Skehan (1999) has recently observed that measures of language aptitude (hereafter, aptitude) developed in the late 50s and early 60s (e.g., the MLAT) were far ahead of their time. They have survived virtually unchallenged for 40 years and have consistently shown strong correlations with general second language proficiency1 (GSLP). Despite this endurance, Skehan calls for a reanalysis of aptitude and aptitude tests in light of our better understanding and changing views about language and second language acquisition (SLA). His call was inspired by the wealth of recent research from the fields of discourse analysis, corpus linguistics, applied linguistics, education, cognitive psychology, and testing. He suggests that new findings should be applied in designing new aptitude tests, or, at the very least, redesigning existing instruments.
Given Skehan's call for a reexamination of aptitude tests, one question that arises is whether aptitude tests, validated mainly with native speakers of English (NSE) learning European languages, generalize to all situations (e.g., native speaker of Japanese learning Thai). This question emerges from a growing body of research indicating that different first languages (L1s) give rise to different skills, strategies, and constraints in SLA. Evidence comes from many areas of language including the use of different clues (syntactic, grammatical, world knowledge) in assigning agent and patient roles in sentence parsing (e.g., MacWhinney & Bates, 1989). Likewise, orthographic processing strategies used in reading the L2 have been shown to be constrained by the L1 writing system (e.g., Koda, 1988). At the phonological level, speech segmentation processes also seem to be influenced by the L1 (e.g., Cutler & Otake 1994). This influence extends to L2 listening. These differences in orthographic processing and phonological segmentation bear directly on the construct of phonetic coding, a central factor of most models of aptitude.
The major tests of aptitude all include some measure of phonetic coding ability. Skehan (1999) x to be of great importance in SLA, especially in the early stages of learning. He argues that phonetic coding ability determines the degree to which learners can make use of both oral and written linguistic input. Consequently, the other two central factors of aptitude, grammatical analysis and memory, only come in to play once phonetic coding has occurred. Phonetic coding ability, as measured by the Modern Language Aptitude Test, subtest two (MLAT-II; Carroll & Sapon, 1959), is defined as:
The relevant section in the MLAT-II requires examinees (usually NSE) to identify relationships between English sounds (nonsense syllables2 presented orally) and the Smith/Trager transcription system. This requires a number of steps. (1) Examinees must consistently (though, not necessarily accurately) perceive orally presented nonsense syllables. (2) They must hold these sounds in phonological short-term memory (STM) for the duration of the task and (3) simultaneously consciously segment the syllables into smaller phonological units (onset & rime3, or phonemes). Finally, (4) examinees must identify relationships between these phonological units and the written symbols (graphemes).the ability to identify and store in long-term memory, new language sounds or strings of sounds… It is necessary because the individual must not only learn the identities of the new phonemes of that language, but must also recognize and remember the phonetic sequences represented by the morphemes, words, and intonation contours of a given language (Carroll, 1971, p. 4).
Given that there are at least these four steps, phonetic coding is likely made up of a number of sub-factors. These include phonological STM (step 2), phonological awareness (step 3), and grapho-phonemic awareness (step 4). Phonological STM is STM for language sounds. This has been shown to be distinct from non-linguistic auditory STM (e.g., musical notes) or visual STM (e.g., pictures; Crowder, & Surprenant, 1995). Phonological awareness is the conscious knowledge that words are divisible into smaller phonological units that have no semantic value. This awareness manifests itself in the ability to count, isolate, remove, recombine, and otherwise manipulate these phonological units.
Phonological awareness has been strongly implicated as an important factor in alphabetic4 reading acquisition (see Adams, 1990 for an overview). Similarly, Carroll (1990) has come to believe that phonetic coding difficulties may be closely related to dyslexia. In fact, in his comprehensive survey of factor-analytic language studies, Carroll (1993) includes a study of phonological awareness in a discussion of the construct of phonetic coding. People with dyslexia also have great difficulty learning grapho-phonemic correspondences (step 4). However, most researchers believe that this is purely a result of poor phonological awareness. If you can't understand that words can be decomposed into smaller phonological units, you can't very well match those units to graphemes.
