Working memory

There are a variety of definitions of WM (Miyake & Shah, Reference Miyake and Shah1999). One simple definition, offered via an example by Ortega (Reference Ortega2009), is that WM holds information (a storage function) and integrates new information with known information already encoded in long-term memory (LTM), which is a processing function. In contrast to LTM, which has a relatively unlimited capacity, WM has limited capacity, and there is evidence for individual differences in this capacity (Just & Carpenter, Reference Just and Carpenter1992).

Whether the WM mechanisms implicated in text comprehension are domain general or language specific has also been controversial (Miyake & Shah, Reference Miyake and Shah1999; Seigneuric et al., Reference Seigneuric, Ehrlich, Oakhill and Yuill2000). For example, Miyake and Friedman (Reference Miyake, Friedman, Healy and Bourne1998) stated that “WM resources implicated in adult language processing may be specific to the language and be separated from the central resources that support other nonverbal thinking processes” (p. 342). Baddeley and Logie (Reference Baddeley, Logie, Miyake and Shah1999) also argue that the effect of WM on various cognitive activities is domain specific and even that the central executive may not be a completely unitary construct. However, based on the assumption that WM capacity is distinguishable from a storage function, Engle et al. (Reference Engle, Tuholski, Laughlin and Conway1999, p. 104) defined WM as “a system consisting of those LTM traces active above the threshold, the procedure and skills necessary to achieve and maintain that activation, and limited-capacity, controlled attention.” Engle and colleagues emphasized the unitary nature of controlled attention in WM. The controlled attention theory of WM (Engle & Kane, Reference Engle, Kane and Ross2004; Kane et al., Reference Kane, Bleckley, Conway and Engle2001) suggested that there is a domain-general component of WM responsible for guiding attention and that domain-specific components of WM are responsible for temporarily retaining information. The domain-general, controlled-attention ability is related to both higher-level cognition and lower-level cognition requiring cognitive control (Colflesh & Conway, Reference Colflesh and Conway2007). According to this theory, individuals with higher WM capacity are better able to control attention to resist interference during encoding and retrieval than individuals with lower WM capacity. Kane et al. (Reference Kane, Bleckley, Conway and Engle2001) showed that low-span participants were slower and less accurate than high-span participants in attention control. Kane and Engle (Reference Kane and Engle2003) also showed that there was a significant difference in Stroop task performance between high-span participants and low-span participants only when the task demanded more attention to maintain the task goal. Colflesh and Conway (Reference Colflesh and Conway2007) further investigated the relationship between WM capacity and controlled attention using the dichotic listening task. The results of the study suggested that individuals with greater WM capacity outperformed those with lower WM capacity not only in selective attention but also in divided attention. Colflesh and Conway concluded that WM capacity significantly affected listening task performance only when the tasks placed greater demands on executive attention. In view of these results, the individual differences in WM capacity are arguably associated with the ability to focus attention and avoid distraction to meet demanding task goals.

The distinction made between WM capacity and short-term memory (STM) capacity has some parallels with the distinction made between the central executive system and storage systems such as the phonological loop (Swanston et al., Reference Swanston, Howard and Sáez2011). Conway et al. (Reference Conway, Cowan, Bunting, Therriault and Minkoff2002) claimed that STM capacity, as a simple storage buffer, is determined by practiced skills and strategies, such as rehearsal and chunking, which are domain-specific resources. In contrast, they argue that WM capacity, as a general resource, contributes to cognitive performance in any domain that demands controlled processing. In the present study, WM will refer to Engle et al.’s notion of controlled attention, reflecting the properties of the central executive component of Baddeley’s model.

