| Phonology in Second Language Reading: | Not an Optional Extra |

Lower intermediate learners with L2 reading comprehension problems at-tempted to recall similar and dissimilar sequences of words in L1 French and L2 English.. This possible source is th

Phonology in Second Language Reading: Not an Optional Extra

CATHERINE WALTER

Institute of Education, University of London

London, England

In examining reading comprehension in a second language (L2), I have demonstrated that the prevailing metaphor of transfer of skills is misleading, and that what happens is access to an already existing gen-eral cognitive skill There is evidence in first language (L1) and in L2 that accessing this skill when reading in an alphabetic language involves efficient use of verbal working memory (VWM) This article reports a study of a component of VWM, the phonological loop, which serves to hold recently read material available in a phonological form The study investigated whether the unreliability of learners’ mental L2 phonologi-cal inventories contributed to reading comprehension problems Lower intermediate learners with L2 reading comprehension problems at-tempted to recall similar and dissimilar sequences of words in L1 (French) and L2 (English) Their performance was consistent with their having unreliable L2 phonological inventories; their upper inter-mediate counterparts, who had no L2 reading comprehension prob-lems, had significantly more reliable L2 phonological inventories This finding has important implications for the classroom: Rather than at-tempting to teach components of a cognitive skill that learners already possess, teachers would do better to spend the equivalent time increas-ing exposure to the spoken language, and improvincreas-ing receptive and productive phonology

In examining reading comprehension in a second language (L2), I

have argued that the prevailing metaphor of transfer of skills is mis-leading, and that what happens is better conceptualised as access to an

already existing general cognitive skill (Walter, 2007) This argument is based on the work of Gernsbacher and her colleagues, who have pro-vided evidence that comprehension—the building of a mental structure corresponding to a narrative or expository account—involves the same cognitive mechanisms and processes whether it is based on listening, reading, looking at picture stories, or watching silent films (Gerns-bacher, 1990, 1997; Gerns(Gerns-bacher, Varner, & Faust, 1990; MacDonald & MacWhinney, 1990) In other words, comprehension is not linguistic: It

is not in the first language (L1), so it cannot be transferred to the L2.

Rather, readers of L2 who comprehend well do this because they have reached a point where they can access, from the L2, their existing skill in building mental structures

Since Alderson’s seminal article in 1984, researchers have been inter-ested in the problems that intermediate L1-literate learners have in L2 reading comprehension; these problems do not seem explicable in terms

of learners’ general proficiency, and they have sometimes been charac-terised as a threshold phenomenon, though the evidence for this is not decisive (Lee & Schallert, 1997) In Walter (2004) I provided evidence that lower-intermediate learners with L2 comprehension difficulties have difficulty with the building of mental structures to represent L2 narra-tives, even when they can understand the individual sentences in those narratives and when they encounter no conceptual difficulties in the text In other words, these readers are good decoders of L2 sentences, but poor comprehenders of L2 texts (while being good comprehenders

of the same texts in L1) Further, I have demonstrated a link between this difficulty in building mental structures and the operation of L2-based verbal working memory (VWM; Walter, 2004, 2007) The same phenomenon, and the same link with VWM, has been observed in young L1 readers by Yuill and Oakhill (1991); and the discontinuity between decoding and comprehension has further been corroborated in L1 by a functional magnetic resonance imaging study associating text compre-hension with increased activity in the right frontal lobe and sentence-level understanding with increased activity in the left frontal lobe (Rob-ertson et al., 2000)

In this article, I report on a study which examined one of the possible sources of the problems that good L1 comprehenders have in compre-hending their L2, despite being good L2 decoders This possible source

is the way in which phonological information is represented within a component of working memory, the phonological loop

WORKING MEMORY, THE PHONOLOGICAL LOOP AND READING

Working memory (WM) can be thought of as a (largely unconscious)

men-tal workbench, where information is processed and stored during com-plex cognitive tasks such as playing chess, making a medical diagnosis, or comprehending a story The tools on this workbench have three func-tions: (1) access to information from external input and from long-term memory; (2) temporary storage of the information while it is needed for the task at hand; and, perhaps most importantly, (3) processing of the

information in order to carry out the task The WM workbench is not infinite in size or functionality: If there is too much information, or if too much processing is demanded, something will fall off the workbench, or one of the tools will not have enough power to function well My aim in this study was to examine whether part of this WM problem consists of difficulties in a component of verbal WM called the phonological loop

