Audiologist and Speech Therapist

AUTHORS

MR. MAINAK SANTRA

AUDIOLOGIST AND SPEECH THERAPIST

HEARING PLUS, Opposite Gajkumar Brothers,, 4, Moti Sil St, Dharmatala, Taltala, Kolkata, West Bengal 700013

Ph no- 9831413770, Mail id-

Corresponding Author

MR.INDRANIL CHATTERJEE

LECTURER

ALI YAVAR JUNG NATIONAL INSTITUTE FOR THE HEARING HANDICAPPED

EASTERN REGIONAL CENTER

B.T. ROAD, NIOH CAMPUS

BONHOOGHLY,

KOLKATA-700090.

Ph no-9433102816

Mail id-

MS. ARPITA CHATTERJEE SHAHI

AUDIOLOGIST AND SPEECH THERAPIST

ALI YAVAR JUNG NATIONAL INSTITUTE FOR THE HEARING HANDICAPPED

EASTERN REGIONAL CENTER

B.T. ROAD, NIOH CAMPUS

BONHOOGHLY,

KOLKATA-700090.

Ph no- 8820688172, Mail id-

PSYCHOMETRIC VALIDATION OF PHONEMICALLY BALANCED WORD LISTS IN BENGALI

MAINAK SANTRA,1 INDRANIL CHATTERJEE,2 ARPITA CHATTERJEE SHAHI3

ABSTRACT

Speech audiometry is an essential component in the battery of audiological tests battery. The plethora of research studies advocates that during speech audiometry an individual needs to be tested in his or her own native language. The present study aimed to develop word lists that exhibited familiarity, homogeneity and phonemic balance. Twenty five items were finalized from the parent list. Standardization of the test material was done on thirty native normal hearing subjects. Validity measures were obtained by comparing the CID W-22 lists to thirty native speakers with L1 Bengali and L2 English. The slopes of the mean psychometric functions for wordlist one and two were 3.5%/dB and 3.6%/ dB respectively. The homogeneity of the word lists was examined by evaluating the inter-list equivalence, inter-word variability and inter-subject variability. The inter-word and inter-subject variability for both the list was 4.3dB and 4.5 dB; and 3.5 dB and 3.4 dB respectively. Binomial probability confirmed that the two lists were essentially equivalent at all presentation level. Based on the above findings it can be concluded that the material can be used to assess word recognition scores for the native Bengali speakers.

Keywords: phonemic balance, psychometric functions, reliability, validity.

For the successful rehabilitation for the persons with hearing impairment, the first step is to conduct a comprehensive audiological assessment. As an integral part of assessment speech audiometry had been used traditionally to determine the speech threshold and speech recognition. The dialectical difference between two languages makes it difficult to use it in the nonnative speakers. Even, if two languages have similar sound system they may vary significantly in their physical characteristics "including both acoustic and articulatory characteristics.21 Word recognition testing is further dramatically influenced by the phonetic, melodic, and intonational differences among languages.6

An individual being tested in a non-native language would not be familiar with the majority of the presented words. This could result in the stimulus words becoming nonsense test items.16 The individuals who have learned English as a second language after puberty perform significantly worse on speech audiometry tests than those who are native speakers of English.13 Even individuals, who are considered bilingual, have been shown to perform poorly on speech audiometry tests as compared to monolingual speakers of the test language. 8, 11, 13, 20, 22 So to serve the colloquial population the authors developed a test material in Bengali. This language occupies the second position among all the scheduled 22 languages of India. It is spoken by 83, 369,769 people in India (8.11% of the total population of India).7

Normally, we associate the term phonetic balance (or PB for short) with word recognition scores. But the term phonetic balance had been used erroneously. It has been realized that the test material used to assess word recognition scores should have a phonemic composition equivalent to that of everyday speech “Phonemically balanced” words that is, the different phonemes in the test material should occur with the same relative frequencies as in everyday speech. The rationale is; the lucidity of a listener is more competent with a more frequent phoneme compared to a less frequent one in his daily life.

METHODS

The aim of the present study was the psychometric validation of Phonemically Balanced word lists in Bengali, for the purpose of assessing word recognition scores.

