Use of the NAL-AB Wordlists as a South African English Speech Discrimination Test

South Africa still lacks a South African English specific speech discrimination test. As an alternative, this study investigated the use of the Australian English, National Acoustic Laboratories Arthur Boothroyd (N AL-AB) wordlists to assess the speech discrimination of South African English speakers. Thirty South African English speakers were tested at 0, 5, 10,20 and 25 dBHL (audiometer dial reading) and their performance-intensity functions were compared qualitatively to the NAL-AB wordlist normative data. Results showed three general patterns; similar performance for both groups; poorer performance by the South African English speakers at the low to mid presentation intensities only; and poorer performance by the South African English speakers across most presentation intensities. Use of the NAL-AB wordlists at threshold levels or for site of lesion assessment was therefore concluded to be unwise. Use of these wordlists at supra-threshold levels, however, would provide a valid and reliable option for the speech discrimination assessment of South African English speakers. verskil tussen die twee groepe nie; swakker diskriminasie deur die Suid-Afrikaanse groep by slegs die lae en middel-intensiteite ; en swakker diskriminasie deur die Suid-Atrikaanse Engelssprekendes by die meeste intensiteite. Die NAL-AB-woordelyste is egter geskik om die spraakdiskriminasie Jan Engelssprekende Suid-Afrikaners bo gehoordrempels te toets. .

