This chapter was divided into three main sections, each of which presented the discussions relating to each research question.
In the discussion that follows, each of the three research questions was examined and discussed in further details. Also, the consistence and inconsistency of results in this study with previous studies were illustrated as well. Possible explanations and discussion for the results of this study was presented and developed based on review of relevant literature. The major findings, limitation and pedagogical implications of the present study’s findings for future research and practice were included in the last chapter.
The Influence of WM on Literal Comprehension
One of the goals of the current study was to understand whether L2 WM is significantly correlated to L2 literal RC. The first hypothesis predicted no significant correlation between L2 WM and L2 literal RC. This hypothesis was supported in this study.
Previous studies indicate that constructs of L2 literal comprehension and L2 WMC are different, which leads to the weak correlation between literal RC and WMC (Altepkin
& Ercetin, 2010; Miyake & Friedman, 1998). First of all, when making literal RC, L2 readers tend to focus on decoding the factual and surface information of the reading texts.
On the contrary, the function of WM is to offer adequate cognitive mental space for individuals to complete demanding cognitive tasks. Literal comprehension can be
regarded as readers’ first step when they become engaged in the linguistic messages, and perhaps they would make further relevant propositions from sentences in close proximity so as to formulate a coherent minimally semantic structure (King, 2007). According to Altepkin and Ercetin (2001), L2 readers are prone to generating text-based RC. Therefore, their comprehension is mostly devoted to understanding explicit textual information,
54
instead of implicit or hidden messages behind the linguistic clues. L2 literal RC is built by automatically recognizing information stated literally from texts and less demanding than making inferential RC. WM works as mental space that can temporarily store and process immediate information, which goes beyond the factual and surface understanding of texts.
The construct of literal RC partially corresponds to the information-decoding part of WM.
Therefore the discrepancy between WM and literal RC weakens the correlation between WM and literal RC. Therefore L2 literal RC has no significant correlation with WM.
Secondly, Daneman and Hannon (2007) indicated that WMC contributed little to the performance of RC in adults when the variable of higher-level processes, such as
inferential reading, was excluded from the participants’ overall RC. Namely, WMC accounts for little variation in literal RC because WMC is more related to highly demanding cognitive tasks rather than surface understanding as literal RC. Literature shows that L2 readers with higher WMC can make better reading inference and that WMC has minor influence on literal RC (Chiappe, Hasher & Siegel, 2000). According to Daneman and Hannon (2007), WMC contributes less to RC performance when inferential RC or higher-level process is taken out from comprehension as a whole. In other words, WMC has little relationship with lower-level word process or literal RC. Furthermore, as Miyake and Friedman (1998) indicated, L2 WMC can be regarded as a great determinant for making syntactic RC. As Miyake and Friedman (1998) observed, L2 WMC plays an important role when readers perform complicated and challenging tasks, such as syntactic RC or comprehension behind the sentences. In other words, L2 WMC would demonstrate its influence on L2 comprehension when the cognitive tasks are challenging to the mind and cause cognitive demands on L2 WMC.
When it comes to WMC and the degree of challenge readers have while reading, studies indicate that high-/low- level WMC differ in terms of degrees of cognitive load (Dennis & Barnes, 2001; Sweller, 1994). Higher WM enables readers to comprehend
55
automatically fluently, forming meaningful conceptualization of immediate messages in order to prevent them from overloading the working memory space with overwhelming information. Higher WMC eventually benefits readers in terms of inferential reading. In other words, higher WM readers are able to fluently comprehend the linguistic clues in the reading, and better at manipulating the current information. They could connect the texts at hand to his or her own long-term memory and background knowledge more smoothly and automatically than those with lower-level WMC.
In addition, high-WMC readers have better function of inhibition and also better control of their attention, which enables them to be better at focusing on the current tasks under the manipulation of central executive (Baddeley, 2012). Inhibition is the capability of suppressing irrelevant information while attending to the current task with clear goals.
Such features are quite similar to the cocktail party effect, which refers to an ability to ignore the irrelevant stimuli and effectively direct one’s own attention to necessary information regarding the goal of tasks at hand (Engle et al., 1999).
