subsequently, segmental and prosodic structures of words are assigned to their position in the frame. During these processes, aphasics can often provide, as did A.K., some information about the sound/form of the word (e.g. initial segments). To retrieve the segmental specification of a word, the speaker must operate linking address that is part of the information provided" (target) word. (Garrett, 1984,  "via a  when identifying the underlying  p.187). Again, access to this linking address provides initial  phonetic shape, syllable shape and stress locus (LeDorze & Nespoulos, 1989). However, an aphasic may have access to the linking address, but the use of the available information may be impaired. In instances where A.K. knew the word but could not say it, it is possible that she would have a partially or totally inoperative linking address yielding more or less information about the form of the word. One of three deficits may be associated with the use of the linking  64  address: 1) the processing of segmental interpretation of word forms in the phrasal frame is impaired and results in mis-assignment of segments to their position in the word (e.g. exchanges, shifts in normal sound errors); 2) the phonological specification of the phonetic form is disordered and results in substitutions of phonemes (or sound-exchange errors in normal speakers); or 3) the construction of motor programs based on the phonetic specification of phrases may be impaired and violate language-specific phonetic categories, and even perhaps phonetic sequencing. Conduction aphasics are thought to have the second type of deficit, thus their difficulty does not rest with lexical access of meaning or form but rather in their positional assignment; however, access to the linking address is not sufficient for correct pronunciation. Caplan (1987) reported Garrett's (1984) suggestion that conduction aphasics "have a problem in the process of accessing lexical phonology from lexical semantics" (p.116). Thus conduction aphasics would be seen as having only partial access to the linking address e.g., s/he may recall only the first syllable of words; this partial access is demonstrated by the conduction aphasics' successive attempts/approximations d'approche). This behaviour was observed in A.K..  to the target (i.e. conduite  In other words, the conduction aphasics  may have retained much of the information about the word form, but obviously not all of it since each attempt is only an approximation of the target. Caplan did not entirely accept the notion of an impaired linking address as an explanation for his subject's difficulties because his subject also showed problems with nonwords, repetition, and reading aloud, none of which necessarily require a search of the linking address. As for A.K., she had some difficulty with repetition and no significant problems with nonwords and reading aloud (see Appendix B), hence Garrett's suggestion may explain her speech disturbance. Caplan has proposed changes to Garrett's (1984) production model, adding two features:  65  first, he suggested that there needs to be a phoneme-to-phoneme mapping for the production of single words prior to the phonetic specification, this second stage of phonemic mapping not being necessarily linked to the lexical insertion into the phrasal frame, but rather a part of the production process for single word target-bound production tasks. Second, he suggested that the stage of phoneme-to-phoneme mapping is easier to execute for words than for nonwords. Tasks using nonword stimuli would be more difficult than those using real words since the latter have an underlying lexical representation that the subject can trace to perform a "confirmatory match" between the underlying phonology of the presented word with the underlying phonology in the lexicon/lexical representation. It is my suggestion that a match may allow the subject to " confirm the identity" of the lexical item. However, this process would be lengthy and unnecessary in some tasks, unless the direct phonological route was impaired in which case the subject would have to utilize the lexical route.  