Effects of language specificity and proficiency on WM

The findings of previous research suggested that the cognitive resources underlying WM capacity are shared by L1 and L2 and that the relationship is language independent (Osaka & Osaka, Reference Osaka and Osaka1992; Osaka et al., Reference Osaka, Osaka and Groner1993). L2 processing may share the same pool of WM resources as L1 processing, at least in bilingual-level L2 users (Miyake & Friedman, Reference Miyake, Friedman, Healy and Bourne1998). In this regard, Miyake et al. (Reference Miyake, Friedman, Osaka, Healy and Bourne1998) examined the influence of L1 and L2 WM capacity as measured by L1 and L2 listening span tests (LSTs) on L2 aural syntactic parsing and showed that the measured L1 WM capacity may be a mediator variable directly related to L2 WM capacity but indirectly related to L2 aural syntactic parsing.

While L1 WM capacity directly influences L1 comprehension independent of L1 linguistic knowledge (Jincho et al., Reference Jincho, Namiki and Mazuka2008; Seigneuric et al., Reference Seigneuric, Ehrlich, Oakhill and Yuill2000; Seigneuric & Ehrlich, Reference Seigneuric and Ehrlich2005), WM capacity measured in the L2 is thought to be difficult to disentangle from L2 proficiency because less proficient learners’ decoding and syntactic processing in the L2 is less automatic than in the L1 (Chun & Payne, Reference Chun and Payne2004). For less proficient L2 users, a particular task may involve greater task demand than that for skilled L2 users (Sagarra, Reference Sagarra and Han2008). For example, Ikeno (Reference Ikeno2006) found a lack of significant partial correlation between L2 WM capacity as measured by the reading span test (RST) and L2 reading comprehension after partialling out the effects of L2 processing efficiency, whereas Harrington and Sawyer (Reference Harrington and Sawyer1992) found that the correlation between L2 WM capacity and L2 comprehension remained significant even after the effects of L2 vocabulary and grammatical knowledge were removed. Ikeno argued that the resource pool for L1 processing and L2 processing could be functionally different, at least at the L2 proficiency level of the examined participants, and lexical-semantic and syntactic parsing processing efficiency drives the relationship between L2 WM capacity and L2 reading comprehension. Alptekin et al. (Reference Alptekin, Erçetin and Özemir2014) also suggested that the processing component of WM as measured by L1 (Turkish) and L2 (English) RSTs tended to yield language-sensitive results for late L2 learners. In contrast, Çeçen and Erçetin (Reference Çeçen, Erçetin, Granena, Jackson and Yilmaz2016) suggested that the variance shared by WM capacity and L2 reading comprehension is due to the general processing efficiency of verbal WM capacity as measured by an L2 RST and the operation span test for adult English learners with a B2 level of proficiency. The interpretation of domain-general factors shared by verbal WM and other WM systems, such as visuospatial WM, has been controversial (Shipstead & Yonehiro, Reference Shipstead and Yonehiro2016); however, Shipstead and Yonehiro interpret the shared processing component as central executive attention organizing information between the modalities. More evidence will be required to investigate what drives the significant relationship between WM capacity and L2 listening comprehension across different proficiency levels.

Another important issue is whether the relationship between WM capacity and L2 listening comprehension is mediated by L2 proficiency. Empirical studies by Adams and Shahnazari-Dorcheh (Reference Adams and Shahnazari-Dorcheh2014) and Walter (Reference Walter2004) revealed that WM capacity measured by an L2 RST could well predict L2 reading ability at the lower-intermediate L2 proficiency level. Walter (Reference Walter2004) suggested that the lower-intermediate French EFL participants in the study had failed to transfer the ability to build well-structured mental representations of text and that this ability was linked to L2 WM capacity. The nature of the link between WM capacity and comprehension of L2 text is thought to be influenced by the learners’ second language ability. Thus, it needs to be investigated whether WM capacity gives rise to different patterns of influence on L2 listening comprehension when proficiency levels are less than advanced.

