Applied Linguistics 25/3: 315±339 # Oxford University Press 2004 Transfer of Reading Comprehension Skills to L2 is Linked to Mental Representations of Text and to L2 Working Memory CATHERINE WALTER Institute of Education, University of London Two notions from cognitive psychology were examined in relation to the transfer of reading comprehension skills from L1 to L2: (1) the notion that reading comprehension proceeds by the comprehender's building of a mental structure representing the text and (2) the notion of working memory. Two groups of French learners of English (at upper-intermediate and lowerintermediate pro®ciency levels) participated in the study: members of both groups were pro®cient comprehenders in L1 French, but they di€ered in their ability to comprehend texts in L2 English, even when the lower-intermediate learners had no problem in processing the individual sentences of those texts. Performance in pro-form resolution in two distance conditions provided strong support for the hypothesis that the lower-intermediate group had failed to transfer to L2 the ability to build well-structured mental representations of texts, while the upper-intermediate group had succeeded in transferring this ability. This structure-building ability was in turn linked to the development of working memory in L2. INTRODUCTION The study of the transfer of reading skills from L1 to L2 has a long history. It has sometimes been stated in terms of a threshold e€ect (articulated by Alderson 1984; Clarke 1988) and studied in several L1±L2 pairs (Barnett 1986; Markham 1985; Bossers 1991; Carrell 1991). Lee and Schallert (1997) addressed the question of whether the phenomenon is better characterized as a continuous one, where the contribution of L1 reading ability to L2 reading ability gradually increases as L2 pro®ciency increases, or a discontinuous one, typi®ed by the threshold metaphor. While Lee and Schallert's results were not conclusive, they found more support for a threshold than for a continuous relationship. All of these studies were prompted by the observation that L2 comprehension skills at lower levels of pro®ciency do not seem to keep pace with learners' ability to understand texts sentence-by-sentence. However, none of the studies proposed to de®ne the nature of the comprehension skills that initially fail to transfer and later do transfer. 316 TRANSFER OF READING SKILLS STRUCTURE BUILDING AND L2 READING COMPREHENSION The ®rst question that I will examine is this: is the transfer of mental structure-building skill associated with the level of success in L2 reading comprehension? The idea that comprehension involves the building of coherent mental representations or structures for texts is a familiar one in L1 reading comprehension research (see for example Bransford and Johnson 1972; Carpenter and Just 1977; van Dijk 1977; Johnson-Laird 1983; van Dijk and Kintsch 1983; Schmalhofer and Glavanov 1986; Garnham 1987; Kintsch 1998). In this tradition, I shall adopt Gernsbacher's (1990) `structure building' model. Gernsbacher gives persuasive empirical evidence that structure building during the comprehension of a narrative involves the same processes whether the narrative is communicated orally, in writing or even as a picture story. In other words, there appears to be an underlying mode-independent structure-building skill, separate from the decoding skills used in listening, reading or interpreting graphic information. It would be parsimonious to suppose that the same structure-building skill underlies comprehension in L2. In order to judge the evidence for the transfer of the structure-building skill to the L2 reading comprehension context, let us look in more detail at Gernsbacher's (1990) description of the structure building process: 1 2 3 4 a foundation for a structure is laid at the beginning of a clause or text; subsequent information is mapped onto that foundation so long as the comprehender considers it relevant; this continues in a hierarchical fashion: information relevant to a substructure is mapped onto that substructure; if subsequent information is not considered relevant to an ongoing structure or substructure, then another structure or substructure is initiated. Note the following points: . this process is very rapid, and not available to conscious introspection; . `information' also includes the activation of knowledge in long-term memory; . there may be more than one `good' structure by which di€erent skilled comprehenders encode the same text; . the structure in question is not seen as occupying a place in the brain, but is more likely to consist of a spatially dispersed network of neural activation. In skilled comprehension, readers suppress inappropriate meanings and build large, cohesive, hierarchical structures, in which each new element, as it is integrated, activates preceding related elements. Elements that receive repeated activation are more accessible for recall. In poor comprehension, on the other hand, readers fail to suppress inappropriate meanings and shift CATHERINE WALTER 317 too often to new substructures, building structures that are `bulkier, less cohesive and less accessible' (Gernsbacher 1990: 213). Thus, if a reader is building an ecient mental structure, a new element is ®tted into an appropriate place in the overall framework of that structure, and so activates related elements from earlier in the text. The more frequently an earlier element is activated, the more accessible it remains for recall, so a cohesive structure leads to better recall of earlier elements. In a poor, lesscohesive structure, a new element will activate few previous elements, and earlier elements will be dicult to recall (Gernsbacher 1990: 211±13). What would this mean for readers who were skilled at structure-building when reading in L1 but had not completely transferred this skill to reading in L2? As skilled comprehenders in an L1 situation, they would recall elements from a text soon after reading them; and building a cohesive structure would also allow them to recall these elements later on. In the L2 situation, if the text was well matched to their level of pro®ciency, they would process it well sentence-by-sentence, and so would be able to recall elements soon after reading; but building a poor, less cohesive mental structure would cause problems in recalling elements at a distance. It follows that one way to gauge the skill of a reader in building mental structures is to measure their success in recalling earlier information from the text. Among the di€erent possible indicators of this success, I have chosen coreference resolution. In co-reference, one expression in the text depends on or presupposes another expression (Brown and Yule 1983: 190±222): examples are pronouns and pro-verbs and their antecedents. Suppose a group of readers are able to build hierarchical, cohesive mental structures when reading in L1; and suppose that in L2 they are reading a text corresponding to their syntactic and lexical level of pro®ciency (i.e. a text they can process sentence-by-sentence). Suppose, however, that for some reason they are not able to build sound mental structures based on reading in L2. Gernsbacher's (1990) framework predicts that these readers will resolve coreferring expressions: . well in L1, whether the two expressions are near to or remote from one another; and . well in L2 when the expressions are near to one another; but . poorly in L2 when the expressions are remote from one another. (If, on the other hand, they are having problems with sentence-by-sentence processing, they will have diculty resolving co-referring expressions even when they are near to one another.) In the ®rst part of this study, therefore, I examine whether relative success or failure in structure building in an L2 situation, as measured by co-reference resolution, corresponds to relative success or failure in L2 reading comprehension. 