Symbolic Machine Learning: A Different Answer to the Problem of the Acquisition of Lexical Knowledge from Corpora

Abstract

One relevant way to structure the domain of lexical knowledge (e.g. relations between lexical units) acquisition from corpora is to oppose numerical versus symbolic techniques. Numerical approaches of acquisition exploit the frequential aspect of data, have been widely used, and produce portable systems, but poor explanations of their results. Symbolic approaches exploit the structural aspect of data. Among them, the symbolic machine learning (ML) techniques can infer efficient and expressive patterns of a target relation from examples of elements that verify this relation. These methods are however far less known, and the aim of this paper is to point out their interest through the description of one precise experiment. To remove their supervised characteristic, and instead of opposing them to numerical approaches, we finally show that it is possible to combine one symbolic ML technique to one numerical one, and keep advantages of both (meaningful patterns, efficient extraction, portability).

Keywords

• corpus-based lexical knowledge acquisition
• symbolic machine learning
• inductive logic programming
• numerical approaches
• semi-supervised machine learning