Working memory and L2 listening comprehension

Information processing theories distinguish between automatic processing and controlled processing. Automatic processing requires few cognitive resources, and separate processing threads can run in parallel. In contrast, controlled processing requires more cognitive resources than automatic processing. Controlled processing, handled by the central executive, cannot operate in parallel (Ortega, Reference Ortega2009). From such an information-processing point of view, listening comprehension is also subject to a trade-off between the storage and processing functions of WM. Given the real-time nature of spoken language and the inability of WM to process all the information under timed conditions, less proficient listeners need to focus more strongly and consciously on what they are listening to (Field, Reference Field2008). When processing spoken language requires conscious attention due to storage and processing needs, comprehension breakdowns occur (Buck, Reference Buck2001; Vandergrift & Goh, Reference Vandergrift and Goh2012), and some partial results from WM processing may be forgotten (Just & Carpenter, Reference Just and Carpenter1992).

Some empirical studies (Andringa et al., Reference Andringa, Olsthoorn, van Beuningen, Shoonen and Hulstin2012: Brunfaut & Revesz, Reference Brunfaut and Revesz2015; Vandergrift & Baker, 2015; Reference Vandergrift and Baker2018; Wallace, Reference Wallace2020) have confirmed a very low association between L2 listening and WM capacity as measured by forward and backward digit/word span tasks or non-word recall tasks. For example, Andringa et al. (Reference Andringa, Olsthoorn, van Beuningen, Shoonen and Hulstin2012) investigated the effect of L2 linguistic knowledge and processing skills, intelligence, and working memory on listening ability in a group of 121 native and 113 non-native speakers of Dutch. The results of the study showed that the correlation of memory variables with listening ability was weak when all the other latent variables were partialled out. Wallace’s (Reference Wallace2020) study also showed that memory variables had no direct or indirect effect on L2 listening comprehension when they were examined in cooperation with other variables. These findings are in line with the current language proficiency model of Hulstijin (Reference Hulstijn2015, Reference Hulstijn2019), who argues that linguistic knowledge (e.g., vocabulary, syntax, and phonology) variables are thought to be more important for listening performance than non-linguistic variables (e.g., WM, intelligence, metacognitive strategy use, and world knowledge). Hulstijin suggests that peripheral variables such as WM are thought to exert some influence on L2 listening performance as proficiency increases but are considerably less predictive than linguistic knowledge and processing variables.

On the other hand, Vafaee and Suzuki (Reference Vafaee and Suzuki2020) argued that previous studies have failed to capture a significant role of WM capacity in L2 listening due to the use of only simple WM tasks and simple listening comprehension tasks. Vafaee and Suzuki (Reference Vafaee and Suzuki2020) examined the effect of WM on L2 listening by conducting two spatial WM tasks as complex measures that engage both the processing and storage of spatial information to avoid confounding WM with L2 proficiency. The results showed that WM capacity was a significant predictor of L2 listening ability (IELT test scores) with 263 Iranian EFL learners. The differences in types of WM tasks and L2 listening ability measures may affect the relationship between WM and L2 listening comprehension. In this regard, Masrai (Reference Masrai2020) examined the role of L2 WM as measured by an L2 LST in L2 listening comprehension with 130 EFL learners from different L1 backgrounds. The study found a strong correlation (r = .64) between L2 WM capacity and L2 listening ability. Masrai (Reference Masrai2020) argues that the LST is a domain-specific measure of WM capacity and engages processes closely related to language comprehension. Regarding the effects of differences in the types of L2 listening comprehension tasks and measures on the role of WM, Brunfaut and Revesz (Reference Brunfaut and Revesz2015) showed that WM capacity was more strongly associated with the scores of the listening tasks requiring decoding and maintaining detailed information than those mainly targeting global listening and overall listening proficiency. Further investigation is needed to clarify the nature of the links between WM capacity as measured by complex WM tasks and L2 listening comprehension.

