Sunday, October 17, 2010

Working memory, SLA, and low-educated second language and literacy learners


Alan Juffs, University of Pittsburgh, Department of Linguistics


"The role of memory in language learning has long been of interest to researchers in first and second language acquisition (SLA) (Baddeley, 1999; Ellis, 2001). At an intuitive level, it seems obvious that part of the explanation for individual differences among adults in success at learning a second language (L2) is attributable to differences in memory capacity. In SLA, researchers have focused on short-term rather than longterm memory differences because they think short-term memory is more responsible for differences in language development. The reason for this belief is that short-term memory is an on-line capacity for processing and analyzing new information (words, grammatical structures and so on); the basic idea is that the bigger the on-line capacity an individual has for new information, the more information will pass into off-line, long-term memory. It is an open question whether low-educated second language and literacy acquisition populations (LESLLA) have short-term memory systems that are similar to literate, educated populations, and if so how their working memory capacity can be measured. This paper will survey the literature on this topic, and will make some suggestions about how models of memory (as they have been applied to second language learning) may and may not be applied to LESLLA contexts...

2 Models of Working Memory

In the psychological literature, theories of working memory can be divided into two
main approaches, each with their own constructs (or ways of operationalizing working
memory) and tests that measure those constructs in individuals. The first is called
'phonological working memory' (PWM) (Baddeley & Hitch, 1974; Gathercole & Baddeley, 1993). PWM tests measure the capacity of an individual to remember a series of unrelated items with covert ‘inner speech’ rehearsal (Ellis, 2001:34). This ability is measured by requiring participants to remember lists of unrelated digits, real words, or non-words; in some versions of this non-word repetition test, these non-words have phonemes that are not in the native language (L1). The second is reading span memory (RSM) (Daneman & Carpenter, 1980). Tests of RSM claim to measure the resources available to simultaneously store and process information. RSM tests require participants to read aloud lists of sentences written on cards (or on a computer) and then recall the final word of each sentence without covert rehearsal. The key difference between the tests for PWM and RSM is that the RSM requires both processing and storage, whereas the PWM only requires the participant to repeat polysyllabic words or repeat a string of unrelated words correctly. PWM and RSM are traditionally treated as separate (Baddeley & Hitch, 1974; Carpenter, Miyake, & Just, 1994; Daneman & Carpenter, 1980; Roberts & Gibson, 2003; Sawyer, 1999) because scores on the tests do not correlate. Carpenter, Miyake, & Just (1994:1078) specifically state that ‘traditional’ span measures (digit, word) do not decline with age and do not correlate with sentence comprehension impairment, whereas RSM does decline with age and correlates with sentence comprehension scores. However, debate and speculation remain on the validity of this separation (Ellis, 2005:339).

...reading involves incremental sentence processing. This view holds that a native-speaker reader of an alphabetical script such as English, Dutch, or French does not ‘take in’ a large amount of text (say 7-10 words) and then decides the appropriate syntax for that set of words. Rather, each word is processed rapidly, and the reader makes assumptions immediately about a possible syntactic structure for that word and the ones that follow.

This view accounts for readers being misled by ambiguous sentences, and the
subsequent ‘surprise’ when their reading goes off track because the structure they had assumed turns out to be wrong. This ‘surprise’ is known as the garden path (GP) effect.

An interesting facet of working memory capacity in this model of reading is that the
effects of individual memory differences are not fixed, but task-dependent (Just et al., 1996; Miyake & Friedman, 1998). For example, a high-memory-capacity individual will be more accurate in comprehension and resolve an ambiguity at crucial points in
reading a sentence such as (1) more quickly than a low capacity individual.

(1) The evidence examined by the lawyer convinced the jury.

In (1) the verb 'examined' is temporarily ambiguous between a main verb and a
reduced relative clause structure. Pragmatic information may be used to quickly resolve the parse in favor of a reduced relative clause reading because ‘evidence’ is inanimate and unlikely to be the agent of any ‘examining’. High WM capacity readers are able to resolve this ambiguity more quickly than low WM capacity readers. According to Just and colleagues, this is because high capacity readers are able to combine pragmatic and syntactic information in parsing more efficiently than low span readers. On the other hand, in a sentence such as (2), while high capacity readers are also more accurate in comprehension, they take more time to resolve the parse:

(2) The soldiers warned during the midnight raid attacked after midnight.