those that are unfamiliar (Edgerton & Danhauer, 1979). Secondly, an ideal speech tests should contain an equal phonetic composition representative of everyday speech in the test population (Egan, 1948). In applying the C.LD. W-22 to the South African population, any equal phonetic representation is lost. Tobias (1964) and Campbell (1965) questioned the need for equal phonetic representation, however, by claiming there was overwhelming clinical and experimental evidence that this an interesting, but unnecessary component in speech audiometry testing. Even today, experimental evidence is lacking and the issue remains unresolved (Mendel & Danhauer, 1997)]. Finally, a valid speech discrimination test should contain an equal range and average level of difficulty within and between wordlists. Although this ideal is rarely achieved in reality (Ross & Huntington, 1962), large variations will affect test reliability and sensitivity. These factors all strongly influence, and possibly invalidate, the use of the C.LD. W-22 with its American English normative data on the South Mrican English speaking population.
In response to the problems faced by the C.LD. W-22, many South Mrican audiologists make use of monitored live voice (MLV) presentation. Whilst MLV provides a partial solution to speakerilistener accent mismatch, it does not reduce the dialectal differences inherent in the test materials themselves. Added to this is MLV's wellrecognised poor test-retest reliability (Carhart, 1965;Brandy, 1966;Kreul, Bell & Nixon, 1969;Northern & Hattler, 1974), and the problem of comparing MLV test results to standardisation information obtained using recorded versions of the stimuli (Mendel & Danhauer, 1997). Considering the problems already faced in using non-South African English speech material, the significantly greater variability ofMLV could effectively negate any advantages it may offer in overcoming accent differences.
If MLV is to be avoided (an opinion not held by all clinicians), this returns the South African audiologist to the option of using the C.LD. W-22 in its pre-recorded form on cassette tape or compact disc. In addition to the previously mentioned problems this introduces, closer inspection of the C.LD. W-22 reveals that it may not be the best option for the South Mrican English speaker in the first place.
The C.LD. W-22 was designed as an improvement on one of the first pre-recorded speech tests, the Psychoacoustic Laboratories phonetically-balanced 50-word lists (PAL PB-50). It improved phonetic balance (Lehiste & Peterson, 1959), equality of difficulty between lists (Brewer & Resnick, 1983) and degree of familiarity (Brewer & Resnick, 1983;Hirsch et aI., 1952). The test consists of 4 lists of 50 monosyllabic words with no repetitions. The words were chosen to be familiar to North American English speakers, and the phonetic composition of each list aims to reflect that of English as a whole (still based, however, on North American English). Each word is preceded by a carrier phrase and each list may be .' presented with one of six different word orders. In its recorded form, it is spoken by an adult male with "general North American" accented English and all its normative values are based on North American English speakers (Hirsch et aI., 1952). Despite the improvements shown by the C.LD. W-22, it still has several faults that will only be worsened when applying the test to the South Mrican English speaking Wayne J. Wilson, Beverley Jones and Peter Fridjhon population. Brewer and Resnick (1983) and Ostergard (1983) question the C.LD. W-22's ability to discriminate effectively among the performance of the hearing impaired, claiming that it is too easy. Ross and Huntington (1962) and Studebaker and Sherbecoe (1991) disclaim equality of difficulty between lists. Mendel and Danhauer (1997) have also claimed that the validity and reliability of the test for certain conditions under which it is used, such as in background noise, are poorly documented. Finally, users recognised almost immediately the benefits of reducing the test-time (Edgerton & Danhauer, 1979) by administering 25 word half-lists (Martin & Forbis, 1978). However, these shorter lists are prone to higher variability and therefore poorer reliability and are no longer phonetically balanced (Ostergard, 1983). Even considering Cakiroglu and Danhauer's (1992) claim that listener's and talker's linguistic background has statistical, but not clinical, effects on the performance of non-native English-speaking subjects on the C.LD. W-22 (although they do warn that listener's having less experience with English may not perform as well as native-English speakers), these identified problems warrant a search for a better speech discrimination test for the South Mrican English speaker.
An alternative which claims to combat many of the problems encountered by whole word tests, such as the C.LD. W-22, is the phonemically scored speech test (Edgerton & Danhauer, 1979). Lehiste and Peterson (1959) argued that as the speech signal is not acoustically invariant, it is impossible to achieve phonetic balance in a speech test. Therefore intelligibility (discrimination) measures should use materials that are phonemically balanced [although again, the need for phonemic balancing in the first place has been queried (Mendel & Danhauer, 1997)].
Phonemic scoring offers advantages of general and specific significance to a South African English speaking population. Generally, factors such as increased reliability, abbreviated test lists, and decreased test time (Edgerton & Danhauer, 1979) are all beneficial. Specifically, factors such as a decrease in the effect ofthe listener's familiarity with the vocabulary used, and the controversial claim that phoneme scoring involves less extrinsic redundan'cy and therefore eliminates the effects of linguisti~ bias (Lyregaard, 1987;contested by Keith, Katbamna, Tawfik & Smolak, 1987), suggest that use of normative '~alues supplied with a phonemically scored wordlist could ~rovide greater test reliability and validity when used on !South Mrican English speakers. \ Following the success of the first phonemically scored wordlists developed in the U.S.A. by Lehiste and Peterson in 1959 (the consonant-nucleus-consonant or CNCilists), and Tillman, Cahart and Wilber in 1963 (the Northwestern University Auditory Tests No.4 and No.6), Arthur Boothroyd of the United Kingdom developed the phonemically scored Arthur Boothroyd (AB) wordlists (Boothroyd, 1968). Boothroyd's test contains 15 isophonemic lists, each containing 10 words with 30 scorable phonemes. Each list contains the same 20 consonants and 10 vowels, in varying combinations (Der~ody' & Mackie"; 1987). The phonemes chpsen were those found to be most frequent in Boothroyd's vocabulary, and the only'limitations placed on the original word lists were that neither first names nor obscenities would be included (Boothroyd, 1968

Reproduced by Sabinet Gateway under licence granted by the Publisher ( dated 2012)
impacted on the design of the test's phoneme content, and that the lists did not comprise random samples of phonemes. Despite Boothroyd's reply that the phonemes are altered by changing their word position, and therefore the phoneme arrangements are essentially random (which would appear to undermine his claim of equality of phonetic representation), the AB wordlists were recorded on at least 2 occasions in the United Kingdom using different U.K. accents (Evans, 1987).
The success of the phonemically scored AB lists in the U.K.led to their use in Australia by the National Acoustic Laboratories (NAL) (Dermody & Mackie, 1987). NAL produced a revised version specifically for Australian use, with rearranged word orders and several new words to replace those considered to have low familiarity for Australian English speakers (Travers, 1990). According to Bench (1987), Australian English is based on that of the United Kingdom, but includes North American influences and some uniquely Australian expressions, while some words common to both dialects may carry different meanings. The NAL-AB wordlists (appendix one) quickly became the most commonly used speech discrimination test in Australia (Bench, 1987), resulting mainly from advantages such as short administration time, and the availability of a high number oflists allowing multiple testing to occur in the one session without learning effects or fatigue (Markides, 1987;Mendel & Danhauer, 1997). In view of the success of the pre-recordedAB wordlists in the U.K. and Australia, the possibility that they could offer significant advantages over the present MLV presentation of the C.LD. W-22 wordlists in South Mrica becomes apparent. The problem oflinguistic bias remains, however, and despite claims that phonemic scoring decreases the effect of linguistic bias (Lyregaard, 1987; contested by Keith, Katbamna, Tawfik & Smolak, 1987), and can be used even on clients with limited familiarity with English (Evans, 1987), it is vital to consider the validity of using this test in a multilingual population such as South Mrica's. :

