Speech Assessment Tests for Arab Children Running Head: SPEECH ASSESSMENT TESTS FOR ARAB CHILDREN
Published in : Audiology, 39, 269-277, 2000.
Development of Speech Assessment Tests for Arabic-Speaking children
Liat Kishon-Rabin and Judith Rosenhousea)
Department of Communication Disorders, Sackler Faculty of Medicine,Tel-Aviv University, a)
Department of Humanities & Arts, Technion - I.I.T., Haifa, Israel
Mailing Address: Department of Communication Disorders, Sheba Medical Center, TelHashomer, Israel 52621 Tel: 972-3-535287 Fax: 972-3-5352868 Email:
[email protected]
1
Speech Assessment Tests for Arab Children
RECEIVED: ______________________
2
Speech Assessment Tests for Arab Children
3
Abstract The high incidence of hearing impairment in the Arabic-speaking population in Israel, as well as the will to use advanced aural rehabilitation devices with them, motivated the development of Arabic speech assessment tests for this population. The first step for this goal was to analyze and describe specific features that are unique to the Arabic language and need to be considered when developing such speech tests in Arabic. These include Arabic diglossia (i.e., the sharp dichotomy between Literary and Colloquial Arabic), emphatization, and a simple vowel system. The second goal of the paper is to describe a new analytic speech test that assesses the perception of significant phonological contrasts in colloquial Arabic of Israel. The perception of voicing, place and manner of articulation, in both initial and final word positions, was tested at four sensation levels in 10 normally hearing subjects. A binary forcedchoice paradigm was used. Results show a relationship between percent correct and presentation level that is in keeping with articulation curves obtained with Saudi Arabic and English monosyllabic words. Furthermore, different contrasts yielded different articulation curves: emphatization was the easiest to perceive whereas place of articulation the most difficult. We suggest explanations for these results based on language-specific acoustic features.
Speech Assessment Tests for Arab Children
4
Development of Speech Assessment Tests for Arabic-Speaking children Speech perception assessment by valid, reliable and sensitive tests is an important tool in the prosthetic management of hearing impaired children. These tests are needed to help establish the need for sensory assistance, to select and adjust devices, to quantify outcome, and to guide educational intervention (Boothroyd, 1991). The need for good tests has become particularly acute now that cochlear implants have become an option for prosthetic management. The findings suggesting that auditory intervention in the first twothree years of the hearing-impaired child’s life is crucial for normal speech, language, social, cognitive, and psychological development, provide further incentive for speech perception tests that can be used with (young) hearing-impaired children. Many tests have been developed for English speaking hearing-impaired subjects, of different ages, hearing impairment types and levels of language skills (a review of these tests can be found in Boothroyd 1984; Erber 1974; 1980, Goetzinger, 1972; Iler Kirk, Diefendorf, Pisoni and Robins, 1997; Markelein, 1981; Northern and Downs 1974; Siegenthaler et al., 1954; Sortini and Flake, 1953; Stark, 1969). In Hebrew, the official and dominant language in Israel, several speech perception tests for children were designed over the years (KishonRabin, 1999, Kishon-Rabin et al., 1999). These tests include some that were adapted from English into Hebrew such as ESP (Moog and Geers, 1990), THRIFT (Boothroyd, 1991), SPAC (Boothroyd, 1984), and others that were developed in Hebrew for the first time, such as Hebrew Picture SPAC (HePi☺SPAC; Kishon-Rabin, 1999). In Arabic, the 6th world language (with over 200 million speakers; see Bakalla, 1983; Kaye, 1997) and the second official language in Israel, there are hardly any available tests. The need for speech perception tests for the Arabic-speaking population in Israel is further enhanced by the fact that there is above-than-expected occurrence of deafness and hearing impairment in this population, and by the will to use technologically advanced devices, such as cochlear implants for rehabilitation. The problem of hearing loss in the Arab population in Israel is acute, compared to its scope in the Israeli Jewish population (1.11/1000) and to other countries (1-1.5/1000) (Sela and Weisel, 1992, Rosenhouse, 1995). Sela and Weisel (1992) refer only briefly to the occurrence of deafness in the Arab community in Israel, noting the lack of valid data about the
Speech Assessment Tests for Arab Children
5
