by Rod Ellis
The approach I will adopt in reviewing the research that has investigated tasks is to select representative studies rather than attempt a comprehensive review of the research. Such reviews can be found in R. Ellis (2003), Samuda and Bygate (2008), Skehan (2011), and Robinson (2011). In the sections that follow, I will first examine studies based on input-based tasks and then those that have investigated output-based tasks. My main purpose is to consider each study in relation to the theoretical constructs that informed it and the extent to which the results of the study lend support to these constructs. It should be noted, however, that there is also a rich pedagogic literature on tasks (e.g. Prabhu 1987; Willis and Willis 2007). Indeed, interest in tasks as a vehicle for learning a language originated in proposals for language pedagogy (Candlin 1987).
Input-based tasks
In an input-based task, learners are presented with L2 input (oral or written), which they need to comprehend in order to achieve the outcome of the task. Thus, an input-based task does not require production on the part of the learner. However, learners are not prevented from speaking and, in fact often do when they fail to comprehend. Input-based tasks are based on the assumption that learners will pick up new linguistic forms through exposure to the input providing that (1) they are able to comprehend the input and (2) notice the new forms. Thus, they draw on those SLA theories that emphasize the importance of comprehensible input (Krashen 1985) and positive evidence. Input-based tasks can also help to develop comprehension skills in learners.
Many of the input-based studies involved ‘listen and do’ tasks. Ellis, Tanaka, and Yamazaki (1994), for example, used a kitchen task, which required learners to listen to directions about where to place various objects in a matrix picture of a kitchen. The objects were depicted in small numbered pictures. The outcome of the task was the matrix picture of the kitchen with the numbers of the small pictures marked in different locations. In this experimental study, different groups of high school Japanese learners of English listened to the directions in baseline form—i.e. the input was unmodified—in premodified form—i.e. the input was simplified to aid comprehension—and in interactionally-modified form—i.e. the students were allowed to request clarification when they did not understand a direction. In this way, it was possible to investigate the effects of exposing learners to different kinds of input on their comprehension of the directions. The results were clear. Every direction was comprehended better by the interactionally modified group than by the baseline or premodified group, lending support to the Interaction Hypothesis. However, a likely explanation for this is that the learners had more time to process the input in the interactionally modified condition. Ellis and He (1999) carried out a similar study, but this time took care to ensure that the amount of time allocated to the premodified input and interactionally modified input conditions was the same. In this case, there was no statistically significant difference in the comprehension scores of the two groups.
Ellis, Tanaka, and Yamazaki did not just investigate the effects of the different kinds of input on comprehension, they also embedded a number of words—i.e. the names of different kitchen objects—that were new to the learners. This allowed them to examine whether or not the different input conditions had any effect on vocabulary acquisition. Overall, none of the groups learned many of the new words, as demonstrated in tests administered after the task was completed. However, significantly more words were acquired by the learners who received interactionally-modified input. Again, this may reflect the additional time afforded learners and the fact that they were exposed to the new words more frequently in this condition. In Ellis and He’s study—where time for each condition was controlled—there was no difference in the vocabulary scores of the premodified and interactionally modified groups.
Ultimately, then, what emerges as important for both comprehension and vocabulary acquisition in these studies, is the fact that learners need both time to process the input and multiple exposures to the target words. Interactionally modified input can guarantee both time and multiple exposures. Premodified input can do likewise, but only providing that care is taken to build sufficient repetition into the linguistic material.
These studies investigated vocabulary acquisition. The question arises as to whether input-based tasks also cater to the incidental acquisition of grammatical features. Shintani and Ellis (2010) and Shintani (2015) investigated this. The learners in these studies were young Japanese children with no prior knowledge of English—i.e. they were complete beginners. The input-based tasks consisted of teacher commands that required the learners to identify objects—represented by picture cards displayed in front of them—and to take them to either a supermarket or a zoo, represented by friezes pinned to the walls of the classroom. Thus the learners displayed their understanding of the commands by selecting the correct card and taking it to the correct location. The same tasks were repeated nine times over a five week period. The commands were designed to expose the learners to two grammatical structures—plural -s (a grammatical feature not present in Japanese and therefore difficult to acquire) and copula be. Some of the commands required the students to identify a singular object—for example, ‘Please take the crocodile to the zoo; while others referred to a plural object—for example, ‘Please take the crocodiles to the zoo’. Although the tasks were input based, they inevitably led to interaction as the learners struggled to understand the teacher’s commands. Acquisition was investigated by means of tests that measured learners’ receptive and productive knowledge of plural -s and productive knowledge of copula-beNOTE 3.