For L2 readers with lower level of WMC, it is difficult to retrieve information from the LTM rapidly with less automaticity and to temporarily store the current information from the reading. The reason is that L2 readers with lower level of WMC tend to maintain irrelevant information and thus consume their mental space with high percentages of information loss. It leads to less contribution to forming and manipulating information in the WM. Therefore, lower-WMC L2 readers might fail to surpass high-WMC readers in terms of inferential RC.
To conclude, in this study, WMC measured with a recall and a recognition reading tasks failed to demonstrate significant correlation with literal RC, which was consistent with the previous studies (Altepkin & Ercetin, 2010; Chiappe, Hasher & Siegel, 2000;
Miyake & Friedman, 1998). The result confirms the generally accepted view in the literature that L2 WM has no relationship with L2 literal comprehension.
56
The Influence of L2 WMC on L2 Inferential Comprehension The result of the current study did not detect any evidence for any significant correlation between L2 inferential RC and L2 WMC, whether L2 WMC was measured with a L2 recall-RST or with a L2 recognition-RST. Neither did the study done by Chun and Payne (2004). Considering the numerous studies supporting the positive influence of WMC on inferential RC (Altepkin & Ercetin, 2001; Anderson et al., 1996; Baddeley, 2012; Conway & Engle, 1994; Daneman & Hannon, 2007), the result of the present study should be considered tentative and the power of generalization should be taken with restrictions. However, the following discussion provides possible reasons based on theoretical opinions about WM and RC for the discrepancy from previous studies.
Previous studies have reported that higher-WMC L2 readers are able to make better reading inference than those with lower-WMC L2 readers (Alptekin & Ercetin, 2009, 2011; Anderson et al., 1996; Baddeley, 2012; Chiappe, Hasher & Siegel, 2000; Conway
& Engle, 1994; Pulido, 2009). Inferential comprehension refers to a deeper understanding of the content of the text in comparison with literal comprehension. Higher WMC enables readers to efficiently delegate information through central executive or to attentively focus on immediate tasks. Furthermore, readers with high WMC have better inhibition and can deal with tasks with concentration under the manipulation of central executive (Baddeley, 2012).
Also, with higher WMC, readers’ cognitive space for processing and storing new and old information would be adequate to fluently deal with demanding tasks, such as making reading inferences. Particularly, Koda (2005) indicated that the processing of L2 WMC demonstrates similar characteristics when individuals undertake high-order cognition tasks, which makes L2 WM highly correlated to L2 inferential RC in numerous of previous studies (Dennis & Barnes, 2001).
Nevertheless, according to Daneman and Carpenter, all individuals with high WMC
57
or low WMC, might have the same or similar capacity of WM. In other words, WMC is not a stable or fixed trait for individuals, like gender or nationality, and WMC might change along with time, place or under other situations. Those who make correct reading inferences efficiently in one reading text demonstrate high WMC. But they might
encounter reading difficulty and demonstrate lower WMC in another reading text. This happens from time to time when skilled readers are forced to read texts the content or the topic of which are unfamiliar to his or her background knowledge or irrelevant to his or her interest, their eye movement deteriorates. They might have poor or shallow
comprehension about the very text (Stanovich, 1986). Shallow comprehension means that skilled readers’ WMC space is overly consumed and occupied with literal information, and therefore weakens and deteriorates the WMC of skilled readers. Genres, domain knowledge, readers’ mental and physical situation or other factors would account for the change of WMC for one individual. To be specific, Daneman and Carpenter (1980) argued that individuals possess similar WMC and the significant difference in
comprehension mainly results from reading skills, not from the capacity of WM. Those with good reading skills consume less WMC and they have more cognitive space to process information than readers with inferior reading skills. Therefore, those with better reading skills have much more cognitive space and high capacity of WM. That is, how higher WMC is related to better both literal and inferential RC. From this perspective, having good reading skill is an essential variable when it comes to the relationship between WM and RC.