Learner variables in L2 listening comprehension

Listening is a complex activity that involves physiological and cognitive processing at different levels (Field, Reference Field2008; Vandergrift, Reference Vandergrift2007). Integrating all of these processes involves a great deal of cognitive activity on the part of the listener (Vandergrift, Reference Vandergrift2007). Listening is also affected by individual differences in linguistic and non-linguistic knowledge (Buck, Reference Buck2001). The listener applies different types of linguistic and non-linguistic knowledge through top-down and bottom-up processes. Cognitive processing in L2 listening varies depending on the learners’ language proficiency level (Rubin, Reference Rubin1994).

Quite a few empirical studies have shown that L2 vocabulary knowledge is associated with L2 listening comprehension (Mecartty, Reference Mecartty2000; Stæhr, Reference Stӕhr2009; Van Zeeland and Schmitt, Reference Van Zeeland and Schmitt2013; Vandergrift and Baker, Reference Vandergrift and Baker2015, Reference Vandergrift and Baker2018). For example, Stæhr (Reference Stӕhr2009) showed significant correlations between L2 listening comprehension and the two dimensions of depth and breadth of vocabulary knowledge with advanced Danish learners of English. Noreillie et al. (Reference Noreillie, Kestemont, Heylen, Desmet and Peters2018) replicated Stæhr’s (Reference Stӕhr2009) study by focusing on the intermediate proficiency level and found a significant correlation between vocabulary knowledge and L2 listening comprehension. The study also suggested that learners need to know fewer words (1000–2000) for adequate listening comprehension at an intermediate proficiency level than advanced learners. Milton, Wade, and Hopkins (Reference Milton, Wade, Hopkins, Chacón-Beltrán, Abello-Contesse and Torreblanca-López2010) also examined the effect of phonological and orthographic vocabulary size on L2 listening comprehension. The results showed that phonological vocabulary size was more strongly associated with L2 listening than orthographic vocabulary size. Milton, Wade, and Hopkins emphasized the relevance of aural vocabulary measures to be used with L2 listening comprehension.

Some researchers have suggested that automatized linguistic decoding skills are a prerequisite to improving listening comprehension (Field, Reference Field2008, Vandergrift & Goh, Reference Vandergrift and Goh2012). For example, Lynch (Reference Lynch1998) suggested that phonological modification features such as reduction and assimilation in spoken English may also influence the listener’s performance of chunking and interpreting connected speech in perceptual processing. Other researchers have also emphasized the importance of lexical segmentation for successful listening comprehension (Brown, Reference Brown1990; Buck, Reference Buck2001; Norris, Reference Norris1993). Satori (Reference Satori2010) investigated the effects of reduced forms on listening comprehension with Japanese EFL learners. Participants completed the TOEIC listening test, and a partial dictation test focused on phonological modification features such as reduction, contraction, liaison, elision, deletion, assimilation, and coalescence assimilation. The correlation between the TOEIC listening scores and the dictation results was highly significant. The results suggested that phonological modification features in spoken English may affect Japanese students’ performance in the perceptual processing of L2 listening comprehension.

Previous studies have also shown that prosodic awareness has a positive effect on foreign language learners’ listening comprehension (Derwing et al., Reference Derwing, Munro and Wiebe1998; Kang, Reference Kang2014; Ma, Reference Ma2007). Buck (Reference Buck2001) stated that prosodic features are important characteristics in spoken English and are closely related to comprehension that is unique to listening and “how this works varies from language to language” (p. 35). In English sentences, for example, content words are generally more strongly stressed than function words, which should be reduced in duration, whereas Japanese is a mora-timed, pitch-accented language. Field (Reference Field2008) stated that focal stress contributes to foreground the most important piece of spoken text by providing a basis for forming hypotheses about the text when L2 listeners with limited decoding skills understand very little else. Beginner Japanese EFL learners without the ability to identify words in stressed and unstressed situations possibly fail to utilize the stressed words in spoken English as an important tip in the process of perceiving the message about what they listen to.