scope of deafness there. They assume a higher rate of occurrence of deafness in the Arab population in Israel than reported in the literature, based on comparisons of Jews of Asian and African origin vs. European and American origin. This fact is most probably one of the outcomes of consanguineous marriages which is so widespread in the Muslim population compared with some European Non-Muslim communities (Jaber et al.,1994, and their references concerning Jordan, Lebanon, Saudi Arabia, Egypt and Algeria). Zlotogora (1997) deals with genetic disorders among Palestinian Arabs without focusing on hearing impairment problems, though he mentions them. He mentions about 44.3% rate of consanguinity (based on the results in Jaber et al., 1994) which yields a relatively high rate of incidence of several genetic disorders, some of which are very rare in the general population. Tell et al. (1999) found that of 32 Arab children who studied in a school integrating both hearing and hearingimpaired children in Jerusalem, 21 (66%) were hearing impaired. An analysis of the family background of the Arab students in this school also revealed consanguineous relations between the students’ parents in 66% of the cases. The results reported by Tell et al. (1999) concerning the incidence rate of hearing impairment in this population are considerably higher than those usually reported in the literature. The lack of valid speech perception tests in Arabic, and the fact that many individuals in this population speak Hebrew as a second language, has led to the most undesirable outcome that speech perception assessment of Arab native speakers of Arabic is conducted primarily in Hebrew. Due to the fact that Hebrew and Arabic are genetically related, being Semitic languages, and that they have long similar histories and inter-communal contacts (Blau, 1981), they have many similar features. Their present phonological and phonetic systems, however, are too different to enable a test in Hebrew to serve well the Arabic-speaking population. In addition, the possibility that the Arabic-speaking individuals may not be familiar with the phonology, not to mention vocabulary and syntax of Hebrew, violates a major prerequisite of a valid speech perception test. This prerequisite dictates that the listener will be familiar with the language of the test in order to ensure that the results were not limited by his/her linguistic knowledge (Iler Kirk et al., 1997). The need to assess speech perception in the Arabic speaking population in Israel, the high incidence of hearing-impairment in this population, the differences between the Hebrew and Arabic languages, and the Arabic language specific features, have all been at the
Speech Assessment Tests for Arab Children
6
background of our decision to develop speech tests for the Arabic-speaking population in Israel. In the next sections of the paper, we describe language specific features that are to be considered in the development of speech assessment tests in Arabic and design difficulties. Following that, we describe in some detail one of the tests that were developed and its initial results. Arabic Diglossia: Literary and Colloquial Arabic A major characteristic of Arabic, wherever it is used, is diglossia (Ferguson, 1959a). This term refers to the fact that there are two levels or registers of this language - the oral, colloquial dialects (CA), and the literary - written and read - register (LA). The colloquial dialects are the real mother tongue of native speakers of Arabic, and they vary due to factors such as religion (Muslims, Christians, Druze), geography (e.g., Galilee dialects differ from dialects in Samaria and Judea; Galilee Bedouin dialects differ from Negev Bedouin dialects) and demography (rural and. urban vs. Bedouin populations) (Kaye and Rosenhouse, 1997). Literary Arabic is a unifying force in the Arabic-speaking world, since it is mainly used for reading and writing and is more uniform than the colloquial dialects. Literary Arabic, however, has to be formally studied at school, and therefore only literate individuals use it. Literary Arabic is not only a written form of the language - it differs from the colloquial varieties on all levels of linguistic structure including vocabulary, grammar and phonology. Since it is also the language form in which is written the holy book of the Muslims, the Koran, as well as the entire Muslim cultural heritage, LA enjoys a prestige that no colloquial dialect has. Typically, native speakers of Arabic refer to LA and disregard the colloquial dialects that they do not really consider worth the term “language”. Due to the unifying force of its reading tradition, LA has a more or less stable phonological system, which varies in certain details from the parallel phonological systems used in the varieties of CA dialects. With increased literacy, the two separate registers, LA and CA, become more and more intermingled, affecting among others the phonetic level. Some of the differences between LA and the urban CA dialects spoken in Israel (CAI) can be demonstrated by a brief phonological comparison. Table 1 shows the LA system with 28 consonants and Table 2 shows the six vowels (3 long and 3 short ones). Also shown is the CAI system, which is very similar, but a few differences can be immediately noticed: 1. certain
Speech Assessment Tests for Arab Children
7