In Baddeley and colleagues’ Multi-Component Model (Baddeley,

1986, 2000; Baddeley, Gathercole, & Papagno, 1998; Baddeley & Logie, 1999; Gathercole & Baddeley, 1993) WM consists of a central executive

plus slave systems One of the slave systems is the phonological loop, a

short-term memory mechanism that stores information in phonological form and automatically rehearses that information by unconscious sub-vocalisation (Another slave system is the visuospatial sketchpad, which performs an analogous function for visual information.) The existence

of the phonological loop is widely accepted, even by researchers whose models of WM are somewhat different from that of Baddeley and col-leagues (Cowan 1999; Engle, Kane, & Tuholski 1999; Kieras & Meyer, 1994) Indeed, even frameworks that appear very different from the Multi-Component Model, for example, distributed processing models like the Interactive Cognitive Subsystems framework (Barnard, 1999) and the CAP2 Hybrid Architecture (Schneider, 1999), or the biologically based computational model of O’Reilly, Braver, and Cohen (1999), are compatible with the construct of the phonological loop There was some debate about the characteristics of the phonological loop in the 1990s and early 2000s (Caplan, Rochon, & Waters, 1992; Cowan, Nugent, El-liot, & Geer, 2000; Lovatt, Avons, & Masterson, 2000); but Mueller, Sey-mour, Kieras, and Meyer (2003) re-examined these characteristics using more rigorous metrics than the earlier studies and provided evidence confirming the model’s predictions and disarming the criticisms The phonological loop holds about 2 seconds of speech, which listeners re-cord automatically when they hear speech It is thanks to the phonologi-cal loop that when my partner says I haven’t been listening to what he’s just been saying, I can deny the charge and repeat back his last sentence word for word (even though he was right: I wasn’t listening, but what he said automatically went into my phonological loop)

So has evolution given us a phonological loop in order to keep couples together? Not according to Baddeley et al (1998), who suggest that the phonological loop may principally serve in language acquisition, carrying out “a fundamental human capacity to generate a longer lasting representation of a brief and novel speech event—a new word” (p 158) That is to say, the phonological loop may have evolved so that people could hold the phonological representations of new words in focal at-tention long enough to construct more stable, durable representations

In support of this idea, studies such as Gathercole and Baddeley (1990),

Gathercole, Hitch, Service, and Martin (1997), and Michas and Henry (1994) have shown that children of equal nonverbal ability and spatial memory skill, but differing phonological loop capacity, differed in their

ability to remember phonologically unfamiliar names for toys (Pyemass),

or familiar–unfamiliar pairs of words (fairy–bleximus), but not in their ability to remember phonologically familiar names (Peter) or familiar– familiar pairs of words (table–rabbit) In adults, similar evidence has been

offered: Saying an irrelevant series of words is known to interfere with

the functioning of the phonological loop (articulatory suppression), and

Papagno, Valentine, and Baddeley (1991) found that articulatory sup-pression hindered recall of L1–L2 pairs of words but not of L1–L1 pairs; Papagno and Vallar (1992) found that word length (important because

of the limited time span of the phonological loop) had an effect on recall

of L1–L2 word pairs but not on the recall of L1–L1 pairs From these and other studies, Baddeley et al (1998) conclude that “the long-term learn-ing of the sound structures of novel, phonologically unfamiliar words depends on the availability of adequate representations of the sound patterns in the phonological loop” (pp 162–163)

The phonological loop has been linked to other aspects of L2 acqui-sition besides word learning For example, it has been shown to play a role in the early development of productive narrative skills in L2 (O’Brien, Segalowitz, Collentine, & Freed, 2006) In this article, I exam-ine its role in L2 reading, where it may also be the case that “the avail-ability of adequate representations of the sound patterns in the phono-logical loop” is crucial

