The Effects of Task Complexity for Text Summarization by Korean Adult EFL Learners
- Author: Lee Haemoon, Park Heesoo
- Publish: The Journal of English Language and Literature Volume 57, Issue6, p911~938, Dec 2011
The present study examined the effect of two variables of task complexity, reasoning demand and time pressure, each from the resourcedirecting and resource-dispersing dimension in Robinson’s (2001) framework of task classification. Reasoning demand was operationalized as the two types of texts to read and summarize, expository and argumentative. Time pressure was operationalized as the two modes of performance, oral and written. Six university students summarized the two types of text orally and twenty four students from the same school summarized them in the written form. Results from t test and ANCOVA showed that in the oral mode, reasoning demand tends to heighten the complexity of the language used in the summary in competition with accuracy but such an effect disappeared in the written mode. It was interpreted that the degree of time pressure is not the only difference between the oral and written modes but that the two modes may be fundamentally different cognitive tasks, and that Robinson’s (2001) and Skehan’s (1998) models were differentially supported by the oral mode of tasks but not by the written mode of the tasks.
task-based language teaching (TBLT) , cognition hypothesis , limited cognitive capacity , complexity , accuracy , summarizing , resourcedirecting , resource-dispersing
In Skehan’s (1998) model of how language use influences adults’ second language acquisition, the nature of the tasks that prompts the learners’ language use is critical in channelling the learners’ attention to the language form. However, since he proposed his model, the nature of tasks has hardly been identified and classified conclusively, though researchers have proposed various standards for classification (Berwick 1993; Brown et al. 1984; Candlin 1987; Prabhu 1987). The most systematic and full-fledged classifications upto present are considered as those of Skehan (1998) and Robinson (2001, 2007).
Both of them proposed that tasks could be complex or simple in a few different dimensions and that certain tasks require the learners a more complex language form inherently such that they push the learners to try the upper boundary of or beyond their interlanguage system, driving the learners’ language acquisition. However, in spite of its theoretical importance, the task-inherent properties that facilitate learners’ language acquisition are yet unclear with mixed empirical findings Prabhu (Kuiken&Vedder2007;Revesz2009). Therefore, the classification of tasks, particularly on the dimension of task complexity that pushes learners’ IL system, is at the heart of controversy on task-based language teaching and learning research.
The present study aims at partly resolving this problem by examining the two variables of summarizing tasks which may contribute to task complexity and second language acquisition: the mode of language use (oral and written) as an operationalization of time pressure and the type of texts to be processed (expository and argumentative) as an operationalization of reasoning demand.
Skehan (1998) and Robinson (2001, 2007) are similar and different in classifying tasks. They both tried to distinguish between the factor that channels the learners’ attention to the complex language form directly due to the content of the task and the factor that helps or hinders such channelling effect of the first factor. Skehan (1998) named the former as ‘code complexity’ and Robinson (2001) named it as ‘resource-directing’ variables. However, the components of ‘code complexity’ or ‘resourcedirecting’ variable have been conflicting between the two researchers.
Skehan (1998) agreed with VanPatten (1990) on that language use prioritizes meaning and only when meaning is easily processed, the remaining cognitive capacity is committed to language form. Therefore, according to him, the tasks with high cognitive demands provide less opportunity to notice the language form whereas the tasks with low cognitive demands provide more opportunity to notice the language form and to acquire the language form. However, Skehan (1998) also proposed that apart from the cognitive load of the tasks, tasks could require the learners a complex language form due to task-inherent semantic contents, naming them ‘code complexity’ of the tasks. Furthermore, Skehan added one more dimension, the physical condition of task performance, to the task complexity, naming it ‘communicative stress,’ which resulted in the three dimensions of task complexity: cognitive complexity, code complexity, and communicative stress.