consonants are pronounced in different manners from the LA model. For example, LA /q/ is pronounced as /q, g, ?/ or /k/ in CAI; LA /d3/ can also be pronounced as /3/ in CAI; LA /k/ is pronounced as the affricate /ts/ in some dialects in CAI (and elsewhere); 2. certain LA phonemes merge to form one speech sound in certain CAI dialects. Thus LA /d/ can be either /d/ or /d/, and LA /d/ can be either /d/ or /d/); 3. CAI has a few speech sounds that do not exist in LA, such as the vowels /e, o/ and schwa, and phonemes borrowed from Hebrew and other foreign languages, such as /p, v/. Moreover, the distribution and frequency of speech sounds in LA and CAI is not identical. Clearly, Arabic speech perception assessment tests, being based on investigating subjects’ phonological/phonetic perception, need to tackle these facts. Language Specific Characteristics Since Arabic is a Semitic language, many of its features, including the phonological and phonetic system, differ from those of European languages such as English.i Still, the LA phonological system shares many features with English as shown in Table 1 of the consonant system of Arabic. It has voiced and unvoiced consonants which are bilabial (/b,m/), interdental (/t,d/), labio-dental (/f/), dental (/t,d,n,l/) alveolar (/s,z/), palato-alveolar (/s,z/), palatal (/j/), velar (/k/) and glottal (/h/). In addition it has, however, consonants which do not exist in English, such as the emphatics (or velarized) (/s,d,t,d/), as well as the velar /x, g/, the uvular /q/, the pharyngeal /h, '/ and the glottal /?/. The emphatic consonants are distinct mainly for their lowered second formant (F2) compared to their non-emphatic equivalent consonants (/s,d,t,d/) and partly for their raised third formant (F3) (Obrecht, 1968; Norlin, 1987). The emphatic articulation involves in some dialects movement of the tongue root towards the back of the vocal tract, raising of the back part of the tongue towards the uvula, and contraction of the pharynx due to the retreat of the tongue root, while the tongue tip or blade keeps its position as in the non-emphatic parallel consonants. This movement usually yields a concave tongue body. It should be noted that emphatic consonants affect adjacent preceding or following consonants and vowels, even up to word boundaries, depending on dialect and phonetic environment. (Lane, 1963; Al-Ani, 1970; Norlin, 1987). As to the Arabic vowel system, it is simpler than the English one: LA has only 6 short and long vowels and CA has usually somewhat more vowels, but not as many as in English. A study of the perception of minimal pairs in Arabic cannot, therefore, ignore these features.
Speech Assessment Tests for Arab Children
8
Colloquial Arabic dialects share with LA a similar phonological system but also have some dialect-dependent differences such as allophones for LA phonemes. Nonetheless, communication between speakers of different dialects can be usually maintained. In most cases, this fact follows not from dialect similarity but from other linguistic forces inherent in inter-dialectal communication, namely koineization (i.e., use of common linguistic elements Ferguson, 1959b) and levelling (i.e., deletion of special dialectal features - Blanc, 1960). It is important that clinicians in the fields of audiology and speech pathology be aware of the details of these features when evaluating speech perception. Though English speech perception assessment tests might be considered a good model for Arabic tests, many aspects, as shown above, need to be considered in the development of these tests in order to reduce uncontrolled and undesirable factors to the results. Criteria for Developing Speech Perception Tests It is clear that developing a speech perception test in Arabic in Israel is problematic and needs to take into consideration both Arabic diglossia and the different CA dialects within Israel. The results of such a test should reflect the auditory capabilities of the listener with minimal influence of the perceiver’s knowledge of the test language. Furthermore, there are certain criteria that need to be followed when developing speech perception test regardless of the language in which they are developed. These criteria have been discussed extensively by many (e.g., Boothroyd, 1991; Iler-Kirk et al, 1997; Tyler,1993) and include the need of the tests to be sensitive, valid and reliable. In testing hearing perception, special emphasis is placed on the development and use of tests that are maximally sensitive to the adequacy of sensory evidence and minimally sensitive to the use of context. Since the potential contribution of context to speech perception depends on the perceiver’s linguistic, world and social knowledge, speech perception results using such tests would not reflect true auditory capacity but rather spoken language function. Although this may be important for general communication rehabilitation, it does not provide the appropriate information for prosthetic management. These tests should also be within the cognitive, motoric and attention capabilities of the test population and provide analytical information. In other words, the test should show the kind of sensory information (as opposed to the amount of sensory information) the listener perceives. This
Speech Assessment Tests for Arab Children
9