However, reading strategies exploiting grapho-phonemic correspondences are not universal to readers of all languages (Koda, 1988) or even to all L1 readers of English (Scarborough, Ehri, Olson, & Fowler, 1998). Speech segmentation processes also seem to be influenced by the L1 (e.g., Cutler & Otake 1994). This influence extends to L2 listening. These differences in orthographic processing and phonological segmentation bear directly on the construct of phonetic coding. Consequently, phonetic coding ability, when measured with tests like the MLAT-II, is likely to correlate differently with aptitude for different populations. To clarify this issue, I will review research that examines the role of phonetic coding and its underlying factors in aptitude. I will further consider research into the correlation between phonetic coding and aptitude in learners of English with low or no alphabetic literacy. In addition, I will review some studies indicating that this discrepancy is due to one of the factors underlying phonetic coding -- phonological awareness.
In a summary of these studies, Sasaki (1996) concludes that the major aptitude test batteries have low to moderate correlations with achievement tests and grades (r = 0.2 to r = 0.6). This means that the aptitude tests account for between 4% and 36% of the variation in achievement measures. In studies of the validity and reliability of these instruments, grammatical sensitivity, rote learning ability for foreign languages, inductive language learning ability, and phonetic coding ability have consistently been the strongest predictor variables (Carroll, 1981).
Another study supporting phonetic coding ability as a factor related to aptitude examined the best predictors of overall proficiency in a foreign language (Sparks et al., 1997). Although this study did not directly examine phonetic coding, it did look at word decoding ability5 in L2 alphabetic languages. The authors note that this "is a direct measure of students' skill in the phonological-orthographic (sound and sound-symbol) component of the foreign language" (p. 557). This construct is very similar to that examined in the MLAT-II.
The study involved 60 girls (all NSE) enrolled in second-year high school Spanish, French, and German courses. Nine predictor variables were examined. In addition to foreign language word decoding, the list included first-year (i.e., previous year) English grade, first-year foreign language grade, high school placement test score, score on the MLAT long form, and scores from various tests of L1 reading, spelling, and general language ability. The tests were administered during the first quarter of the first year of L2 study (high school second year). L2 proficiency was measured by three tests developed using ACTFL guidelines. These three tests, a reading test, a writing test, and a listening/speaking test, were administered at the end of the students' second year of L2 study. The strongest relationship between predictor variable and L2 proficiency was the L2 decoding test, which accounted for 46% of the variation (r = .68). No other variables maintained significance after analysis. Sparks et al. concluded that learning the grapho-phonemic system of the L2 may help students learn to speak and comprehend new words.
Thus, the role of phonetic coding ability in aptitude research is well supported. All the major aptitude batteries include tests of phonetic coding ability, and there is no evidence challenging its role in aptitude, at least not for English L1 examinees. However, while Carroll (1981) and Sparks et al. (1997) have demonstrated that phonetic coding ability predicts GSLP, it is likely that more basic underlying factors are implicated in this relationship.
Figure 1. Cross-lagged partial correlations between non-word repetition
ability and vocabulary scores.
Bold lines represent significantly greater correlation than the corresponding cross-lagged partial correlation (adapted from Gathercole, Willis, Emslie & Baddeley, 1992).
Service (1992) tested the non-word repetition ability of nine-year-old Finnish children just before they began learning English, then compared the results to their ESL and mathematics grades two and a half years later. Repetition scores accounted for 44% of the variance in future English grades (r = 0.66, p < 0.001), though no significant correlation was found with mathematics grades. Echoing Gathercole et al. (1992), Service contends that the strong correlation may be mediated by the effect of phonological memory on vocabulary acquisition.