AIMS
The aims of this study were as follows: I 1) To determine if first language South Mrican English speakers performed differently to first language, Australian English speakers, bn the pre-recorded NAL-AB wordlists.

I /'
2) To determine if the first ten (of fifteen) pre-recorded NAL-AB wordlists were of equal difficulty for first-language, South African English speakers. 3) To investigate the feasibility of using the pre-recorded NAL-AB wordlists, with the' NAL Australian English normative data, to assess the speech discrimination abilities of first-language South Mrican English speakers in South Africa.

METHODOLOGY
Exact replication of the NAL-AB wordlist normative study proved to be difficult. The normative data on the performance of Australian English speakers on the prerecorded NAL-AB wordlists is documented in a clinical manual entitled "AB Word Lists: NAL Protocols" prepared by Anne Travers, NAL, Victoria, August, 1990. This document is supplied with theNAL-AB Word List compact disk and outlines graphically all normative data for the 15 wordlists separately, and as a total average, and contains full recommended procedures the test's clinical use. On close inspection, however, the document provides little detail as to the exact procedures used to obtain the normative values. Details of the NAL-AB wordlist study were not published in the literature and the original data has since been removed from NAL archives. The only fact that is clearly stated in the Travers document is that the data was obtained using 120 normally hearing subjects. With this limited information in mind, the following methodology was used.

SUBJECTS Subject selection criteria
Subjects were selected using a convenience sampling technique according to the following criteria: (i) Subjects should be adults between 17 and 40 years. This controlled for the well-recognised effects of paediatric and geriatric age ranges on performance on speech discrimination tasks (Hall, 1983). (ii) Subjects should be female. This criterion was included for ease of subject selection, as it was deemed unlikely that a sample balanced for gender could be easily obtained. (iii) Subjects should be first-language speakers of South African English. (iv) Subjects should be resident in Gauteng, South Africa as according to Fuller (1987), subjects for research in the area of speech audiometry should be native to the local area. (v) As determined by a pretest audiogram, the test ear should have hearing thresholds no greater than 5dBHL at 250, 500, 1000, 2000, 4000 and 8000 Hz. (v) As determined by self-report on a pretest questionnaire, subjects should have no significant history, past or present, of: hearing impairment; speech or language impairment; tinnitus; ear infections; noise exposure; speech/articulation disorders; the presence of any disorders of language or; family history of hearing problems. (vi) Subjects should have no knowledge or experience with the NAL-AB wordlists specifically. (vii) Subjects should have little or no knowledge or experience of Australian English.

Subject description
An initial sample of 32 female, first language South African English speaking subjects, all resident in the Gauteng region was collected. Two subjects were excluded when pure tone screening revealed unacceptable hearing thresholds leaving a final 30 subjects ranging in age from 17.5 years to 26.0 years, with a mean age of20.5 years, a standard deviation of 1.8 years, and a median age of21.0 years. Education levels included two subjects with secondary level, and 28 with, or receiving, tertiary level education. 26 ofthe 28 tertiary level subjects were studying in the field of speech pathology and audiology. On the basis of selecting the best hearing ear as the test ear, 12 right and 18 left ears were tested.

