Evaluation of analogies through Kolmogorov complexity Meriam Bayoudh, Henri Prade, Gilles Richard Abstract Recognizing analogies and/or making analogies are largely considered to be at the core of the human cognition process, allowing the paradigm shift essential to creativity. Analogies (or analogical proportions) like “2 is to 4 as 5 is to 10” or “read is to reader as lecture is to lecturer” can be precisely understood within a sim- ple algebraic framework. In that case, the interpretation of the analogical ratio fits with the expected logical properties (symmetry, central permutation, etc.). When it comes to concepts expressed in natural language as “engine is to car as heart is to human” or “wine is to France as beer is to England”, such an algebraic framework is missing. In that context and in order to assess a numerical value to the analogical ratio, we start from the works of Kolmogorov on complexity theory and we present a new computational approach to define, and ultimately to try to detect, analogical proportions in natural language. We investigate the link between a purely abstract view of analogical proportion and our definitions based on Kolmogorov theory. De- spite the apparent difference between a purely abstract view of analogical proportion and the complexity based definition, it is quite clear that the two viewpoints can be related. Finally, this theory is used as a backbone to experiment a classifier of natu- ral language analogies whose results are consistent with the abstract setting, and are compared with another approach to a similar problem. Meriam Bayoudh IRIT, Universit´ e Paul Sabatier, 31062 Toulouse Cedex 09, France e-mail: bayoudh@irit.fr, Henri Prade IRIT, Universit´ e Paul Sabatier, 31062 Toulouse Cedex 09, France e-mail: prade@irit.fr, Gilles Richard British Institute of Technology and E-commerce, Avicenna House 258-262 Romford Road London E7 9HZ, e-mail: grichard@bite.ac.uk