Scheffler, P., & Cinciała, M. (2010). Explicit grammar rules and L2 acquisition.ELT journal, ccq019.
The experiment in this study consists two parts. The first part is a production test, during which the experimenter collected data on participants’ target-like usage. The second part asks participants to pick a correct sentence out of 2 given choices, and say why the other one is not correct.
The two parts of this experiment require two different cognitive activities. The production test requires one to recall L2 knowledge; the pick-the-correct-sentence task requires only the recognition of L2 structures. Apparently, it is easy to recognize things than to recall things, because “recall” involves two stages: memory retrieval and information processing, while recognition requires only the latter stage. That is probably why most people achieved 100% accuracy in the “pick-the-correct-sentence” task. The researchers made things even easier by contrasting the two sentences with only one grammar point, which practically gives away the answer as to why one is correct and the other is not. That is, the participants might be simply analyzing how the two sentences differ from one another, rather than what the grammar rules are. Or, maybe the apparent difference between the two sentences facilitated the retrieval of L2 knowledge. In other words, the correct production during the interview might be triggered by implicit knowledge, while the correct metalinguistic responses were only stimulated during the pick-the-correct-sentence task by the apparent difference between the 2 given sentences.
I do not quite understand the rationale behind picking grammar points with more than 75% productive accuracy. Could correct L2 knowledge be associated with production errors? That is, could one understand the rules but still make errors in production? In order to show explicit grammar instruction is not counterproductive, the research should show that lower productive accuracy is associated with wrong metalinguistic knowledge, or the absence of metalinguistic knowledge. For example, the researchers could choose the ones with less than 25% accuracy in production, and see how participants perform in the “pick-the-correct-sentence” task. Maybe the participants can pick the right one if given only two possible choices, and they may even provide correct L2 rules even though they cannot apply the rules in their production.