MDL-based Acquisition of Substitutability Relationships between Discourse Connectives

Ben Hutchinson

Abstract: Knowledge of which lexical items convey the same meaning in a given context is important for many Natural Language Processing tasks. This paper concerns the substitutability of discourse connectives in particular. This paper proposes a datadriven method based on a Minimum Description Length (MDL) criterion for automatically learning substitutability of connectives. The method is shown to outperform two baseline classifiers.