Three types of task planning on Fluency, Accuracy and Complexity in L2 oral production

"The Differential Effects of Three Types of Task Planning on the Fluency, Complexity, and Accuracy in L2 Oral Production"- Paper by Rod Ellis, published in Applied Linguistics 30/4, 2009. Available at this LINK to download.

This paper by Rod Ellis is an excellent summary of research on the effects of planning time in L2 oral production till 2009. Ellis puts the research till date in perspective and outlines questions for future research. A summary with comments is given below. I am adding my comments in Italics wherever appropriate. Relevant references are copied from the original paper  and given at the end of this post for your quick reference. 

Research on the effects of planning time on L2 oral production informs the methodology of task-based teaching for improvement.

Three kinds of planning are distinguished by Ellis (2005). They are Rehearsal planning, Strategic planning and Within-task planning. 

Rehearsal and Strategic are both Pre-task planning which is done before the learner actually does the task.

Rehearsal planning is when the learner rehearses the entire task before performing the task a second time. 

Strategic planning lets the learner plan what language and content to use during the task, but doesn't let the learner rehearse the task.

Within task planning is when planning takes place during the performance of the task. There are two kinds- Pressurised and Unpressurised. Pressurised planning has a specific time limit. Therefore, learners must strive to finish the task within this constraint. This is otherwise known as pressurised online planning. Unpressurised planning provides unlimited time to complete the task. This leads to careful online planning. 

Three aspects of language production are looked at to see the the effect of planning time on them. They are, accuracy, fluency and complexity. They constitute a learner's language proficiency. These three terms are defined by Skehan (1998) and Skehan and Foster (1999). Fluency is the capacity to use language in real time, complexity is the capacity to use more advanced language and Accuracy is the ability to avoid error in performance. In practice, however, researchers develop and use operational definitions to serve the goals of their research. This is problematic when it comes to comparison of different studies. 

The framework developed by Skehan (1998) speaks of two systems possessed by language speakers. They are rule-based and memory-based systems. Rule-based system consists of the underlying rules and patterns of language. Memory-based system consists of chunks of language or formulaic sequences which are readily accessible for use. The difference between these two is in terms of processing load. The former requires large amounts of attentional resources to process, while the latter requires much less resources. Language performance involves the use of both in a variable manner depending on the requirements and demands of language processing at the time of production. This has three implications. Firstly, this implies that proficiency is a flexible phenomenon which adapt according to the conditions available during performance. Secondly, if one learner lacks in one of the systems, he/she can compensate that shortcoming using the other. Thirdly, when the situation demands greater dependence on one system, some aspects of performance might be affected. For example, when rule-based system is heavily used during a performance for the sake of structural accuracy, fluency of speech might get affected. Individual differences can also affect performance in this manner. 

1. Studies in Rehearsal Panning Time

Rehearsal planning give the learner an opportunity to perform the entire task prior to actual performance that is counted. For the same reason, rehearsal planning time is not practical in testing practice. Ellis cites three studies: Bygate (1996), Gass et.al. (1999) and Bygate (2001). Two major questions are asked in these studies.
i. Does task repetition have any effect on the performance of the same task?
All three studies produced evidence for beneficial effect of task repetition on the performance of the same task.
ii. Does task repetition have any effect on the performance of a new task?
All studies reported that that there is no transference of effects to new tasks, even when the new tasks are of the same type as the rehearsal task.
Which aspect of performance is affected by rehearsal?
Fluency and complexity are the most influenced aspects as reported by all three studies. Bygate used 10 week gaps to see if the effect is due to immediate recall, which is not. Accuracy was found to have no/little effect on performance.
Acquisition-Performance connection
Since the studies showed that there is no transference of the effect to new tasks, it implies that performance in L2 does not lead to acquisition. Or, to lead to acquisition, learners need to get feedback on the initial performance to enable 'noticing' and successive acquisition. An interesting study reported is Sheppard (2006)- an unpublished PhD thesis from the University of Auckland. This study reports the feedback intervention mentioned above.

Thus we could conclude that task repetition affects performance. But when used along with other tools like feedback, it could even lead to acquisition, and positive influences on all three aspects of performance. 

2. Studies in Strategic planning time

Ellis cites 19 studies here. He uses four parametres to talk about the results of these studies. They are: learners, settings, tasks and planning.