Cognitive complexity of the tasks refers to complexity of the thinking required for the tasks. Complexity of required thinking is of the two kinds: cognitive familiarity and cognitive processing. If the topic, genre and task format itself are familiar to the learners, the task would require less thinking. Cognitive processing involves the information load of the task. If the information for the task is provided in an efficient way, processing of the information would require less thinking. For example, if the information is organized in a natural chronological order in the narrative text, it requires less thinking than when it is not. But if the information has to be transformed or manipulated or processed multiply at the same time, or requires logical reasoning, then the task requires more thinking. Also indirect, insufficient, abstract, or dynamic information was proposed to require more thinking in comparison to the information that is static, declarative instead of being contextual. On the other hand, code complexity refers to the complexity of the language required for the semantic content of the task. It is measured by the linguistic complexity and variety including vocabulary load, redundancy and density required by particular tasks. For example, narrative tasks require past tense forms of verbs, which are more complex linguistic forms than present tense verbs. Finally, the third dimension, communicative stress, involves performance condition. If there is a time limit, and a large number of participants to interact with, and little opportunity to control interaction, then the task is more complex. According to Skehan (1998), whereas code complexity directly promotes IL complexity, the other two dimensions affect the role of code complexity only as secondary to code complexity, either negatively or positively. Skehan’s classification is summarized below in Table 1.
Based on Skehan’s (1998) model, Robinson (2001) reorganized the features Skehan proposed and added new features. Robinsons’ (2001, 2005, 2007) cognition hypothesis proposed that tasks could be classified by three factors which affect the learners IL differently. The three factors for the classification of tasks are task complexity (cognitive factor), task conditions (interactional factor) and task difficulty (individual factor). As a newly added factor, task difficulty deals with the learners’ individual differences such as affective variable and ability variable, which could nullify the objective complexity classified by the first two factors due to the individual learners’ perceptive and affective state. Task conditions are the interactional demands that include participation variables (e.g., one-way or two-way) and participant variables (e.g., familiar or unfamiliar). They are overlapping partially with Skehan’s third dimension, communicative stress. The first dimension of Robinson, task complexity, is merging Skehan’s (1998) first two dimensions, code complexity and cognitive complexity along with some other variables that were included in communication stress dimension by Skehan (e.g., time limit and time pressure).
However, Robinson made an important distinction within his first factor, cognitive complexity, that is, between resource-directing and Dahl, Osten. “On the Complexity of Complexity.” Paper Presented at Complexity, Accuracy and Fluency in Second Language Use, Learning and Teaching Conference, Brussels, Belgium. (2007a). cognitively complex in two ways. If the content of the tasks has the cognitively complex concept, then the tasks require a complex language form. Based on Givon’s (1985, 1995) functional approach, Robinson proposed that mental demands on cognitive/conceptual level drives the learners to attend to the aspects of an L2 system in order to accurately understand or produce the complex language form (Robinson 2002; Robinson & Gilabert 2007). The conceptual aspect of the task content has been named as resource-directing dimension. Therefore, Robinson’s cognition hypothesis predicts greater accuracy and complexity along the resource-directing dimension of tasks. The other dimension by which tasks could be cognitively complex is the resource-dispersing dimension. Tasks could be cognitively complex without requiring particular complex language form if there is demand on the performance dimension such as taking away planning time, or making dual or multiple simultaneous task demands. Then the accuracy and the complexity of IL production are expected to decrease on such complex tasks. Robinson’s three dimensions and the features are summarized in comparison to those of Skehan (1998) in Table 2.
As a result of different classifications of task complexity, the four Skehan’s secondary components were relocated as direct factors for IL complexity and accuracy by Robinson:
concrete/abstract, many items simultaneous, if-then needed, and cognitive familiarity. Therefore, the two models have opposing predictions regarding the role of these four moved items for the learners’ IL performance. On the other hand, one of Skehan’s secondary components, time pressure, moved to Robinson’s resource-dispersing dimension such that both scholars categorized time pressureas a secondary variable to IL complexity and/or accuracy. The present study targeted at the two variables among the moved items: the reasoning demand (Skehan’s if-thenas cognitive complexity, secondary to code-complexity) and the time limit. The two researchers have opposing predictions for the role of reasoning demand but they have the same prediction for the role of time pressure.
