working memory measures and l2 research

Revised conceptualizations of memory constructs such as that of Miyake and Shah have informed the field of L2 research and led to the development of new ways of assessing memory. Such measurements include “span” tests of verbal working memory and phonological STM (PSTM) measures.

4.3.1.“span” tests of verbal working memory

The theoretical shift from STM to WM was motivated in large part by the need to capture the processes involved in the performance of complex cogni-tive tasks such as language processing. Several research paradigms have verified the importance of WM in the performance of complex cognitive tasks.

Hence, designs in SLA research in recent years have included a “span” test of verbal WM that places simultaneous demands on processing and storage, thus emulating the demands theorized to be placed on WM during completion of a complex cognitive task. A span test typically consists of reading or listening to sets of sentences and, in more recent versions, making judgments regarding their grammaticality and/or semantic sense while trying to remember the final word of each sentence. After a given set of two to six sentences is presented, participants recall aloud or in writing the final words for the sentences in that set. Participants in such research are generally classified into different groups based on their performance on the span test or else their scores are correlated with performance on a target language task.

The test that set the standard for research was Daneman and Carpenter’s (1980) reading span test. In this test, participants read sentences aloud as they were presented one at a time on cards; this task of reading aloud was consid-ered to satisfy the processing component of WM, but it was not quantified in any way and thus was not taken into account in the ultimate determination of participants’ WM spans. After reading a particular set of two to six sentences, participants were required to recall in writing the final words of sentences in that set. Limitations of the reading span test included the questionability of measuring processing with a read-aloud task and the fact that the scoring of

the test took only the recall component of WM into account. While the con-struct of WM was posited to include both processing and recall components, its operationalization included only recall. Furthermore, the reading span test was found to be somewhat unreliable over time (Waters and Caplan, 1996).

Waters and Caplan (1996) developed a sentence span test based on the work of Daneman and Carpenter (1980) in which participants were required to judge the acceptability of sentences while simultaneously trying to remem-ber the final words of sentences. The WM score was a composite z-score based on accuracy of sentence processing, recall of sentence-final words, and reac-tion time.

Results of SLA research using reading or listening span tests and their adap-tations are intricate and at times contradictory. Harrington and Sawyer (1992) found a strong correlation between a reading span task administered in the L2 to native Japanese English speakers and participants’ TOEFL grammar and reading subtest scores as well as between scores on an L1 reading span task and an L2 reading span task. These findings indicate that a relationship indeed ex-ists between the variables, but they do not specify the direction of the relation-ship, thus making it impossible to make any claims that one variable causes another. Therefore, the results must be interpreted with caution.

Miyake and Friedman (1998) reported on a study in which native Japanese speakers of English performed a listening span task in both their L1 and L2.

Their data pointed to strong causal links between L1 WM and L2 WM as well as between L2 WM and syntactic comprehension in the L2. The researchers cautioned, however, that their results could be attributed to the advanced L2 proficiency of their participants.

In their study of WM, interactional feedback, and L2 development in na-tive Japanese learners of English, Mackey, Philp, Egi, Fujii, and Tatsumi (2002) used three different measures of memory ability. Following previous arguments that the strong predictive validity of aptitude tests such as the MLAT can be attributed to their composite nature (see, for example, Skehan, 1998, 1989), they evaluated memory with measures of L1 and L2 verbal WM as well as with a test of PSTM. Basing themselves on measures used in previ-ous research (Harrington and Sawyer, 1992; Turner and Engle, 1989; Waters and Caplan, 1996), the researchers created listening span tests for the L1 and L2, and operationalized PSTM as the ability to recall sets of nonwords imme-diately after auditory presentation. In their analysis they considered the L1 and L2 verbal WM and PSTM test scores both independently of one another and as a composite z-score. When they analyzed the L1 and L2 verbal WM test results, they found a high positive correlation between L1 and L2 scores, and thus they decided to combine L1 and L2 results in a composite verbal WM score. Participants were classified in groups of low, medium, and high

WM capacity, and the low and high WM groups were compared in terms of their ability to notice and respond to feedback that occurred during an inter-active language task. Results showed a marginally significant relationship be-tween composite WM scores and noticing.