Ellis and Beaton (1993) found additional support for phonological STM's role in L2 vocabulary learning. They examined the psycholinguistic factors contributing to German L2 vocabulary learning in 47 undergraduate university students (NSE). In individual learning trials, students were told to use a specific learning strategy (repetition, keyword, own) depending on what group they were assigned to. Students were given an immediate posttest and retested one month later. Results were analyzed based on not only strategy use, but also word characteristics such as the relative position of phonemes in the words, pronounceableness, part of speech, imageability, and orthographic factors. In their conclusion, Ellis and Beaton stated that, "representation of the novel sound sequence of a new word in phonological short-term memory promotes its longer-term consolidation both for later articulation and as an entity with which meaning can be associated" (p. 599). Ellis (1996) argues convincingly that the same processes that underlie the connection between phonological STM and vocabulary learning also explain idiom learning and the acquisition of grammar.
This is supported by results from an experimental study by Daneman and Case (1981) which looked at the ability of six NSE children between the ages of two and six to learn both novel (non) vocabulary and syntactic structures that are not used in English. Daneman and Case found that the participants’ age was significantly related to ability to both comprehend and produce the new structures. However, when the results of a word span test of PSTM was entered into a regression equation first, age no longer accounted for a significant proportion of the variance. In contrast, PSTM still accounted for a significant proportion of the variance when the effects of age were removed first, F(4, 25) = 4.86, p < .01. Daneman and Case concluded, “STM may be the only age-related factor that has a significant impact on children’s language learning, once the effect of prior specific experience is removed.” (p. 377).
Studies using other tests have also strongly implicated phonological
STM as a factor in SLA (e.g., Harrington & Sawyer, 1992, but see Skehan,
1999). Taken together, this evidence suggests a major role for phonological
STM as a factor in phonetic coding, and relatedly, in vocabulary and grammar
acquisition. Consequently, phonological STM is also likely to factor directly
Vellutino and Scanlon (1986) conducted a study that tested the trainability of both phonemic awareness and phonological STM. The study compared groups of poor and normal readers (English L1) in second and sixth grade on verbal response learning and code acquisition tasks with or without training in phonemic awareness. The experiment involved three treatment and two control groups. However, only the results of the training group (Vellutino & Scanlon's PSTRA group) are of interest here. The training involved exercises in phonemic segmentation, practice in remembering orally presented phonemically similar nonsense syllables (sij, duj, dif, sug), and practice in identifying grapheme-phoneme relationships in printed pseudowords. Training took place over five or six days with one half-hour session each day. Vellutino and Scanlon subsequently tested the children on their ability to recall orally presented nonsense syllables (different from those used in training). They also tested the children's ability to learn the names (nonsense syllables) of unfamiliar cartoon characters. Lastly, they tested the children's ability to learn grapheme-phoneme correspondences.
The training had no significant effect on oral recall or name learning ability. However, the training group performed significantly better in learning the grapho-phonemic correspondences than they did on a pretest, and significantly better than the control groups on either the pre- or posttest. A post-experimental test showed that the training had also had a positive effect on phonemic awareness. Thus, Vellutino and Scanlon's research suggests that phonemic awareness is trainable, and that such training has a facilitory effect on abstracting grapho-phonemic correspondences. However, phonological STM did not prove susceptible to training in this experiment.
While these results lend more support to the idea that the phonological STM factor in phonetic coding also factors in aptitude, the trainability of phonemic awareness indicates that it should not be included in a model of aptitude. Convergent evidence comes from studies of phonemic awareness in populations with low or no alphabetic literacy.
In order to test the hypothesis that knowledge of an alphabetic orthography is necessary to develop deep levels of phonological awareness (phonemic awareness), Read, Zhang, Nie, and Ding (1986) replicated a study by Morais, Bertelson, Cary, and Alegria (1986). In the original study, illiterate adults were asked to delete phonemes (e.g., say stop without the /t/) and reverse phonemes (e.g., what does cat sound like if you say it backwards?). They were unable to perform these tasks. In the replication study, the subjects were literate in their L1 Chinese, but had never learned to read an alphabetic language or pinyin (a romanization of Chinese). They were as unable to perform the tasks as the illiterates were. These two studies demonstrate that identifying individual phonemes (phonemic awareness) is an artificial ability that results from having learned to read an alphabetic orthography.8 Consequently, measures of aptitude which include phonemic awareness as a factor are likely to have low validity and reliability when applied to populations with low or no alphabetic literacy.