Learner Variables:

1. Second and Foreign language learners 
2. Proficiency level of learners (most studies looked at intermediate level learners). Wigglesworth (1997), Kawauchi (2005) and Tavakoli and Skehan (2005) manipulated proficiency as a variable. 
3. Learner's orientation to planning 
4. Individual learner difference factors

Settings Variables:

1. Classroom
2. Laboratory
3. Testing

These variables are connected to length of planning time. Compared to testing context, the other two settings provided longer planning time to learners. 

Task Variables:

1. Interactive Variables (Monologic vs. Dialogic tasks)
2. Task Complexity (Simple vs. Complex tasks)

All the testing studies included in Ellis' list used monologic tasks in order to control the multitude of influences that come in with an interlocutor. All initial testing studies used monologues. But there are dialogic testing studies available. 

Task complexity or task difficulty is a difficult topic due to many variables involved in it. The factors affecting complexity, and used in the studies listed here are: degree of familiarity with the task context, the degree of structure in the information to be communicated, the number of distinct referents to be encoded and temporal reference (Here-and-Now vs. There-and-Then).

Planning:

1. Length of planning time
2. Guided vs. Unguided planning
3. Form-focus vs. Meaning-focus

Most teaching research used ten minute planning time. Testing research uses lesser duration. Wigglesworth (1997) used 1 minute, Elder and Iwashita (2005) used 3 minutes, and Tavakoli and Skehan (2005) used 5 minutes. Mehnert (1998) studied the effect of varied planning times and found that longer planning time had more effect on performance. 

Guided planning asks learners to do particular things while planning, while unguided condition doesn't. Kawauchi (2005) studied three kinds of guided planning: writing what was planned, rehearsal of what is to say, and read a model of what to say. 

How does the above variables affect Fluency, Complexity and Accuracy?

Fluency

Operationalisation of Fluency: a. Measure of temporal aspects of fluency (number of words/syllables per minute), b. measure of repair phenomena (false starts, repetitions, reformulations). 

General finding in teaching studies is that strategic planning has a positive effect on fluency in both temporal and repair fluency aspects. But in testing, this is different. Elder and Iwashita (2005) found no effect, and Wigglesworth (2001) found negative effect. 

Second and foreign language learners experienced positive effects of planning on fluency.

Influence of proficiency level on the effect of planning on fluency was found to be varied. Wigglesworth (1997) found greater effects on the fluency of high proficiency test-takers. Kawauchi (2005) reports positive effect on low and high proficiency learners, but not on advanced learners (probably because they did not need planning time to speak well). Tavakoli and Skehan found that high proficient learners performed better than low proficient learners. Thus with the available data, no conclusion can be made on the effect of planning time for learners of different proficiency levels.

Learner's attitude towards the opportunity to plan found varied results in different studies. 

Learner's Memory played a role in fluency in some studies. Guara-Taveres (2008) found that learner's working memory and measures of fluency correlated well when planning time was available. 

Effect of Settings have clear conclusions. In laboratory and classroom settings, planning time has positive effects on performance. But in testing context, it seems to have lesser or no effect. 

With the available studies, one cannot make a conclusion regarding fluency about the participatory structure of tasks. Majority of studies involving dialogic tasks showed positive effect. Only two studies in monologic tasks did not show an effect. 

Task complexity was well studied in Foster and Skehan's studies. They used personal information, narrative and decision making tasks. The first was deemed easy because no external information was needed. The third was deemed more difficult because of the unfamiliarity of information to be communicated, and its lack of availability of structure. Mehnert (1998) found more effect on fluency in more structured tasks. Tavakoli and Skehan (2005) found more effect on fluency on structured tasks. In short, planning interacts with task complexity/difficulty, and the effect on fluency is more in case of less complex tasks. 

Length of Planning seems to have a clear effect on fluency. Mehnert (1998) found that longer the planning time, greater the effect. 

Type of Planning has an effect. Guided planning had more effect on fluency than unguided planning. But more studies are needed to find when guided planning works better than unguided planning- whether it depends on the nature of the task, proficiency of learners, etc.