TEST ENVIRONMENT, EQUIPMENT AND RECORDED MATERIAL
Testing took place in a two-room sound treated booth at the University of the Witwatersrand Hearing Clinic in Die Suid-Afrikaanse Tydskrifuir Kommunikasieafwykings, Vol. 45, 1998 Reproduced by Sabinet Gateway under licence granted by the Publisher ( dated 2012) Johannesburg, Gauteng, SouthMrica. The booth included a one-way mirror to allow the tester to observe the subjects during testing. The NAL-AB wordlists were presented via a Sanyo compact disc player CDP-41 coupled with a Grayson-Stadler GSI16 audiometer. Presentation was by means ofTDH-50 headphones with MXAR41 cushions. The NAL Speech Recognition Materials (Disc 1) compact disc with intensity levels equated using an A-weighted Long Term Equivalent Level (Leq) procedure provided the speech stimuli.
The NAL-AB Wordlist CD provides 15 pre-recordedAB wordlists (appendix one) arranged to be as representative as possible of 'typical' Australian English (Travers, 1990). Lists are spoken by an adult male Australian general English speaker. Each list contains the same 20 consonants and 10 vowels, arranged into various combinations of 10 CVC monosyllables per list. The lists are ordered according to their reliability as measured on the Australian population (Dermody, cited in Travers, 1990). The order reflects a pairing of those lists with the most similar performance-intensity functions. The last three of the 15 lists demonstrate the most widely variant functions and NAL recommends that these be used at the top end of the PI function only. Each of the word lists is preceded by a short warning pip, the inter-stimulus interval is 4 seconds which serves as the subject response period (there is no carrier phrase), and each list is recorded on a separate track.

MEASUREMENT PROCEDURES AND DATA COLLECTION
On passing the initial hearing and case history questionnaire screening, subjects were assessed on their speech discrimination performance on the first 10 of the 15 pre-recorded NAL-AB wordlists in their better hearing ear. Administration of all 15 lists to both ears was considered to be impracticable for 3 reasons: (1) the duration of testing would be increased to unacceptably high levels possibly leading to subject fatigue; (2) the additional repetitions of each list would induce a greater degree of practise/learning effects; and (3) as lists 13-15 were shown to be more variable in the Australian sample (Dermody, cited in Travers, 1990), it was felt that lists 1-10, the most reliable and commonly used lists, were more pertinent to the aims of this study.
An identical set of instructions was given to all subjects, via the headphones, as per NAL recommendations (Travers, 1990, p8). The instructions read as: "You are going to hear some single words. Mter each word there will be a pause for you to repeat the word. Listen carefully and repeat whatever you hear. Some of the words will be very soft and it is important that you have a guess at each word even ifit doesn't make sense or you only hear a part of the word such as 'ee' or 'ch'. Just repeat what you hear." Once subjects had been instructed, list 11 was presented . at 30dBHL (audiometer dial reading), in order to familiarise subjects with the test procedure. The wordlist presentation was pseudorandomised with the initial ten list order being randomly selected, and then repeated five times. This was done to prevent a particular list from occurring several times in succession, or in close proximity. The intensity presentation was then randomised throughqut all 50 presentations such that each of the ten lists was presented once at five different intensity levels.

Wayne J. Wilson, Beverley Jones and Peter Fridjhon
The whole process of pseudorandom is at ion was conducted separately for each individual subject. Intensity levels of 0, 5, 10, 20 and 25 dBHL (as per the audiometer dial reading) were used, emulating the normative data provided by NAL.
All subject's responses were recorded and scored. online by a single tester, a student of Speech and Hearing Therapy at the University of the Witwatersrand, under the supervision ofa full member of the Audiological Society of Australia, in fulfilment of that Society's requirements for use of materials. Responses were transcribed using a combination of orthographic and phonetic transcription. Each word was then assigned a score of 0%, ifno phonemes were correctly repeated; 3% for 1 correct phoneme; 7% for 2 correct phonemes; or 10% if the response was completely correct; as per the scoring protocols, suggested by NAL (Travers, 1990).
Subject responses were considered to be correct only if they were pronounced exactly as per the Australian English pronunciation on the NAL-AB wordlist CD. It was noted, however, that none of the 15 NAL-AB wordlists contained words where a predicted difference in pronunciation between Australian English and South African English should occur (for e.g., there were no words such as "dance", which would be pronounced as /dams/ in Australian English, but as /dans/ in South African English). In cases where single phonemes or nonsense syllables were given as a response instead ofa whole word, the phonemes were scored based on their acoustic similarity to the relevant Australian English phonemes in the stimulus word. In the case of more than one response being given for a single stimulus, the first response was taken irrespective of accuracy. Subjects were allowed the opportunity to rest at any stage during the test procedure.