Thus, subjects could have a deficit in processing from the underlying  phonological representation to the superficial phonological representation, the underlying representation being the permanent lexical representation of a word linked to the semantic and syntactic features of the word, from which the superficial representation is formed during speech production. Shattuck-Hufnagel's (1987) analysis of normal speech error data revealed that the lexical item is not the only processing unit, but that individual phonemes are also represented and manipulated during the various stages of speech production. Based on her data, she developed two hypotheses: First, "The processing of sublexical information in speech production planning makes use of representational units that correspond to single phonemic segments". Second. "This planning process includes a mechanism for the serial ordering of individual phonemic segments, during whose operation ordering errors can occur. The mechanism is referred to ... as sublexical serial ordering mechanism " (Shattuck-Hufnagel, 1987, p.19). On the basis of her analyses, she suggested that in the speech production mechanism,  66  segments and their slots are independently represented, and that an association process links the two, all of which would take place during a pre-execution planning process (much similarly to Kohn's, 1984, pre-articulatory programming stage). Consequently, following Dell's (1986) theory, an error would signify a misselection between similar target segments since these systems/processes are mostly sensitive to structure, i.e., segments that share many features. The additions, omissions, shifts, substitutions, exchanges, blends and other errors produced by aphasics, including A.K., can be explained by a breakdown in the segment-to-slot association mechanism, the check-off monitor, or both.  The association  process (or scan-copier) copies the segmental information onto the slots while the checkoff monitor deletes the copied segments fron the set of candidate segments. A failure in the scan-copier can result in the selection of an incorrect segment from elsewhere in the utterance, or incomplete or incorrectly tranferred information to the slot. A failure in the checkoff monitor can result in an anticipatory or persevatory error: in either case, the segment which has been used has not been deleted from the candidated set. The error monitor scans the utterance for repetitions and similar sequences (i.e. sequences with many similar segments such as in tongue twisters). The monitor may also mistakingly erase a sequence because it seems erroneous, i.e., a sequence is too similar to one that has recently been articulated. This monitor may also act as an "in-the-head" detector for a speech error that is noticed by the speaker before s/he articulates it. In other words, it edits out what would resemble an exchange or persevatory error, whether it is an error (normal functioning) or not (malfunctioning). Nespoulos et al. (1987) also hypothesized that the impairment demonstrated by conduction aphasics is at the level of the scan-copier mentioned in Shattuck-Hufnagel (1987). The scan-copier would not scan only the segments of a single word but rather those of an entire phrase, and then copy each segment into its respective syllable loci (if the subject is normal).  67  The checkoff monitor and the error monitor (to prevent unacceptable phonemic sequences) are in operation during the copying process. It is necessary to re-order the segments of words even though the information about their order is already available as part of their definition in the lexicon (Shattuck-Hufnagel, 1987) possibly because in one step of the speech planning process phonemes must be transferred/copied from one processor (the lexicon) to another (the word/phrase marker). Shattuck-Hugnagel commented that: "This process presumably involves many steps, among them one that is subject to single-segment errors at any position in the word" (Shattuck-Hufnagel, 1987, p.47). She suggested that the slot-and-filler model include five logically ordered steps: 1) the selection of a cohort of open or closed-class words from the lexicon by transferring lexical items to a short-term memory processing store, or by marking their lexical representation temporarily, their form, i.e., segments and order, are provided; 2) the syllabic structure is constructed as are other processes to assign lexical stress, processes which incorporate the word minus the onset; 3) non-onset portions are associated with their (non-onset) slots; 4) onset portions are associated with their (onset) slots; 5) the representation is transformed with hierarchical organization into a string of fully specified individual segments, which will be eventually translated into motor commands. Sublexical errors can occur during the transfer of