The Phonological Loop in L2 Reading

The phonological loop comes into play in a somewhat counterintui-tive way in the reading of alphabetic languages Strange as it may seem, L1 readers of languages with alphabetic writing systems store the most recently read material (about as much as the reader can say in 2 seconds)

in their phonological loop rather than in their visuospatial sketchpad L1 readers of these languages do not mentally see what they have just read: They hear it One kind of evidence for this is that activities that interfere with the phonological loop (like counting out loud), performed while reading, interfere with the understanding of sentences (Baddeley, El-dridge, & Lewis, 1981; Waters, Caplan, & Hildebrandt, 1987) Note that what is in question is not how written words are decoded (which is probably from multiple cues at letter, letter-sequence, letter-cluster, and word level) but how they are temporarily stored, and the evidence is that the visual trace is lost in favour of the phonological product

In earlier work (Walter, 2004, 2007), I found that low L2-based VWM

corresponded to problems with reading comprehension, so I wondered

if, inter alia, this problem with comprehension might arise because L2 readers had a problem with phonological storage: If phonological rep-resentations of just-read material are unreliable, it may be difficult to use these representations for higher level mental structure building Mate-rial in the phonological loop is subject to decay, and if the elements (sounds, or words, or onsets and rimes, or whatever units the reader is using) are not linked quickly to long-term representations (e.g., pho-nemes, however specified, or exemplars, or lexemes), decay may make it difficult to associate the sounds with meanings There is evidence that underspecified long-term phonological representations are related to the problems of poor L1 readers (Boada & Pennington, 2006; Elbro, 1996; Elbro & Jensen, 2005; Griffiths & Snowling, 2002; Swan & Gos-wami, 1997) For L2 readers, difficulties in building a mental structure corresponding to a read text may in part result from the lack of a well-elaborated L2 phonological inventory in long-term memory This is what the current study sought to determine If phonemes, as idealised forms,

or as prototypes, or even as multiple exemplars of variants, are not well distinguished from one another, even a slight decay in the phonological loop record may cause enough information loss to make words hard to recognise This is not to say that the L2 reader has to have internalised

a standard pronunciation of L2 phonemes, but only that she or he needs

to be able to distinguish words from one another when they differ by a phoneme; if the reader cannot do this, basing comprehension on the word becomes difficult

Flege and MacKay (2004) found that some L1 Italian speakers, long-time residents of Canada and proficient users of English, still had prob-lems aurally in discriminating \ɑ\ from \\, \e\ from \æ\, and \ i \ from \\,1 identifying both members of each pair as the same vowel when listening

If the same sort of lack of differentiation occurs in the phonological loop during reading, it might be part of the reason for difficulties in building structures corresponding to texts

However, lack of differentiation between two neighbouring phonemes may not be the only source of problematic storage in the phonological loop Lange and Oberauer (2005) give evidence for a model based on competition within a distributed memory representation whereby fea-tures of a phoneme can be “overwritten,” which degrades the memory representation If features of a phoneme could be overwritten, then not only similar phonemes in a one-to-two relationship between L1 and L2, but other phonemes as well, might be poorly represented in the phono-logical loop

1 Where English is represented phonetically here, British conventions for representing standard southern British in the International Phonetic Alphabet are used.

Note that I am focusing on learners of English whose L1 has an alphabetic orthography There is evidence that learners whose L1 or-thography is ideographic may store the immediate products of decoding

in the visuospatial sketchpad, rather than in the phonological loop; and that they may transfer this decoding practice to the reading of an alpha-betic L2, at least initially (Chikamatsu, 1996)

Grapheme-to-Phoneme Conversion

Another possible source of phonemic confusion in L2 is an insuffi-ciently developed grapheme-to-phoneme conversion system If this were the case, a distinctive phonological representation would be available in long-term memory, but faulty grapheme-to-phoneme conversion would lead to activation of the wrong phoneme The wrong phoneme might be activated because an L2 reader applies L1 grapheme-to-phoneme

con-version rules to L2, as in the encoding of white as \wi t\ by a French learner of English, so that white and wheat would be indistinguishable in

the phonological loop

RESEARCH QUESTION

The study reported in this article aimed to answer the question: Is approximate or erroneous phonemic representation one of the factors that distinguishes unskilled from skilled L2 readers?

