The acquisition of unrestricted feature-based conceptual representations from corpora

Methods for the automatic acquisition of feature-based conceptual representations from text corpora have the potential to offer valuable support for theoretical research on conceptual representation. A major challenge for automatic feature aquisition is the unconstrained nature of the concept-relation-feature triples occurring in human-generated norms (e.g. flute produce sound, deer have antlers). Many existing methods focus on concept-feature tuples (e.g. flute–sound) or on triples involving specific relations only (e.g. is-a or part-of relations). We have investigated the challenges that need to be addressed in both methodology and evaluation when moving towards the acquisition of more comprehensive conceptual representations from corpora. In particular, we investigated the usefulness of three types of knowledge in guiding the extraction process: encyclopedic, syntactic and semantic. We introduce a novel method for large-scale feature extraction which involves extracting candidate triples consisting of concept, relation and feature terms (e.g., flute produce sound) from corpus data parsed for grammatical dependencies. The candidate triples are then re-weighted on the basis of conditional probabilities calculated from a class-based semantic analysis of existing property norms. We apply this method to an automatically parsed Wikipedia corpus which includes encyclopedic information and evaluate its accuracy using a number of different methods: direct evaluation against the McRae norms in terms of feature types and frequencies, human evaluation, and a novel evaluation in terms of conceptual structure variables. Our investigation highlights a number of issues which require addressing in both methodology and evaluation when aiming to improve the accuracy of unconstrained feature extraction further