information from the lexical processor to the phrasal processor. The phrasal framework specifies that there are onset consonants/portions and the rest of the word, but it does not specify the number of segments in either part. In lists, and possibly in single word production tasks, the processor goes from the second to the fifth step where all consonants, and possibly all segments for that fact, are susceptible to participate in an error,  "it appears that we have neuropsychological evidence that ordering and selection  mechanisms are separate from the mechanisms that constrain sequence of segments" (Buckingham, 1980, p.210). Caplan disputed the idea that a failure in the scan copier would  68  account for single word productions. Rather, he suggested that the copier copied sequences onto a phrase marker. Non-environmentally (non-contextually) determined additions or substitutions in which a segment not present in the word is inserted, and errors such as additions and omissions, are also due to the failure of the checkoff monitor or the error monitor. In such cases, the syllabic structure is also changed (altered) since a segment is added into a slot that apparently did not exist. Such errors may be due to: 1) an incorrect copying or incomplete or incorrectly transferred information to the slot; or 2) because a word is intruding in the planning segments by mistake, but is associated to the target word or phrase in the mind of the speaker because of what he is hearing or seeing while he speaks. It is not yet known whether the  syllabic structure of language invariably remains permissible following additions and  omissions. Nespoulos et al. hypothesized that the monitoring of psycholinguistic processes would be helped by permanence of visual stimulus and linear phonological representation would be more stable because of the permanent visual stimulus in reading. "In reading aloud, stimuli are presented through the visual pathway, and are thus permanently available and capable of constantly reinforcing the target representation" (Joanette et al:, 1980, p.39) which is an internal representation, i.e., the subject has an awareness of how a planned utterance is to be executed.  This representation is at the positional level, from which was derived the  representation from the lexico-phonological level. The two monitoring systems, even though they may be impaired, are counter balanced (if they are impaired) by the compensatory effect of the permanent stimulus. Hence the fewer errors, observed when A.K. was reading aloud, have at least two possible explanations: 1) stimulus permanence, 2) "grapheme-phoneme correspondence, ... helps ... to copy more efficiently the different segments required for the construction of linear phonological representation at the positional level" (Nespoulos et al.,  69  construction of linear phonological representation at the positional level" (Nespoulos et al., 1987,  p.78). According to Nespoulos et al. (1987), the 4 types of errors observed (i.e., substitution,  addition, displacement, omission) may arise following a disruption at the lexical level or at the phrasal level. A disruption at the lexical level would be based on the premise that segments are inserted into their syllabic position before lexical insertion into the syntactic frame, whereas a disruption at the phrasal level suggests that segment insertion into the syntactic frame would need to occur at the phrasal level (Nespoulos et al., 1987). The existence of both levels of disruption is not improbable: the choice of level for segment insertion would depend on the nature of the response in the production task, i.e., single word or phrase length. Hence, displacements as well as environmentally conditioned additions and substitutions can arise where a correctly chosen segment is assigned to an inappropriate slot. The hypothesis of the scan-copier is a strong argument, and explains A.K.'s errors, provided the checkoff monitor and the error monitor are also evoked in the model. As Caplan has stated: "Not all aphasic perfomances are deficits; hence lesions may create new functions which may or may not be localized, and which do not bear on normal mechanism" (Caplan, 1981, p.128).  