Machine learning under felicity conditions: exploiting pedagogical behavior
In task instruction a knowledgeable and well-intentioned teacher will provide focusing information, highlight decision points, indicate structure, and so forth. Recognition of these communicative acts is critical for practical learning. This paper describes an extension to explanation-based learning that augments domain knowledge with knowledge of constraints on the structure of instructional discourse. VanLehn’s felicity conditions have formalized some of the constraints guiding a teacher during the presentation of examples. In this paper we weaken his condition that each lesson be an optimal set of examples, instead assuming that the teacher provides an explanation. Instruction is viewed as planned explanation, and plan recognition is applied to the problem at both domain and discourse levels, and extended to allow the learner to have incomplete knowledge. The model includes (a) a domain level plan recognizer and (b) a discourse level plan recognizer that cues (c) a third level of plan structure rewriting rules. Details are given of an example in which a robot apprentice is instructed in the building of arches. Tractable learning and instructability are our goals.
KeywordsDomain Knowledge Pedagogical Behavior Discourse Context Abstract Operator Explanatory Information
Unable to display preview. Download preview PDF.
- Cohen, P. & Levesque, H. (1990). Rational interaction as the basis for communication. In P. Cohen, J. Morgan, & M. Pollack, (Eds.), Intentions in Communication (pp. 221–256). Cambridge, MA: MIT Press.Google Scholar
- Cohen, P., Perrault, R., & Allen, J. (1981). Beyond question-answering. (Technical Report 4644.) Cambridge, MA: BBN Inc.Google Scholar
- Coulthard, R. M. (1970). An Empirical Investigation of Restricted & Elaborated Codes. PhD thesis, University of Birmingham.Google Scholar
- Fikes, R. E. (1972). Monitored execution of robot plans produced by STRIPS. In C.V. Freiman, (Eds.), Information Processing 71: Proceedings of the IFIP Congress 71, Vol. 1, Foundations and Systems, 23–28 August 1971, Ljubljana, Yugoslavia (pp. 189–194). Amsterdam, ND: North-Holland.Google Scholar
- Lewis, J. D. & MacDonald, B. A. (1992). Task learning from instruction: An application of discourse processing techniques to machine learning. Proceedings of Applications of AI X, 168–179.Google Scholar
- Littlestone, N. (1988). Learning quickly when irrelevant attributes abound: a new linear-threshold algorithm. Machine Learning, 2(4), 285–318.Google Scholar
- Pollack, M. (1986). Inferring Domain Plans in Question Answering. PhD thesis, University of Pennsylvania.Google Scholar
- Rivest, R. & Sloan, R. (1988). Learning complicated concepts reliably and usefully. AAAI-88, Proceedings of the Seventh National Conference on Artificial Intelligence, St. Paul Minnesota. Volume 2, 635–640.Google Scholar
- Sinclair, J. M. & Coulthard, R. M. (1975). Towards an Analysis of Discourse: The English Used by Teachers and Pupils. Oxford: Oxford University Press.Google Scholar
- VanLehn, K. (1983). Felicity conditions for human skill acquisition: validating an AI-based theory. (Interim Report P8300048). Palo Alto, CA: XEROX Palo Alto Research Centers.Google Scholar
- VanLehn, K. (1990). Mind Bugs: The Origins of Procedural Misconceptions. Cambridge, MA: MIT Press.Google Scholar
- VanLehn, K. (1991). Personal communication, October.Google Scholar