Syntactic Pattern Recognition from Observations: a Hybrid Technique

This paper presents a novel technique for automated learning from observations. The technique arranges in a row four traditional pattern recognition approaches (numeric, logic, statistical and finally syntactic) within a unifying framework. Each processing step is conceived as a transformation of the input dataset from one state to another. The proposed technique considers measurable observations as inputs and produces a set of formal rules, i.e., a grammar, as final output. To this end, a four-state grammar induction process is described in detail by means of a step-by-step example. As a proof-of-concept for the feasibility of the proposal, references to early experimental validations are given. Finally, possible comparison with other well-known approaches are discussed.