Conclusion  In conclusion, the speech errors produced by the subject support previous findings that errors made by aphasics are mostly phoneme substitutions, one feature removed from the target phoneme. It was apparent that in most cases the subject was able to retrieve the meaning of the target word; her difficulty, on the other hand, was hypothesized to reside at a phonological level of speech production. This hypothesis was supported by the observation that the subject was aided by the permanence, of a stimulus in oral reading. In this task, she may have used a strategy by which each phoneme was converted (when possible) to its grapheme  70  cognate. The permanence of graphemic representation may have compensated for problems in monitoring or in using Shattuck-Hufnagel's (1979) hypothesized "scan-copier".  Her deficit  was assumed to be somewhat due to a partially inoperative linking address, i.e., when retrieving a word, the subject was at times able to access only part of the information about the form of the word. This was evidenced in instances when the subject clearly did not retrieve all the segments of the target word. Other errors were explained by a malfunctioning of the scan-copier mechanism, the check-off monitor, or both (Shattuck-Hufnagel, 1979).  71  BIBLIOGRAPHY Basso, A. Casati, G. & Vignolo, L. A. 1977. Phonemic identification defect in aphasia. Cortex, 13, 85-95. Baker, E., Blumstein, S. E. & Goodglass, H. 1981. Interaction between phonological and semantic factors in auditory comprehension. Neuropsychologia, 19, 1-15. Baum, S. R., Blumstein, S. E., Naeser, M. A. & Palumbo, C. L. 1990. Temporal dimensions of consonants and vowel production: An acoustic and CT scan analysis of aphasic speech. Brain and Language, 39, 33-56. Benson, F. 1979. Neurological correlates of anomia. In H. Whitaker & H. A. Whitaker (Eds.), Studies in neurolinguistics , vol. 4. New York: Academic Press. Blanken, G. 1990. Formal paraphasias: A single case study. Brain 534-554.  and Language,  38,  Blumstein, S. E. 1973. Some Phonological Implications of Aphasic Speech. In H. Goodglass & S. Blumstein (Eds.), Psycholinguistics and Aphasia (pp. 123-137). Baltimore: The Johns Hopkins University Press. Blumstein, S. E., Baker, E. & Goodglass, H. 1977. Phonological factors in auditory comprehension in aphasia. Neuropsychologia, 15, 19-30. Blumstein, S. E„ Cooper, W. E., Goodglass, H., Statlender, S. & Gottlieb, J. 1980. Production deficits in aphasia: A voice onset time analysis. Brain and Language, 9, 153-170. Blumstein, S. E., Cooper, W. E., Zurif, E. B. & Caramazza, A. 1977. The perception and production of voice onset time in aphasia. Neuropsychologia, 15, 371-383. Blumstein, S. E., Tartter, V. C, Nigro, G. & Statlender, S. 1984. Acoustic cues for the perception of place of articulation in aphasia. Brain and Language, 24, 128-149. Buckingham, H. 1980. On Correlating aphasics errors with slips of the tongue. Neurolinguistics,  Applied  1, 199-220.  Buckingham, H. W. 1982. Perseveration in aphasia. In S. Newman & R. Epstein (Eds.), Current  Perspective  in Dysphasia,  Churchill Livingston: London.  Buckingham, H. W. 1983. Neurolinguistic and philosophical implications of electrical stimulation mapping of the human brain. The Behavioral and Brain Sciences, 2, 209-21 0. Burns, M. S. & Canter, G. J. 1977. Phonemic behaviour of aphasic patients with posterior cerebral lesions. Brain and Language, 4, 492-507.  i  72 Butterworth, B. 1979. Hesitation and the production of verbal paraphasias and neologisms in jargon aphasia. Brain and Language, 8, 133-161. Caplan, D. 1981. On the cerebral localization of linguistic functions: Logical and Empirical issues surrounding deficit analysis and functional localization. Brain and Language, 14, 120-137. Caplan, D. 1987. Phonological representations in word production. In E. Keller & M. Gopnik (Eds.), Motor  and sensory  processes  of language.  New Jersey: Lawrence Erlbaum  Associates. Caplan, D. 1987. Neurolinguistics and linguistic aphasiology: An introduction. Cambridge: Cambridge University Press. Correia, L., Brookshire, R. H. & Nicholas, L. E. 1990. Aphasic and non-brain damaged adults' descriptions of aphasia test pictures and gender-biased pictures. Journal of Speech  and Hearing  Disorders,  55, 713-720.  Dell, G. S. 1986. A spreading-activation theory of retrieval in sentence production. Psychological  Fromkin, V. F. 1991.  Review,  93:3, 283-321.  Speech production. In Gleason &. Ratner (Eds.),  Psycholinguistics  today, MS to appear.  