Like Daneman and Carpenter, Ericsson and Kintsch (1995) suggest that readers’
WMC changes along with immediate reading tasks. The changes of individuals’ WMC result from their background knowledge about the texts at hand (Kintsch & Ericsson, 1995). Kintsch (1995) suggests that individuals’ WMC might be expanded or consumed based on how much background knowledge readers have about the immediate reading
58
tasks. WM, a rather limited cognitive space, in other words, would be expanded with resources from LTM and there would be enough cognitive space for immediate cognitive tasks. L2 readers are likely to encounter reduced working memory spans when involved in L2 input, and this phenomenon of reduced WM results from the lack of adequate LTM contributions to L2 WMC (Brown & Hulme, 1992).Thus, long term WM could be defined “as the part of the LTM that can be accessed rapidly and reliably by means of retrieval cues in WM” (Ericsson & Kintsch, 1995).
From this perspective, the predictive power of WMC on cognitive tasks depends on at least two factors: one is the short-term working interface of storing and processing information and the other is the resources from individuals’ background knowledge of LTM. While WM stores and processes incoming information through sensory stimuli, from reading texts for one thing, it mediates among perception, LTM and actions. And resources from LTM has great influence on how well the immediate information is processed (Baddeley, 2012). In other words, WM varies with individuals’ background knowledge about texts and also with the difficulty level they encounter in every different text. Readers’ resources in LTM-working memory reduce the consumption of readers’
WMC and more cognitive space is saved to deal with information. Therefore, readers’
background knowledge and reading skills are strong factors when it comes to the influence of WM on RC. Also according to Hannon (2012), L2 WMC sometimes has indirect influence on L2 inferential comprehension. L2 WMC serves as an efficient
workspace for readers to combine and formulate comprehensive understanding toward the reading texts; however, lack of other important resources, such as background knowledge or trained reading speed, would severely deteriorate the comprehension performance, especially when individuals deal with demanding comprehension tasks.
Another possible explanation for the inconsistent finding regarding L2 WMC and L2 inferential RC might result from the difference between the characteristics of L1 and L2
59
readers. Alptekin (2004) suggests that L2 readers tend to perform similarly like
insufficient L1 readers for L2 readers usually build up their comprehension literally. That is, it is not the first instinct or habit for L2 readers to infer beyond the linguistic
information during the reading process. It takes guided attention or trainings of reading strategies to assist L2 readers to make correct reading inference. In this study, as the descriptive analysis showed, the participants found the reading texts difficult to
understand literally, which made it even harder to make correct reading inferences beyond the texts. It could be imagined that in this study readers only stored linguistic information, leaving short of deeper understanding. Therefore, the participants’ L2 inferential
comprehension performances were similar to their L2 literal comprehension performance.
L2 readers tend to have “shallow textual representation" of the reading, and therefore they tend to make false reading inferences (Alptekin & Ercetin, 2011, p.237). When it comes to L2 literal comprehension, L2 WMC has no significant influence on L2 literal
comprehension (Altepkin & Ercetin, 2010; Miyake & Friedman, 1998). In other words, the results of L2 inferential comprehension could not fully represent the participants’
ability to generate correct inferential comprehension. The result might have shown a strong correlation between L2 WMC and L2 inferential RC if the participants had less reading difficulty. This flaw of research design, such as reading difficulty of the given RC test, was elaborated in the limitation section of the next chapter. Relevant factors resulting from the research design for the result about L2 WMC and L2 inferential RC in this study was further included in the following.
RST Types and WMC
Finally, the result showed that the recall-WMC and recognition-WMC had no relationship with RC. It was assumed that L2 recall-WMC had a stronger relationship with L2 inferential comprehension than L2 recognition-WMC does. First, literature shows that L2 WMC is more related to L2 inferential comprehension (Alptekin & Ercetin, 2011;
60
Kane et al., 2001; Nassaji, 2002; Unsworth & Engle, 2007). Secondly, recall-RSTs are better to evaluate and represent L2 WMC than recognition-RSTs do (Unsworth & Engle, 2007). Therefore, it was hypothesized that L2 WMC has positive correlation with L2 inferential RC when WM is measured with a L2 recall-RST, rather than with a L2 recognition-RST.