DATA ANALYSIS
Comparison between the South Mrican English speaker scores and the NAL-AB wordlist Australian English normative data was done graphically with all NAL normative data values being extrapolated from the graphs provided in Travers (1990) for reanalysis. Plots lof the extrapolated NAL normative data and the South African English speakers data were completed using Mi~rosoft Excel for Windows, Version 5.0©.

Analysis for differences within the South African
English speaker discrimination scores, between p~esen tation intensities, was conducted separately fot" each wordlist using Kruskal-Wallis ANOVA by Ranks (~cross all intensities at p<0.05) and Mann-Whitney U (between adjacent intensities only, at p<O.Ol to allow for the multiple comparisons) analyses. Non-parametric statistics were used as a result of the non-homogeneity of variances (determined by the F test at the p<O.Ol level) and the significant correlation of mean scores to variances present in the data (

Reproduced by Sabinet Gateway under licence granted by the Publisher ( dated 2012)
speaker discrimination scores between wordlists was conducted separately for each presentation intensity using Pearson's Product Moment Correlation Co-efficient analyses (at the p<0.05 level). Similar analyses for correlations between presentation intensities was also conducted separately for each wordlist. All statistical analysis was conducted using Statistica for Windows, Version 4.5 software from Statsoft Inc, 1993©. Figure 1 shows the South Mrican English speakers' speech discrimination results, superimposed on the extrapolated NALAustralian English normative data, for each individual NAL-AB word list. Overall, the NAL Australian English speakers' mean scores tended to be better than the South African English speakers' mean scores with greater differences occurring at the lower presentation intensities. Extrapolated mean values for the NAL Australian English normative data generally fell within 1 standard deviation above the South African English speakers' means. Only list four at 5 dBHL, list eight at 0 dBHL and list nine at 5 dBHL, showed the NAL Australian English scores to be greater than one standard deviation above the mean South African English scores on visual inspection.

SOUTH AFRICAN ENGLISH SPEAKERS' VERSUS AUSTRALIAN ENGLISH SPEAKERS' RESULTS
Specific comparison between the mean South African English speakers' speech discrimination scores and the extrapolated NAL Australian English normative data showed three basic (qualitative) categories: 1) Similar mean scores across four or more presentation levels (lists one and five). 2) South African English mean scores falling below the NALAustralian English mean scores in the mid and/or low presentation levels only (lists three, four, seven, ei~ht and nine). i 3) South African English mean scores falling below the NAL Australian English Imean scores across four or more presentation levels (~ists two, six and ten). Table 1 shows the mean arid standard deviation speech discrimination scores for the 30 South Mrican English speakers on all ten NAL-AB wordlists, at all five presentation intensities. Table 2 shows Kruskall-Wallis ANOVA by Ranks and Mann-Whitney U analysis results for differences in the scores between presentation intensities, conducted separately for each wordlist. Significant differences (p<O.Ol) were shown between all adjacent presentation intensities in each wordlist except between 20 and 25 dBHL for lists 1, 2, 5, and 6. Tables 3  and 4 show one-way repeated measures ANOVA and Tukey Honest Significant Difference test results for differences in scores between wordlists, conducted separately for each presentation intensity. Results showed multiple significant differences between wordlists at all five presentation intensities, but predominantly at the higher intensities.

I
Pearson's Product Moment correlation co-efficient analyses (p<0.05) between scores at each presentation intensity, conducted separately for each wordlist, showed mild (0.4) to strong (0.9) correlations between intensities for all wordlists. A small number (<10% ofthe total number of correlations) of poor «0.4) or non-significant (p>0.05) correlations occurred within the wordlists between low and high presentation intensities only. Similar analyses between scores across wordlists, conducted separately for each presentation intensity, showed mild (0.4) to strong (0.9) correlations between scores for all presentation intensities except for 25 dBHL, which showed large numbers (24.4%) of poor «0.4) or non-significant (p>0.05) correlations between wordlists.
F Test analysis for differences in the variance of the South African English speakers' discrimination scores between the different presentation intensities, conducted separately for each wordlist, showed the variance at 25 dBHL to be significantly different (p<O.01) to the variance at 0, 5 and 10 dBHL in lists 1, 2, 3, 4, 6, 9 and 10; to the variance at 0 and 5 dBHL in list 5; and to the variance at 5 and 10 dBHL in list 8. The variance at 20 dBHL was shown to be significantly different (p<O.01) to the variance at 0 and 5 dBHL in list 3 and to the variance at 0 dBHL in list 5. List 7 showed no significant differences (p<O.Ol) in variance at any presentation intensity.

