A PROPOSED MODEL OF SPEECH PERCEPTION SCORES IN CHILDREN WITH IMPAIRED HEARING

Transcription

1 A PROPOSED MODEL OF SPEECH PERCEPTION SCORES IN CHILDREN WITH IMPAIRED HEARING Louise Paatsch 1, Peter Blamey 1, Catherine Bow 1, Julia Sarant 2, Lois Martin 2 1 Dept. of Otolaryngology, The University of Melbourne 2 The Bionic Ear Institute, Melbourne ABSTRACT: Open-set word and sentence perception test scores are commonly used as a measure of hearing abilities in children and adults using cochlear implants and/or hearing aids. In the case of children, it is difficult to interpret speech perception scores without considering the effects of lexical knowledge and speech production abilities on their responses. This study postulated a simple mathematical model to describe the effects of hearing, lexical knowledge and speech production on the perception test scores for monosyllabic words by children with impaired hearing. Thirty-three primary-school children fitted with hearing aids and cochlear implants were evaluated using speech perception, speech production and language measures. The mathematical model demonstrated that the levels of performance in an open-set word perception test in the auditory alone mode is strongly dependent on residual hearing levels, lexical knowledge and speech production abilities. Further applications of the model provided an estimate of the effect of each component on the overall speech perception score for each child. INTRODUCTION Open-set word and sentence tests have become a standard method of speech perception evaluation for children and adults with hearing impairments using cochlear implants or hearing aids. Overall scores on speech perception tests have been used to measure the amount of sensory evidence available from the damaged ear beyond the person s pure tone threshold levels (Boothroyd, 1978) and have demonstrated the effectiveness of particular devices in developing speech perception abilities (Boothroyd, 1995; Dawson et al., 1992). Results from such tests have also been used as part of the selection criteria for cochlear implantation (Dowell, 1997), and have measured change in perception abilities after the fitting of sensory devices (Dawson, et al., 1992). Reporting of results on speech perception tests have led to the planning of appropriate habilitation programs for children and adults with hearing impairments (Erber, 1979; Paatsch, 1997). It is usually assumed that language and speech production do not influence speech perception results in tests that are based on age-appropriate vocabulary for the person being tested (Tyler, 1993). In the case of adults where the hearing loss is acquired postlinguistically, it is usually safe to assume that they have adequate language to perform the task, therefore eliminating the factors unrelated to the person s hearing abilities. In the case of children, however, the interpretation of an open-set speech perception test score is complicated by the fact that errors may arise from more than one process: hearing, speech production and lexical processing. An open-set word perception test requires an individual to listen to, and repeat, a set of stimulus words in either the auditory or auditory-visual condition. The probability of a correct response relies heavily on an individual s ability to identify or discriminate phonetic contrasts, and will also depend on knowledge of the lexical items in the test (Boothroyd & Nittrouer, 1988), and the ability to pronounce all phonemes within the stimulus word. Spoken word recognition also depends on the structural relations among the sound patterns of words in memory, in particular, the effects of word frequency and word neighbourhoods in the recognition of the stimulus word (Luce & Pisoni, 1998). Practically, therefore, it is difficult to determine the effects of hearing on speech perception scores in children with hearing impairments without considering the effects of language and speech production abilities. A speech perception score may underestimate a child s ability to make auditory discriminations between speech sounds. A significant relationship between language, speech perception and speech production abilities of children with a hearing loss was reported in Blamey et al. (2001). Scores on the open-set word and Accepted after abstract review page 421