The three aspects of language performance—complexity, accuracy and fluency — not only reflect the learners’ degree of language acquisition, but also they differently contribute to language acquisition. Whereas language complexity, emphasized as the most important aspect for language acquisition, contributes to extending the current boundary of the learners’ IL system, accuracy is the aspect of control of already existing IL knowledge. Fluency is the aspect of automatization of already existing IL knowledge (DeKeyser 2007). Due to such a nature of fluency, when the learners are concentrating on the meaning of what they say, their fluency increases because they mostly resort to already existing knowledge without attending to accuracy and complexity of their language (Ellis & Barhuizen 2005). The tripartite relationship is summarized in Figure 1.
However, Skehan (1998) and Robinson (2001, 2007) differ in terms of the tension among the three aspects. Based on the limited cognitive capacity of human being, both scholars agree that form and meaning are competing for limited human attention. However, regarding the tension between accuracy and complexity, the two scholars differ. Whereas Skehan (1998) proposed that there is also a tension between accuracy and complexity in language performance in any task due to the limited cognitive capacity of human being, Robinson (2007) proposed that accuracy and complexity are both subject to the conceptual complexity of the task such that cognitively complex task on the resource-directing dimension requires the learner language to be both accurate and complex simply by nature of the task content. Robinson proposed that they do not compete for the limited attentional capacity based on the multiple resource theory of Wickens (2007) which argued that attention is not limited if the cognitive loads are required by different dimensions simultaneously like visual and auditory dimensions.
A summarizing task has been chosen as a target task for the study. Unlike the tasks designed for the past empirical studies that have been brief and fit into the lab setting (Bygate 2001; Bygate & Samuda 2005; Forster & Skehan 1996; Skehan & Forster 1997), the present study aimed at examining the tasks comparable to the natural real-world activities. The texts to read and summarize have been chosen from real world as they typically belong to different types of texts. Also the two modes of summarization, oral and written, are the typical two natural modes of summarization in real world, rather than the results of the artificial manipulation of time management as in the lab setting. Therefore, summarization tasks are important and natural tasks of IL performance that have rarely been studied in the literature.
Summarization is one of the most important academic and highly cognitive activities that involve selecting, deleting and reorganizing the information in the given text (Brown & Day 1983). Since its cognitive load may vary depending on the complexity of the text to be summarized, the two types of the texts were chosen depending on the amount of their reasoning demand. Expository texts that describe certain cultures or phenomena, for example, require the readers relatively little reasoning process to understand and summarize in comparison to the argumentative texts that lead to a conclusion through logical inferencing and reasoning. Regarding how the different reasoning demands of the two text types affect the learners’ IL performance, the two models have opposing predictions: Robinson predicts that the higher reasoning demand require the learners more complex and accurate IL whereas Skehan predicts the opposite due to the non-linguistic cognitive load of the reasoning demand.
The other variable for the present study, time pressure, is predicted by both models in the same way: the more time-pressed the learners are, the more all three aspects of language performance would be negatively affected. Time pressure was operationalized by oral and written mode of summarization. Summarizing verbally to an interactional partner without any planning time was hypothesized to have more time-pressure in terms of its production speed compared to the summarization in the written mode.
Based on the above review of the two theoretical models, the present study established the following research questions:
(1) Is the language used in the summarizing task affected by the types of the texts?
(2) Is the language used in the summarizing task affected by the mode of performance, oral and written?
(3) Do complexity and accuracy of learners’ IL performance have trade-off effects?
The data for the present study have been collected from the two separate studies: the oral summary data was taken from the previous study of repeated oral summarization (Lee et. al., 2007), and the written summary data was taken from the study of repeated written summary study three years later (Park, 2011). Within the limitation that the two data set have not been collected under the ideally controlled experimental conditions in complete parallel, the data have been analyzed for comparison in consideration of all such limitations.