318 TRANSFER OF READING SKILLS WORKING MEMORY AND READING COMPREHENSION The second question to be asked in this study is this: is the development of L2 verbal working memory linked to the transfer of reading comprehension skill (and hence structure-building skill) from L1 to L2? Working memory (henceforth WM) is a system of mechanisms by which humans process the information they need for the performance of complex cognitive tasks and maintain it in an accessible form; that is to say, it is a combination of a processing system and a storage system. What is processed and stored can be information from long-term memory, or new information, or both. Within the general de®nition of WM, detailed descriptions di€er from one model to another, but, as shown in the admirable 1999 volume edited by Miyake and Shah, there is more consensus than disagreement on major issues (Anderson et al. 1996; Baddeley and Hitch 1993; Wilson and Baddeley 1988; Conway and Engle 1994; Cowan 1995; Ericsson and Kintsch 1995; Howes and Young 1996; Kieras and Meyer 1994; O'Reilly et al. 1999; Schneider 1999; Teasdale and Barnard 1993). There is ample support in the literature for the involvement of WM in L1 reading comprehension (Daneman and Carpenter 1980; Baddeley et al. 1985; Daneman and Tardif 1987; Engle et al. 1992; Cantor and Engle 1993; Ericsson and Delaney 1999). In the studies of Yuill and Oakhill (1991) in the UK and Cornoldi et al. in Italy (1996) comparing groups of children who were either good L1 comprehenders, or good decoders but poor comprehenders, the researchers found consistently lower WM capacity for the poor comprehenders. Swanson and Berninger (1995) studied 91 skilled and less-skilled child readers in English-speaking Canada and found that low WM (but not low phonological short-term memory) correlated with poor reading comprehension, supporting the hypothesis that WM de®cits contribute to reading comprehension problems. These are only a few examples of a large body of literature (see also Whitney et al. 1991; Oakhill and Yuill 1996; Perfetti et al. 1996; Light and Capps 1986). In L2, Geva and Ryan (1993) studied 10- to 12-year-old English±Hebrew bilinguals and found that their performance in reading comprehension in L2 was accurately predicted by performance on static and working memory storage tests, in combination with `underlying intelligence and L2 oral pro®ciency'. Their subjects had learnt to read in L1 at the same time as they were learning both to speak and to read in L2. Geva and Ryan did not ®nd any e€ect of years of L2 instruction on L2 reading comprehension. However, it is dicult to compare this study with others focusing on the transfer of reading comprehension skills from L1 to L2, since the latter consider individuals who have learnt to read pro®ciently in L1 before beginning L2 instruction. Also in L2, Harrington (1992) and Harrington and Sawyer (1992) studied WM in Japanese±English bilingual university students, and found a strong correlation between their L2 WM measure and reading comprehension as CATHERINE WALTER 319 measured by the TOEFL. Note, however, that this work focused on learners who were classi®ed as bilingual. The focus of the present study is the transfer of reading comprehension skills from L1 to L2, at a much lower level of L2 pro®ciency than the bilinguals in the Geva and Ryan or Harrington and Sawyer studies. METHODOLOGY Participants In order to control the number of variables in the study, two groups of participants were selected from the same L1 cultural and educational background. Choosing to keep these linguistic and sociological variables constant while varying the level of L2 pro®ciency meant that the mean age and the educational attainment of the two groups were di€erent, and this can be seen as a weakness in the design. The older, more academically advanced group is virtually guaranteed to comprehend somewhat more skilfully than the less advanced group, even in L1, and this needed to be taken into account in the analysis of results. However, the alternatives (selecting groups from di€erent L1, educational or cultural backgrounds in order to control for age and/or educational attainment) would almost certainly have led to even more serious threats to the design. The participants were French native speakers from a middle school (ColleÁge) and an upper school (LyceÂe) in a provincial town in the Savoie department of France. They came from a range of socioeconomic categories and lived either in the town or in one of the nearby mountain resorts. Four possible candidates were excluded after initial interviews, since they were bilingual from birth and spoke both languages at home. Thus all participants in the study were native speakers of French from monolingual households, who did not speak any other language on a regular basis outside of a classroom setting. Two other participants (one from each of the two groups) were excluded when they left school for psychosocial reasons near the beginning of the study. This left two groups of 19 and 22 participants respectively. The lower-intermediate group of participants consisted of 19 adolescents (13 girls and 6 boys, mean age 14;6 years, with a range of 13;5 to 15;11 years) from a single English class in the ®nal year of ColleÁge (middle school). A wide range within a year group is not unusual in the French school system, where little stigma is attached to repeating a year (Royal 1999). The class was of mixed ability but it did not include any pupils with special educational needs. They were in their fourth year of English classes (3  50 minutes a week), and in their second year of another foreign language (usually German or Italian). The upper-intermediate group consisted of 22 young people (18 young women and 4 young men, mean age 17;10 years with a range of 16;11 to 19;6 years) from a single English class in the ®nal year of LyceÂe (upper school). 320 TRANSFER OF READING SKILLS Only pupils with suciently high marks on the examinations at the end of middle school are admitted into upper school, so overall academic ability in this group will have been more uniform than in the younger group. The upper-intermediate participants were in their seventh year of English classes, and had studied or were studying at least one other foreign language. Participants also took part in three other experiments besides those described and referred to here; each participant received a present at Christmas (a book for upper-intermediate participants and a bag of sweets for lower-intermediate participants) and a book voucher at the end of the series of experiments. Materials and procedures Three measures were administered, each in both languages: a baseline comprehension assessment, a pro-form resolution test and a working memory measure. Baseline comprehension measure The baseline comprehension measure was undertaken to establish that both groups were skilled comprehenders in L1 and to con®rm that a di€erence in L2 comprehension skill could not be attributed to decoding or sentence-level processing di€erences. For this measure, a summary completion task, that is a gapped summary to complete, was chosen. Taylor (1996: 83±8) argues persuasively that summary completion o€ers a reading comprehension assessment format which takes account of readers' natural processing of a text. This distinguishes it from multiple choice questions or discrete-item comprehension questions, whose basis in an analysis of reading as a set of subskills has been called into serious question (by, for example, Alderson 2000: 93±7). Moreover, summary completion, unlike question-answering, does not require the reader to understand additional material besides that which is already included in the text: a particularly important point in L2 comprehension testing. The English language stimuli were taken from graded readers designed for learners of English as a foreign language, chosen from the list of the 600 books judged to be the best in print by the Edinburgh Project for Extended Reading (EPER 1998). Short (124- to 150-word) self-contained texts were chosen from sixteen graded readers which corresponded to the pro®ciency level in English of the lower-intermediate group as determined by (1) the EPER level; (2) their teacher's assessment of the texts, (3) comparison with their course book, and (4) observation of the class. Most texts were from the very beginning of each book. Texts were chosen for similarities in style and topics as well as in level and length. Some stories focused on action and plot and some on exploring feelings and relationships, so as not to favour boys' or girls' preferred narrative styles (Millard 1997). Care was taken that the stories did not depend on CATHERINE WALTER 321 cultural schemata that the French participants would not share (Ste€ensen et al. 1979). The sixteen French texts were translations of the English texts, checked for naturalness by two French native speaker linguists. As a ®nal guarantee of the accessibility of the English texts for the lowerintermediate