2 sentence perception tests were strongly related to language levels and speech production scores in both the auditory-visual and auditory alone test conditions. This paper used a linear addition regression model to express speech perception scores as a function of pure-tone average hearing loss, equivalent language age, and percent correct production of consonants in conversational speech samples. The limitations of the linear additive model were discussed in the paper. The present analysis of similar data investigates a non-linear model in which the factors are multiplicative rather than additive. Theoretically, separation of the contributions of sensory processes, lexical processes, and speech production processes would make the results of a speech perception test more specific and more reliable. It would also be useful in evaluating the effects of habilitation such as auditory training, vocabulary training, or speech production training on speech perception. The aim of this study was to separate the contributions of sensory processes from language and speech production in perception tests of monosyllabic words by children with impaired hearing. If this can be done reliably, it will be possible to see how each of the components changes over time, and which component(s) are most easily improved by specific habilitation. The separation was effected by testing in two different sensory conditions: auditory alone and reading of the word lists. A mathematical model was postulated to describe the combined effects of the three components. This model predicts that there are three independent contributions to speech perception a sensory component, a lexical component, and a speech production component. It also hypothesized that the contributions of lexical knowledge and speech production would be the same in the two conditions of auditory alone speech perception and reading. METHOD The participants in this study were thirty-three children (16 M, 17 F) aged between 6 and 14 years (mean 10.8 years) with a range of aetiologies (19 unknown, 7 genetic, 3 rubella, 2 CMV, and 2 meningitis). Twenty-one of the children were fitted with behind the ear hearing aids with a mean pure-tone-average (PTA) threshold of 81 db HL averaged over the three frequencies 500, 1000 and 2000 Hz. Twelve children, with a mean pre-operative PTA of 109 db HL, were cochlear implant users. Twenty-one of the children had a hearing loss greater than 90 db HL, seven had a severe hearing loss (70 db HL to 90 db HL) and five had a moderate hearing loss (40 db HL to 70 db HL). All 33 children attended mainstream primary and/or secondary schools where an oral/aural method of communication was used. Each school provided additional support for these children during mainstream classroom activities and within small group and/or individual sessions. None of the children in the study had an uncorrected visual impairment or a known sensory dyslexia and all had the basic reading skills necessary to read lists of monosyllabic words. A test battery to assess speech perception, speech production and vocabulary was administered to the thirty-three children who took part in this study. Speech perception abilities of the children were assessed using the Consonant-Nucleus-Consonant (CNC) monosyllabic word test (Peterson & Lehiste, 1962). This is an open-set speech perception test, consisting of phonetically balanced lists of fifty words. Each word was presented one at a time without a carrier phrase using live voice, by a female speaker, at a distance of approximately one metre in a quiet room. Two different lists of the CNC word test were presented in the auditory alone condition. These two lists, plus a further two lists of CNC words, were read aloud by the participants to provide data for comparisons between the read lists and those in the auditory alone condition. All participants verbal responses were videotaped then scored and transcribed phonetically offline, by two linguists using broad transcription (without the use of diacritics). A word was scored as correct if the child s response word matched the stimulus test word. This required all phonemes to be produced correctly with no omissions, additions or substitutions. Inter-list variability was assessed by comparing the scores for the pairs of CNC word lists in each condition, for each child. The standard deviation of differences across all conditions was 5.45% for lists of 50 words. Speech production skills were measured by obtaining conversational speech samples from each participant. These samples were elicited by an audiologist or teacher of the deaf using prompting questions about familiar topics. All conversations were videotaped, and on average, a total of 60 utterances were transcribed phonetically by two linguists. All utterances were transcribed using broad phonetic transcription so that productions of phonemes could be compared with those represented in Accepted after abstract review page 422

