(Pre-)Linguistic Problem Solving Based on Dynamic Semantics.

Based on the famous monkey-banana problem, we propose a general solution mechanism for alloplastic coping using dynamic and database semantics, dynamic programming and minimalist grammar as a pre-linguistic device. We show that the generation of a solution plan (i.e. a program) for a given problem - “fetch the banana from the ceiling” - involves solving the frame problem of artificial intelligence in terms of generalization and unification operators, where the problem space is represented by locally interpreted feature-value relations. Using dynamic programming for the backward planning, a sequence of mentally represented instructions is obtained under the constraint that antecedent and consequent states of successive actions must be compatible to each other. These conditions can be naturally represented by the feature-matching mechanism of minimalist grammars. We discuss the crucial difference of this kind of pre-linguistic inference with proper linguistic derivations.