The results of these studies are interesting. The learners were successful in acquiring receptive knowledge of plural -s, but only a few of the them developed productive knowledge of this structure and none of the learners acquired productive knowledge of copula-be. The learners’ successful acquisition of receptive knowledge of plural -s can be explained by the fact that the tasks created a functional need to process this feature: they could only achieve a successful outcome if they were able to distinguish singular and plural nouns in the input. In the interaction that resulted from the tasks, the learners sometimes checked whether the noun was plural or singular—for example by asking ‘one? two?’—and they also received feedback on whether they had selected the right card(s). However, only a few of the learners ever produced the plural form of a noun during the tasks. This might be due to the fact that because they did not have to produce plural nouns the learners did not develop productive knowledge of them. In the case of copula be, ample exposure in the input did not lead to the ability to produce this structure, probably because this feature is redundant and non-salient in the input or—again—because the learners did not produce it.
Input-based tasks are well suited to low-proficiency learners of the kind that figured in these studies because they can be designed to expose learners to the kinds of input that research has shown can assist comprehension. They can also incorporate specific target features—both lexical and grammatical—making it possible to investigate if acquisition takes place. When implemented in the classroom, learners can be given—or just take—opportunities to request clarification, leading to the negotiation of meaning and form that has been shown to facilitate both comprehension and acquisition (see Chapter 7). Input-based tasks can facilitate incidental acquisition of vocabulary and of grammatical items that are functionally important for performing the task.
However, the studies I have considered in this section suggest that input-based tasks have their limitations. They showed that the learners developed receptive knowledge of the target features, but not productive knowledge. To acquire productive knowledge, learners may need opportunities to produce the target features. Also, the studies make it clear that sheer exposure to target features may not be enough. Acquisition becomes evident only when tasks create a functional need for the learners to attend to form. For this reason, redundant and non-salient features may not be attended to and not acquired.
<
br /> Output-based tasks
Output-tasks require learners to speak or write. However, by and large, the research that has investigated output-based tasks has focused on oral production. Output-based tasks can be monologic or dialogic. An example of a monologic task is when learners are asked to produce (or reproduce) a narrative, as in Skehan and Foster (1999) or Yuan and Ellis (2003). Dialogic tasks that require learners to interact require information-exchange and are typically of the information-gap or opinion-gap kind. For example, Foster and Skehan (1996) included a decision-making task where the learners had to work in pairs to reach agreement about suitable prison sentences for a number of offenders.
Monologic and dialogic tasks differ in some fundamental ways. In the former, learners have to rely entirely on their own resources, while in the latter, they work collaboratively and can scaffold each other’s production. Dialogic tasks also allow for online corrective feedback as in Lyster’s studies (for example, Lyster 2004), which we considered in Chapter 7. For some researchers, tasks are necessarily dialogic. McDonough and Mackey (2000), for example, claimed that the aim of a task is to ‘provide learners with opportunities to engage in meaningful interaction and to direct their attention to form’ (p. 83). This, however, is an over-narrow view of a task.
Central to all the research involving output-based tasks is the identification of the design and implementation variables that influence how learners perform a task. Design variables refer to features of the task-as-workplan—i.e. the task materials. Implementation variables refer to ways in which the task is carried out and thus relate to the methodology of task-based teaching. A large number of variables have been identified and investigated. Table 11.2 lists some of the major ones.
Design variables: task workplan Implementation variables: task performance
1 Contextual support
1 Learners’ role (i.e. listener vs interactive participant)
2 Number of elements to be manipulated
2 Pre-task planning
3 Topic familiarity
3 Time pressure
4 Shared vs split information
4 Rehearsal (task-repetition)
5 Dual vs single task
5 Post-task requirement
6 Closed vs open outcome
7 Inherent structure of the outcome
8 Discourse mode (e.g. description vs narrative)
9 Here-and-now vs there-and-then
Table 11.2 Selected task design and implementation variables
The extent to which it is possible to influence the way in which learners perform a task by manipulating design and implementation variables is a matter of some controversy. Sociocultural theorists have argued that it is not possible to predict the ‘activity’—i.e. the process—that results from a ‘task’ as learners are likely to interpret the task-as-workplan differently in accordance with their own goals and motives. Coughlan and Duff (1994) reported a study that showed that the ‘activity’ that took place when the same task was performed by different learners varied considerably. This study also showed that the ‘activity’ varied when the same learner performed the task on different occasions. Seedhouse (2005) argued that the discrepancy between the predicted and actual language use resulting from a task was so great that a task should only be defined with reference to the language processes that arose in its performance. However, as we will see, there is plenty of evidence to show that specific design and implementation variables are predictive of the language use that results from performing a taskNOTE 4.
I will begin by considering design features of tasks: first in monologic tasks and then in dialogic tasks. I will then turn my attention to an examination of a number of implementation variables that have attracted the attention of researchers, such as pre-task planning. In the process, I will examine the competing claims of Skehan’s Trade-off Hypothesis and Robinson’s Cognition Hypothesis.
Effect of task design features on L2 production
Monologic tasks
The research that has investigated monologic tasks has examined how the complexity of a task affects CAF in learners’ production. As you may recall, there are different theoretical positions concerning this. Skehan (2009) argued that because learners’ processing capacity is limited, they will prioritize one aspect of production over others—for example, either accuracy or complexity, but not both—depending on the difficulty of the task. Robinson (2007) proposes that a complex task will have an effect on both accuracy and complexity.