group, the content, vocabulary, and grammatical structures in the English versions of the stories were checked by their English class teacher, who felt comfortable that the children would not have any signi®cant diculty with these. Additionally, she included all less frequent words for productive use in class work shortly before the trials. For each text, a straightforward summary of about ®fty words was written, containing ®ve gaps. Seven native English-speaking volunteers (three linguists and four adolescents of the same age as the lower-intermediate participants) attempted to complete the English summaries without reading the original texts, in order to make certain that no answers were unequivocally internally predictable. The adolescents were paid £5 an hour. The texts and their summaries were entered into the Psyscope programme (Cohen et al. 1993) on a Macintosh PowerBook.1 The languages of the texts were counterbalanced, so that half of the participants read each text in one language and half of the participants read it in the other language. The order of the texts was rotated. To balance the risks of other-language activation,2 familiarity e€ect, and fatigue e€ect, the language was changed every two texts: two texts in one language were followed by two texts in the other language and so on. Each text was preceded ®rst by a screen with a row of asterisks, then by a screen stating the language of the following text (`English'/`francËais'), and ®nally by a screen with the title of the text. Each summary appeared on a single screen after its text, from which it was separated by a screen with a row of asterisks and then a screen with the title of the text and instructions in the appropriate language. Paper answer sheets gave an entire line for each blank in the summary. A sample text and its summary are given in Appendix 1. Piloting was conducted in English with nine native English-speaking girls (mean age 13;7 years) and in both languages with eleven French learners of English (mean age 17;2 years). The aims of this piloting were (1) to con®rm that the basic stories were clear and interesting to this age group; (2) to check the summaries for appropriate diculty and for clarity (answers were discussed with each volunteer after the trial); (3) to use the answers in order to test and ®ne-tune the marking system; and (4) to con®rm that the texts presented no cultural diculties for French participants. In the study, trials were conducted with participants individually. After receiving a spoken explanation in French and before beginning the trials, each participant practised self-paced reading on the computer with a story in French; and ®lled out a sample answer sheet. The participants were told that neither spelling nor grammar was important in ®lling out the summaries; that they could put as many or as few words as they wished for each blank; and 322 TRANSFER OF READING SKILLS that they could write answers in French for blanks in English summaries if they so wished. `Meaning is all that counts' was the ®nal instruction for summary completion. The participant pressed a key to move from one screen to the next, but could not go back to a previous screen. If a screen was skipped because of faulty handling of the computer, however, the experimenter told the participant what had been on the skipped screen or showed it to them on paper (this happened three times). As a further check that decoding or sentence-level processing problems did not interfere with comprehension, participants were invited to ask the meaning of any words they did not understand in English, either in the text or in the summary. No participant asked about more than a total of six words for all of the eight English stories, and the questions were overwhelmingly to con®rm correct knowledge, for example, `Window, c'est feneÃtre?'. In scoring, one point was awarded for each of the ®ve gaps in each summary. Half-points were allowed for answers which, while not complete, gave an indication that the participant had a partial understanding of the information required. Criteria for scoring had been established on the basis of the piloting operation. They were then adjusted using the ®rst ten answer lea¯ets from each group as feedback, that is a few additional acceptable answers occurred in the ®rst ten lea¯ets, and these answers were added to the list of acceptable answers. These twenty lea¯ets were then re-marked, and all the subsequent lea¯ets were marked using the same criteria. Pro-form resolution measure As in Yuill and Oakhill's (1991: 85±95) study of young L1 comprehenders' resolution of co-referring expressions, the co-referring expressions were embedded in naturalistic stories. This procedure was chosen rather than using several passages which would require participants to set up a new context for each pro-form. Two stories (`Burglars', 875 French/865 English words; and `McCarthy', 951 French/938 English words) were written. The stories contained both action and exploration of feelings, in order not to privilege girls' or boys' reading styles (Millard 1997). To ensure that the English texts did not present decoding or sentence-level processing diculties for the lower-intermediate group, the same procedures were followed as for the comprehension assessment. To check for naturalness of discourse, ®ve educated native speakers with no training in linguistics were asked to read the story and say `if there was anything strange about the way they were written'; three Englishspeaking and two French-speaking linguists were asked to read the story and say if the discourse seemed to proceed unnaturally at any point. No problems were signalled by any of the readers. Instances of pronoun and pro-verb co-reference were chosen (16 for the `Burglars' story and 15 for the `McCarthy' story) and when the stories were CATHERINE WALTER 323 printed on paper these words were printed in red to contrast with the ordinary black typeface of the rest of the story.The beginning of the `Burglars' story is given in Appendix 3. The experimenter and three other linguists made independent judgements about the pro-forms in the two stories, putting them into one of two distance categories: . immediate, where the ®rst mention is in the same or previous clause; . remote, where the ®rst mention is more than two clauses distant, with any clear mediating mention also two or more clauses distant. In each of the two stories, ten pro-forms for which all four linguists agreed on classi®cation were used for scoring (four pronouns and one pro-verb in each distance condition). Thus, in each language, each participant was scored on the resolution of ®ve pro-forms for each distance condition (immediate, remote). There was always another plausible ®rst mention in the story for the scored item. Trials were conducted with participants individually. After an explanation in French and a practice session in French and English on several twosentence passages, the participants each silently read one story on paper in each language. They were told to stop when they got to a word in red, read the word aloud, and tell the experimenter `what the word meant', in either language, in as few or as many words as they wished. Half the participants in each group read `Burglars' in French and `McCarthy' in English, and the other half read with the languages reversed. Within these groups, half the participants read the French story ®rst and half read the English story ®rst. When they encountered an expression printed in red, they read the word aloud and then tried to identify the ®rst mention. If the response was incorrect, the experimenter ®rst indicated merely that something was not right; if this did not elicit the correct answer the experimenter gave some guidance, recalling the context of the story. This procedure was chosen for several reasons: . to allow for the possibility of marking the results either strictly or liberally, to avoid ceiling or ¯oor e€ects (in the event, however, only the spontaneous responses were used in the scoring) . to prevent an early error from in¯uencing later items. The decision had been made to embed the trial items in one story for each language rather than writing a new mini-story and obliging participants to set up a new context for each item. However, a risk of this approach is that earlier errors can in¯uence later decisions; this risk was minimized by the procedure described. Note that the instrument was measuring pro-form resolution skill, and not reading comprehension, which had been assessed in the baseline study. . to put the participants at their ease, and to avoid their leaving the trials with a sense of failure. 