Complexity

Operationalisation or measurement of complexity was done through: a. amount of subordination, number of different verb forms used, type-token ration and number of different word types. Results are more mixed than those of fluency studies, although there is plenty of evidence that strategic planning helps up complexity of production. 13 of the 19 studies reported here showed positive effects, while the rest (6) did not find any effect. Overall, strategic planning appears to have a greater effect on grammatical complexity. 

Both second and foreign language learners recorded benefit from planning on complexity. 

Proficiency: Advanced learners may not benefit from planning in terms of complexity (Kawauchi, 2005).

Working memory was found to be significantly related to complexity in planning group, not in no-planning group (Guara´-Tavares, 2008).

Complexity doesn't depend on laboratory/class setting. In testing, there is very less or no effect on complexity.

Task factors interact with planning time to have an effect on complexity, but 'how' is not clear. Most dialogic and monologic tasks showed a positive effect of planning on complexity. 

Task complexity: Foster and Skehan (1996) found that more complex decision making tasks did not have an effect on complexity while personal information and narrative tasks have higher grammatical richness with planning condition. But other studies give different results. More studies are needed to reach conclusions. 

Planning time variable: Mehnert's (1998) is the only study in planning time variable reported no effect on complexity in all of her different conditions. Different types of planning also did not have any effect on complexity. The degree of guidance might have an effect on the outcomes, however. 

Accuracy 

Accuracy also showed mixed results in various results. Thirteen of the studies found that planning enhanced accuracy but six reported no effect. 

Learner’s proficiency has an effect on accuracy. Kawauchi (2005) found less effect on advanced level learners than low proficiency learners. Learner proficiency needs to be controlled while investigating effects of planning on accuracy. 

Learners’ attitudes towards planning also have an effect on accuracy. 

Working memory wasn't found to be related to accuracy.

Task type is a potential influence factor. Interaction between task type and planning where accuracy is concerned. Foster and Skehan (1996) reported that planning led to greater accuracy in personal and decision making tasks (low and high difficulty tasks). No conclusion is really possible about how task type influences the effect that planning has on accuracy. How learners orientate to the task is also an important task. 

Type of planning failed to find effects in the studies listed here. Mochizuki, N. and L. Ortega. (2008) showed that when the guided planning is focused it can have an effect on accuracy. 

General Comments

Overall, strategic planning has clear effects on task performance. In the light of research, it is possible to say that conclusions about learner and task variables' influence on performance can be made. But we need more data and more studies that carefully control and manipulate these variables. 

Ellis questions: The three dimensions of performance we chose are not perfect descriptors of performance. Are ratings based on these these dimensions sensitive to the influence of strategic planning? This doubt is natural since no study reported an effect of planning on the ratings even when there were differences in discourse analysis results! It may thus seem that ratings are sensitive to influence of strategic planning on fluency in learning context, not in testing. 

Thus testing context seems insulated to the influence of strategic planning in terms of fluency. Only Wigglesworth (1997) showed that there is a significant effect (in discourse analysis). Why such insulation? Ellis speculates that it is because the test-taker knows that he/she is being tested, which leads to focus on accuracy at the expense of fluency and complexity. That is, testing context neutralizes the influence of planning otherwise available in learning contexts. Or may be the length of planning must be longer (althouth Mehnert (1998) proves this wrong). 

But in learning studies, strategic planning is found to have consistent influence on fluency throughout almost all studies. Prior planning of what to say saves on attentional resources enabling learner to speak fluently. Skehan proposes this trade-off between aspects of performance when learners focus on one aspect of performance. Depending on what is prioritized, the other aspects suffer. This explains the variable results of many studies. More studies are needed to say definite things about the influence of learner characteristics and task characteristics. 

Proficiency is one such variable that must be studied. Except in testing context, it has shown variations. 

Task variables like degree of structure need to be studied in detail. Existing studies have shown effects. But what is less clear is how task structure interacts with planning. 

Other considerations are: 

• Whether planning is guided or unguided.
• Whether focus is invited to form or meaning.