PARTICIPANTS

The study included four groups of participants. The first group of participants included five native speakers with L1 Bengali. They independently rated the lists of monosyllabic words for familiarity. The second group consisted of thirty native speakers with L1 Bengali and L2 English (13 male, 17 female), they participated for the evaluation of monosyllabic words. Further these subjects were screened for no positive history of ear diseases and normal hearing. The third group was another 30 normal hearing subjects who apart from being native speakers of Bengali were also proficient in speaking English. The final group had three subjects with hearing impairment; the first of them was diagnosed as Bilateral Moderately Severe Sensori-Neural hearing loss, the second diagnosed as having Moderately Severe Mixed hearing loss in right ear and Profound hearing loss in left ear and the third subject was diagnosed as Bilateral Moderate Conductive hearing loss. Consent letter from the participants were taken for ethical consideration.

INSTRUMENTATION

The 30 subjects were tested in a double room audiological testing suite. MAICO MA 53 Dual Channel Diagnostic Audiometer calibrated using ANSI (2004) specifications.1 The stimulus (both pure tones and speech) were delivered monoaurally through the Telephonics TDH – 39 supra – aural headphones.

PROCEDURE

PHASE – I: Development of the test material

The test material consisted of commonly used monosyllabic words in Bengali, with no duplicate words across the two lists. The CVC syllable structure was used. The lists consisted of words with same relative frequencies as in everyday speech. The average difficulty and the range of difficulty of each list was equivalent to that of the other list. Monosyllabic words from two sources.9,18 Familiarity was tested by 5 adult native speakers of Bengali. Judgment was done on a 3 point rating scale. The familiarity ratings was given weight ages such as – most familiar (+3), fairly familiar (+2), very unfamiliar (+1).

For the purpose of consonant balancing the number of times a consonant should appear was calculated using the following e.g.: the phoneme /r/ appears 8.24 times out of 100 phonemes spoken or written, so when the number of phonemes is 150, the occurrence of /r/ is:

Number of occurrence for a phoneme (in %)100 ×150

So, each word list should have twelve (12) /r/ phonemes approximately. After having calculated the occurrence of each phoneme out of 150, fraction or decimal value was rounded off using to the next higher number. Finally, fifty PB wordlist were alphabetically developed.

PHASE – II: Standardization of the test material

For the psychometric or performance – intensity functions for the word lists. The 30 normal hearing subjects evaluated the monosyllabic words by listening and responding to the words. They were provided with two to three rest periods of about 5 minutes in between. The subjects were presented with the word list at various intensity levels ranging from 10 to 70 dBHL. Each subject listened and responded to approximately 1200 presentations (600 words for each ear). The order of the words and the presentation levels were randomized. They were not made familiar with the word list prior to the commencement of the test. The word lists were presented using monitored live voice (MLV) through the stereo live speech microphone. Speech microphone was kept constant at six inches. A thin paper separator was used between the nose and the mouth of the tester/speaker to prevent the unnecessary noise generated due to nasal emissions. Each word was preceded by the carrier phrase “|apni bolun|” (say the word).

PHASE – III: Correlation of live vs. recorded voice sample

Adobe audition CS6 and Nuendo 6 softwares were used to record and edit the live voice and converted to recorded voice for analysis. The recorded speech was presented via Sennheiser- HD 202 II headphones with microphone.

Scoring

The tester marked the subject’s response as either correct or incorrect on a response/score sheet. A sample of the response sheet is provided at Appendix – I. Test retest and validity was measured.

Half List formation

In order to develop the half list, the threshold (or the 50% recognition point) in dBHL for each word in both the wordlist 1 and 2 was calculated. The words were then arranged in descending order from most difficult to the easiest. The original wordlist 1 and 2 was thus divided to form two half lists.

STATISTICAL METHOD

Mean and standard deviation at which the decade interval percent correct recognitions were obtained for the wordlist one and two was calculated for the subject and for the words. It was hypothesized that There will be no significant difference between the word recognition scores obtained between the CID W-22 and the developed PB word list (H0: μ Wordlist 1 & 2 = μ CID W - 22 against the alternative hypothesis Ha: μ Wordlist 1 & 2 ≠ μ CID W - 22). Two tailed tests were used to test the hypotheses. The null hypothesis in all the three cases was rejected if the value of the test statistics (S) was greater than 1.96 at 5 % level of significance.

RESULTS AND DISCUSSION

This study was done to develop and standardize the materials to be used to assess word recognition scores of the native speakers of Bengali. Test materials in other languages used with the native speakers of Bengali, will not give an exact estimation of the subject’s speech recognition ability. Hence, there was a need to develop and standardize test materials in Bengali to assess the speech recognition ability of an individual speaking Bengali.