324 TRANSFER OF READING SKILLS As a further check that decoding or sentence-level processing problems did not interfere with comprehension, participants were also encouraged to ask for the meaning of any words or expressions that they did not understand in English. There were very few problems with lexis, and participants' questions were overwhelmingly con®rming in nature. A correct answer given without any prompting was scored 1 and all other responses were scored 0. (More liberal scoring did not give di€erent results.) Verbal working memory measure The verbal WM measure used in this study is fully described in Walter (2000) and Walter and Williams (in preparation). This measure is based on Waters and Caplan's (1996) modi®cation of the classical Daneman and Carpenter (1980) reading span measure. The great advantage of a measure of this type over the Daneman and Carpenter measure is this: WM comprises both processing and storage; but while the Daneman and Carpenter measure requires subjects both to process and to store, it only measures storage. Like Waters and Caplan's (1996) instrument, the WM measure in the present study measures both processing and storage. In each language, the stimuli consisted of sets of simple declarative sentences, six to eight words long, in which the argument requirements of the verb are either respected (`logical' sentences) or not (`illogical' sentences). That is to say, in half the sentences (`illogical' sentences) an inanimate subject is used with a verb which requires an animate one (e.g. Several of the ¯utes play my cousins). The sentence-®nal words are all di€erent concrete nouns. Sentences were randomly ordered for `logical'/'illogical' conditions. They were then divided into four series of ®ve increasingly longer sets of sentences: the ®rst series contained ®ve two-sentence sets and the last series contained ®ve ®ve-sentence sets. Adjustments were then made (respecting the `logical'/ 'illogical' order) to ensure that, in any set, the sentence-®nal words were not phonologically similar and did not have obvious semantic links. There were four versions of the materials, varying along two parameters: 1 2 In each language, the original randomized list was reversed and re-divided into sets, and small adjustments were made to avoid phonological similarities and semantic links. Two of the versions began with a series of ®ve two-sentence French sets and two began with a two-sentence English series. In all versions, the language was changed after the initial two-sentence series and then after every two series. The sentences were entered into the Psyscope programme (Cohen, MacWhinney, Flatt, and Provost 1993) on a Macintosh PowerBook, to appear in the centre of the screen, one sentence at a time. Each set of sentences was followed by a screen showing a line of asterisks. Each group of ®ve sets was preceded by a screen announcing the language and length of the next set. CATHERINE WALTER 325 Participants judged whether each sentence was logical or illogical (processing measure); the time taken for this judgement was recorded (processing measure); and, at the end of a set, participants attempted to remember the last word of each sentence in order (recall measure). Each participant was allowed to choose which of 1 or 0 on the computer keyboard they preferred for `logical' and `illogical' choices, and practised with sample material until they were at ease with the procedure. Pressing the button to signal `logical' or `illogical' made the next sentence or a row of asterisks appear on the screen. When the asterisks appeared, the participant attempted to say the sentence®nal words of the set in order and the result was recorded on paper by the experimenter. Success in an item was de®ned as correct recall of all sentence®nal words in a set, in order, with this exception: if a participant said the sentence-®nal words correctly but in the wrong order, the experimenter asked (depending on the language in use) `Dans l'ordre?/In order?', and a correct response to this prompt was counted as a success. The experimenter did not otherwise indicate whether individual responses were right or wrong. The trial continued until the participant could no longer reliably recall ®nal words. Working memory scores in each language were calculated for each participant, based on three ®gures: one measure of storage (reliably remembered sentence-®nal words, or `recall span'), and two measures of processing (reaction time in milliseconds for correct answers at recall span; and percentage of correct logicality judgements). Recall span, as in Daneman and Carpenter (1980) and Waters and Caplan (1996), was expressed as the highest level (sentences per set) at which the participant correctly recalled the sentence-®nal words in three out of ®ve sets, with an extra half-point given for correct recall in two of the ®ve sets at the next higher level. For example, a participant who recalled all sentence-®nal words correctly on three out of ®ve four-sentence sets, and then recalled all sentence-®nal words correctly on two out of ®ve ®ve-sentence sets, would obtain a recall span score of 4.5. Reaction time was measured as the time in milliseconds between the appearance of a sentence on the screen and the pressing of the button to signal the logicality judgement. Only reaction times at reading span for correct logicality judgements were taken into account. To obtain an overall WM score, z-scores for the three components (recall span, percentage of correct logicality judgements, reaction time) were calculated by taking both French and English components into account. In other words, to take recall span as an example, each subject contributed two values (recall span in French and recall span in English) to the sample population on which the recall span z-scores were based. The z-scores for reaction time at span and percentage of correct logicality judgements were calculated in the same way. The reaction time z-scores were multiplied by ±1 (higher reaction times indicating less good processing, whereas higher scores on the other two measures indicate better processing and recall). Overall WM scores were calculated as the means of the three z-scores. 326 TRANSFER OF READING SKILLS RESULTS Baseline reading comprehension assessment Mean scores out of 40 and standard deviations are given in Table 1. Figure 1 shows the summary completion scores by language and group. From Table 1 and Figure 1, showing the results of the baseline reading comprehension assessment, it can be seen that in L1 French, while both groups showed good evidence of comprehension, at 80 per cent and 70 per cent of correct summary completion replies, there was a small di€erence between them. This di€erence was con®rmed by a one-way ANOVA in which there was an e€ect of group, F(1,39) = 7.041, p < .05. This di€erence was not surprising, given the disparity in maturity and academic attainment between the two groups. However, at 70 per cent of correct judgements, the lowerintermediate participants were comprehending well in L1. The next question was whether the between-group di€erence in comprehension scores in L2 paralleled the di€erence in L1; or whether the lower-intermediate group's L2 Table 1: Means and standard deviations of summary completion scores for all participants (out of 40), by language and group Language French English Lower-intermediate (N = 19) Upper-intermediate (N = 22) Mean SD Mean SD 28.07 13.84 4.18 7.21 32.00 27.73 4.30 6.28 Figure 1: Mean summary completion scores (out of 40), by language and group CATHERINE WALTER 327 scores were much further from their L1 scores than their upper-intermediate counterparts' scores. In order to investigate whether there was a di€erence in the degree to which the two groups were transferring their comprehension skills from L1based comprehension to L2-based comprehension, a 22 mixed-design ANOVA was carried out on the summary completion scores, with one between-subjects factor: group (upper-intermediate, lower-intermediate), and one within-subjects factor: language (French, English). Both main e€ects were signi®cant. There was a main e€ect of group, with upper-intermediate participants completing summaries more successfully than lower-intermediate participants: F(1,39) = 47.20, p < .0001. There was a main