Given the above evidence, a strong argument exists for including phonological STM in a model of aptitude. However, the trainability, and language specific nature of phonological awareness suggest that it is a skill to be learned. Thus, phonological awareness does not belong in models of aptitude, and may confound the results of aptitude tests that rely on it (e.g., the MLAT-II). In situations where all the examinees are NSE, this may not be serious, as most of the examinees can be expected to have high levels of phonological awareness. However, in situations where the examinees' first language does not use an alphabetic orthography, problems may arise.
Sasaki (1996) examined the relationship between GSLP, aptitude, and intelligence in 160 Japanese9 university students who had studied English for an average of 7.3 years. The aptitude test used was the Language Aptitude Battery for the Japanese (LABJ), developed especially for the study. The LABJ was modeled after the short version of the MLAT and included a test of phonetic coding similar to the MLAT-II. Structural equation modeling was used to analyze relationships between 26 variables. A best-fit model was identified and examined for its theoretical implications. This model was then subjected to a Lagrange Multiplier test. In the final model that resulted from the analysis, the phonetic coding test had no significant relationship with aptitude. This is in stark contrast with the findings outlined above for studies done with NSE. As a possible reason, Sasaki suggests that "the original tests might have been valid aptitude tests, but their Japanese counterparts * might have lost their validity as aptitude tests because of language differences" (p. 121). Alternatively, she suggests that both the MLAT-II, and the LABJ tests may be poor tests of the studied group's aptitude. Both conclusions are in line with the findings reviewed above (Morais, Bertelson, Cary, & Alegria, 1986; Read, Zhang, Nie, & Ding, 1986; Vellutino & Scanlon, 1986).
The major implication of these conclusions is that aptitude tests employing measures of phonetic coding should be revised to test phonological STM independently of phonological awareness. Unfortunately, current tests of phonological STM are more difficult and time consuming to administer than the existing phonetic coding tests. For this reason, the existing tests will likely continue to be used for large testing groups. However, aptitude tests need to take advantage of emerging technology, particularly computer speech recognition. It seems likely that a computerized version of the non-word repetition test employed by Gathercole, Willis, Emslie, and Baddeley (1992) could be employed in the near future.
In situations where either the L1 or L2 employs a non-alphabetic orthography, aptitude tests should avoid measures that depend on phonological awareness. If a paper and pencil test were required in such situations, a possible compromise would be to use a test like the MLAT-II, but with a syllabic transcription system, instead of the (alphabetic) Trager-Smith system.
1 GSLP is here defined rather traditionally as a four-skills,
multi-dimensional model, including morphology, vocabulary, syntax, and
2 These syllables are all well-formed, according to the phonological rules of English.
3 This spelling is commonly used in psychology and refers to the linguistic unit. Thus, a rhyme for boy is toy, but they share the same rime, /oi/.
4 Alphabetic writing systems encode the language at the phoneme (individual sound) level and include English, Spanish, Thai, Hebrew, and many others. Syllabic orthographies encode syllables and include the Japanese kana (hiragana & katakana) and syllabaries for certain North American Indian languages. Finally, there are logographic orthographies like the Japanese kanji, which represent single- or multi-syllable morphemes.
5 Decoding is used to mean deriving the spoken form of a word from its written form.
6 Cunningham and Stanovich (1998) argue for a small but significant role for orthographic knowledge independent of phonological factors in L1 English reading, but they admit this is controversial. Though orthographic knowledge may also factor marginally in measures of phonetic coding, a discussion of its role is beyond the scope of this paper.
7 Non-word repetition tests phonological STM by having a participant listen to and immediately repeat progressively longer non-words. Real words are not used because of confounding factors with long term memory.
8 Most researchers now believe that phonemic awareness and learning to read an alphabetic orthography are reciprocally causal (Ehri, 1998).
9 Japanese uses two kinds of orthographies, but neither are alphabetic.