With regard to the relationship between L2 syntactic knowledge and L2 listening, Vafaee and Suzuki (Reference Vafaee and Suzuki2020) showed that L2 syntactic knowledge contributed to L2 listening comprehension independent of the other pertinent factors (L2 vocabulary, WM capacity, metacognitive knowledge, etc.). The relative significance of L2 syntactic knowledge in L2 listening was also confirmed by Joyce’s (Reference Joyce2008) multivariate approach that provided a balanced perspective on the impact of L2 linguistic knowledge and processing variables on L2 listening with 443 Japanese university-level EFL learners. The results indicated that L2 syntactic knowledge made the greatest independent contribution to L2 listening comprehension. L2 lexical segmentation, L2 sentence stress awareness, and metacognitive listening strategy were also statistically significant but weaker predictors. Joyce (Reference Joyce2008) suggested that less proficient listeners are highly reliant on the linguistic and psycholinguistic sub-skills closest to the surface of the message. Joyce’s study focused only on the storage aspect of WM, and the results showed that neither L1 nor L2 WM storage capacity predicted L2 listening comprehension. In light of the importance of the processing aspect of WM, the present study partially replicates this study by reflecting the effects of L1 and L2 WM measured with L1 and L2 LSTs on L2 linguistic knowledge and processing skills as well as L2 listening comprehension.

The current study

Most of the previous studies have examined the role of WM in L2 comprehension with proficient L2 users and overlooked the effect of language proficiency on the relationship between WM capacity and L2 comprehension. The research inevitably provides a limited view of the influence of WM capacity on L2 reading and listening. The present study, therefore, sought to clarify whether WM capacity influences L2 listening comprehension independent of L2 linguistic knowledge and processing skills and whether these effects operate differently across different L2 proficiency levels. To focus on the role of WM in the lexical decoding and syntactic parsing processing involved in L2 listening comprehension for less proficient listeners, the present study included a total of four explanatory L2 processing skill variables. These four sub-skills were L2 lexical segmentation, L2 syntactic knowledge, L2 vocabulary breadth, and L2 sentence stress awareness. The rationale for choosing these four variables is that they are thought to tap surface-level or text-based processing involved in L2 listening and were also statistically significant predictors of L2 listening comprehension at an intermediate proficiency level in the Japanese EFL environment in Joyce’s (Reference Joyce2008) study. Walter (Reference Walter2004) suggested that the relationship between WM capacity and comprehension of L2 text may be influenced by learners’ L1 backgrounds and their L2 proficiency, such as syntactic parsing skills. The participants in the present study were also university-level Japanese EFL learners who ranged from lower-intermediate to mid-intermediate proficiency levels. In this regard, these four variables may represent the possible L2 linguistic knowledge and processing skills that place a heavy cognitive load on WM capacity in less-skilled listeners. Thus, it may be worth investigating the relationship between WM capacity and these four variables. These variables affect L2 listening performance by interacting with each other at the different levels of processing involved in L2 listening. Lexical segmentation and sentence stress awareness tap metalinguistic processing and play a role in perceptual processing encoding of the acoustic signal. At this stage, listeners segment the continuous speech stream into phonemes and retain the aural input in the echoic memory. Vocabulary breadth and syntactic knowledge measures reflect knowledge of linguistic forms and mainly play a role in parsing. At this stage, the words recognized at the perceptual stage are converted into a mental representation of the combined message, while an utterance is segmented with the help of syntactic structures and cues to meaning. WM capacity is assumed to be involved with each L2 processing skill differently according to proficiency level. It would be contributive to examine the relationship between WM capacity and L2 processing sub-skills and the extent to which WM capacity accounts for unique variance in L2 listening comprehension after removing its variance common to these L2 sub-skill variables.

While the controlled attention component identified by Engle and colleagues (2004) is acknowledged to be language independent (Osaka & Osaka, Reference Osaka and Osaka1992; Osaka et al., Reference Osaka, Osaka and Groner1993), Chun and Payne (Reference Chun and Payne2004) argued that using a WM task in L2 makes it difficult to disentangle WM capacity as an independent resource from L2 proficiency, especially in the case of less proficient listeners. Sagarra (Reference Sagarra and Han2008) also argued that L2 WM tasks might measure a part of L2 proficiency, especially in the case of less proficient learners with limited L2 linguistic knowledge. To address this issue, the present study sought to investigate the role of both L1 and L2 WM capacity in L2 listening comprehension. The following research questions were posed.

