Abstract: The origin for the classification of analogy is a comparison procedure we use to make a conclusion about similarity between the source and the target. Whenever the comparison procedure is not clarified, the analogy remains ambiguous. But if this procedure is formalized the analogy may allow to formulate and check conditions of its (still partial, but often very high) confidence. Below I discuss some ideas, which grounded certain algorithms allowed to calculate different kinds of similarity between texts, in particular of natural languages. The literal similarity of two words, as well as the lexical similarity of two given texts, can be estimated using so called “Theory of Finite Sequences Similarity”. The structural similarity can be measured by fixing certain name groups in the source text and check the cohesion (proximity) of names belonging to corresponding groups in the target. Special logical formalism called “The Language of Ternary Description” can provide good templates for source text structuring when comparing texts of natural languages. It was demonstrated statistically that algorithms proposed for texts’ analogy estimation provide “practically trustworthy” conclusions in the knowledge testing area. I argue also that high degree of confidence for that type of analogy is connected with the background of analogy (the notion which was discussed by G. Polya) though that background need not to be evidently formalized.

Keywords: analogy, texts’ similarity, knowledge testing, free answers

ACM Classification Keywords: I.2.6 Artificial Intelligence – Learning – Analogies; K.3.1 Computers and Education - Computer Uses in Education

Link:

ANALOGIES BETWEEN TEXTS: MATHEMATICAL MODELS AND APPLICATIONS IN COMPUTER-ASSISTED KNOWLEDGE TESTING

Leonid Leonenko

http://foibg.com/ibs_isc/ibs-19/ibs-19-p14.pdf