THE STUDY

The methodology used in this study was a partial replication of Bad-deley (1966) BadBad-deley found that phonological similarity has a large adverse effect on recall of word sequences, but semantic similarity has no comparable effect In other words, people have much more difficulty remembering a list of words when the words are phonologically similar than when they are phonologically different, and this finding is true whether the words are heard or read; similarities in word meaning do not interfere with recall in the same way This finding has frequently been replicated, recently by Mueller et al (2003) and Nimmo and Rood-enrys (2005) The current study uses the same methodology, not to explore the characteristics of the phonological loop, but to examine its possible role in L2 reading Remember that readers of alphabetic lan-guages store the immediate products of decoding in the phonological

loop (Baddeley et al., 1981; Waters et al., 1987) If one reason for poor comprehension by good decoders of written L2 is a poorly elaborated L2 phonological inventory in long-term memory, these learners will have problems rehearsing lists of words for which they have no long-term memory representation, and they can be expected to show a stronger phonological similarity effect in L2 trials than their counterparts who are good L2 comprehenders Although the task of recalling lists of words is not a reading task, there is no reason to assume that the encoding in the phonological loop that takes place for this recall task should be different from the encoding that takes place when words are read for compre-hension

Participants

A cross-sectional study was designed, with three groups of participants: two groups of French learners of English and a control group of L1 English (L1Eng) speakers French learners of English were chosen be-cause French is an alphabetic language, like English: In both these lan-guages readers store the immediate products of decoding in the phono-logical loop, rather than in the visuospatial sketchpad The L1 French learners of English were the members of one class each from the middle and upper state secondary schools of a provincial French town; they were from monolingual French families and were accustomed to southern British English accents in their classrooms The poor L2 comprehender (PoorC) group comprised 21 adolescents: 13 girls and 8 boys with a mean age of 14 years, 7 months, in their fourth year of studying English The good L2 comprehender (GoodC) group comprised 23 young people: 19 young women and 4 young men with a mean age of 17 years,

10 months, in their seventh year of studying English

The 21 L1Eng participants (12 girls, 9 boys, mean age 14 years, 6 months) were from schools in two southern English counties and their families were monolingual English speakers

In Walter (2004), both L2 groups had scored equally well in baseline L1 comprehension measures; in L2, however, only the GoodC group did well, and the PoorC group had the good decoder–poor comprehender profile observed in many earlier studies from Alderson (1984) to Lee and Schallert (1997), and this despite the summary-completion baseline comprehension measure’s being carefully designed to the L2 level of the PoorC group In the 2004 study, both groups did well in resolving anaph-ora in L1, whether the proform was near to or far from its antecedent In L2 the GoodC group performed nearly as well as in L1 in anaphora resolution; but although the PoorC group could resolve anaphora when the proform was near its antecedent, they had difficulty when the

ante-cedent was remote In other words, in L2 (on identical stories to L1, rotated for language across the group) the PoorC group were under-standing sentence by sentence, but were not able to build a mental representation of the text In Walter (2007), the PoorCs (but not the GoodCs) had problems in L2 (but not in L1) in detecting when a main point in a narrative was contradicted by a later statement, and the per-formance of the PoorCs (but not the GoodCs) in detecting these con-tradictions correlated with their L2-based verbal WM So a small increase

in L2-based VWM gave a poor comprehender a correspondingly better chance of comprehending a text well enough to detect a contradiction

Materials

To construct the word sequences for the task, four pools of eight words each were composed All words had the pattern consonant–vowel– consonant:

FrDiss List: Eight French words, maximally differentiated for French

speakers (i.e., eight different initial consonants, eight different vowels,

eight different final consonants, or 8 by 8 by 8): benne, chaude, danse,

file, gaz, loupe, noir, cinq.