issue has clinical applications in prosthetic management. For example, the type of information provided by different sensory aids is important for the selection of an optimal device. Previous Arabic Speech Tests Few speech assessment tests are known to us to exist in Arabic-speaking countries (Alusi et al., 1974; Ashoor and Prochazka, 1982; Messouak, 1956; Michelian, 1976 cited in Hanna, 1976). Messouak’s test used Colloquial Maghrebin Arabic, whereas the tests by Alusi et al. (1974) and Ashoor & Prochazka (1982) used Modern Standard Arabic. These three latter tests used phonetically-balanced words. Alusi et al. (1974) designed a test comprising monosyllabic and disyllabic words for pan-Arabic use. Ashoor and Prochazka’s (1982) monosyllabic lists yielded an articulation curve for testing Arabic SRT. Dialect differences (mainly concerning the phonological system) between different countries and the relative isolation of Israel from Arab countries motivated us to develop our own tests. Several speech perception assessment tests for Arabic were developed at the Dept. of Communication Disorders at Tel-Aviv University. Disyllabic word lists were generated to determine speech reception thresholds (SRT) for adults and for children (the latter being a closed-set picture-naming test for children as young as 3 years of age), and the Arabic versions of the World Intelligibility by Picture Identification (WIPI) test and the Early Speech Perception (ESP) test. These tests were found to be of much clinical value due to the absence of other available tests. These tests have, however, two main drawbacks. One relates to the fact that the information that these tests provide on the auditory capabilities of the hearing impaired is general and not analytical. The second drawback is related to the fact that in the development of these tests, the complexities of the Arabic language in Israel were not taken into account. Note that this problem is somewhat minimized in the closed-set tests where the tester can verify that the listener is familiar with the words prior to testing on the expense of prolonging the time allocated for auditory evaluation. Speech Pattern Contrast Test The need to develop speech perception tests that are sensitive to auditory capabilities but are minimally influenced by language and world knowledge, meet the cognitive demands of both children and adults, provide analytical information and use speech stimuli, has led Boothroyd (1984) to develop the Speech PAttern Contrast (SPAC) test. This
Speech Assessment Tests for Arab Children
10
test was designed to provide information about an individual’s access to phonologically significant changes in the speech stimulus. The test uses a binary forced-choice identification paradigm, and the response alternatives are presented in writing. For the scores to reflect perceptual rather than linguistic capabilities, listeners must, therefore, be familiar with the phonology of the language and be able to read. The test includes ten sub-tests: eight segmental and two suprasegmental contrasts. The segmental contrasts are: vowel height (e.g., “fool” vs. “fall”), vowel place (e.g., “fool” vs. “feel”), initial consonant voicing (e.g., “Sue” vs. “zoo”), final consonant voicing (e.g., “neat” vs. “need”), initial consonant continuance (e.g., “Sue” vs. “too”), final consonant continuance (e.g. “spies” vs. “spied”), initial consonant place (e.g., “seat” vs. “sheet”), and final consonant place (e.g., “cake” vs. Kate”). The suprasegmental contrasts are word stress and pitch rise vs. pitch fall. In each of the segmental sub-tests, there are 12 different pairs of CVC monosyllabic words that differ along one contrastive dimension. Thus, the test requires the detection of phonologically significant contrasts presented in varying phonetic contexts. This feature of the test enhances its validity as a predictor of the potential for speech perception performance at a conversation level (Boothroyd, 1984). The SPAC test has been used successfully in several studies of auditory and prosthesis evaluation in English speaking adults and children (Boothroyd, 1984, 1991; HnathChisolm and Kishon-Rabin, 1988) and in Hebrew speaking individuals using the Hebrew version of the SPAC test (Kishon-Rabin, 1999; Kishon-Rabin et al, 1999). The success of the test has led the authors of this paper to believe that a speech pattern contrast test in Arabic may meet the various requirement discussed above for the native speakers of Arabic in Israel. The fact that it requires only linguistic knowledge of the phonology of the language may make it appropriate for speakers of CAI. The purpose of the remainder of our paper is therefore to describe a speech pattern contrast test in Arabic, hence named ArSPAC test, while emphasizing the differences in the phonology of the English and Arabic languages and their effect on the development of the test. The Arabic Speech Pattern Contrast (ArSPAC) Test Table 1 presents, in addition to LA phonemes, the consonant system of the Colloquial Arabic dialects spoken in Israel and Table 2 the vowel system of these dialects. The Tables
Speech Assessment Tests for Arab Children
11
show that in addition to the phonological contrasts of English, Arabic has another contrast category, namely, emphasis. The test includes two monosyllabic word lists, each with eight sub-tests examining the four phonological contrasts of Arabic in the two locations: syllable onsets and syllable offsets. Each sub-test includes 12 contrast pairs. (See Appendix A). Sub-test 1 and 2 tested, respectively, initial and final voicing in the following minimal pairs: /s,z/, /t,d/, /t, d/, /s, d3/, /t, d/, /h, ‘/, /x, g/. Sub-test 3 and 4 examined, respectively, initial and final place of articulation in the following minimal pairs: /b,d/, /f,s/,/x,h/, /s,sh/, /f,h/, /s,h/, /h,s/, /d,g/, /x,s/, /t,x/, /3,d3/, /s,t/, /f,x/, /z,d/, /f,h/, /h,sh/, /s,x/, /h,s,/, /g,d3/, /g,z/, /h,h/, /n,m/, /d3,d/, /t,s/, /s,t/, /h,t/, /s,f/, /h,x/. Sub-tests 