Participants for the oral summarization were 6 first-year student volunteers from an intact required English class majoring in unspecified humanities at a large university in Seoul. The participants for the written summarization task were 24 paid volunteers majoring in humanities from the same university, regardless of their years of study. The written summary data was collected three years later. The written group was further divided into two groups for the two types of texts, expository and argumentative, whereas the oral group read the two types of text in two treatment sessions without being further divided.
All the participants were required to have the standardized English proficiency test scores within the past one year. All three groups had no experience of living abroad. Their English proficiency test scores were converted into a TOEIC scale according to the conversion table provided by ETS. The highest mean score, 713.8, was achieved by the group who summarized the expository text in the written mode, followed by the written argumentative text group as 690.4 and by the oral summary group who achieved 603.3 as summarized in Table 3. There was a significant between-group difference (
F=8.18, p=.001). The significant difference was found between the oral group and the written expository group ( p=.002) and between the oral group and the written argumentative group ( p=.013) but not between the two written groups ( p=.699) by ANOVA. In sum, the two written groups were more proficient than the oral group. This difference was considered in the statistic procedure.
The two texts for the oral summary and the other two texts for the written summary were taken and adapted from ESL textbook (Adelson-Goldstein, 2004). ; Barton & Sardinas, 2004; Beatty, 2004; Malarcher, 1999). Adaptation of the texts was minimal and its purpose was mainly to reduce the memory load. The proper nouns have been reduced and the extremely low frequency vocabulary was replaced by the higher frequency synonyms. The four texts were comparable in terms of length. They were between 292 and 308 words in total and the number of T-units were between 23 and 29 as shown in Table 4. The length of the texts was intentionally controlled in the process of selecting the texts. However, the structural complexity was neither controlled nor adjusted, because it was considered to be inherently related to the cognitive complexity of the text content as discussed above (Givon, 1985; 1995, Robinson, 2002). Interestingly, the structural complexity measured by the number of subordination per T-unit was not equal in all 4 texts. Complexity ranged between 0.31 and 0.61 across the 4 texts. The complexity of two expository texts was low and close to each other as 0.35 and 0.31. The complexity of the two argumentative texts was 0.61 and 0.46, both higher than that of the expository texts, though it is not as close to each other as that of the expository texts. The average complexity of the argumentative texts was 0.54 whereas that of the expository texts was 0.33. In sum, the argumentative text was structurally more complex, in support of Givon’s (1985, 1995) and Robinson’s (2002) proposal of the linear relationship between cognitive complexity and language complexity from their functional perspective. As a result, though the oral and written summary data were taken from different studies, the four texts were generally comparable other than their structural complexity. The full four texts are in the Appendix.
The 6 oral summarizers read the article and when they looked up from the text they summarized the article to a partner, without looking at the article. The subjects summarized the argumentative text to the first partner and then after a short break, summarized the expository text to another partner. All the partners were from the same intact class as that of the subjects who had never read the target texts before. They only participated as the subjects’ partners for the purpose of understanding the content. Though interaction was allowed, little interaction occurred.
Among the 24 written summarizers, 12 subjects were given an expository text and then were instructed to write a summary of their own length without looking at the text again. The other 12 subjects were given an argumentative text and were instructed in the same way. Reading and summarizing were considered as one task because, even though summarization has two sub-steps, reading comprehension and producing the content in reorganization, because in real-world situation the two steps are often combined into one cognitive event. Such comparability of realworld activities is one of the defining properties of task (Skehan 1998). For the same reason, time and length of the summary were not controlled. For the oral summary group, reading took 2 to 3 minutes and summarization took 1 to 8 minutes. For the written summary group, reading took about 5 minutes and summarization took 10 to 15 minutes. These differences were considered as inherent and legitimate in the two modes of summarization.
