Text Coherence & Self-Explanation III
In the first post, we looked at how Ainsworth and Burcham defined the concepts they were working with--text coherence and self-explanation. Text coherence was defined, in general, as "the extent to which the relationships between the ideas in the text are made explicit" and self-explanation was defined as "additional knowledge generated by learners that states something beyond the information they are given to study." In addition, we learned that both text coherence and self-explanation can produce gains in students' learning from text. But they seemed to do so in opposing ways: Whereas text coherence advantages learners by "repairing" (i.e., removing) inferences, self-explanation often produces gains even when—and perhaps especially when—text remains minimally coherent.
For their experiment, four groups of undergraduates were created: (1) a group who was assigned "maximally coherent" text and received self-explanation training, (2) a group who was assigned maximally coherent text and received no self-explanation training, (3) a group who was assigned "minimally coherent" text and received self-explanation training, and (4) a group who was assigned minimally coherent text and received no self-explanation training. Each group was given a three-section pretest and a matching three-section posttest, which assessed students' knowledge of the "explicit propositions in the text," or textbase. Participants were also given (at posttest only) "implicit questions" and "knowledge inference questions" (so, a two-section additional posttest), both of which assessed students' situation models.
Two hypotheses were proposed:
(a) The minimal text condition when accompanied by self-explanation training will present the optimal conditions for learning. Minimal text is hypothesized to increase self-explaining, and self-explanation is known to improve learning. Consequently, low knowledge learners who self-explain will not only be able to overcome the limitations of less coherence but will actively benefit from it as they will have a greater chance to engage in an effective learning strategy.
(b) Maximally coherence [sic] text accompanied by self-explanation will present the optimal condition for learning. Although maximal text is hypothesized to result in less self-explanation than minimal text, when learners do self-explain they will achieve the benefits of both text coherence and self-explanation.
If we stop here, we would be justified in concluding that (a) was the winning hypothesis here. It would seem that self-explanation has a more robust positive effect on learning outcomes than does text coherence. And since the literature tells us that minimally coherent text produces a greater number of self-explanations than does maximally coherent text, minimizing text coherence is desirable for improving learning.
Luckily, Ainsworth and Burcham went further. They coded the types of self-explanations made by participants and analyzed each as it correlated with posttest scores. While they did find that students who read minimally coherent text produced significantly more self-explanations, they also noted this:
Whilst using a self-explanation strategy resulted in an increase in post-test scores for the self-explanations conditions compared to non self-explanation controls, there was no signficant correlation within the self-explanation groups between overall amount of self-explanation and subsequent post-test performance. Rather, results suggest that it is specific types of self-explanations that better predict subsequent test scores.
Now, each of those correlations seems pretty ridiculous. They all seem to point in one way or another to the completely unsurprising conclusion that understanding a text pretty well correlates highly with doing well on assessments about the text.
What is interesting, however, is the researchers' observation that the surplus of self-explanations in the "minimal" groups could be accounted for primarily by three other types of self-explanation, none of which, in and of themselves, showed a signficant positive correlation with total posttest scores: (1) goal-driven explanations ("an explanation that inferred a goal to a particular structure or action"), (2) elaborative explanations ("inferr[ing] information from the sentence in an elaborated manner"), and (3) false self-explanations (self-explanations that were inaccurate).
To put this in perspective, there were only two other types of "self-explanation" coded that I did not mention here. Out of the remaining six, three showed no significant positive correlations with posttest scores (or, in the case of false self-explanations, a significant negative correlation), yet those were the self-explanations that primarily accounted for the significant difference between the minimal and maximal groups.
Or, to put it much more simply, the minimal groups had significantly more self-explanations, but those self-explanations were, in general, either ineffective at raising posttest scores or actually harmful to those scores. It is possible that the significant positive main effect for self-explanation in the study could, in fact, have been greatly helped along by the better self-explanations present in the maximal groups. All of this leads to this conclusion from the researchers:
This study suggests that rather than designing material, which, by its poverty of coherence, will drive novice learners to engage in sense-making activities in order to achieve understanding, we should design well-structured, coherent material and then encourage learners to actively engage with the material by using an effective learning strategy.
Labels: research