Garrett, M. F. 1984. The organization of processing structure for language production: Application to aphasic speech. In D. Caplan & A. R. Lecours (Eds.), Biological perspectives on language (pp. 172-193). Cambridge: MIT Press. Garnsey, S. M. & Dell, G. S. 1984. Some neurolinguistic implications of prearticulatory editing in production. Brain and Language, 23, 64-73. Goodglass, H., Berko Gleason, J., Ackerman Bernholtz, N. & Hyde, M. R. 1972. Some linguistic structures in the speech of a Broca's aphasic. Cortex, VIII: 2, 191-212. Goodglass, H. & Kaplan, E. 1972. Boston  Diagnostic  Aphasia  Examination.  Philadelphia: Lea  & Febiger. Guyard, H., Sabouraud, 0. & Gagnepain, J. 1981. A procedure to differentiate phonologial disturbances in Broca's aphasia and Wernicke's aphasia. Brain and Language, 13, 19-30. Henaff Gonon, M. A., Bruckert, R. & Michel, F. 1989. Lexicalization in an anomic patient. Neuropsychologia,  27:4, 391-407.  Itoh, M., Tatsumi, I. F., Sasanuma, S, & Fukusako, Y. 1986. Voice onset time perception in Japanese aphasic patients. Brain and Language, 28, 71-85. Joanette, Y., Keller, E. & Lecours, A. R. 1980. Sequences of phonemic approximations in aphasia. Brain and Language, 11, 30-44. Kaplan, E., Goodglass, H. & Weintraub, S. 1976. 7?7e Boston Naming Test. Lea & Febiger.  73  Katz, W. F. 1988. Anticipatory coarticulation in aphasia: Acoustic and perceptual data. Brain  and Language,  35, 340-368.  Katz, W., Machetanz, J., Orth, U. & Schonle, P. 1991. Anticipatory labial coarticulation in the speech of German-speaking anterior aphasic subjects: Acoustic analyses. Neurolinguistics,  to appear.  Kohn, S. E. 1984. The nature of the phonological disorder in conduction aphasia. Brain and Language,  23, 97-115.  Ladefoged, P. 1982. A course in phonetics. San Diego: Harcourt Brace. Jovanovich. 2  n d  ed.  Lecours, A. R. & Rouillon, F. 1976. Neurolinguistic analysis of jargonaphasia and jargonagraphia. In H. Whitaker and H. A. Whitaker (Eds.), Studies in Neurolinguistics, vol.2 (pp. 95-144). New York: Academic. LeDorze, G. & Nespoulos, J.-L. 1989. Anomia in moderate aphasia: Problems in accessing the lexical representation. Brain and Language, 37, 381-400. MacKenzie, C. 1982. Aphasic articulatory defect and aphasic phonological defect. Journal of Disorders of Communication, 17:1, 27-46.  British  Macneilage, P. F. 1982. Speech production mechanisms in aphasia. In S. Grille/ et al. (Eds.), Speech Motor Control, (pp 43-60). Oxford: Pergamon. Marshall, R. C. & Tompkins, C. A. 1982. Verbal self-correction behaviors of fluent and nonfluent aphasic subjects. Brain and Language, 15, 292-306. Monoi, H., Fukusako, Y. & Sasanuma, S. 1983. Speech sound errors in patients with conduction and Broca's aphasia. Brain and Language, 20, 175-194. Nespoulos, J.-L., Joanette, Y., Ska, B., Caplan, D. & Lecours, A. R. 1987. Production deficits in Broca's and conduction aphasia: Repetition versus reading. In E. Keller & M. Gopnik (Eds.), Motor and sensory processes of language. New Jersey: Lawrence Erlbaum Associates. Nicolosi, L., Harryman, E. & Kresheck, J. 1983. Baltimore: Williams & Wilkins. 2nd ed.  Terminology of communication disorders.  O'Grady, W. & Dobrovolsky, M. 1987. Contemporary linguistic analysis: An introduction. Toronto: Copp Clark Pitman Ltd. Perecman, E. & Kellar, L. 1981. The effect of voice and place among aphasic, nonaphasic right-damaged, and normal subjects on a metalinguistic task. Brain and Language, 12, 213-223. Ryalls, J. H. 1981. Motor aphasia: Acoustic correlates of phonetic disintegration in vowels. Neuropsychologia,  19:3, 365-374.  Saffran, E. 1982. Neuropsychological approaches to the study of language. British  Journal  74 of Psychology,  73, 317-337.  Schlenck, K.-J., Huber, W. & Willmes, K. 1987. "Prepairs" and repairs: Different monitoring functions in aphasic language production. Brain and Language, 30, 226-244. Shattuck-Hufnagel, S. 1979. Speech errors as evidence for a serial order mechanism in sentence production. In W. Cooper & E. C. T. Walker (Eds.), Sentence processing: Psycholinguistic studies presented to Merrill Garrett, (pp. 295-342). Hillsdale, NJ: Lawrence Erlbaum Associates. Shattuck-Hufnagel, S. 1987. The role of word-onset consonants in speech production planning: New evidence from speech error patterns. In E. Keller & M. Gopnik (Eds.), Motor  and sensory  processes  of language.  New  Jersey: Lawrence Erlbaum  Associates. Shinn, P. & Blumstein, S. E. 1983. Phonetic disintegration in aphasia: Acoustic analysis of spectral characteristics for place of articulation. Brain and Language, 20, 90-114. Tuller, B. 1984. On categorizing aphasic speech errors. 547-557.  