e€ect of language, with summaries of French stories being completed more successfully than summaries of English stories: F(1,39) = 59.70, p < .0001. There was a signi®cant two-way interaction between language and group: F(1,39) = 17.95, p = .0001. This demonstrates that the lower-intermediate participants' comprehension scores in the two languages were much further apart than those of their upper-intermediate counterparts. (Stricter and more liberal scoring regimes gave the same results.) Pearson product±moment correlations were carried out between the mean French and English summary completion scores, to investigate whether French L1 comprehension skill predicted English L2 comprehension skill. There was an overall correlation between the two sets of summary completion scores, r = .31, p < .05, but no signi®cant correlations when the scores were broken down by group. Pearson product±moment correlations were also carried out between the English mean summary completion scores and English ®nal school marks from the previous academic year as based on the schools' pro®ciency examinations. Here, in contrast, there were correlations not only for both groups taken together (r = .39, p < .05), but also for each group taken on its own (lower-intermediate: r = .73, p < .001; upperintermediate: r = .61, p < .01). Note that these school marks were based on di€erent examinations for each group, but that the examinations were of the same nature and based on the same scales. Within each group, pro®ciency in English, as measured by school marks, predicted success in English summary completion better than did L1 French reading comprehension skill measured as success in summary completion; this pattern is current in the literature about reading at lower levels of L2 pro®ciency (in, for example, Bernhardt and Kamil 1995; Carrell 1991). Recall that the comprehension texts were speci®cally written and tested so that the lower-intermediate group could process the language at a sentence level. Yet the di€erence between lower-intermediate L1 and L2 text comprehension scores (36 percentage points) was signi®cantly larger than the upper-intermediate L1/L2 di€erence (10 percentage points). It is unlikely that the di€erence between the L1 scores of the two groups would cause such a large disparity between their L2 scores without the intervention of some other factor. It can thus be said with con®dence that these two groups of 328 TRANSFER OF READING SKILLS readers were (a) processing well at sentence level and comprehending well at text level in L1; (b) processing well at sentence level in L2; and (c) showing patterns of transfer (upper-intermediates) or diculty in transfer (lowerintermediates) of reading comprehension skills from L1 to L2 texts, similar to the patterns found in previous studies. Pro-form resolution measure Mean scores out of 5 and standard deviations are given in Table 2. Figure 2 shows the pro-form resolution scores by language, distance, and group. Table 2: Means and standard deviations of pro-form resolution scores (out of 5), as a function of language, distance, and group Language French English Distance Immediate Remote Immediate Remote Lower-intermediate (N = 19) Upper-intermediate (N = 22) Mean SD Mean SD 4.10 3.73 3.74 2.26 0.87 1.03 1.00 1.33 4.45 3.90 4.23 4.00 0.74 1.23 0.81 1.23 Figure 2: Pro-form resolution scores (out of 5), by language, distance, and group CATHERINE WALTER 329 A mixed-design 222 ANOVA was carried out, with one betweensubjects factor: group (upper-intermediate, lower-intermediate), and two within-subjects factors: language (French, English) and distance (Immediate, Remote). All main e€ects were signi®cant: a main e€ect for group, F(1,39) = 10.58, p < .005, indicating that the upper-intermediate participants resolved pro-forms more successfully than the lower-intermediate participants; a main e€ect for language, F(1,39) = 8.53, p < .01 indicating that L1 pro-forms were resolved more successfully than L2 pro-forms; and a main e€ect for distance, F(1,39) = 27.47, p < .0001, indicating that Immediate pro-forms were resolved more successfully than Remote pro-forms. There was a signi®cant two-way interaction between language and group, F(1,39) = 6.39, p < .05, indicating that the upper-intermediate participants' scores in L2 were reliably closer to their scores in L1 than those of their lower-intermediate counterparts. There was a signi®cant two-way interaction between distance and group, F(1,39) = 4.33, p < .05, indicating that the upper-intermediate participants' scores for Remote pro-forms were reliably closer to their scores for Immediate pro-forms than those of their lower-intermediate counterparts. Most importantly, there was a signi®cant three-way interaction between language, distance and group, F(1,39) = 6.91, p < .05, indicating that the di€erence in L1 and L2 proform resolution between the two groups in the Remote condition was reliably di€erent from that in the Immediate condition. (Note that an ANOVA with `Assignment of story version' nested within group was carried out, but its results were identical to those of the 22 ANOVAs reported here.) Pearson product±moment correlations were performed between L2 baseline comprehension and L2 resolution of Remote pro-forms for each group. The results are given in Table 3. The upper-intermediate participants were resolving Immediate and Remote pro-forms similarly well in L1 and in L2. The lower-intermediate participants resolved Immediate and Remote pro-forms well in L1, and resolved Immediate pro-forms well in L2. This is consistent with their building e€ective structures corresponding to the L1 text, and with their processing the L2 text well at sentence level. However, the lower-intermediates were signi®cantly worse at resolving Remote pro-forms in L2 than in L1. If they had experienced diculty in sentence-level processing in L2, this would have been re¯ected in their Immediate L2 pro-form resolution scores, and this was not the case. The patterns of the lower-intermediate scores correspond to what was predicted if they were building e€ective structures corresponding to L1 texts, but failing to build e€ective structures corresponding to L2 texts. In other words, these results support the hypothesis that the transfer of mental structure-building skill is associated with the level of success in L2 reading comprehension. In the baseline comprehension measure, we saw two groups of learners at two di€erent levels of L2 pro®ciency. The upper-intermediate group showed via the summary completion measure that they comprehend reading texts similarly well in L1 and in L2. The lower-intermediate group results in 330 TRANSFER OF READING SKILLS summary completion showed that they comprehend well in L1; but that they have problems comprehending L2 reading texts, even when these are geared to their level of L2 pro®ciency. Now we see that, in the Remote pro-form resolution measure, designed to gauge structure building skill, the lowerintermediate group performs well in L1 and poorly in L2. This is consistent with success in transfer of structure building skill being associated with success in L2 comprehension. In this regard, the pattern of correlations in Table 3 is interesting. There are no correlations between summary completion and Remote pro-form resolution in L1, and this may indicate that both groups were suciently skilled in structure building in L1 so that individual di€erences were not re¯ected. In L2, the lack of correlation for the lower-intermediate group is unsurprising given their relative lack of success in both measures. What is interesting is the signi®cant correlation in L2 for the upper-intermediate group, whose members were successful in L2 in both comprehension and summary building measures. The better an upper-intermediate participant was at structure building, the better s/he was at comprehension. This provides further support for the hypothesis that the transfer of structure building skill is associated with success in L2 comprehension. Table 3: Pearson product-moment correlations between summary completion scores and remote pro-form resolution scores, as a function of language and group French L1 English L2 Lower-intermediate (N = 19) Upper-intermediate (N = 22) Total (N = 41) ns ns ns .63** ns .65*** ** p < .01; *** p < .0001 Working memory measure Means and standard deviations for percentage of correct logicality judgements, number of ®nal words reliably recalled and reaction time for correct