Monday, June 30, 2014

Taxonomies for teaching and learning

(This post is joint with Pat Shafto and Noah Goodman; Pat is the first author and you can also find it on his blog).

Earlier this year, we read an article in the journal journal Behavioral and Brain Sciences (BBS) by Kline, entitled How to learn about teaching: An evolutionary framework for the study of teaching behavior in humans and other animals. The article starts out from the premise that there are major debates about what constitutes teaching behavior, both in human communities and in ethological analyses of non-human behavior. Kline then proposes a functionalist definition of teaching, that teaching is "behavior that evolved to facilitate learning in others," and outlines a taxonomy of teaching behaviors that includes:
  • Teaching by social tolerance, where you let someone watch you do something;
  • Teaching by opportunity provisioning, where you create opportunities for someone to try something;
  • Teaching by stimulus or local enhancement, where you highlight the relevant aspects of a problem for a learner;
  • Teaching by evaluative feedback, which is what it sounds like; and
  • Direct active teaching, which is teaching someone verbally or by demonstration.
We were interested in this taxonomy because it intersects with work we've done on understanding teaching in the context of the inferences that learners make. In addition, Kline didn't seem to have considered things from the perspective of the learner, which is what we thought made the most sense in our work – since adaptive benefits of teaching typically accrue to the learner, not the teacher. We wrote a proposal to comment on the Kline piece (BBS solicits such proposals), but our commentary was rejected. So we're posting it here.

In brief, we argue that evolutionary benefits of teaching are driven by the benefits to learners. Thus, an evolutionary taxonomy should derive from the inferential affordances that teaching allows for learners: what aspects of the input they can learn from, what they can learn, and hence what the consequences are for their overall fitness. In our work, we have outlined a taxonomy of social learning that distinguishes three levels of learning based on the inferences that can be made in different teaching situations:
  • Learning from physical evidence, where the learner cannot make any inference stronger than that a particular action causes a particular result;
  • Learning from rational action, where the learner can make the inference that a particular action is the best one to take in order to obtain a particular result, modulo constraints on the actor's action; and
  • Learning from communicative action, where the learner can infer that a teacher chose this example because it is the best or maximally useful example for them to learn from.
Critically, our work provides a formal account of these inferences such that their assumptions and predictions are explicit (Shafto, Goodman, & Frank, 2012).  Our taxonomy distinguishes between cases where an individual chooses information intended to facilitate another’s learning (social goal-directed action in our terminology, or “direct active teaching” in Kline’s taxonomy) and cases where an individual engages in merely goal-directed behavior and is observed by a learner (“teaching by social tolerance” in Kline’s taxonomy). We've found that this framework lines up nicely with a number of results on teaching and learning in developmental psychology, including "rational imitation" findings and Pat's work on the "double-edged sword of pedagogy" as well as other empirical data (Goodman et al., 2009; Shafto, et al., 2014)

The remaining distinctions proposed by Kline – teaching by opportunity provisioning, teaching by stimulus enhancement, and teaching by evaluative feedback – also fit neatly into our framework for social learning. Opportunity provisioning is a case of non-social learning, where the possibilities have been reduced to facilitate learning. Stimulus enhancement is a form of social-goal directed learning where the informant chooses information to facilitate learning (much as in direct active teaching). Teaching by evaluative feedback is a classic form of non-social learning known as supervised learning.

On our account, these distinctions correspond to qualitatively different inferential opportunities for the learner. As such, the different cases have different evolutionary implications – through qualitative differences in knowledge propagation within groups over time. If the goal is to understand teaching as an adaption, we argue that it is critical to analyze social learning situations in terms of the differential learning opportunities that they provide, and any taxonomy of teaching or social learning must distinguish among these possibilities by focusing on the inferential implications for learners, not just through characterization of the circumstances of teaching.


Bonawitz, E. B., Shafto, P., Gweon, H., Goodman, N. D., Spelke, E. & Schulz, L. (2011). The double-edged sword of pedagogy: Instruction affects spontaneous exploration and discovery. Cognition, 120, 322-330.

Goodman, N. D., Baker, C. L., and Tenenbaum, J. B. (2009). Cause and intent: Social reasoning in causal learning. Proceedings of the Thirty-First Annual Conference of the Cognitive Science Society.

Shafto, P., Goodman, N. D., and Griffiths, T. L. (2014). Rational reasoning in pedagogical contexts. Cognitive Psychology.

Shafto, P., Goodman, N. D. & Frank, M. C. (2012). Learning from others: The consequences of psychological reasoning for human learning. Perspectives on Psychological Science, 7, 341-351. Kline, M. (2014). How to learn about teaching: An evolutionary framework for the study of teaching behavior in humans and other animals. Behavioral and Brain Sciences, 1-70 DOI: 10.1017/S0140525X14000090

No comments:

Post a Comment