DISCUSSION
The aims of this study were: (i) to determine if first language, South African English speakers performed differently to Australian English speakers on the prerecorded NAL-AB wordlists, (ii) to determine if the prerecorded NAL-AB wordlists were of equal difficulty for first-language, South African English speakers and (iii)    to investigate the feasibility of using the pre-recorded NAL-AB wordlists, with the NAL-Australian English normative data, to assess the speech discrimination abilities of firstlanguage South African English speakers in South Africa. To achieve these aims, NAL-AB wordlist speech discrimination scores were obtained from 30 first language South African English speaking, female adults with normal hearing. These scores were then compared to the NAL Australian English speaker normative data (as provided by Dermody, in Travers, 1991). In general, the SouthMrican scores were lower, with three qualitative patterns being observed; similar performance between the two groups (lists one and five); South African scores occurring below the Australian scores at the lower and/or mid presentation levels only (lists three, four, seven, eight and nine); and South African scores occurring below the Australian scores across four or all five presentation intensities (lists two, six and ten). Comparisons of South Mrican scores within and between NAL-AB wordlists showed multiple differences and correlations, depending on which lists and presentation intensities were being compared.
The qualitative finding that South African scores were generally worse than those of the Australians, indicates the direct llse of the pre-recorded NAL-AB wordlists with their associated normative values cannot be applied to the South African English speaking population without significant modification.
The three differing patterns of South African versus Australian results indicates the different NAL-AB wordlists were not of equal difficulty for the two populations. The first category of results showing similar mean scores for the two groups, suggests these lists were equally difficult. The second category showing differences at the low and/or mid-presentation intensities only, suggests both groups were able to hear these wordlists equally well at the higher presentation intensities, but the Australians were able to extract more information at the lower presentation intensities, due most probably to their  2,3,4,5,6,7,8,9,10 150 <0.005* Sig. diff's (p<O.01*) between all adjacent presentation levels except between 20 and 25 dBHL on lists 1, 2, 5, and 6. familiarity with Australian English. The third category showing a poorer South African English speaker performance across four or all five presentation intensities, suggests that even the advantage of increased presentation intensity did not allow the South Africans to overcome the difficul ties encountered in listening to Australian English. This "three category ranking" of Australian versus South African scores places the ten NAL-AB wordlists in a "valid" (category one) through to "use with caution" (category three) ranking for South African English speakers. This supports Markide's 1987 disagreement with Boothroyd's 1968 andKeith's 1987 claims that phoneme scored speech discrimination tests generally, and the AB wordlists specifically, can be used accurately with non-English speakers. It also questions Evans' 1987 suggestion that subjects with minimal knowledge of English may be tested with AB lists by a clinician whose own experience with the ,test is extensive. Evans stated the PI functions are likely to be less steep, approximating those for nonsense syllables, but maximum scores may still reach 100%. This study showed equal or steeper sloped performance intensity functions for the South African English speakers, with the steeper functions seen in category two.
The 4:1-Australian:South African ratio of sample sizes used in this study prevents any comment on whether there was a true difference in the variance of the speech discrimination scores between the Australians and South Africans. A notable observation, however, was the significant decrease in the variance of the South African scores as the presentation intensity increased. This finding appears to be similar (on visual inspection only) to the Australian plots (Travers, 1990), and only partly agrees with Ostergard's (1983) findings that variability in speech discrimination scores tend to decrease for extreme scores and increase for midrange scores.
Difference analyses within the South African scores themselves, further indicated that the NAL-AB wordlists were not of equal range and difficulty for the South African English speakers (although a similar conclusion was found for the Australian English speakers by NAL themselves). Overall, the multiple significant differences seen indicate that the performance ofthe South Africans was dependent on which wordlist and presentation intensity was being used. Interpretation ofthe significant differences must be ,approached with caution, however, as close inspection shows many of these to be more dependent on differences in variance, rather than any true clinical differences in mean scores. Even so, the South Africans did show a steady improvement in scores on all wordlists with increases in presentation intensity, with evidence ofthe beginning of a ceiling effect at the 25 dBHL level.
Correlational analyses within the South African scores showed a high number of significant positive correlations between different lists at the different presentation intensities, particularly for the mid-presentation intensities. This indicates a good relationship between .NAL-AB wordlists for the South Africans in that a high score on one wordlist at one intensity, will generally predict a high score on the same or another wordlist at the same or different intensity (and vice versa). The exception to this was at 25 dBHL where a large number (24.4%) of poor «0.4) or non-significant (p>0.05) correlations occurred between wordlists. This finding supports the earlier suggestion of the beginning of a ceiling at 25 dBHL presentation intensity as scores in a ceiling range should Wayne J. Wilson, Beverley Jones and Peter Fridjhon be more random and therefore have fewer correlations.
Minor limitations to this study include the all young, adult female composition of the sample, and the lack of control over right versus left ear selection. With no reports, to the authors knowledge, of significant female/male or right/left ear performance differences in the literature for any ofthe monosyllabic word tests, these points were not considered to have had any significant impact on this study's findings.
A more significant limitation was the sample's English first language prerequisite, and predominance of tertiary level education, much of which was in the field of Speech Pathology and Audiology. Davis (1983) (cited in Lutman, 1987) demonstrated a relationship between socio-economic status and type of occupation, and performance in speech audiometry with higher education level implying a certain level of linguistic competence and even sophistication. It might therefore be predicted that a more representative sample ofthe South African English speaking population would not have performed as well as the more educated sample used in this study.
Finally, it must be noted a significant factor affecting any interpretation of speech discrimination results is the large amount of variability inherent in speech testing. Mendel and Danhauer (1997) warn that a margin of 16-20% should be allowed for erroneous scoring alone, because of errors in the scorer's perception. Similarly, Thorton and Raffin (1978), Ostergard (1983) and Green (1987) state that a single score obtained for a particular wordlist is only an indicator of a range of scores in which the true score is likely to be. This inherent variability in speech audiometry diminishes its accuracy in all uses generally, and the strength of the descriptive results of this study specifically.