answers at best recall span by language and by group are given in Table 4. The means and standard deviations of overall WM scores, by language and group, are given in Table 5. In order to examine the overall relationship of WM with the baseline comprehension (summary completion) scores, both French and English WM scores were plotted on the x-axis of the same graph, against the means of the corresponding summary completion scores. The resulting correlations are given in Table 6. Correlations were compared using Fisher's (1921) method for testing the di€erence between two independent rs, and this showed CATHERINE WALTER 331 Table 4: Means and standard deviations of scores for recall span (maximum 5.0), logicality judgement (percentage of correct judgements) and reaction time for correct answers (in milliseconds, at span) for all participants, by group Measure Language Lower-intermediate (N = 19) Mean Recall span French English Logicality French judgement English Reaction time French English SD 3.73 2.05 Upper-intermediate (N = 22) Total (N = 41) Mean Mean 0.79 0.95 4.50 2.84 SD 0.77 0.58 4.14 2.47 94 83 4 12 96 91 5 8 95 87 6144 11785 1615 3704 5278 6522 1148 1401 5679 8961 SD 0.86 0.86 5 11 1434 3777 Table 5: Means and standard deviations of WM scores (expressed as means of the z-scores of the three contributing measures, with the reaction time zscore polarities reversed) for all participants, by language and group Language French English Lower-intermediate (N = 19) Upper-intermediate (N = 22) Total (N = 41) Mean SD Mean SD Mean SD 0.35 ±1.01 0.25 0.59 0.73 ±0.02 0.26 0.30 0.55 ±0.48 0.32 0.67 Table 6: Correlations between all WM scores and summary completion scores All scores French scores English scores Upper-intermediate Lower-intermediate .76** .73*** .46** .79*** .33* * p < .05; ** p < .01; *** p < .0001 signi®cant di€erences between the French and English correlations (p < .05) and the lower-intermediate and upper-intermediate correlations (p < .01). In other words, Table 6 shows that having a higher verbal WM (ability to process and store complex information simultaneously) corresponds to being 332 TRANSFER OF READING SKILLS better at reading comprehension, for both groups. It also shows that this is signi®cantly more the case for the lower-intermediate group. Small advantages in WM give signi®cant advantages in comprehension, especially at the lower level of pro®ciency. This supports the hypothesis that there is a link between the development of verbal WM in L2 and success in L2 reading comprehension. When correlations were carried out between WM and summary completion scores for each language by group, neither group yielded a signi®cant correlation in French, and nor did the upper-intermediate group yield a correlation in English. The only signi®cant correlation in English was between the lower-intermediate group WM scores and their baseline comprehension scores, r = 0.49, p < .05. This single signi®cant correlation is similar to ®ndings in other studies of WM. Observation of WM di€erentiation `when the going gets tough' is common in the literature, and is well accounted for in various models (e.g. Baddeley and Logie 1999; Ericsson and Kintsch 1995; Ericsson and Delaney 1999; Engle et al. 1999). Neurophysiological support for this observation is provided by D'Esposito et al. (1995), who performed fMRI studies on patients who were simultaneously carrying out a spatial task and a verbal task; D'Esposito's group found that the dorsolateral prefrontal cortexÐthe part of the brain that has been found to be associated with WM executive control processesÐwas activated (as were the domain-speci®c areas associated with the two tasks). It is notable that the prefrontal cortex was not measurably activated when the two tasks were performed independently, but only when there was a higher overall demand on WM. In the present study, the correlation between WM scores and comprehension scores corresponds well to the lower-intermediate group's experiencing high demand on their L2 WM in the summary completion tasks. In order to examine the relationship of verbal working memory with structure building, both French and English WM scores were plotted on the xaxis of the same graph, against the means of the corresponding Remote proform resolution scores. The resulting correlations are given in Table 7. Detailed correlations between English WM and English Remote pro-form resolution scores by group were performed, but these did not, however, yield signi®cant correlations. The Table 7 ®gures do tend to support the hypothesis of higher demands being made on WM in the English pro-form resolution condition. They also correspond to higher WM demands on the lower-intermediate group generally. However, the predicted correlation between Remote pro-form resolution and WM for the lower-intermediate group did not emerge, and this merits further study. CATHERINE WALTER 333 Table 7: Correlations between all WM scores and Remote pro-form resolution scores All scores French scores English scores Upper-intermediate Lower-intermediate .45** .52* ns . 53** ns * p < .001; ** p < .0001 DISCUSSION The ®rst hypothesis that this study addressed was: The transfer of mental structure-building skill is associated with level of success in L2 reading comprehension. This hypothesis was supported: the baseline reading comprehension measure showed that both groups of participants were skilled comprehenders in L1, but di€ered in their ability to transfer their comprehension skill to L2. Next, the pro-form resolution measure showed that both groups were building sound mental structures in L1, but that there was a signi®cant di€erence between the groups' ability to do so in L2, even when they understood texts on a sentence-by-sentence basis. Patterns of correlation supported the notion that the more successful structure builders were operating similarly in L1 and L2, and that the better they were at L2 structure building, the better they were at L2 reading comprehension. The second hypothesis was: The transfer of reading comprehension skill (and hence structurebuilding skill) from L1 to L2 is linked to the development of verbal working memory in L2. This hypothesis also received some support: verbal WM correlated with the baseline reading comprehension measure precisely in that `tough' circumstance (lower-intermediate, L2 Remote pro-form resolution) predicted by models of WM. In addition, L2 verbal working memory tended to correlate with L2 structure building as measured by Remote pro-form resolution. However, there was not a signi®cant correlation between Remote pro-form resolution and WM for the lower-intermediate group speci®cally, and this deserves further study. So a crucial element in transferring reading comprehension skill from L1 to L2 appears to be the transfer of the structure-building ability; and this transfer appears to be linked to the development of WM in L2. These ®ndings provide a possible explanation for the perceived threshold nature of the transfer. If successful structure building is accomplished in L1 but not in L2, it is not the ability to build mental structures that is absent; what is lacking is the 334 TRANSFER OF READING SKILLS attainment of some level of L2 ability which acts as a precondition for the structure-building skill to operate. Recall the nature of the structure-building skill as described by Gernsbacher (1990). If some element or set of elements in L2 WM were insuciently developed for the comprehender to link incoming material to a previously established structure, the comprehender would fall back on initiating a new structure. The result of this process occurring repeatedly would be the sort of bulky, inaccessible mental representation that less-skilled L1 comprehenders produce (Gernsbacher, 1990). A failure to link appropriately even once or twice at crucial early points in the building of a mental representation might be enough to prevent the building of a cohesive structure, and hence to prevent recall of earlier information. The L2 comprehender would continue to have diculties in recall for a time, even as the underlying capacity continued to develop progressively. Success in building reliable mental representations based on L2 would only improve when the underlying capacity had developed to a point where quite a large number of the unconscious decisions involved in building a representation were successful. This would account for the perceived threshold nature of the L1-to-L2 