• RQ1: Do the effects of L1 and L2 WM capacity on L2 listening comprehension and L2 linguistic knowledge and processing skills operate differently across different L2 proficiency levels?

• RQ2: Are the effects of WM systems on L2 listening comprehension independent of language or domain specific for language?

Participants

The participants in this study were 150 Japanese first- and second-year students from a technical college in Japan. The majority were 19–22 years of age and homogeneous in their social and educational background. The participants had completed 6 years of compulsory English language education before entering the school. Their major was the English language, and they received on average 9 hr of English language instruction per week. Their level of English proficiency ranged from 205 to 970 on the TOEIC test (average = 444.6, SD = 149.8). Based on their performance on the TOEIC listening test, the participants were then divided into three proficiency levels by the same grouping technique as Joyce’s (Reference Joyce2008) study: higher (50 students who scored 300 and above), middle (51 students who scored between 230 and 295), and lower (49 students who scored 225 and below). Based on the TOEIC-CEFR level mapping provided by the Educational Testing Service (ETS), 30% of participants in the higher-proficiency group were at the CEFR B2 level, and the remaining 70% of participants were at the CEFR B1 level. On the other hand, all participants in the lower-proficiency group were at the CEFR A2 level.

Instruments

Procedure

The research instruments were administered in the week following the TOEIC and split across two lessons to reduce fatigue. The data were collected at a Japanese technical college by the researcher herself. Most of the paper-based tests were group-administered as part of English language classes. Along with the participants’ identification numbers, their responses to the paper-based test items were manually entered into spreadsheet files after the data had been fully compiled and prepared using Excel.

Analyses

After the initial preparation, the statistical analyses were performed using SPSS for Windows Version 19. The present study focused on the effects of WM capacity on L2 listening comprehension and L2 linguistic knowledge and processing skills across different L2 proficiency levels (RQ1) and the effects of language specificity and proficiency on the relationship between WM and L2 listening comprehension (RQ2). The analytical procedures for addressing these questions included Pearson’s correlation analysis, t-test analysis, internal consistency analysis, and multiple regression analysis, which are established statistical methodologies for these types of RQs.

Results

The descriptive statistics are presented in Table 1. TOEIC listening test forms are designed to measure the listening comprehension sub-skills identified in four sections. Sections 1 and 3 are associated with comprehension in shorter passages. Sections 2 and 4 are associated with comprehension in longer passages. Sections 1 and 2 are thought to reflect global comprehension in L2 listening, and sections 3 and 4 reflect literal comprehension in L2 listening. As seen in Table 1, sections 2 and 4 (comprehension in the longer passages) had the lowest percentages of correct answers. On the other hand, sections 1 and 3 (comprehension in the shorter passages) had the highest percentages of correct answers.

Table 1.

Descriptive statistics for scores on the variables

The correlations between the variables (see Table 2) indicate that scores for L1 and L2 WM capacity yielded significant correlations with scores for the TOEIC listening test performance and L2 linguistic knowledge.

Table 2.

Summary of correlations between the variables (n = 150)

Note. *p < .05, **p < .01, ***p < .001

An important aspect of this study concerns the influence of L2 proficiency on the role of WM. To address this issue, based on their performance on the TOEIC listening test, the participants were then divided into three proficiency levels: higher (50 students who scored 300 and above), middle (51 students who scored between 230 and 295), and lower (49 students who scored 225 and below). In the present study, the higher group and the lower group were compared. Based on the TOEIC-CEFR level mapping provided by ETS, 30% of participants in the higher-proficiency group were at the CEFR B2 level, and the remaining 70% of participants were at the CEFR B1 level. On the other hand, all participants in the lower-proficiency group were at the CEFR A2 level. Table 3 shows the descriptive statistics for the higher-proficiency group and the lower-proficiency group. The results of t-tests showed significant differences in the mean scores of all the TOEIC listening test performances, L2 linguistic knowledge, and LSTs between the two groups. There were no significant differences in the mean DST scores between the two groups.