FrSim List: Eight French words, minimally differentiated for French

speakers (i.e., two initial consonants, two vowels, two final consonants,

or 2 by 2 by 2): bal, belle, cale, quelle, basse, baisse, casse, caisse.2

EngDiss List: Eight English words, maximally differentiated for English

speakers (8 by 8 by 8): cheese, job, fine, yes, soup, nose, date, rich.

EngSim List: Eight English words, minimally differentiated for English

speakers (2 by 2 by 2): men, mine, met, might, when, wine, wet, white.

All English words were among the 3,400 most common words in L1 English (Sinclair, 1995) Consulting the respondents’ textbooks and

teachers indicated that only the word might, though common in L1

English, risked being less familiar to the PoorC group, so this word was reviewed in class shortly beforehand

Ten four-word sequences were drawn at random from the FrDiss list, with the constraint that no word appeared more than once in a

four-2French dictionaries traditionally record different phonemes for the vowels in bale/cale (/a/) and basse/casse (/ɑ/) However, this difference has disappeared in many French

dialects, and dictionaries now record /a/ for both cases (Rey-Debove, Rey and Robert

1993, p XXI ) In a week’s observation of the PoorC class, I observed no uses of /ɑ/ An elicitation test with 20 young and middle-aged inhabitants of the participants’ town, fo-cused on meaning and not pronunciation, yielded two examples per respondent of these words, all pronounced /a/.

word sequence, and 10 sequences were drawn from the FrSim list in the same way A randomly ordered list of these 20 sequences was then con-structed Likewise, a randomly ordered list of 20 four-word sequences was constructed on the basis of the two English lists (Baddeley (1966) had used five-word sequences in his study, but in the current study, adolescents who piloted the task in L1 with five-word sequences were at floor in recall of the L1 dissimilar sequences.)

Design and Procedure

Participants were tested in groups of 6 to 12 individuals The L1 English participants worked with the English lists only, and the L1 French participants with both the English and the French lists The words were presented in written form, to allow the participants to encode them in the same way as they would if they were reading Before trials in each language began, participants were shown all the words in that language; in the case of the English words, the French participants were reminded of the meanings and pronunciation Large-type versions of each pool of words in the relevant language were posted beside the viewing area throughout each language trial, and participants were in-vited to consult them for spelling if they wished (but were told they would not be penalised if their spelling was not perfect) It was explained that any sequence of four words would contain words from only one of the pools, and that no word would appear twice in the same sequence One word at a time was presented on an overhead transparency at the rate of 2 seconds per word After each four-word sequence, the screen was darkened and participants were given time to write down their recall

of the sequence on an answer sheet They were told that they would have

30 seconds to write down each sequence, but that the experimenter would say “Ready?” or “Prêts?,” depending on which language they were working in, before showing the next sequence, so that anyone who had not finished could ask for more time In fact, with this size of group it was possible to see that everyone had finished writing before going on to the next sequence, and no participant asked for more time Half the French participants saw the English sequences first, and half saw the French sequences first Trials were preceded by a practice session of four se-quences to confirm that all participants had understood the task and could see the display clearly

Results

Participants received a score of 1 or 0 for each sequence A score of 1 was awarded if all the words in a sequence were correct and in the correct

order Words were counted as correct as long as the spelling made it

clear that the correct word was intended (so, e.g., whine was counted as correct for wine) The means and standard deviations for the

word-sequence recall scores are given in Table 1 Figure 1 shows the recall scores by phonological similarity category, language, and group

A 3-by-2 mixed-design ANOVA was carried out on the results of the L1 trials (i.e., French for the GoodC and PoorC groups and English for the L1Eng group), with one between-subjects factor, group (GoodC, PoorC,

TABLE 1 Means and Standard Deviations of Word Sequence Recall Scores (Out Of 10) By Language,

Phonological Similarity, And Group

Lang PhonSim

Mean

SD

Mean

SD

Mean

SD

Mean

SD

N = 44 N = 21 N = 23 N = 21

Note PhonSim = phonological similarity, Diss = dissimilar, Sim = similar, SD = standard

devia-tion.

FIGURE 1 Number of sequences (out of 10) correctly recalled by language, phonological similarity

category, and group.