transfer. The question now arises: what is the nature of the link between L2 WM and L2 reading comprehension? One way of investigating this question is to examine empirically the possible sources of demand on WM capacity, for example less well elaborated L2 phonological representations, or more e€ortful syntactic processing. Another is to undertake a longitudinal study in which WM, L2 reading comprehension skill, and various elements of L2 pro®ciency are tracked. This would eliminate the problems associated with any experimental design that involves matched groups, more pronounced in the present case where it is very unlikely that groups can be matched both for ®rst language background and for age. Further studies involving learners from di€erent L1 backgrounds, including languages outside the Indo-European family and languages whose orthography is not phonological, would throw additional light on this subject. (Final version accepted April 2004) ACKNOWLEDGEMENTS I am most grateful to the participants from the ColleÁge Jean Rostand and the LyceÂe ReÂgional Polyvalent et Professionel, both in MouÃtiers, France; to Michael Swan and John Williams for their continuing advice and support; to Julie Norton, Susan Mandala, Dominique Flandin-Granget, and Catherine CobeÁs for serving as linguist informants; to the members of the Working Memory Workshop, and especially Alan Baddeley and Robert Logie, for their advice on working memory matters; and to ®ve anonymous reviewers for their very helpful comments on earlier versions of this article. CATHERINE WALTER 335 APPENDIX 1 Project Omega Charles Hat®eld Baker III was a very rich man. There wasn't a richer man in all of New York City. He was the head of a large number of important companies. He had land and buildings and houses in the country. He had boats and horses and an aeroplane. But he was a simple man. He liked the simple things of life: a good cup of co€ee, a kiss from a pretty girl, a walk through the park on a ®ne autumn day. He had all these things on this Wednesday morning. He drank two cups of very good Italian co€ee for breakfast. His daughter, Julia kissed him goodbye. `Have a nice day, daddy,' she said. And he left his apartment to walk to his oce. He always walked through Central Park to his oce. (Reprinted by permission of Addison Wesley Longman Limited from Project Omega by Elaine O'Reilly, 1991.) Summary Charles Hat®eld Baker III was _____A_____ New York City. He controlled a lot of _____B_____. He had land, buildings, houses, and expensive possessions. He liked _____C_____ things. On this Wednesday he had two _____D_____, his daughter kissed him, and then he left his apartment to _____E_____. APPENDIX 2 (The beginning of one of the stories from the pro-form resolution test. The words which are underlined here were printed in red.) BurglarsÐEnglish It was a very dicult day at the oce. Monica's secretary was ill, and a temp from an agency was taking her place. She wasn't very ecient, and Monica had to explain the same things to her again and again. Then there was a problem in another part of the building that Monica had to go and sort outÐ she was away from her desk for more than an hour, and while she was out she missed an important phone call. [. . .] 336 TRANSFER OF READING SKILLS NOTES 1 The texts were entered into the Psyscope program so as to appear in one of four segmentation conditions: sentence-by-sentence; word-by-word; by meaningful phrases; or cut in the middle of syntactic units. This was done to test hypotheses about possible di€erences in reading performance based on segmentation condition. In the event, there was no signi®cant di€erence between any of the conditions. 2 To some extent, both languages are always activated in bilinguals. But excessive otherlanguage activation can interfere signi®cantly with processing in tasks where the participants have no indication of which language is going to come next. However, studies have shown that bilinguals can exercise a great deal of conscious control over excessive other-language activation, especially when it is clearly indicated which language is called upon for each part of a task, and also when languages are not switched too often. (See Scarborough et al. 1984; Durgunoglu and Roediger 1987; Basden et al., 1994; Costermans and Galland 1980; Doctor and Klein 1992; and Durgunoglu and Hancin's 1992 review of the subject.) REFERENCES Alderson, J. C. 1984. `Reading in a foreign language: a reading problem or a language problem?' in J. C. Alderson and A. H. Urquhart (eds): Reading in a Foreign Language, Burnt Mill, Harlow: Longman Group Limited. pp. 1±24. Alderson, J. C. 2000. Assessing Reading. Cambridge: Cambridge University Press. Anderson, J. R., L. M. Reder, and C. Lebiere. 1996. `Working memory: activation limitations on retrieval.' Cognitive Psychology 30: 221±56. Baddeley, A. D. and G. J. Hitch. 1993. `The recency effect: implicit learning with explicit retrieval.' Memory and Cognition 21: 146±55. Baddeley, A. D. and R. H. Logie. 1999. `Working memory: the multiple-component model' in A. Miyake and P. Shah: Models of Working Memory: Mechanisms of Active Maintenance and Executive Control, Cambridge: Cambridge University Press. pp. 28±61. Baddeley, A. D., R. Logie, I. Nimmo-Smith, and N. Brereton. 1985. `Components of fluid reading.' Journal of Memory and Language 24: 119±31. Barnett, M. A. 1986. `Syntactic and lexical semantic skill in foreign language reading: importance and interaction.' The Modern Language Journal 704: 343±9. Basden, B. H., J. L. Bonilla-Meeks, and D. R. Basden. 1994. `Cross-language priming in word-fragment completion.' Journal of Memory and Language 33: 69±82. Bernhardt, E. B. and M. L. Kamil. 1995. `Interpreting relationships between first language and second language reading: consolidating the linguistic threshold and the linguistic interdependence hypotheses.' Applied Linguistics 161: 15±34. Bossers, B. 1991. `On thresholds, ceilings and short-circuits: the relation between L1 reading, L2 reading and L2 knowledge' in J. H. Hulstijn and J. F. Matter (eds): Reading in Two Languages. AILA Review 8: 45±60. Amsterdam: Free University Press. Bransford, J. D. and M. K. Johnson. 1972. `Contextual prerequisites for understanding: some investigations of comprehension and recall.' Journal of Verbal Learning and Verbal Behavior 4/2: 717±26. Brown, G. and G. Yule. 1983. Discourse Analysis, Cambridge: Cambridge University Press. Cantor, J. and R. W. Engle. 1993. `Workingmemory capacity as long-term memory activation: an individual-differences approach.' Journal of Experimental Psychology: Learning, Memory and Cognition 19: 1101±14. Carpenter, P. A. and Just, M. A. 1977. `Integrative processes in comprehension' in D. Laberge and S. J. Samuels (eds): Basic Processes in Reading: perception and comprehension, Hillsdale, NJ: Lawrence Erlbaum Associates. pp. 217±41. Carrell, P. L. 1991. `Second language reading: reading ability or language proficiency?' Applied Linguistics 122: 159±79. CATHERINE WALTER Clarke, M. A. 1988. `The short circuit hypothesis of ESL readingÐor when language competence interferes with reading performance' in P. L. Carrell, J. Devine and D. E. Eskey (eds): Interactive Approaches to Second Language Reading, Cambridge: Cambridge University Press. pp. 114±24. Cohen, J. D., B. MacWhinney, M. Flatt, and J. Provost. 1993. `Psyscope: a new graphic interactive environment for designing psychology experiments.' Behavioral Research Methods, Instruments and Computers 25: 252±71. Conway, A. R. A. and R. W. Engle. 1994. `Working memory and retrieval: a resourcedependent inhibition model.' Journal of Experimental Psychology: General 123: 354±73. Cornoldi, C., R. De Beni, and F. Pazzaglia. 1996. `Profiles of reading comprehension difficulties: An analysis of single cases' in C. Cornoldi and J. Oakhill (eds): Reading Comprehension Difficulties: Processes and Intervention, Mahwah, NJ: Lawrence Erlbaum Associates. pp. 113±36. Costermans, J. and J. Galland. 1980. `Sur l'accessibilite du lexique chez les bilingues.' Canadian Journal of Psychology/Revue canadienne de Psychologie 34: 381±7. Cowan, N. 1995. Attention and memory: an integrated framework, New York: Oxford University Press. Daneman, M. and P. Carpenter. 1980. `Individual differences in working memory and reading.' Journal of Verbal Learning and Verbal Behavior 19: 450±66. Daneman, M. and T. Tardif. 