CONCLUSIONS
The similarities observed between the Australian English and South African English speakers on the prerecorded NAL-AB wordlists makes this test a valid option for the speech discrimination assessment of South African English speakers, under certain conditions. The NAL-AB wordlists are most su.itable for use at suprathresholdilevels (where the fewest dIfferences were observed) where the advantages of shorter test times, and more available wordlists, can be accessed without critically compro~ising test reliability or validity. Use ofthe NAL-AB wordli'sts at threshold, or near threshold, levels (where the Imost differences were observed) should be approached! with caution, however, and reliance on these wordlists fo'r site , I of lesion purposes should be avoided. General test reliability and validity could be improved by changing the wordlist presentation order to (5, 1), (3, 9), (7,6), (8, 4), (2, 10) (where brackets indicate interchangeable lists) to suit South African English speakers, and by collecting larger South African normative data bases where possible. Despite this study's findings, the over-riding need for a South African English specific speech discrimination test remains. Whilst the NAL-AB wordlists offer/advantages over the C.I.D. W-22 wordlists such as ph~emic s~oring and, based on this study's results, at least 'some idea of how South African English speakers vary from the normative data (information which is not currently published for the C.I.D. W-22 wordlists), the cdntinued use of non-South African tests must be seen ~s an interim measure only.
In view of similarities between the performance of South The South African Journal of Communication Disorders, Vol. 45, 1998 Reproduced by Sabinet Gateway under licence granted by the Publisher ( dated 2012) African English and Australian English speakers seen in this study, the option of modifying the NAL-AB wordlists to suit South African English (as per NAUs modification of the original U.K. AB wordlists to suit Australian English) could provide the starting point needed to develop the long awaited South African English specific speech discrimination test.