Still other studies (e.g., Juffs, 2003; Sagarra, 2000) have found no signifi-cant relationship between WM and the higher order cognitive processes im-plicated in language learning and performance.

4.3.2.pstm tests

Still other researchers claim a particularly important role for PSTM in SLA.

PSTM is closely related to the functioning of the phonological loop subsystem of Baddeley’s model of WM and has been operationalized as the ability to re-peat phonological input accurately immediately after presentation. One such measure, used by Gathercole, Service, Hitch, Adams, and Martin (1999), is the nonword pairs recall test. Pairs of nonwords (i.e., strings of syllables that are phonologically possible in the participants’ L1) consisting of two to five syl-lables each are presented aurally followed by a two-second pause. After the pause, participants are cued to repeat aloud the pair of nonwords.

PSTM capacity has been linked to L2 vocabulary acquisition (e.g., Gather-cole and Thorn, 1998; Papagno and Vallar, 1992; Service and Kohonen, 1995) as well as grammar learning (e.g., N. C. Ellis and Schmidt, 1997; N. C.

Ellis and Sinclair, 1996; Williams, 1999; Williams and Lovatt, 2003). Pa-pagno and Vallar (1992) conducted a series of experiments to investigate the effects of various types of interference on adult participants’ ability to learn pairs of known words and novel pseudowords. Results suggested that the ca-pacity to temporarily store and rehearse verbal material is implicated in the learning of novel words but not of real words for which lexical-semantic repre-sentations already exist in LTM. The results must be interpreted with caution, however, in light of the fact that, as Papagno and Vallar acknowledged, the ex-perimental conditions were quite different from real vocabulary learning, which is far less systematic and occurs more gradually over time.

Service and Kohonen (1995) undertook a longitudinal study to investigate whether vocabulary learning at the initial stages of L2 development relies on PSTM, which they operationalized as the ability to accurately repeat pseudo-words that followed the phonotactics of the L2 (English) their child partici-pants were starting to learn. Participartici-pants were tested once a year for four years on their ability to accurately repeat L1 and L2 pseudowords immediately after their auditory presentation. Results supported the hypothesized relationship between PSTM and the acquisition of foreign-language vocabulary.

N. C. Ellis and Sinclair (1996) undertook research to examine the relation-ship between PSTM function and the acquisition of L2 syntax. In their study,

adult L1 English speakers completed a computerized lesson in Welsh vocabu-lary and soft mutation, a complex phonological rule system in Welsh. The les-son consisted of four learning trials in which participants attempted to learn 30 target items. Computerized corrective feedback was provided throughout the learning phase and participants completed the lesson in one of three con-ditions: one in which they were required to repeat aloud each Welsh item ev-ery time it was presented, one in which they remained silent, and one in which articulatory suppression (AS) in the form of constant whispered repetition of the numbers 1 through 5 prevented rehearsal of the Welsh items to be learned.

Results of posttests revealed no effect of learning condition on accuracy of grammaticality judgments; however, the repetition and silent conditions per-formed significantly better than the AS condition on both a rule structure test and speech production test. There was no significant difference between per-formance on the structure test by participants in the repetition and silent con-ditions, but significant differences in performance between these two groups emerged in the speech production data. As the researchers acknowledged, the findings of their experiment cannot speak to whether the locus of the observed advantages is at the input or output phase of learning, but regardless of the lo-cus of the advantages, the results of this study suggest that rehearsal of L2 ut-terances is a crucial component of learning. The researchers elaborated on this point and claimed that the rehearsal function of PSTM facilitates the accumu-lation of words and sequences of words in LTM. According to N. C. Ellis and Sinclair, acquisition of grammar entails the abstraction of regularities from these stored sequences, thus implicating PSTM not only in vocabulary learn-ing but in grammar acquisition as well.

Other researchers, however, deny a direct role for PSTM in the acquisition of L2 grammar. Adams and Willis (2001), for example, claim that while peo-ple with greater PSTM capacity are better able to produce compeo-plex syntactic structures, this does not necessarily imply a relationship between PSTM and grammar learning.

4.4.conclusions and future research