Analogies Between Sentences: Theoretical Aspects - Preliminary Experiments

Analogical proportions hold between 4 items a, b, c, d insofar as we can consider that “a is to b c d”. Such are supposed obey postulates, from which one derive Boolean or numerical models relate vector-based representations of making a proportion. One basic postulate the preservation proportion by permuting central elements and c. However this becomes debatable in many cases when words sentences. This paper proposes weaker set postulates based on internal reversal, new derived. The system used extend finite examples machine learning perspective. By embedding whole sentence into real-valued vector space, tested potential these for classifying analogical sentences valid non-valid proportions. It advocated identifying may be interest especially checking discourse coherence, question-answering, argumentation computational creativity. proposed theoretical setting backed with promising preliminary experimental results also suggests possibility crossing an ontology-based representation words. hybrid approach might provide some insights automatically extract natural language corpora.