3 the CNC word test. Percentage consonant correct scores (PCC) used only singleton consonants from the conversation samples. The Peabody Picture Vocabulary Test (PPVT III; Dunn & Dunn, 1997) was used to measure each participant s receptive vocabulary. The raw scores were converted into age equivalent scores which are based on the age at which the average score of normally-hearing children is equal to the score of the child being evaluated. The PPVT III age equivalent scores for the thirty three children in this study ranged from 4 years 10 months to 14 years 9 months (mean 7.07). The mathematical model and statistical analyses It was assumed that there is a probability of correct hearing, correct lexical access and correct speech production that are independent of one another such that the probability of making none of the three types of errors can be expressed as the following equation. P T = P S x P L x P P (1) P T is the total probability of a correct response, P S is the probability of sensory information being transmitted to the lexicon, P L is the probability of identifying the stimulus in the lexicon, and P P is the probability of correct speech production. This simple model does not allow for the possibility that the child may use a purely phonological process that does not involve lexical access. Equation (1) can be further expressed in the following equations to take into account the two different sensory conditions measured in this study (auditory alone (A), and reading (R)). P TA = P SA x P L x P P (A = Auditory alone) (2) P TR = P SR x P L x P P (R = Reading) (3) It should be noted that the probabilities of lexical correctness (P L ) and speech production correctness (P P ) are hypothesised to be the same in both conditions. Since none of the children had an uncorrected visual impairment, it is also hypothesized that there are no sensory errors in the reading condition so that P SR is equal to one (P SR = 1). These hypotheses are specifically tested using the experimental data in this study. It should also be noted that P TA and P TR are the overall performance scores in the two conditions. Using the results from the reading condition, it is therefore possible to calculate the sensory probability in equation (2) (P SA ) by: P SA = P TA /P TR (4) Paired t-tests were used to test the model predictions that the total scores in the reading condition should be greater than the overall scores in the auditory alone condition (P TR > P TA, ) and that P SR is equal to one (P SR = 1). Regression analysis was used to test the model predictions that: the overall scores in the auditory alone condition (P TA ) should be significantly associated with measures of hearing, vocabulary, and speech production; the total scores in the reading condition (P TR ) should be significantly associated with measures of vocabulary and production but not measures of hearing; and the scores from the calculations of sensory probabilities in the auditory alone condition (P SA = P TA /P TR ) should be significantly associated with measures of hearing but not with measures of speech production or vocabulary. The mathematical products in equations (2) and (3) were transformed to sums by taking logarithms, expressed in the following equations. log P TA = log P SA + log P L + log P P (5) log P TR = log P L + log P P (6) This transformed model predicts that linear regression analyses of log P TA and log P TR should have significant factors related to sensory measures (log P SA ), and speech production measures (log P P ). It should be noted that the hypothesis that P SR = 1 in the multiplicative model is equivalent to the hypothesis that log P SR = 0 in the transformed model. In order to assess these factors, the measure of hearing used was pure-tone average (PTA) thresholds for hearing aid users, with an equivalent hearing loss value of 78 dbhl for the cochlear implant users. This value was derived from the recent study by Blamey et al. (2001) in which it was shown that a group of forty-seven children using Accepted after abstract review page 423

4 cochlear implants performed similarly to a group of forty children using hearing aids with an average of 78 dbhl on a broad range of speech perception and language measures. This corrected hearing measure is labelled PTAC. The value of 78 dbhl takes into account the average improvement in hearing provided by the cochlear implant. The vocabulary measure used in the analyses was derived from the equivalent age from the PPVT III. Blamey et al. (2001) showed that there was a non-linear relationship between language age and speech perception scores such that language ages up to seven years were associated with increasing perception scores but further increases in perception were not observed above the language age of seven years. To allow for this non-linearity, the present study uses a corrected PPVT equivalent age, PPVTC, which was calculated from the PPVT III equivalent ages, by imposing a maximum value of seven years. The speech production measure is the percentage of consonants correct (PCC) from the transcription analyses of conversations. Because PCC is actually a percentage measure, like the P P probabilities in equations (5) and (6), log PCC was used in the regression equation. This means that the coefficient of log PCC in the regression analysis has a simple interpretation: P P = (PCC) C where C is the regression coefficient. The regression analyses were performed and the hypotheses and predictions were tested using t-tests on the regression coefficients. RESULTS The mean scores for percentage CNC words correct indicate that the mean A-word score (P TA = 47.8 sd 21.2) is less than the mean R-word score (P TR = 78.5 sd 16.8), (t (32)= 9.39, p < 0.001) as predicted by the model assumption that P SR = 1. Correlations for hearing (as measured by PTAC), lexical knowledge (as measured by PPVTC), and speech production abilities (as measured by log PCC) show that none of these measures were significantly correlated with one another at the 95% confidence level in this group of children. As a result, PTAC, PPVTC and log PCC can be used as independent variables in later multiple regression analyses to estimate the relative size of their contributions to speech perception. Table 1. Regression analyses of word scores as a function of PTAC, PPVTC and log PCC. PTAC PPVTC log PCC log P TA Coef St Dev t p log P TR Coef St Dev t p log P SA Coef St Dev t p Table 1 shows the results of regression analyses for logarithmically transformed word scores as a function of pure tone average hearing loss (PTAC), lexical knowledge (PPVTC) and logarithmically transformed percentage consonants correct (log PCC). P values for coefficients that are significantly different from zero at the 95% confidence level are shown in bold. The units for the PTAC regression coefficients are db 1. The units for the PPVTC regression coefficients are years 1. The coefficients of log PCC are dimensionless numbers because they are the ratio of one percentage compared to another. As predicted by the model, the coefficient for PTAC is not significantly different from zero for log P TR. Non-significant results are also apparent for the coefficients for PPVTC and log PCC for log P SA as predicted by the model. The coefficients for PPVTC in both conditions are significantly different from zero, and are well within two standard deviations of one another, indicating that they are not significantly different (p > 0.05). Similarly, the coefficients for log PCC are significantly different from zero but not significantly different from one another as hypothesised in the model. Accepted after abstract review page 424

5 DISCUSSION The contributions of hearing, language and speech production to the perception of monosyllabic words were separated through the application of the proposed non-linear mathematical model. All the predictions of the model were confirmed by the data analysis in both conditions. The model showed that the levels of performance in an open-set word perception test in the audition alone mode was strongly dependent on residual hearing levels, lexical knowledge, and speech production abilities. Higher overall scores were evident in the reading condition showing a strong dependence on speech production abilities and lexical knowledge, but no dependence on residual hearing levels. In both the auditory alone and reading conditions, the child is required to perform a distinct number of information-processing tasks. Levelt (1989) describes the retrieval of a word from the lexicon as being based on the word s meaning, with the addition of syntactic, morphological, and phonological information. Levelt also suggests that speakers construct the framework of a word or utterance without a great deal of regard for the phonology of words. In the reading task, the child is required to identify letters, retrieve the word s phonological form, access the semantics of the word, and produce the sounds of the word as they are represented by the letters. Coltheart & Leahy (1996) describe a dual route theory of converting print to sound; ie, using lexical and non-lexical procedures. Both these procedures must be accessed when reading words aloud. In the auditory alone condition, the dual route theory may also be applied. In this case the dual routes are a lexical procedure, which requires auditory input rather than visual, followed by a retrieval of lexical items stored in the mental lexicon. The non-lexical procedure involves the imitation of phonemes. In the case of the present study, the meanings of CNC words were not tested. As a consequence, the child may not have had all words within their lexicon and may only be accessing the non-lexical route when trying to read the unknown words. Results of the present study and a number of findings discussed in the study by Blamey et al. (2001), showed that speech perception scores were significantly influenced by language, speech production performance, and hearing levels. Another similarity is that no significant correlation was found between the measures of hearing, speech production abilities, and lexical knowledge, allowing each to be used as independent variables in the data analysis. In contrast to the