Tavakoli and Foster (2011) investigated a narrative task performed by two separate groups of learners—a group of 60 EFL learners in Tehran and a group of 40 ESL learners in London. The study involved four tasks that differed in terms of whether the storyline had a tight or loose structure, and whether the story did or did not include background events. They hypothesized that (1) the narrative with tight structure would lead to learners prioritizing accuracy; and (2) the more complex narrative that involved encoding background events as well as foreground events would result in learners prioritizing complexity—both lexical and grammatical—but not accuracy.
The results showed that the design of the tasks did indeed affect the production of both groups with the tight narrative structure promoting accuracy and the two story lines greater syntactic complexity. Tavakoli and Foster then went on to note:
As a result, a narrative with loose structure and only foreground events elicits a performance of relatively low accuracy and low syntactic complexity, whereas performance in a narrative of tight structure and both foreground and background events elicits relatively higher accuracy and complexity.
(Tavakoli and Foster 2011: 57)
As Tavakoli and Foster then pointed out, parts of these results fit both Skehan’s Trade-off Hypothesis and Robinson’s Cognition Hypothesis. Both tasks with a clear ordering of narrative elements were performed more accurately than the tasks with a loose structure, thus supporting Skehan’s model and replicating the results of previous studies—for example, Foster and Skehan 1999. In terms of Levelt’s speaking model, when learners have to spend less effort on conceptualization, they are free to focus more on formulation and—hence—accuracy increases. However, there was also some support for the Cognition Hypothesis as the task which had a tight structure and two story lines and which therefore incorporated both a source-dispersing and a resource-directing variable resulted in both greater accuracy and complexity, as predicted by the hypothesisNOTE 5.
This study also reported on the effect of the learning context (i.e. Tehran or London). It had no effect on accuracy or fluency, but it did affect lexical complexity although not grammatical complexity. They suggested that London learners’ opportunity to use the target language in everyday tasks outside the classroom primarily benefits lexical development.
There have been a number of studies that have investigated the claims of the Cognition Hypothesis. Jackson and Suethanpronkul (2013) reported a meta-analysis of nine studies that investigated simple and more complex narrative tasks. Overall, the results of the analysis did not support Robinson’s hypothesis. The more complex tasks resulted in measurable greater accuracy. However, they were not found to have a significant effect on complexity whether of the grammatical or lexical kind. In other words, task complexity was not shown to have a joint effect on both accuracy and complexity. Jackson and Suethanpronkul concluded that resource-directing variables, presumed to enhance task complexity, appear to only affect accuracy.
It is difficult to come to any firm conclusions about the effects of task-design variables on different aspects of learner production in monologic tasks. Overall, increasing task complexity does appear to result in greater accuracy and also has a negative effect on fluency, which in part supports the Cognition Hypothesis. But there is limited evidence to show that it also has an effect on complexity. Rather—as claimed by the Trade-off Hypothesis—simpler tasks have been shown to support production complexity. However, the studies that have investigated these theoretic
al positions have operationalized task complexity in very different ways and also have employed different measures of CAF. This makes any attempt at synthesizing the results of the studies problematic.
Dialogic tasks
Researchers have investigated the effects of design features of dialogic tasks by examining how they influence (1) the quantity and quality of the interactions that arise when the tasks are performed, and (2) CAF.
In Chapter 7, I examined features of the interaction—for example, the negotiation of meaning, the negotiation of form, and modified output—that arise when learners perform dialogic tasks that have been hypothesized (and have indeed been shown) to assist acquisition. In this chapter, I will focus on a number of design features of tasks that have been found to affect the incidence of these interactional features.
In Ellis (2003), I summarized the findings of some of the early work on tasks by identifying a number of design variables that had been shown to influence interaction and that constituted ‘psycholinguistically motivated dimensions’ of tasks (Long and Crookes 1987). Two key variables were (1) whether the information to be exchanged was required or optional, and (2) whether the outcome of the task was closed—i.e. there was only one possible outcome—or open—i.e. there were many possible outcomes. The results of the studies that investigated these tasks are quite mixed, but overall required information-exchange as in information-gap tasks was found to be more likely to promote the negotiation of meaning than optional information-exchange, as in opinion-gap tasks (Pica, Kanagy, and Falodun 1993). Closed tasks were also found to be more likely to result in negotiation of meaning than open tasks. Thus—from the perspective of the Interaction Hypothesis—tasks that require information exchange and have closed outcomes are more likely to promote acquisition. However, my review of this early research also showed that different task variables impacted differently on different aspects of interaction. For example, some studies showed that optional information-exchange tasks with an open outcome could lead to greater complexity in learner output. I concluded that the research was only suggestive of what design features promoted the kinds of interactions hypothesized to be important for acquisition.