The transcripts of the 12 oral summaries and 24 written summaries were coded for their complexity and accuracy as suggested by Kuiken & Vedder (2007). As one of the most reliable measures of L2 writing,1 complexity was measured by the number of subordinations per T-unit and accuracy was measured by the rate of error-free T-units to the total number of T-units.2 The kinds of errors counted were errors of word order, subject-verb agreement, word choice and other morphosyntactic errors. Fluency was not considered for the present study because fluency in oral and written performance is of the different nature and more importantly because the present study is focused on complexity issue first followed by accompanying accuracy issue.
1Though the measures of oral performance may be more varied, the summaries from both oral and written modes were analyzed by one and the same measure involving T-unit, because it is a more general measure and the type of oral performance in the present study was little interactive and the topic or the nature of task were not typically verbal interaction-oriented, for which AS unit is more suitable. 2Kuiken and Vedder (2007) considered ‘clause’ as an important component of language development and reported that the three measures of syntactic complexity that involve clauses —the number of clauses per T-unit, the number of dependent clauses per T-unit and the number of dependednt clauses per total number of clauses —were proven to increase linearly according to the learners’ proficiency level. Regarding the measure of accuracy, the number of error-free T-units and the number of errors per T-unit was proposed as the best measures, the former for the advanced and the latter for the beginner and intermediate level learners (Kuiken & Vedder, 2007; Wolfe-Quintero, Inangaki & Kim, 1998), The present study chose the ratio of error-free T-units because summarizing itself requires relatively advanced level of proficiency, though the more varied and detailed number of error analysis may provide the more precise analysis in the future study.
Text type and performance mode were two independent variables and the dependant variables were complexity and accuracy of the language in the subjects’ summaries.3 For the oral summary group, the text type was a within-subject variable and it was measured by paired t test. For the two written summary groups, the text type was a between-group variable and was measured by independent t test. The comparison between oral and written summaries, however, were conducted by ANCOVA with the standardized English proficiency test scores set as a co-variate because of its significant difference between the two groups. All the statistical analyses were run by SPSS 18 package.
Twelve oral summaries and 24 written summaries were examined for their length in terms of the total number of words. In case of oral summaries, meaningless backchannels or discourse markers were excluded. As a result, as shown in Table 5, the mean length of oral summaries was 196.8 words and that of written summaries was 105.4 words with a significant difference (
F=.257, p=.000). However, there was no significant difference between the text types. The mean length of the expository text summaries was 139.6 and that of the argumentative text was 132.2 ( F=.047, p=.829). There was no interaction effect ( F=.193, p=.663). In sum, oral mode produced the greater amount of summary output than the written mode in both text types.
Both in oral and written summaries, the text type did not result in difference either in complexity or accuracy as shown in Table 6. In the oral mode, the complexity of the expository text summaries was .25 and that of the argumentative text summaries was .48 (
t=-1.556, p=.180). The accuracy of the expository text summaries was .42 and that of the argumentative text summaries was .28 ( t=1.284, p=.225). Therefore there was no significant difference either in complexity and accuracy, though the two text types showed symmetrical distribution of complexity and accuracy scores: expository text summaries were low in complexity but high in accuracy whereas the argumentative text summaries were the opposite, high in complexity and low in accuracy.
In the written mode, the complexity of the expository text summaries was .32 and that of the argumentative text summaries was .19 (
t=1.923, p=.067). The accuracy of the expository text summaries was .31 and that of the argumentative text summaries was .33 ( t=-.277, p=.784). Therefore, though there was no significant difference either in complexity and accuracy in written mode, the two types of text were almost the same in accuracy but there was a tendency ( p=.067) that complexity was higher in the expository text than in the argumentative text summaries.
In sum, the text type effect was shown only as a tendency in the complexity of the written summaries but it was in the opposite direction from Robinson’s model which predicts that the reasoning demand as a resource-directing dimension would elicit more complex IL and in support of Skehan’s model which predicts that the reasoning demand would reduce IL complexity due to its cognitive load.