linear regressions used to determine the effects of the three variables on the overall speech perception score in the study by Blamey et al. (2001), the present study used non-linear functions and multiplication of factors rather than addition of factors to separate the three components. This should provide a more plausible theoretical background and is likely to give a more reliable estimate of the size of each component in the equation for the overall speech perception score. A further refinement was applied in the present study through the introduction of the reading condition. It can be assumed that the effect of hearing has been eliminated in this condition, and what remains is a measure of speech production and lexical knowledge. This approach provides a further application to the proposed model by calculating the sensory component (as presented in Equations (5) and (6)) for individual children. This direct measurement provides a better estimate of the effects of hearing on the overall score of speech perception. The proposed model in this study can be applied in the future to predict speech perception performances of hearing impaired children. This may lead to better analysis of performance in openset word perception tests for children being considered for cochlear implantation, and lead towards establishing further criteria for selection. The model will also provide a more analytical way of comparing the effectiveness of listening devices based on children s performance on open-set speech perception tests. The separation of lexical knowledge and speech production abilities from hearing levels will enable a more accurate evaluation of the amounts of sensory gain provided by specific listening devices. The model can also be applied to evaluate the success of specific habilitation methods by measuring the relative size of each of the three components after training. ACKNOWLEDGEMENTS The authors wish to acknowledge the children, parents and staff from Mountview Primary School and St. Mary s School for Children with Impaired Hearing Inc. for their cooperation and assistance in the collection of data. Financial support for the study was provided by the Australian National Health and Medical Research Council project grant # and fellowship # Accepted after abstract review page 425

6 REFERENCES Blamey, P. J., Sarant, J. Z., Paatsch, L. E., Barry, J. G., Bow, C. P., Wales, R. J., Wright, M., Psarros, C., Rattigan, K., & Tooher, R. (2001). Relationships among speech perception, production, language, hearing loss, and age in children with impaired hearing. Journal of Speech, Language, and Hearing Research, 44 (2), Boothroyd, A. (1978). Speech perception and sensorineural hearing loss, in M. Ross & T. Giolas (Eds.), Auditory Management of Hearing-Impaired Children (pp ). Baltimore: University Park Press. Boothroyd, A. (1995). Speech perception tests and hearing impaired children. In G. Plant & K.-E. Spens (Eds.), Profound Deafness and Speech Communication (pp ). London: Whurr Publishers Ltd. Boothroyd, A., & Nittrouer, S. (1988). Mathematical treatment of context effects in phoneme and word recognition. Journal of the Acoustical Society of America, 84(1), Coltheart, M., & Leahy, J. (1996). Assessment of Lexical and Nonlexical Reading Abilities in Children - Some Normative Data. Australian Journal of Psychology, 48(3), Dawson, P. W., Blamey, P. J., Rowland, L. C., Dettman, S. J., Clark, G. M., Busby, P. A., Brown, A. M., Dowell, R. C., & Rickards, F. W. (1992). Cochlear implants in children, adolescents, and prelinguistically deafened adults: speech perception. Journal of Speech, Language and Hearing Research, 35(2), Dowell, R. C. (1997). Preoperative audiological, speech, and language evaluation, in G. M. Clark, R. S. C. Cowan & R. C. Dowell (eds), Cochlear Implantation for Infants and Children: Advances, CA: Singular Publishing Group Inc, pp Dunn, L. M., & Dunn, L. M. (1997). Peabody Picture Vocabulary Test. (Third ed.). Minnesota: American Guidance Service. Erber, N. (1979). An approach to evaluating auditory speech perception ability. The Volta Review, 81(1), Levelt, W. J. (1989). Speaking: from intention to articulation. Cambridge, MA: Massachusetts Institute of Technology. Luce, P. A., & Pisoni, D. B. (1998). Recognising spoken words - The neighbourhood activation model. Ear & Hearing, 19(1), Paatsch, L. (1997). The effectiveness of the Auditory Skills Program in developing auditory skills in severe to profound hearing impaired children. Unpublished Master of Education, The University of Melbourne, Melbourne. Peterson, G. E., & Lehiste, I. (1962). Revised CNC lists for auditory tests. Journal of Speech and Hearing Disorders, 27, Tyler, R. S. (1993). Speech perception by children, in R. S. Tyler (Ed.), Cochlear Implants: Audiological Foundations, CA: Singular Publishing group, Inc, pp Accepted after abstract review page 426