Fluency

Fluency of performance is a hallmark of robust learning. A perceived contrast between learning activities that isolate knowledge components and those that bring them together in authentic performance has been at the heart of non-productive arguments in education, including mathematics, reading and second language acquisition. The research opportunity is to demonstrate the value of a more integrated approach to learning that links knowledge component learning, particularly achieving feature validity, and fluency, achieving knowledge component strength. Second language learning provides an excellent domain to explore fluency because fluency is such a clear instructional objective of language learning, whereas there is debate in mathematics education regarding the role of fluency and basic skills. Moreover, some of our research benefits from the Unified Competition Model, which was developed by one of our team (MacWhinney & Bates, 1989; MacWhinney, 2005). This is one of the most fully developed models in the field of second language processing, and it appears to have novel, unexplored applications in other domains as well. A key goal for the second year is to encourage fluency research in a math or science domain. We intend that such studies will draw on results and emerging understanding from our language studies.

A key hypothesis is that fluency is the outcome of learning procedures that both explicitly support the acquisition of knowledge components, their feature validity and strength particularly as they are integrated in practice with other components. This hypothesis guides our approach to certain difficult and contentious issues, like the effectiveness of explicit vs. implicit learning (DeKeyser, 2003), the relationship between production and comprehension (e.g. Izumi, 2003; VanPatten, 2002) and the nature of effective practice (DeKeyser, in press).

The fluency cluster combines researchers interested in second language instruction with researchers interested in the fundamental architecture of human memory (where "automaticy" is very related or identical concept). Fluency is intimately associated with repetition and context effects, which are key concepts in human memory research. Designing instruction that increases fluency translates into manipulating the encoding context across repetitions to increase feature validity and thus, the probability of retrieval in a new context. But what is context? How do the students' innate linguistic predilections and their first language affect how they perceive context and even the to-be-remembered item itself? The two disciplines are essentially approaching the same questions from very different perspectives, so by bringing them together in the fluency cluster, we expect to foment bi-directional payoffs.

One group of studies within the fluency cluster test the hypothesis that the addition of explicit instruction (descriptions) in addition to implicit learning opportunities (examples and practice) produces more robust learning. The studies are:

• If the knowledge component to be learned is a Chinese character composed of radicals, and one radical is semantically related to the meaning of the character (Liu & Perfetti, study 3) or phonologically related to the pronunciation of the character (Liu & Perfetti, study 4), is it better to point this out explicitly to the student?
• If the knowledge components to be learned are two similar French grammatical constructions (Tokowicz, study 4) or two French words that map onto a single English (L1) linguistic unit (Tokowicz, study 5), is it better to explicitly distinguish the two while teaching them together, or to teach them at different times without pointing out their relationship?
• If the knowledge component to be learned is the meaning of a novel English word as it is applied in a specific text, do ESL students learn more robustly if the instruction just presents the word in bold to draw attention to it, or also provides a definition of the word as well? (Eskenazi, Juffs, et al, study 1).

These studies all vary the explicitness of the instruction while controlling for the unique knowledge events students' experience. That is, they provide a knowledge event for each knowledge component in both the implicit and explicit conditions, but in the explicit instruction condition, they also present a description of the knowledge component.

A causal theory that predicts the results of these experiments starts by noting that the explicit condition is intended to force the learner's attention to a specific component of a complex stimulus with lots of features. If the student attends to the explicit instruction, the prediction is that the critical feature (the instructed feature) of the stimulus becomes part of the learner's encoding of the stimulus, thus increasing feature validity.

These considerations lead to certain predictions:

• If the student is unable to infer the knowledge component and critical feature(s) from context, then explicit instruction is necessary at least once and possibly more times per component.
• The student must eventually be able to apply the knowledge component alone or to infer it without the aid of the explicit instruction. This associates the appropriate features with the knowledge component.
• If a knowledge component is instructed explicitly (i.e., the students attends to the depiction of it), then an implicit instruction episode should follow soon thereafter. The retrieval cues of the implicit context do not match those of the explicit one, so successful retrieval depends on the strength of the knowledge component, which decays rapidly.

The general prediction is that for knowledge that is not readily inferred from context (implicitly), the best training schedule is a few rounds of explicit instruction followed almost immediately by a few rounds of implicit instruction, followed by less frequent implicit instruction.