5 and 6 examined, respectively, manner of articulation differences in initial and final consonants, as follows: /d,z/, /s,t/, /d,l/, /b,m/, /d,d/, /d,n/, /s,t/, /d,r/, /q,x/. Sub-tests 7 and 8 examined emphasis in initial and final consonants, respectively, as follows: /d,d/, /t,t/, /s,s/, /d,d/. Note that the vowels were not tested because the influence of emphasis on adjacent vowels could not be avoided. Development of a test that will include the vowel contrasts would need to consider breaking the vowel system to allophone sets, phonetic environment etc. Another unique feature of the ArSPAC test is in the structure of the test items. The typical English test items are CVC. Although CVC monosyllabic words exist in Arabic, they are rare, especially in CAI. This is a language constraint of Literary Arabic and of the structure of its lexicon. Thus, it was not possible to provide a sufficient number of CVC minimal pairs to test all the necessary contrasts. For this reason the test items were either CVCC or CVVC patterns. Within each minimal pair, however, identical word structure was maintained. Experiment The validation of a speech perception test is usually a lengthy process that does not end after one or two studies. Usually, the information that is being accumulated over several studies contributes to the validity of the test. As part of this process, we conducted a study with the purpose to show the relationship between the perception of speech pattern contrasts in Arabic and presentation level.
Speech Assessment Tests for Arab Children
12
Several investigators have shown that there is a strong relationship between the ability to perceive speech stimuli and changes in intensity (Reviewed by Penrod in Katz, 1985). By plotting a graph with percentage of items correct for every presentation level, an articulation gain function can be generated. In general, speech perception performance varies from minimum to maximum values over a 30-40 dB range. The steepness of the articulation gain function was found to vary depending upon the particular stimulus material used. Articulation function curves of speech stimuli with high linguistic redundancy were found to be steeper than stimuli with low linguistic redundancy. One would also assume that different articulation functions would be derived for different phonologically significant contrasts and that these differences could be explained on the basis of the acoustic cues that are believed to underlie their perception. Our study set out to test this latter assumption. Method Subjects Ten normally hearing young adults (five males and five females), 17-21 years of age, participated in the study. All subjects were native speakers of Arabic, came from the same region in Israel and spoke the same dialect of CAI. Subjects were either seniors or high school graduates. None had known reading or other learning problems. Speech Stimuli The stimuli used in the test included real and mainly frequent monosyllabic words selected from a standard Arabic-Arabic LA dictionary (Al-Munjid). Though most of the words were familiar to the subjects, some were rather rare. The words contained the phonetic features described above in the section “The Arabic Speech Patterns Contrast (ArRSPAC) Test" (see Appendix A). Procedure The words were recorded using a female speaker who was a native speaker of Arabic and spoke the same CA dialect as the subjects. Recordings were made in a sound treated room onto a tape recorder (Sony TM 5000). The sub-tests were recorded in a sequence. There was a 3 second pause between words to allow the subject time to mark the required word. A VUmeter controlled recording volume.
Speech Assessment Tests for Arab Children
13
Subjects were tested individually in a sound treated room. Stimuli were presented from a tape recorder (Sony TC-129) to a portable audiometer (Amplaid 171) and via headphones (TDH-49) to one ear. Five of the subjects were tested via the left ear and five subjects via the right ear. For each subject the two versions of the Arab SPAC test were administered in four sensation levels: 5, 15, 25, and 35 DB SL. Order of presentation of the levels, contrasts and test versions was counterbalanced across subjects. Subjects responded by circling their answer on a pre-prepared response sheet with the two available choices for each test stimulus. Each subject was tested on two occasions that lasted 30-40 minutes each. Overall, each subject was tested four times, at each of the four presentation levels and two lists were administered at each presentation level. Eventually, 24 minimal pairs were tested in each level, contrast and subject. Altogether, 768 (8x12x2x4) stimuli words were examined for each subject. The scores were transformed into percents and analyzed.
Results Table 3 shows the group mean scores and standard deviations for each of the contrasts as well as the composite score at each presentation level. The articulation gain function of the composite Arab (CAI) SPAC results is shown in Figure 1. As expected, performance increased monotonically (from 78.12% to 98.12%) with increase of presentation level (from 5 dB SL to 35 dB SL, respectively). Analysis of variance of contrast variables on the arcsine transformed data revealed that performances at the various presentation levels were significantly different from each other (P