1987. `Working memory and reading skill re-examined' in M. Coltheart (ed.): Attention and performance, XII, Hove: Lawrence Erlbaum Associates Ltd. pp. 491±508. D'Esposito, M., J. A. Detre, D. C. Alsop, R. K. Shin, S. Atlas, and M. Grossman. 1995. `The neural basis of the central executive system of working memory.' Nature: 378: 279±81. Doctor, E. A. and D. Klein. 1992. `Phonological processing in bilingual word recognition', in R. J. Harris (ed.): Cognitive Processing in Bilinguals, Amsterdam: Elsevier Science Publishers. pp. 237±52. Durgunoglu, A. Y. and B. J. Hancin. 1992. `An overview of cross-language transfer in bilingual reading' in R. J. Harris (ed.): Cognitive Processing in Bilinguals, Amsterdam: Elsevier Science Publishers. pp. 391±411. 337 Durgunoglu, A. Y. and H. L. Roediger III. 1987. `Test differences in accessing bilingual memory.' Journal of Memory and Language 26: 377±91. Ehrlich, M.-F. 1996. `Metacognitive monitoring in the processing of anaphoric devices in skilled and less skilled comprehenders' in C. Cornoldi and J. Oakhill (eds): Reading Comprehension Difficulties: Processes and Intervention, Mahwah, NJ: Lawrence Erlbaum Associates. pp. 221±49. Engle, R. W., J. Cantor, and J. J. Carullo. 1992. `Individual differences in working memory and comprehension: A test of four hypotheses.' Journal of Experimental Psychology: Learning, Memory and Cognition 18: 972± 92. Engle, R. W., M. J. Kane, and S. W. Tuholski. 1999. `Individual differences in working memory capacity and what they tell us about controlled attention, general fluid intelligence and functions of the prefrontal cortex' in A. Miyake and P. Shah. Models of Working Memory: Mechanisms of Active Maintenance and Executive Control, Cambridge: Cambridge University Press. pp. 102±34. EPER (Edinburgh Project on Extensive Reading). 1998. List of Graded Readers: Recommended Titles. Edinburgh: Institute for Applied Language Studies, University of Edinburgh. Ericsson, K. A. and P. F. Delaney. 1999. `Long-term working memory as an alternative to capacity models of working memory in everyday skilled performance' in A. Miyake and P. Shah. Models of Working Memory: Mechanisms of Active Maintenance and Executive Control, Cambridge: Cambridge University Press. pp. 257±97. Ericsson, K. A. and W. Kintsch. 1995. `Longterm working memory.' Psychological Review 102: 211±45. Fisher, R. A. 1921. `On the probable error of a coefficient of correlation deduced from a small sample.' Metron 1: 3±32. Garnham, A. 1987. Mental Models as Representations of Discourse and Text. Chichester: Ellis Horwood. Gernsbacher, M. A. 1990. Language Comprehension as Structure Building. Hillsdale, NJ: Lawrence Erlbaum Associates. Geva, E. and E. B. Ryan. 1993. `Linguistic and cognitive correlates of academic skills in first 338 TRANSFER OF READING SKILLS and second languages.' Language Learning 43: 5±42 Harrington, M. 1992. `Working memory capacity as a constraint on L2 development' in R. J. Harris (ed.): Cognitive Processing in Bilinguals, Amsterdam: Elsevier Science Publishers. pp. 123±35. Harrington, M. and M. Sawyer. 1992. `Second language working memory capacity and second language reading skills.' Studies in Second Language Acquisition 141: 25±38. Hirst, W. and G. A. Brill. 1980. `Contextual aspects of pronoun assignment.' Journal of Verbal Learning and Verbal Behavior 19: 168± 75. Howes, A. and R. M. Young. 1997. `The role of cognitive architecture in modelling the user: Soar's learning mechanisms.' HumanComputer Interaction 12: 311±43. Kieras, D. E. and D. E. Meyer. 1994. The EPIC Architecture for Modeling Human Informationprocessing and Performance: A brief introduction. EPIC Report N 0 1, TR-94/ONR-EPIC-1, Ann Arbor: University of Michigan. Kintsch, W. 1998. Comprehension: A Paradigm for Cognition. Cambridge: Cambridge University Press. Johnson-Laird, P. N. 1983. Mental Models: Towards a cognitive science of language, inference and consciousness. Cambridge: Cambridge University Press. Lee, J.-W. and D. L. Schallert. 1997. `The relative contribution of L2 language proficiency and L1 reading ability to L2 reading performance: a test of the threshold hypothesis in an EFL context.' TESOL Quarterly 31/4: 713±39. Light, L. L. and J. L Capps. 1986. `Comprehension of pronouns in young and older adults.' Developmental Psychology 22: 580±5. Mandler, J. M. 1984. Stories, Scripts and Scenes: Aspects of Schema Theory. Hillsdale, NJ: Erlbaum. Markham, P. 1985. `The rational deletion cloze and global comprehension in German.' Language Learning 35: 423±30. Millard, E. 1997. Differently Literate: Boys, Girls and the Schooling of Literacy, London: Falmer Press. Miyake, A. and P. Shah. 1999. Models of Working Memory: Mechanisms of Active Maintenance and Executive Control, Cambridge: Cambridge University Press. Oakhill, J. and N. Yuill. 1996. `Higher order factors in comprehension disability: processes and remediation' in Cornoldi, C. and J. Oakhill (eds): Reading Comprehension Difficulties: Processes and Intervention, Mahwah, NJ: Lawrence Erlbaum Associates, Inc. pp. 69±92. O'Reilly, R. C., T. S. Braver, and J. D. Cohen. 1999. `A biologically based computational model of working memory' in A. Miyake and P. Shah. Models of Working Memory: Mechanisms of Active Maintenance and Executive Control, Cambridge: Cambridge University Press. pp. 375±411. Perfetti, C. A., M. A. Marron, and P. W. Foltz. 1996. `Sources of comprehension failure: theoretical perspectives and case studies' in C. Cornoldi and J. Oakhill (eds): Reading Comprehension Difficulties: Processes and Intervention, Mahwah, NJ: Lawrence Erlbaum Associates. pp. 137±65. Royal, S. 1999. `La mutation des colleÁges pour tous et pour chacun: vers un colleÁge plus juste', ConfeÂrence de presse, Documents, DeÂbat national sur le colleÁge, http://hebergement. ac-poitiers.fr/college/etudes/confpres.htm Scarborough, D., L. Gerard, and C. Cortese. 1984. `Independence of lexical access in bilingual word recognition.' Journal of Verbal Learning and Verbal Behavior 23: 84± 99. Schmalhofer, F. and D. Glavanov. 1986. `Three components of understanding a programmer's manual: verbatim, propositional, and situational representations.' Journal of Memory and Language 25: 279±94. Schneider, W. 1999. `Working memory in a multilevel hybrid connectionist control architecture CAP2' in A. Miyake and P. Shah. Models of Working Memory: Mechanisms of Active Maintenance and Executive Control, Cambridge: Cambridge University Press. pp. 340±74. Shohamy, E. 1984. `Does the testing method make a difference? The case of reading comprehension.' Language Testing 1: 147±70. Snow, D. 1996. More than a Native Speaker: An introduction for volunteers teaching abroad. Alexandria, VA: TESOL, Inc. Steffensen, M. S., C. Joag-Dev, and R. C. Anderson. 1979. `A cross-cultural perspective on reading comprehension.' Reading Research Quarterly, XV/1: 10±29. Swanson, H. L. and V. Berninger. 1995. `The role of working memory in skilled and less CATHERINE WALTER skilled readers' comprehension.' Intelligence 21: 83±108. Taylor, L. B. 1996. An investigation of textremoved summary completion as a means of assessing reading comprehension ability. Unpublished doctoral dissertation, University of Cambridge. Teasdale, J. D. and P. J. Barnard. 1993. Affect, Cognition And Change: Re-modelling depressive thought, Hove, UK: Erlbaum. van Dijk, T. A. 1977. `Semantic macrostructures and knowledge frames in discourse comprehension' in M. A. Just and P. A. Carpenter (eds): Cognitive Processes in Comprehension, Hillsdale, NJ: Lawrence Erlbaum Associates. pp. 3±31 van Dijk, T. A. and W. Kintsch. 1983. Strategies of Discourse Comprehension. Academic Press: New York. Walter, C. 2000. The involvement of working memory in reading in a foreign language. View publication stats 339 Unpublished doctoral dissertation, University of Cambridge. Walter, C. and J. Williams. `A working memory measure for intermediate learners of a foreign language'. In preparation. Waters, G. S. and Caplan, D. 1996. `The measurement of verbal working memory capacity and its relationship to reading comprehension.' The Quarterly Journal of Experimental Psychology 49A (1): 51±79. Whitney, P. B., G. Ritchie, and M. B. Clark. 1991. `Working-memory capacity and the use of elaborative inferences in text comprehension.' Discourse processes 14: 33±145. Wilson, B. and A. D. Baddeley. 1988. `Semantic, episodic and autobiographical memory in a post-meningitic patient.' Brain and Cognition 8: 31±46. Yuill, N. and J. Oakhill. 1991. Children's Problems in Text Comprehension. Cambridge: Cambridge University Press.