Development of the test material

Collection of the familiar monosyllabic words and familiarity rating – The commonly used monosyllabic words was collected from two sources. The first source18 had a pool of words arranged in the descending order of frequency. From this pool of words, only the monosyllabic words with a CVC structure were taken. The count of words collected from this pool was only 119. This pool of 119 words was not sufficient to form two phonemically balanced word lists, hence a second source was chosen. The second source9 provided a huge database of words also arranged in descending order of frequency. From this second source 377 monosyllabic words with a CVC syllable structure were drawn. Thus, we had a pool of 496 CVC words. The words that appeared common to both the lists and having English origin (but spoken in Bengali) were eliminated. This reduced the pool of CVC monosyllabic words to 299 words. Among all the 299 words, 255 words were rated as most familiar, forty three as fairly familiar. The word /hɔn/ was rated 'unfamiliar' by all the raters and was therefore eliminated from the list. From the final pool of 298 words two separate lists were made taking into consideration the frequency of distribution of phonemes in Bengali language.18 Each list consisted of 50 monosyllabic CVC words. The following Table 1 and 2 shows sample of the final distribution of phonemes of the two lists. The two phonemically balanced word lists are provided in Appendix – II.

Table 1

Distribution of phonemes for Wordlist 1

Phonemes / Frequency of occurrence of the phonemes in Bengali Language / Number of times the phonemes should appear in List 1
(calculated value) / Number of times the phonemes appear in List 1
Consonants
/k/ / 4.74% ~ 5 / 7.5 ~ 8 / 8
/kʰ/ / 1.04% ~ 1 / 1.5 ~ 2 / 2
/ɡ/ / 1.27% ~ 1 / 1.5 ~ 2 / 2
/ɡʰ/ / 0.18% ~ 1 / 1.5 ~ 2 / 1
Vowels
/ɔ/ / 4.99% ~ 5 / 7.5 ~ 8 / 9
/a/ / 9.86% ~ 10 / 15 / 23

Table 2

Distribution of phonemes for Wordlist 2

Phonemes / Frequency of occurrence of the phonemes in Bengali Language / Number of times the phonemes appear in List 1
(calculated value) / Number of times the phonemes actually in List 2
Consonants
/k/ / 4.74% ~ 5 / 7.5 ~ 8 / 8
/kʰ/ / 1.04% ~ 1 / 1.5 ~ 2 / 2
/ɡ/ / 1.27% ~ 1 / 1.5 ~ 2 / 2
/ɡʰ/ / 0.18% ~ 1 / 1.5 ~ 2 / 2
Vowels
/ɔ/ / 4.99% ~ 5 / 7.5 ~ 8 / 6
/a/ / 9.86% ~ 10 / 15 / 19
/e/ / 9.28% ~ 9 / 13.5 ~ 14 / 4

In Bengali, the phonemes /s/ and /ʃ/ are allophonic variations of the same phoneme /ʃ/. Hence, while preparing the phonemically balanced word lists they were not included separately. Similar inclusionary criteria were applied for the phoneme /r/ also; which again has two allophonic variants flap /ɾ/ and trill /r/. The allophonic variant of /h/ i.e. /ɦ/ was not considered in the list because /ɦ/ always occurs in combination with other sound. It occurs initially as the first member in combination with /r/ and /l/ and medially with /r/.4 The consonants in each lists closely approximates to the frequency of occurrence of the consonants.18 The present study showed good agreement between the speech recognition and pure tone thresholds. Further live voice correlated with recorded voice at α 0.82 at p=0.05.

Psychometric Functions for the wordlist 1 and 2

The data (both threshold and supra-threshold) that was obtained in speech audiometry is represented as points on a psychometric function. The graphic plot that relates some aspect of patient performance (output) to a stimulus dimension (input). Typically, patient performance is plotted on the ordinate (y – axis) and the level of the signal expressed in dBHL is plotted on the abscissa (x – axis). The word recognition psychometric function is described by two characteristics; the first is the location of the function in the Cartesian co – ordinate system and second by the slope of the function. 27,28 Table 3 and Table 4 illustrate the mean [percent correct recognition (and standard deviations) for Wordlist 1 and Wordlist 2 from 30 listeners with normal hearing. The mean psychometric functions for the Wordlist 1 and 2 obtained from the 30 normal hearing subjects are shown in Figure 1 and Figure 2 respectively.

Table 3

The mean percent correct recognition (and standard deviations) for Wordlist 1 from 30 listeners with normal hearing