These studies illustrate how processes being explored in other research clusters may be relevant, in this, the co-training processes. An alternative explanation for why adding explicit instruction to implicit instruction may improve learning is that having both implicit and explicit sources affords co-training (in the version generalized from the original co-training theorem). The relevant co-training notion is that input sources that yield uncorrelated errors will enhance learning when used together in co-training. In these studies, learners' processing of explicit descriptions may lead to interpretation errors that are uncorrelated with the errors generated in processing implicit examples and practice opportunities. Thus, a learner may be able to engage in sense making by disambiguating their interpretation of one form of instruction using the other form.

A second group of studies in the fluency cluster focuses on the interaction of prior knowledge (and especially, prior perceptual skills) with instruction. The key idea is that fluency creates cognitive headroom, that is, when a knowledge component is strong, its retrieval is highly likely, so the student does not have to cope with retrieval failure or delay when that knowledge component is being used while learning a new knowledge component. For instance, if the instruction is illustrating a new grammar construction and the student cannot remember what one of the words in the example means, and then acquisition of the new grammar construction is harmed. It can be harmed either because the student fails to correctly comprehend the grammatical knowledge component being taught, or the student struggles so much when doing so (e.g., they have to look up the unknown word in the dictionary) that the cues associated with the new grammatical knowledge component are not appropriate for retrieval. The current studies in this group are:

• Extensive practice on word dictation (given a French word orally, type it, spelling it correctly) may increase both the acquisition of other aspects of the word (e.g., its meaning) and of other words and grammatical constructions. (MacWhinney, study 1).
• Drill on two difficult grammatical aspects of French (gender and conjugation) may increase learning of related grammatical constructions, and perhaps even words (MacWhinney, study 2).
• For cued vocabulary training (French word + cue ' definition in English), strong associative cues work better than flashcards but flashcards work better than weak associative cues.
• Pictographic Chinese characters can be learned faster than non-pictographic ones (Liu & Perfetti, study 5)
• Spoken Chinese includes sounds (called "tones") that are not part of English. The lack of prior training on recognizing these sounds makes it difficult to learn characters that depend crucially on them. However, if English speakers are trained using artificial and natural sounds to discriminate among tones, then character learning is faster (Wang & Perfetti, Studies 1, 2 and 3)

For all the studies above, the increased strength of relevant prior knowledge should cause, according to the theory, more robust learning of the knowledge components that involve them. Many of these predictions are consistent with MacWhinney's (2005) Unified Competition Model. This model holds that fluency is best established through practice that continually challenges the student to retrieve and produce forms that are neither impossible to retrieve nor trivially easy to produce. This work echoes both earlier suggestions from Pimsleur {} regarding optimal spacing in practice spacing and recent work by Pavlik, Pashler, and Rovee-Collier. MacWhinney furthers argues that these effects represent the impact of neural circuits for language learning that require resonance between the hippocampus, the anterior cingulate, and at least two cortical areas.

However, even in second language acquisition, sense-making is just as critical as foundational skill building. For instance, one PSLC study tests the hypothesis that making learning easy does not always pay off in the long run:

• Although some French words have English synonyms, others do not. Some English words are related to them, but not complete synonyms. Training with synonyms leads to initially rapid learning, as predicted by the relevance of prior knowledge. However, when non-synonymous words are introduced, students infer the wrong knowledge component-they think the word is synonymous with an English word when it is not. Thus, they learn the wrong meaning. (Tokowicz, Study 1).

This study pits fluency against sense making. That is, a practice that is good for fluency may be bad for sense making. This study illustrates how the classification of studies into foundational skills and sense making is not intended to exclude studies that address both.

The last group of studies in the fluency cluster investigates other factors that could influence the robustness:

• When vocabulary words are learned from a list, sorting the list by semantic categories creates an encoding context that is less likely to match retrieval contexts than sorting the list randomly (Tokowicz, study 2)
• When vocabulary words for concrete objects are learned by presenting pictures of the objects, the most robust learning occurs when the images are presented upside down (because this lengthens the amount of time the vocabulary pair is in attention) as opposed to right-side up or as English words instead of images (Tokowicz, study 3)

In short, fluency studies address how the conditions that surround a knowledge event affect the ability of the student to accurately (feature validity) and quickly (strength) recall the knowledge later. Currently, all the fluency studies involve second language learning, in part because words and grammatical structures make such good stimuli for these kinds of studies. However, second language courses also value fluency highly, so in vivo experimentation on fluency is particularly welcome in them. We are going to make a special solicitation for project plans for fluency studies in math or science in Year 2.