[ CogSci Summaries home | UP | email ]

Falkenhainer, B. (1990) A unified approach to explanation and theory formation. In Shrager, J. & Langley, P. (eds.) Computational Models of Scientific Discovery and Theory Formation. Morgan Kaufman: San Meteo, CA. pp 157--196.

  ALTauthor = 	 {Falkenhainer, B.},
  ALTeditor = 	 {Shrager, J. and Langley, P.},
  title = 	 {A unified approach to explanation and theory formation},
  chapter = 	 {6},
  publisher = 	 {Morgan Kaufman},
  year = 	 {1990},
  OPTpages = 	 {157--196},

Author of the summary: Jim Davies, 2001, jim@jimdavies.org

Cite this paper for:

SYSTEM: Phineas

Explanatory similarity is the common core of explanation and analogy. [p158]

Abduction: inference to the best explanation.

The search for similarity between the situation to be explained and some understood phenomenon suffices as the process model for all explanation tasks. [p161] Deduction: explain with a complete match with features in memory
Assumption: Assume some facts in addition to partial match
Generalization: uses knowledge-based refinement to assume the truth of a higher level fact above a false fact that's getting in the way, when a sibling fact will work. Requires matches between features having a close generalization.
Analogy: a range of matches between different features and scenarios

Note that all involve matching to things in memory. The distinction between deduction and analogy is based on how well existing knowledge supports the explanation. [p164]

envisionment: description of the possible behaviors generated by a qualitative simulator. [p164] It describes states and transitions between them. States represent an interval of time where nothing (qualitatively) is changing.

Behaviors are indexed in memory with behavioral abstractions. [167]

A theory consists of entities, behaviors, and relevent facts.[168]

The steps of Phineas: access, mapping/transfer, qualitative simulation, and revision. For details see (Falkenhainer 1988).


Phineas searches memory for things under the same behavioral abstractions. It's further refined by matching behaviors and structure. SME does this. Thus, it's done based first on behavioral similarity. Each behavioral state knows what was relevent for it.

Skolem object: an unknown object. In this case, it's what must map to something, even though we haven't found out what it is yet. [171] Caloric is a skolem object that's generated. [p174]

Summary author's notes:

Back to the Cognitive Science Summaries homepage
Cognitive Science Summaries Webmaster:
JimDavies (jim@jimdavies.org)
Last modified: Thu Apr 15 11:07:19 EDT 1999