Chi, M. T. H., Bassok, M., Lewis, M. W., Reimann, P., & Glaser, R. (1989). Self-explanations: How students study and use examples in learning to solve problems. Cognitive Science, 13(2), 145–182

As far as I know, this is the paper which introduced the term Self-explanation. It’s introduced as a means to study how students construct understandings while reading texts—in particular, worked examples, where lots of inferences are necessary to understand much of the implicit explanation. The authors hypothesize that differences in problem-solving capacity may arise from differences in engagement with the text; they use self-explanation to evaluate that theory.

In their small (N=8) experiment, students study classical mechanics from a text. They answer declarative and qualitative questions about the material, repeating with help as necessary until they can answer every question. Then the students study worked examples while making self-explanations. Finally, the students solve isomorphic and far transfer problems involving the same material.

The students are separated into Good and Poor students by their performance on the problems (82% vs 46% averages). Observations about these groups:

• Good students generate far more idea statements (52 vs 18) and spend far more time (13 vs 7 min) when studying examples, but similar amount of speech and timings when solving problems.
• Of those statements, many more actually relate to the physics content for Good students.
• Good and Poor students performed ~equally on declarative/qualitative questions, but Good students articulated additional components of Newton’s Laws (missing from their earlier explanations) when explaining the worked examples.
• Good students were much more likely to note comprehension failures
• When solving problems, Good students referred to strategic snippets of example problems to compare or map; Poor students re-read example problems from the first line, looking for solutions they could co-opt.
  • This isn’t just a case of just-in-time processing: Poor students verbalized far fewer explanations while re-reading in this way than Good students had in their prior study session.
  • (This behavior is pretty familiar from my university experience…)

The authors suggest that self-explanations help problem solving by creating ACT-R-style production rules (“inference rules”) which allow the student to know the conditions where certain actions or ideas should be applied.

This paper doesn’t yet demonstrate the Self-explanation effect—just a correlation that more self-explanation corresponds with more understanding.

Q. How do the authors’ assumptions about bottlenecks in learning processes differ from John Anderson’s?
A. Anderson’s theories focus on the problem of making encoded knowledge become “smooth, fast, skillful”. The authors think that completeness of initial encoding is a more important issue.

Q. Why are self-explanations originally invoked in this study?
A. As an instrument to observe differences in encoding processes in students.

Q. How were the Good and Poor groups defined?
A. Post-hoc separated by their strikingly different performance in problem-solving.

Q. How did the Good and Poor groups differ in their performance on the declarative and qualitative questions?
A. They didn’t!

Q. In what sense did self-explanation during worked examples seem to make the Good students’ understanding of Newton’s Laws more complete?
A. They articulated principles of the Laws during their self-explanations which they’d omitted during their earlier explanations.

Q. How did the Good and Poor groups differ in their comprehension monitoring (two details)?
A. Good students were much more likely to mention comprehension failures; Poor students’ comprehension failures were only at quantitative (not conceptual) loci.

Q. Why do the authors conjecture that detecting comprehension failures is important?
A. Verbalizations of confusion are often followed by explanations (i.e. they instigate inference, deeper processing)

Q. How did Good and Poor students differ in how they referred to examples when problem-solving?
A. Good students referred to strategic lines of example problems to compare or map; Poor students re-read example problems from the first line, looking for solutions they could co-opt.