Neuropsychologia,  22:5,  75  Appendix A Transcription of tasks 1-6* target/ attempt tree comb  txn kae  saw  kom so  helicopter  ae  broom  brA  he brum oka  oksopAS  mushroom mar  rriAj  rriAj-rumz  clothes  kel  kozl haertkar  JoctojDuj;  hanger  haer  haengl  mask  ma  massk  jrjretzel[  phu  bench sne  sneil  atlas  aetl-las  aet  mouth joj^an_ acorn  par  par-rezal  ae  aetlas  p/ wezal  bents  snail  muskrat  -koptar 1  mAskare-  m  mAskraet  mous  m  mau9  m  mau0 gan  ek  ek-a-rAn  ke-grAn  e-ra-ron  e-kar  krast or-  ekarAn  76 igloo  i  igxou  igarai  escalator  e  ekstsr  eks  harp  h  harp/  harp  asparagus  a  asskparagus  scroll  s skol  „  ig  iglu  eks  eks-s-... kaertaerlar  1  yoke  _  „„  sk  ,,  skA  sk-te  es-k-r-krol  °  jo/ok  easel  i  Tgal  neck  n  ne  Chinese  tje  tjainiz  ke  kontirj  dishes  d  difaz  girl-boy  g-  boij  taking  t  to  SkAS  skar skorl  sku-wal  •  Tzal  TEf nek  —*  counting  1  —  1  —  tekn ...  some  so  som ™  other  5or  they  5o  8e  tipping  P  tipirj  sneak  o0ar  slik  is  s  IZ  kite  ka  kaut  some  s  som  girl-lady  g  1  8ar  snik  leidi  wmwMSfvvvvvvvwwwwwwwvvwwm  Smith  s  samiO  all  1  al  grass  g  gaeraes  cries  ki  kraiz  bites  b  ba-  bone  > b " " " " " be  teeth  "8  !  bais b  boun  tie  toy  tae  toj  smoke  s  smok  g  gaun  afraid  ae-f  fard  aefreid  story  s  F.  sai  red  r  these  h  red hiz  children  t  tfildran  stsri  ...L...LJ.J  ...... nvwvwwvwmMW.  r  Rick children  k  rrk tfildran  wedding  Wll  widm  wede  married  me  mard  m  her  fh  maridi  - - - - '™~*-»>'v>~>o^vvvvvvvv  hsr airin  Irene [  her (his)  f ti  hsr  girlfriend  jgsr  garl-frend  ..  79 no  n  no  no  no  wedding  w  widirj  cut  k  kot  Mike  ma  maik  this  a  5is  nine  n  nam  husband  h  hAsbond  went  w  Ment  to  t  tu  together  t  tu-garSar  should  s  Jud  to  t  tu  Quebec  k  kw  kwab-bek  camp  kor  kau  koump  1  --  r r  P  -  nPnPf  „„„„„„„, . 1rl  Prrrrn(  i  """""""  ••ft,,, - , - . . , - r . w , ™ , m «  Borden  m  muar  camp  k  kaemp  on  0  on  between province Strong  burdan  pitwid  L5L mwQ  pro  —  prov/bms  strorj  is  "haed moved floor  ..„,„. ,„„  mi  if  rn  rrnr  ..  r  mud fwar  far-br  l  80  Qer  somebody (there)  SAm  Japanese  °5  cfeaspaeniz  she  J  Ji  very  f  fori  has  h  spine  SAm-bsSs  LUULLLUUUU.... 1 .1 . 1. .  .  _  hsez  the  a  5e  ^et (?)  f  find  both  b  bo3a  didn't  d-A  didnt  gymnasiu m  t  g  ^ qsim-n-ne-  ae  dining  d  dainirj  renovated  re  r  back  ba  bask  rem  8a  the  Peace  Spain  ziArn  they  ye?  SAmbsda  [j pit  J  renso-fetsd  1  e s  P  1  i  ™  phis  *Note: only those targets in which A.K. demonstrated a "conduite d'approche" were included in the appendices.  81 Appendix B Transcription of task 7 target / attempt somebody yoke merote P'd mushroom pult rath smoke perashil neck taking blait renovated fepeet story married pafade hanger rapeenal Mbmy. camp atlas winute muskrat mouth fidety peace octopus yard other sheparty sprickater igloo Borden members escalator backyard sprit  1  12  SAm-bAdi  |  4  3  juk msr-Po-ot p i d  mAj-rum  [  p A l t  j  JraeG  r smok  l.  pair-a-jfi-71  nek tekxin bket nen-o-fenn f stoni m paef-ed haengk*raep:inasl ga-bini khaemp ?aet-lae/as wintfut mAskl maof f pis ok-topAS jae/srd o6&-  ! par-d-Ji/ii | >  I j feb-p-pit j j maerid  —  1  —  1 j f ,mmmnnii.mm..*vmmmr*mmwt*  1 f | mAsk-raet | maoG fidel-iti | I ....<  r*mmn  ,,.,.„  •wwmmmm*..,,..  ,  \  jfe-par-fi pr-nk-kae-tae-sr ba^-dn mem^z esksl-eta P spr-rA/it 1  [ pnkaetsr i | | | paeg-jog-jord \ spa-t  K"  m,n  mm*  irmmm n  • mrnrm  **r*mm*mm*mm .  L  1 II 1  bites pab ropabite fide snail cursonal dedew pretzel tree province harp pirrel floor perimto rad bench spine wedding shovel meview acorn pereet carible Quebec afraid scroll bone easel bebuse neff dog greep  82  bed  baits  paeb r faid sm kV-AS-sn ded-j?u pretSAl  roba-bait  r-roba-bait sn-neil ka'-osenasi ded-ju  nrP  ka'-s-en-Al  tjyi pa-vms  p^vis-f-f haea-p parrel  pa^vis 1  pa<~imp-to rasd bentf s wedirj  pro-vms  —  Spain  JAVSI  ven-nu  vel-nu as-korn  k  venu  pa^-it  k  kar-ibl kxibik  kxi ae-freid sj  moon ez beb nef dog gr  kxe-bek-bek  sk boon izol baeb-AS  sknol  gr-ri7ep  grip  baebus  . „,„, „  rnr ri b  racroon gymnasium sneak helicopter lady clothes  n  raek-r <%im-naest/-JAm sn/ik  \ raek-run j d5im-nezJAm |  hei-ik-kbpta>-  |  leidi  klo8  1  [  ull  uulluuJUU  „ . „ „ _ „ ,  83  Appendix C Confusion matrix of substitution errors in tasks 1-7  


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



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


Related Items