The performance mode did not make any difference in complexity or accuracy in expository text summaries but it made a significant difference in the complexity of argumentative text summaries. As shown in Table 7, in the expository text summaries, the complexity was .25 in the oral mode and was .32 in the written mode (
F=.166, p=.690). The accuracy was .42 in the oral mode and was .31 in the written mode ( F=2.83, p=.113). However, in the argumentative text summaries, the complexity was .48 in the oral mode and was .19 in the written mode with a significant difference ( F=4.579, p=.049). The accuracy was .28 in the oral mode and was .33 in the written mode ( F=.131, p=.722). Therefore, the effect of performance mode was found only in the complexity of argumentative text summaries. However, the result showed the opposite to both models of Skehan and Robinson, because the complexity was higher in the more time-pressed oral mode than in the written mode.
Neither the trade-off nor the linear relationship between complexity and accuracy were observed significantly in the present study. Only in the descriptive statistics, there was a pattern of trade-off effect such that the IL complexity was higher and the IL accuracy was lower in the argumentative text summaries than in the expository text summaries within the oral mode. Other than that, trade-off between complexity and accuracy of IL were hardly noticed either through the text-type comparison or through the comparison of modality even in the descriptive statistics.
3However, two-way ANOVA or ANCOVA was impossible because there were only three groups of subjects, one group for oral summaries of the two types of text and the two groups for the written summaries of the two types of text.
The text type, representing the reasoning demand, did not affect either complexity or accuracy in either mode. Only a tendency was found in the written mode: the written expository text summaries resulted in higher complexity .32 than the written argumentative text summaries .19 (
p=.067). It is interpreted that the reasoning demand rather demoted the IL complexity as predicted by Skehan’s model in the written mode at least. On the other hand, the mode of performance affected IL complexity in the argumentative text type: the oral argumentative was structurally more complex as .48 than the written argumentative summaries as .19 ( p=.049). This finding is against the both models because both models predicted that the reduced time pressure of the written mode would promote the IL complexity. A further interpretation of the results shows that descriptively the complexity of written argumentative summaries was lowest among the four types of summaries (written argumentative, written expository, oral argumentative, oral expository). And statistically the complexity of written argumentative text summaries .19 was lower than its expository counterpart ( p=.067) and also was lower than its oral counterpart ( p=.049). Therefore, the reason why the written argumentative text resulted in such a low IL complexity score needs to be discussed.
First, regarding the effect of text type, Skehan’s limited capacity model seems supported in the written mode. However in the oral mode, an opposite pattern is observed regarding IL complexity and the text type. Oral summaries showed a higher IL complexity in the argumentative text than in the expository text summaries in support of Robinson, though the difference was not significant. In contrast, lower IL accuracy in the argumentative and higher IL accuracy in the expository text summaries were found, though not significant, in support of Skehan’s trade-off effect (1998). Moreover, considering that the number of subjects in the oral mode was as small as six, the argumentative text has the potential to increase IL complexity probably in the larger data set.4 more in support of Robinson (2001, 2007). Therefore, oral summarization seems to have shown the potential effect of text type as a reasoning demand which directly affects IL complexity, though it did not affect IL accuracy together.
In contrast, in the written mode the IL accuracy was almost the same in the two text types, as .33 for the argumentative and .31 for the expository text. Whereas the accuracy was very close to each other in the two text types, IL complexity tended to be different, resulting in no pattern of trade-off effect between IL complexity and accuracy. The two modes of IL performance, therefore, seem different in being affected by the text type. In the oral mode, the reasoning demand of the text type may directly promotes IL complexity but it may not in the written mode.
In consideration that both Skehan’s and Robinson’s theories put their bases on oral tasks, findings from the written mode require interpretation from new perspectives other than from TBLT. Also it should be considered how the between-group difference in the proficiency level affected. Considering that the written group was the higher proficiency group, it is not clear whether TBLT model is better fit to the lower level learners or to the oral mode only.