The account of this difference in processing speed between (1) and (2) for high WM
capacity readers is that in (1) high WM individuals are able to make rapid use of
pragmatic information, whereas in (2) the ambiguity of ‘warned’ sets up three purely
syntactic possible parses: a main verb reading, an intransitive verb reading, and a reduced relative reading. Just and colleagues argue that high WM individuals in this case are able to maintain all three possible parses active in parallel, and hence take longer to process them. Ultimately, however, they are more accurate with comprehension probes, whereas low WM capacity individuals are faster, but less accurate. Low WM individuals allow the parse to crash, and therefore read more quickly. However, the cost is that they reject these sentences as implausible or fail to understand the relationships among the noun phrases...

3.2 Working Memory and Second Language Sentence Processing

Juffs & Harrington (1995) were the first L2 acquisition researchers to use a self-paced reading paradigm to look at real-time L2 processing of syntax, although some studies had investigated the lexicon using reaction time data (for a review, see Juffs, 2001). Based on this 1995 study, and further research (Juffs, 1998a,b; Juffs & Harrington, 1996), the indications are that L2 learners process their L2 word-by-word in a similar but not identical way to native speakers. (For literature reviews see Clahsen & Felser, 2006; Fender, 2001.)

The similarities between L1 and L2 processing are that the profiles of decisionmaking
at the word level during processing seem to depend on argument structure, i.e. the number of noun phrases and prepositional phrases that are required by the meaning of the verb. The evidence for this comes from Garden Path (GP) sentences. Recall that a conscious GP effect occurs when the hearer or reader cannot interpret the clause without an effort that brings the structure to his or her conscious attention. The situation in (4a) presents such a processing challenge because ‘the socks’ is initially interpreted as the object of ‘mended’, but must later be reanalyzed as the subject of the verb ‘fell’. In (4b), in contrast, no surprise effect occurs.

(4) a ¿After Mary mended the socks fell off the table.
b After Mary mended the socks they fell off the table.

Non-native speakers seem to be ‘Garden-Pathed’ in the same way native speakers are (Juffs & Harrington, 1996; Juffs, 2004); they do not seem to accumulate ‘chunks’ of text before deciding on a parse, but (like native speakers) decide on a structure as soon as possible and then go back and revise it if it is necessary...

Moreover, there is a hint from data in Juffs (1998a,b) that speakers of head final languages (Subject-Object-Verb order, e.g. Japanese and Korean) appear to slow down on processing verbs and objects, which may suggest an effect of L1 word order. Fender (2003) has subsequently reported that Japanese learners were superior to speakers of Arabic in simple word recognition, whereas Arabic speakers were superior to Japanese in syntactic integration. These results suggest that Japanese learners are at a particular disadvantage in processing head-initial syntax, despite their superior ability to recognize words...

Similar to findings for native speakers of English reported by Just and his colleagues, some of the intra-group differences are as great as the between-group differences in studies of second language speakers (Juffs, 1998a,b)...

4 Working Memory and Less-Educated Second Language Learners

In one of the few papers to emerge from the literature, Loureiro et al. (2004, p. 502) report on a study of 97 Brazilian illiterate [sic] and semi-literate adults. They found that phonological memory (as measured by real word and non-word repetition tasks) was very low in the population they term ‘illiterate’ (68 out of their total 97 participants). The scores for real words were much higher than for non-words. They also report that this memory ability was unrelated to letter knowledge. They therefore conclude that phonological memory, phonemic awareness and phonological sensitivity are not related in this population. In another study, Petersson et al. (2000) published brain-imaging results that suggest a reason for poor performance on non-word tests of working memory in non-literate populations. Petersson et al. (2000:365) report that ‘learning to read and write during childhood alters the functional architecture of the brain’. The result that is particularly