With regard to constructs behind these two RSTs, first of all, recall-RST requires participants to do dual-task simultaneously—judging the grammaticality of presented sentences and memorization of ending words. In this case, participants do what they do to process and store information in WMC—to store and process information at the same time. On the contrary, when taking the other kind of WMC RST, the recognition-RST, participants judge sentence grammaticality while reading sentences and recognize ending words from a list of provided words. To complete a recognition-RST, participants,
recognizing ending words instead of memorizing and spelling them, simply store the immediate information without the process function in recall-RST tasks. Making reading inferences and completing the recall-RST both require participants to do the similar cognitive tasks—to temporarily store information and process, linking new information to knowledge withdrawn from LTM. Hence, some researchers have considered the influence of WM on RC is strong when WMC is measured with a recall-RST (Altepkin & Ercetin, 2009; Unsworth & Engle, 2007).
Apart from the original hypotheses and expectations, the regression analysis of the current study showed that L2 WMC, measured with a recall-RST or with a
recognition-RST, did not show significant power on predicting the performance of overall L2 RC (R2=1.7%). Although the results of regression analysis showed higher
accountability of L2 recall-WMC than that of L2 recognition-WMC, the accountability power was still not statistically significant.
Previous studies concerned with RST types and the influence of WMC have different
61
opinions.
First, WMC measured with both a recall-RST and a recognition-RST demanded similar cognitive loads on participants. Due to the lack of a significant correlation between WMC and RC in this study, both recall-WMC and recognition-WMC were not significantly correlated to RC. According to Turner and Engle (1989), RST of all kinds could tap the same variation in reading comprehension regardless of WMC tools (a recall-RST and a recognition-RST). The results showed that the WMC of these participants in this study did not have strong correlations either with their literal or inferential RC. If both the recall-RST and recognition-RST could tap the same variation in RC when WMC was not highly related to RC, it was reasonable that the recall-WMC and recognition-WMC had no relationship with RC.
However, some researchers with different opinions from Turner and Engle suggest that different constructs are assessed in different RSTs (Alptekin & Ercetin, 2009;
Unsworth & Engle, 2007). The recognition-RST measures the participants’ abilities to recognize the ending words from the provided options or “externally presented retrieval cues” (Unsworth & Engle, 2007, p.112). The recognition of externally presented cues might only activate short-term memory.
To complete a recall-RST, participants recall the stored item, the ending words, from their internally generated cues. It requires the long-term memory of participants to
memorize or link all the ending words. Though differences exist in terms of cognition demands in both RSTs, others still think both a recall-RST and a recognition-RST require knowledge from LTM (Alptekin & Ercetin, 2009). It seems that recognizing ending words from the provided list does not activate LTM knowledge. But it would be difficult and impossible for the participants to temporarily keep the ending words in mind for the later test of word recognition without knowing the words in the first place. Thus, the operation of both a recall-RST and a recognition-RST consumes and requires LTM
62
knowledge to certain extent. One of the research questions was concerned with whether or not these two RSTs demonstrate difference in assessing the influence of WM on RC.
Judging from the result, still more studies are needed in order to understand the distinction between these two RSTs or among other RSTs.
Even though it was uncertain whether different RSTs assess different or the same cognitive variation in terms of WMC operation, the non-significant correlation might not result from this uncertainty. While participants were taking the RST, they stored and processed information under clear guidance and fixed time periods. Their attention was focused on comprehending the sentences presented at presence while they judged the grammaticality and memorized the ending words at the same time. To complete the RST, participants had a simple mission and a clear goal. To complete the recall-RST and the recognition-RST, special reading skills or background knowledge about the sentences were not needed in order to fully comprehend the sentences since the 42 sentences were all simple narrative sentences. RSTs are designed to examine participants’ abilities to comprehend texts and process information, not to measure their size or scope of memory (Baddeley, 2002). To complete the later test of RC, they deduced or inferred explicit and implicit information. Their RC performance involved not only their WM ability to store and process information but also their reading skills and background knowledge. When participants took the RC test and tried to comprehend the given text, they acted
independently without clear guidance to seek out the mission goals and to control their attention. WM, as suggested to have great influence on RC, supports readers in
processing and storing information. But more variables are essential for making correct and deeper understanding about texts beside the variable of WM.
processing and storing information. But more variables are essential for making correct and deeper understanding about texts beside the variable of WM.