Second, regarding the time pressure, the significantly higher complexity of oral argumentative summaries than the written argumentative summaries against the prediction suggests that mode was not a proper operationalization of time-pressure. In the previous empirical studies, 1 to 10 minutes of planning time before oral production led to higher complexity and accuracy operating on the resource-dispersing dimension (Bygate 2001; Bygate & Samuda 2005; Forster & Skehan 1996; Skehan & Forster 1997). One possible interpretation is that the difference in modes, oral and written, is not only a matter of more or less time available. Planning time followed by regular speedier oral production tasks may be of different nature from writing process in which time management is controlled entirely by the writers throughout the given time.
The findings regarding both text type and mode suggest that summarizing task in the written mode is largely deviant from the two models, because written summaries of the argumentative text were not complex enough relatively to oral counterparts and to the expository counterparts. The adult learners in the present study might have secured a certain level of accuracy in writing as a priority — accuracy of the two text type of summaries was very close to each other as .33 and .31 for argumentative and expository text each. And then the larger reasoning demand of the argumentative text could have led the learners to attempt less complexity in the argumentative text summaries than in the expository text summaries in support of Skehan (1998). On the contrary, in the oral mode, without fixing accuracy constant, the learners seem to have attempted the complex structure to convey complex thinking, i.e., reasoning. Such a high risk-taking behavior in the oral mode may be supported by the larger total amount of production in the oral mode than the written mode.
Based on the discussions so far, it may be possible to conclude that the text type seems to function as a resource-directing dimension of task complexity, playing the role of pushing the learners’ IL boundary for restructuring in the oral mode in the group of lower proficiency level. Accuracy, however, does not seem to behave together with complexity. Accuracy seems to compete for the resource with complexity in the oral mode. On the other hand, written mode may not simply add extra time to oral mode as a feature of resource-dispersing dimension. Unlike planning time for the oral production task, the extra time available during writing seems to play a more complex role. Learner-writer can move back and forth at the wanted speed any time while writing. Therefore, writing seems to proceed through the fundamentally different process of controlling and balancing within the learners’ IL knowledge and the limited cognitive resource.
Still remaining a question is that written summaries were not better than oral summary in any aspect of IL performance or in any text type, particularly if it is considered that the subjects in the written group were significantly higher proficiency group. Though the present study suggested the priority of accuracy in writing, even accuracy was not significantly higher than in oral summaries. Writing involves probably more resource-dispersing demand other than such a positive factor as extra time. It may probably include writing mechanics and more importantly information organization which decides the quality of writing in terms of its content. The effect of extra time on IL performance might have been overridden by those demands. The quality of the summary might be examined most importantly in the further study as a possible destination of the extra time resource.
The present study is not free from its limitations. First of all, the operationalization of reasoning variable through the type of the text has not been widely validated in the literature. Obviously expository texts include less causal reasoning process both in reading and its summary writing whereas argumentative texts do more. However, summarization itself involves reasoning process to some extent in selecting, deleting and reorganizing the information in the text, regardless of the text type. The present study assumed the amount of required reasoning would be different, which still needs to be verified. In relation to this, as a second limitation, the present study combined reading comprehension and summary production tasks into dual and sequenced tasks without checking the effect of the first to the second step, which remains to be improved in the future study without damaging the property of ‘task’ which should be comparable to the real-world activities. Instead of checking the comprehension level of the target text, the present study only accessed the learners’ general proficiency level and conducted a pilot study in the assumption that the given texts would be readable and comprehensible enough for the subjects. Third, due to the different data sources for oral and written summaries, the number of subjects and their proficiency levels were not the same, though this problem was partially resolved by using differential statistical procedures, i.e., setting their proficiency level as a covariate. It was also meaningful that the written group, the higher proficiency group, were not any better than the oral group because the uncontrolled proficiency level did not confound the results and it supports the authors’ interpretation that oral and written modes are very different kinds of IL production such that they did not proportionately reflect the learners’ proficiency level. Forth, because the oral data was taken from the repeated measure design whereas the written data was not, statistic analysis was not efficiently conducted with 2 by 2 ANOVA design. Fifth, though the reading texts used for the oral and written groups were similar and intended to be comparable, the identical texts might have produced more reliable results. The sixth controversial issue is the time-control. The time given to the subjects in each condition and the length of summaries in each condition were not tightly controlled deliberately for the purpose of collecting the natural oral and written data. Instead of controlling the time, the present study intended to control the input and to leave the reading and summarizing process as natural as possible. The results showed that in the oral mode, the learners produced the larger amount of IL production during the shorter time, which seems natural. However, time and the amount of IL production may eventually need to be examined more systematically in the future study because they are both sensitive in terms of cognitive load. Finally, the quality of the summaries needs to be examined in the future study.