Table 3.

Descriptive statistics for scores on the variables in each proficiency group

Note. **p < .01, ***p < .001

The correlations between the variables in each proficiency group are shown in Tables 4 and 5, respectively. There were some marked differences between the two sets of correlations. Most notably, the correlations between scores for L1 and L2 WM capacity and L2 listening performance in the lower-proficiency group were greater than the corresponding correlations in the higher-proficiency group. Similar patterns emerged for L2 lexical segmentation, L2 sentence stress awareness, and L2 syntactic knowledge.

Table 4.

Summary of correlations between the variables in the higher-proficiency group (n = 50)

Note. *p < .05, **p < .01, ***p < .001

Table 5.

Summary of correlations between the variables in the lower-proficiency group (n = 49)

Note. *p < .05, **p < .01, ***p < .001

To identify the memory variables predictive of L2 linguistic knowledge in each proficiency group, multiple regression analysis was performed with scores for L2 linguistic knowledge as the dependent variables and scores for LSTs and DSTs as the independent variables in each proficiency group. Regarding the higher-proficiency group, neither LST nor DST scores were significant contributors to L2 listening comprehension and L2 linguistic knowledge. On the other hand, regarding the lower-level group, L1 LST scores accounted for 9.4% of the variance in scores for L2 sentence stress awareness. L2 LST scores accounted for 9.6% of the variance in scores for L2 syntactic knowledge. L2 DST and L2 LST scores accounted for 38.1% of the variance in scores for L2 lexical segmentation, as shown in Table 6.

Table 6.

Summary of regression analysis in the lower-proficiency group (n = 49)

Note. *p < .05, **p < .01, ***p < .001

Next, multiple regression analysis was performed with scores for the TOEIC listening test performance as the dependent variables and LST scores and L2 linguistic knowledge scores as the independent variables in each group. In the higher-proficiency group, the regression results are presented in Table 7. In the lower-proficiency group, for further investigation of the role of WM in L2 listening comprehension after adjusting for the effects of L2 linguistic knowledge, a hierarchical regression analysis was performed. L2 linguistic knowledge variables were entered into the model in Step 1, and L1 and L2 WM capacity were entered in Step 2 using the stepwise method. The results are presented in Tables 8, 9, 10, 11, and 12 for the lower-proficiency group. Regarding the higher-proficiency listeners, L2 syntactic knowledge, L2 vocabulary knowledge, and L2 lexical segmentation were the significant predictors of L2 listening performance. On the other hand, L1 WM capacity was a significant predictor of TOEIC listening test scores, listening performance in sections 3 and 4 (literal comprehension), and listening performance in sections 1 and 2 (comprehension in the shorter passages), as shown in Tables 8, 10, and 11, respectively, for the lower-proficiency listeners.

Table 7.

Summary of regression analysis in the higher-proficiency group (n = 50)

Note. *p < .05, **p < .01, ***p < .001

Table 8.

Summary of hierarchical regression in the lower-proficiency group (n = 49): TOEIC listening test

Note. *p < .05, **p < .01

Table 9.

Summary of hierarchical regression in the lower-proficiency group (n = 49): Sections 1 and 2

Note. **p < .01

Table 10.

Summary of hierarchical regression in the lower-proficiency group (n = 49): Sections 3 and 4

Note. *p < .05, **p < .01

Table 11.

Summary of hierarchical regression in the lower-proficiency group (n = 49): Sections 1 and 3

Note. **p < .01

Table 12.

Summary of hierarchical regression in the lower-proficiency group (n = 49): Sections 2 and 4

Note. *p < .05, **p < .01