CAF

CAF — Complexity, Accuracy, and Fluency — is the dominant analytic frame for measuring L2 production in task-based research, oral testing, and writing assessment. The three dimensions partition learner output into how elaborate the language is (complexity), how target-like (accuracy), and how smoothly produced (fluency), giving researchers a multidimensional view of performance that single-construct proficiency scores cannot capture.

Origins

Skehan's Individual Differences in Second Language Learning (1989) introduced the three-way distinction in passing, but the framework took its working shape in Skehan (1996) "A framework for the implementation of task-based instruction" (Applied Linguistics 17/1) and Skehan's A Cognitive Approach to Language Learning (1998, Oxford University Press), which proposed the Limited Attentional Capacity model: working memory cannot service all three dimensions simultaneously, so attention given to one inevitably comes at the expense of the others. Foster & Skehan's (1996) "The Influence of Planning and Task Type on Second Language Performance" (Studies in Second Language Acquisition 18/3) gave the framework its first large empirical study, showing that pre-task planning lifted fluency and complexity but not, in linear fashion, accuracy. Ellis & Barkhuizen's Analysing Learner Language (2005, Oxford) consolidated the operational measures and made CAF the standard reporting template across SLA journals.

The trade-off vs cognition debate

CAF sits at the centre of a long-running theoretical disagreement. Skehan's Trade-Off Hypothesis takes attentional capacity to be a single, scarce pool: tasks that load conceptual difficulty force learners to sacrifice form for meaning, complexity for accuracy, or accuracy for fluency. Robinson's Cognition Hypothesis (Robinson 2001, 2003, 2005), built on the Triadic Componential Framework, counters that learners draw on multiple, non-competing attentional resource pools. Increasing resource-directing complexity (reasoning demands, here-and-now vs there-and-then) should push complexity and accuracy upward together, while resource-dispersing variables (planning time, prior knowledge) trade off in the way Skehan predicts. The two positions generate opposite predictions about how to sequence pedagogic tasks, and the empirical record across two decades remains genuinely mixed.

Operationalisations

Each dimension fragments into competing measures, which is the framework's main methodological headache.

Complexity splits into syntactic complexity (mean length of AS-unit or T-unit, clauses per AS-unit, subordination ratio) and lexical complexity, where lexical diversity is most often indexed by MTLD or D (vocd-D / HD-D) since the older Type-Token Ratio is sample-size dependent. AS-units (Analysis-of-Speech units, Foster, Tonkyn & Wigglesworth 2000) are standard for spoken data; T-units (Hunt 1965) for written.

Accuracy is typically operationalised as percentage of error-free clauses, errors per 100 words, or target-like use of a specific feature (for example obligatory plural marking). Holistic ratings appear too, but suffer from rater dependence.

Fluency decomposes further into speed fluency (pruned and unpruned speech rate, articulation rate), breakdown fluency (filled and unfilled pause counts and durations), and repair fluency (false starts, reformulations, self-corrections). For writing, fluency is words per minute of composing time or burst length between pauses.

Use in research and assessment

CAF anchors the dependent-variable side of task-based research: any study that manipulates task complexity, planning time, repetition, or interaction conditions reports CAF outcomes as standard. In oral testing, examiner band descriptors implicitly track CAF — IELTS Speaking's Lexical Resource and Grammatical Range and Accuracy bands map onto complexity and accuracy, with Fluency and Coherence mapping onto fluency. The framework also underwrites automated writing evaluation tools that pipe learner texts through measures like L2SCA (Lu 2010) for syntactic complexity and TAALES for lexical sophistication.

Critiques

Pallotti's (2009) "CAF: Defining, Refining and Differentiating Constructs" (Applied Linguistics 30/4) is the most cited critique. Pallotti argues that the framework suffers from definitional plurality — "complexity" alone has been instantiated as cognitive complexity, structural complexity, and developmental complexity, often without distinction — and that without construct unity, comparison across studies collapses. He pushes for a clearer separation between performance descriptors (CAF) and developmental claims, and argues that CAF needs a fourth dimension, adequacy, since a learner's response can be complex, accurate, and fluent yet still fail to meet the task's communicative requirement. The adequacy dimension has since been picked up in functional adequacy research (Kuiken & Vedder 2017, 2018).

A second critique targets the framework's psycholinguistic assumptions: if attentional resources are not in fact unitary, much of Skehan's predictive apparatus loses its grounding. A third targets measurement: subordination is a poor index of complexity at advanced levels, where coordination and clause-internal elaboration carry more of the load (Norris & Ortega 2009).

None of these critiques displace CAF in practice. They argue for cleaner definitions, better measures, and the recognition that no three (or four) numbers exhaust what learner production is doing.

References

• Ellis, R. & Barkhuizen, G. (2005). Analysing Learner Language. Oxford University Press. https://elt.oup.com/catalogue/items/global/linguistics/oxford_applied_linguistics/9780194316347/
• Foster, P. & Skehan, P. (1996). The influence of planning and task type on second language performance. Studies in Second Language Acquisition, 18(3), 299–323. https://www.cambridge.org/core/journals/studies-in-second-language-acquisition/article/influence-of-planning-and-task-type-on-second-language-performance/1C705F1926E1530572258D3477DD9034
• Foster, P., Tonkyn, A. & Wigglesworth, G. (2000). Measuring spoken language: A unit for all reasons. Applied Linguistics, 21(3), 354–375.
• Norris, J. & Ortega, L. (2009). Towards an organic approach to investigating CAF in instructed SLA. Applied Linguistics, 30(4), 555–578.
• Pallotti, G. (2009). CAF: Defining, refining and differentiating constructs. Applied Linguistics, 30(4), 590–601. https://academic.oup.com/applij/article-abstract/30/4/590/225725
• Robinson, P. (2001). Task complexity, cognitive resources, and syllabus design: A triadic framework for examining task influences on SLA. In P. Robinson (Ed.), Cognition and Second Language Instruction (pp. 287–318). Cambridge University Press.
• Skehan, P. (1989). Individual Differences in Second-Language Learning. Edward Arnold.
• Skehan, P. (1996). A framework for the implementation of task-based instruction. Applied Linguistics, 17(1), 38–62.
• Skehan, P. (1998). A Cognitive Approach to Language Learning. Oxford University Press.