4This interpretation is partly proved in support of Robinson’s prediction about the relationship between the reasoning demand and the complexity of IL, though indirectly, by the finding from the earlier study of repeated oral summaries in which the difference in complexity between the two text types turned statistically significant when the three repetitions of the same task were compared as a total (Lee et al., 2007).
Cognitive complexity is a fancy and exciting concept vitally interwound with language but it seems yet to be more systematized. Moreover, the distinction between the cognitive complexity that directly requires certain linguistic complexity and the one that does not seems even less systematized. One such variable, reasoning variable was examined in the present study. As Dahl claimed that it is a complex task to classify task complexity (2007a, 2007b), it is hard to decide whether reasoning demand is linguistically involved such that it requires a more complex language form than no reasoning tasks, or whether on the contrary reasoning/thinking is conducted in separation from the linguistic complexity and rather leaves little resource left for linguistic processing in support of Skehan. In the present study, the oral summary was better explained by Robinson’s classification, whereas the written summary was better explained by Skehan’s classification. That is, reasoning demand of the argumentative text was a cognitive burden to IL complexity in the writing mode but it was a driving force for IL complexity in the oral mode. The systematical rationale is warranted that includes the mode of IL production.
Regarding the trade-off between complexity and accuracy, in the oral mode, though not significant, the symmetrical distribution of complexity and accuracy in the two text types supports Skehan’s trade-off effect, though such a pattern disappeared in the written mode. In the written summary, the accuracy level was constant across the two text types, with lower complexity in the argumentative text summaries. In sum, oral and written modes may be different not only in terms of time pressure but in some other cognitive ways too.
Another consideration is that learner factor - task difficulty as named by Robinson - was not considered in the present study. That is, it is not clear whether the text types and summarizing tasks match the level of the learners such that the relationship between complexity and accuracy is realized as Robinson expected. It is very probable that not all levels of tasks show such pattern of relationship because both Givon’s (1985, 1995) functional linguistic argument and Wicken’s (2007) multiple resource theory are not modelling the learners on their developmental stage but the people who are fully functioning in the area of performance, i.e., native speakers. In case of learners’ performance, learners’ repeated performance of the same type of the tasks, for example, may eventually show the predicted relationship among the task types, complexity and accuracy rather than the beginning level learners’ one-time performance (Bygate 2001; Bygate & Samuda 2005).
Even with all these critical issues unresolved yet, the cognitive model of language use and acquisition proposed by Skehan and Robinson seems to be the core question in language acquisition research to be resolved in cooperation with cognitive psychology.
[Table 1] Skehan’s (1998) Three Classificatory Dimensions of Tasks
[Table 2] Robinson’s (2001, 2007)Three Classificatory Dimensions of Tasks
[Figure 1] Skehan’s Three Aspects of Task Performance (Ellis and Barhuizen, 2005, p. 143)
[Table 3] Properties of Target Texts
[Table 4] Properties of Target Texts
[Table 5] Length of the Summaries
[Figure 2] Length of Summaries Across Mode and Text Type
[Table 6] Effects of Text Type for Complexity and Accuracy
[Figure 3] Effects of Text Type for Complexity and Accuracy
[Table 7] Effect of Performance Mode for Complexity and Accuracy
[Figure 4] Effect of Performance Mode for Complexity and Accuracy