3. Studies in Within-task planning time

In fact there is only few studies that report this. Yuan and Ellis (2003) compared the performance of learners in pressurised (no planning) and unpressurised (within-task planning) planning time conditions. Learners with the within-task planning condition spoke longer. Fluency was not significantly different for both groups. Within-task planning resulted in greater syntactical complexity (without much differences in syntactical and lexical variety). It also generated more accurate speech in terms of error-free clauses and verbs. Surprisingly, there was no effect on fluency, probably due to the pressure to speak without prior planning. Thus the study concludes that when given ample time to plan, we can expect more accurate and complex speech in learning contexts. 

One problem with this kind of planning is that we do not know what the learners were doing during the planning stage. Other studies have shown that the effect of unpressurised planning is most beneficial in the initial stages of the task performance. 

The summary of another study by Ahmadian and Tavakoli (2010) can be accessed at this link. (Language Teaching Research, Vol. 15, Issue 1). 

Summary

Strategic planning is the most researched planning type so far. This preference is not theory-based. The other two types of planning must be examined with equal vigour. 

Rehearsal results in greater fluency and complexity. But these effects do not transfer to the performance of a new task unless there is some kind of additional intervention. That is, simple repetition of a task may not have a measurable impact on acquisition.

Strategic planning clearly benefits fluency. Results are mixed with complexity and accuracy. The possible trade-off between these two aspects must be the reason (i.e. learners will tend to prioritize either complexity or accuracy). Other variables which have an impact on the effect of strategic planning are the learners’proficiency (the effects are less evident in very advanced learners),the degree of structure of the information in the task and working memory. 

Within-task planning may benefit complexity and accuracy without having a detrimental effect on fluency.

Theoretical Perspectives

Theories of variability

Levelt's model of speaking

Skehan's theory

Robinson's theory

Levelt's Model 

There are three overlapping processes: conceptualisation, forumlation and articulation. The utterances can be monitored prior to and after the production. 

Two characteristics of speech production: a) controlled and automatic processing, b) incremental production. Conceptualiser and monitor operate under controlled processing. Formulator and articulator operate under automated conditions. However, the cases of native speakers and learners might be different in terms of formulation and articulation. 

Putting Levelt's model along with the limited attentional processes of learners explains many findings of the studies discussed above. The relationships between planning and aspects of performances can be understood with this theoretical background. 

• Rehearsal and strategic planning are likely to assist conceptualization and thus facilitate fluency
• They may have effects on formulation and articulation as well since the linguistic resources necessary are already accessed during conceptualisation. 
• Fluency oriented learners may not benefit in terms of complexity and accuracy. In other words, since only limited resources are available, focus on one aspects leads to problems in other aspects of performance. 
• Advanced learners who have lesser problems in formulation and articulation will have an effect on fluency, but much less on complexity and accuracy. 
• In unpressurised within-task planning, complexity and accuracy may increase due to benefits to formulation. 

These accounts are helpful in understanding how things work. But we also need to know how individual differences interact with different aspects of performance and variables we study. All learners do not engage in all the processes to the same extent. Orientation, working memory, language aptitude, willingness to communicate, and anxiety are a few such individual differences. 

A framework suggested by Ellis has four sets of variables. 

TTT

The model hypothesises that task and individual variables influence how learners plan and mediate the effect planning has on production. We need to answer the question 'how does planning assist development of fluency and acquisition of linguistic knowledge'. This has yet to be studied. Fluency development and acquisition must be two different phenomena. Like Skehan says, may be fluency is an outcome of development of exemplar-based system. So fluency can develop independent of acquisition. Rehearsal and strategic planning helps learners to develop exemplar-based system. Therefore, such planning helps build fluency. Within-task planning might influence automatization of grammatical knowledge. 

There are three senses of acquisition: a) acquisition of new linguistic features, b) restructuring of existing linguistic resources and c) development of greater control or accuracy over existing linguistic features. This understanding is necessary to theorize the relationship between planning and acquisition. The studies have showed that planning has very less influence on first kind of acquisition. Second and third kind of acquisition experiences effects of planning. Planning affects restructuring by its effect on complexity (Skehan, 1998). These assumptions are based on the condition that more complex production leads to acquisition. 

Limitations and Directions for Future Research

There is lack of information on what learners do during planning. 

Within-task planning and the combined effects of within-task and pre-task planning haven't been studied yet. 

Longitudinal study of the effects of planning has not been done yet. The maximum duration so far is ten weeks!

Studies listed did not collect baseline data of native speakers performing the same tasks. 

Stuidies did not give proficiency level data of the learners. 

No study has investigated the extended performance of learners on a task (not just the early stage of the task, but the entirety of the task performance). 

How individual learner factors affect performance is not studied yet.

References

Bygate, M. (1996). ‘Effects of task repetition: Appraising the developing language of learners’ in J. Willis and D. Willis (eds): Challenge and Change in Language Teaching. Heinemann.

Bygate, M. (2001). ‘Effects of task repetition on the structure and control of oral language’ in M. Bygate, P. Skehan, and M. Swain (eds): Researching Pedagogic Tasks, Second Language Learning, Teaching and Testing. Longman.

Elder, C. and N. Iwashita. (2005). ‘Planning for test performance: Does it make a difference?’ in R. Ellis (ed.): Planning and Task Performance in a Second Language. John Benjamins.

Ellis, R. (2005). ‘Planning and task-based research: theory and research’ in R. Ellis (ed.): Planning and Task-Performance in a Second Language. John Benjamins.

Foster, P. and P. Skehan. (1996). ‘The influence of planning on performance in task-based learning,’ Studies in Second Language Acquisition 18/3: 299–324.

Gass, S., A. Mackey, M. Fernandez and M. Alvarez-Torres. (1999). ‘The effects of task repetition on linguistic output,’ Language Learning 49: 549–80.

Kawauchi, C. (2005). ‘The effects of strategic planning on the oral narratives of learners with low and high intermediate proficiency’ in R. Ellis (ed.): Planning and Task-Performance in a Second Language. John Benjamins.

Mehnert, U. (1998). ‘The effects of different lengths of time for planning on second language performance,’ Studies in Second Language Acquisition 20: 52–83.

Mochizuki, N. and L. Ortega. (2008). ‘Balancing communication and grammar in beginninglevel foreign language classrooms: A study of guided planning and relativization,’ Language Teaching Research 12: 11–37.

Sheppard, C. 2006. The Effects of Instruction Directed at the Gaps Second Language Learners Noticed in their Oral Production. Unpublished PhD Thesis, University of Auckland.

Skehan, P. (1998). A Cognitive Approach to Language Learning. Oxford: Oxford University Press.

Skehan, P. and P. Foster. (1999). ‘The influence of task structure and processing conditions on narrative retellings,’ Language Learning 49/1:93–120.  

Tavakoli, P. and S. Skehan. (2005). ‘Strategic planning, task structure, and performance testing’ in R. Ellis (ed.): Planning and TaskPerformance in a Second Language. John Benjamins.

Wigglesworth, G. (1997). ‘An investigation of planning time and proficiency level on oral test discourse,’ Language Testing 14/1: 21–44.

Yuan, F. and R. Ellis. (2003). ‘The effects of pre-task and on-line planning on fluency,complexity and accuracy in L2 monologic oral production,’Applied Linguistics 24/1: 1–27.

Measurement of Performance in Task-Based Assessment

Introduction
Task-based assessment (TBA) uses tasks to gather information about the linguistic abilities of test-takers. It employs either performance-referenced or construct-based approach to do this. Performance-referenced TBA looks at target language use (TLU) tasks to identify what aspects of performance to be tested, so that the test is useful for its intended purposes. This is called work-sample approach. Criterion-based approach selects a theory of language learning and use to define the underlying language abilities to be measured in the test. Whatever is the approach, after the test is conducted, we need to interpret or assess the performance to get statements regarding the test-taker's abilities to predict future performance, or to take decisions. For this, we have three different ways.

1. Direct assessment of task outcomes
Direct assessment is useful for closed tasks. Closed tasks are tasks that result in a solution that is either right or wrong. Scoring is dichotomous- either right or wrong. Such criterion-referenced tests directly let us judge whether the test-taker has passed or failed in the test.
Advantages: There is very little subjectivity involved. The test result very clearly states the outcome. It is easy. It is quick.
Disadvantages: Direct tests need direct observation of task performance. That is, a rater must observe each test-taker individually. This is time consuming and almost impossible. In case of direct written tests, this problem can be easily solved by evaluating the test scripts later. Another disadvantage is the lack of clarity regarding such test's ability to measure language ability as against non-linguistic or general knowledge. This problem more real for direct performance-referenced tests where work sampling approach is used. Construct-based tests of the direct nature do not face this problem.

2. Discourse Analytic Methods
As the name indicated, this method analyses the discourse produced in the test. It counts specific linguistic features appearing in a task discourse. There are many ways one can do this. First is to look for the test-taker's linguistic competence by looking for complexity, fluency and accuracy in the discourse. Second is to look for sociolinguistic competence- like the use of different strategies for eliciting information from interlocutors. Third is to look for discourse competence- like appropriate use of connectors, topic changers, etc. Fourth is to look for strategic competence- like use of different strategies to negotiate meaning, build discourse, or to overcome breakdowns in communication.

Discourse analytic methods are generally objective. It can be considered direct measurement. Some do not consider this as direct since in the real world, we do not analyse discourse to see if someone has succeeded in communicating something.

This method is used widely in task-based research. It is a time consuming process since the discourse requires to be transcribed for analysis. Therefore, this is not used widely in task-based language teaching. But in research, this is a very useful tool. Most of the times, one uses discourse analysis along with external rating in order to compare the assessment afforded by both.

3. External Ratings
External ratings make use of an external rater/observer to assess performance. Judgment comes from this rater. But this is not direct assessment in the nature of judgment made. Here, judgment is made by the rater subjectively, while in direct assessment, judgment is objective since it comes as part of the performance itself.

The most common external rating type uses rating scales. Both performance-referenced and system-referenced tests make use of rating scales. What do rating scales do? They specify the ability/competency being measured, and provide different levels of performance as bands of performance. Usually this scale is on a spectrum ranging from 0 to 5. The highest level usually is native-like proficiency.

Another kind of scale used makes use of checklists. This checklist lists different abilities/aspects of performance. The rater observes the performance and checks off those that are present in performance.

In TBA, there are two ways of specifying competence.
1. In Behavioural terms. 2. In Linguistic terms. Choice of behavioural or linguistic specification depends on whether you want to know about the general language proficiency or specific language proficiency of the test-taker.

Behavioural specification of competence uses rating scales that provide us with levels of test-taker performance. For example: instructions are correctly given with clarity. Such scales are essentially task dependent. Every task needs its own rating scale.

Linguistic specification of competence is problematic. How do we specify competence? What aspects of linguistic competence must be measured? There are two choices a) holistic measure which looks at general language proficiency to fulfill functions required, without attention to specific linguistic features. b) analytic measure which looks for specific linguistic abilities in terms of the four language skills. Many tests uses both holistic and analytic measures. Analytic scale requires us to specify what aspect of performance are we looking at. These aspects need to be connected to an underlying theory of language. Skehan's framework suggest 'accuracy, fluency and complexity' as three dimensions. One can also make use of functional definitions of ability instead of this. The basic rule is to use a theory of communicative competence or performance to determine the competencies to be measured.

How does one determine the checklists or performance levels used in rating scales? McNamara proposes two approaches- theoretically driven and empirically driven. Theoretically driven approach uses context-driven descriptors. Criterion behaviours are specified. Performance is observed to see if criterion behaviour is satisfied. Here, descriptors are developed in terms of linguistic skill displayed. In empirically driven approach,  Rasch scaling procedure is used to statistically study the relationship between different items in a test and general abilities to be deduced to form a scale. Such scales are not very common since the process involved is complex. An alternative way is to develop scales using the insights received from discourse analysis.

Criterion level in a scale has to be defined. This is a problematic process. Most often, this decision is arbitrary or relative. For example, some universities choose IELTS score 6 as criterion for admission
criterion, while some others choose 7 or 7.5. Such decisions are not theoretically or empirically explained.

A very important aspect of rating scale is its interaction with the rater. This is another field of study altogether.

4. Self Assessment
Self assessment has many problems and advantages. Advantages are that they are easy, less time consuming, less expensive, helps learners to be self-regulated learners, develop reflective learning, etc. Disadvantages are about the reliability and validity of self assessment. How able are learners to judge language proficiency. Bachman has found that self assessment is more meaningful when questions are about actual performance or needs, rather than about linguistic abilities. According to some studies, it lacks predictive and concurrent validity (criterion-related validity). Reliability is high when it is about internal consistency. But it is not the same case with test-retest reliability.