relevant for PWM is that literates do not differ in word and non-word repetition tasks, but illiterates do differ. Petersson et al. (2000:373) interpret the patterns of brain activity to indicate that ‘literates automatically recruit a phonological processing network with sufficient competence for sublexical processing and segmentation during simple immediate verbal repetition, whether words or pseudowords, while this is not the case for the illiterate group.’ The implication is that knowing an alphabetic system allows literates to process phonological segments (sublexical elements) of unknown words, whereas this is not possible for illiterates. Moreover, Kosmiris et al. (2004)’s findings that suggest level of literacy is a factor in phonological tasks is an important confirmation of suggestions made by Petersson. In their study, Komiris et al. (2004, p. 825) compare semantic and phonological processing in three groups: high and low educated literates and non-literates. They found that semantic processing was unaffected by literacy, but augmented by schooling; in contrast Komiris et al. (2004, p. 825) state that: ‘explicit processing of the phonological characteristics of material
appeared to be acquired with literacy or formal schooling, regardless of the level of
education attained: those who had attended school and had acquired symbolic representation could perform the task, but those who had not, did very poorly’.
Exploring the implications of this research for non-literate adult learners of a
second language awaits further research. A pessimistic view might be that if we assume a critical period for language (DeKeyser, 2000; Johnson & Newport, 1989), then
learning a new language will be particularly hard for non-literate adults because they will find the L2 especially challenging because by definition it consists of ‘pseudo-’ or ‘non’ words for them. However, some caution is in order before one becomes too pessimistic. First, debate on the critical period continues, even for phonology (e.g., Birdsong, 2005; Flege et al., 2005), and it may be that other factors such as motivation, exposure, and culture play an even greater role than age in predicting success. One must also take care in how one defines success in a second language, since success probably goes beyond a definition based narrowly on morpho-syntactic and phonological features to one based on the ability to participate meaningfully in another culture. In addition, evidence exists that some illiterates can become literate in their L2 as adults; this is an achievement that should not be possible if a true neurally based critical period exists. Finally, differences among children in non-word repetition capacity exist, and differences do predict vocabulary size and growth in these children.

Since children are not literate at age 3, and can learn language, the implication is that the phonological loop for non-literates might still be a useful measure to explore. In general, the results in this literature suggest that establishing a test of working memory for non-literates will be difficult, because non-literates are likely to perform at floor level with non-word repetition tests. Without a range in scores, there can be no correlation with other language proficiency measures, not even those that are not related to literacy. Since pseudo-words are not processed in the same way in illiterates as they are in literates, real word and digits in the L1 could possibly be used exclusively. Overall, given that some researchers (e.g. Pappagno & Vallar, 1995; Williams & Lovatt, 2003) have used span tasks successfully, the span tasks hold out the most promise for preliminary research with illiterates.

Finally, Baddeley’s construct of the ‘episodic memory buffer’ may have some promise as a test for the ability to relate long-term knowledge and memory. Differences may exist in the ability to recall characteristics that are associated with known words
and construct imaginary situations with those words. For example, Baddeley (2000b)suggests that when accessing long-term memory for use on-line, one could imagine an
exercise that would require a participant to think about how an elephant would perform as an ice-hockey player. This novel situation would require the participant to hold in memory the characteristics of elephants (large, ungainly, long trunk) and ice hockey (slippery surface, fast, violent) to construct a scenario: an elephant might play well in goal, be slow, and able to ‘body-check’ effectively. Differences in the ability to access such knowledge and construct ‘new’ or imaginary situations with that knowledge might be used to predict language learning outcomes. This task may be particularly promising because some researchers report that the participants who are most successful at the RSM task are those participants who covertly construct a story with the words that are the target of recall, even though they are not supposed to engage in covert rehearsal (Osaka & Osaka, 1992; Juffs, 2004). Hence, episodic memory may mediate between visual spatial long-term memory and long-term memory for language...

5 Conclusion

The role of working memory in explaining individual differences in L2 learning has a
history of less than twenty years. Many problems remain in replicating the relationships between PWM, RSM, language proficiency and reading even when experimental participants are literate L2 learners. The role of the L1 appears more important than differences in working memory in explaining performance on some on-line processing and reading tasks (c.f. Marinis et al., 2005). Moreover, the little research that does exist with non-literate populations suggests that they perform poorly on such tests and that literacy may change brain architecture to the extent that non-word tests may not be useful as a measure of working memory. Given the cultural assumptions that decontextualized psychometric tests make, and the problems that LESLLA populations have in understanding such tests, extreme caution is necessary before any predictions or conclusions about the abilities of non-literate and low-educated learners’ ability to succeed in acquiring proficiency in an L2 can be made on the basis of current tests of working memory."

No comments: