Catalogue of Artificial Intelligence Techniques
Categories: Natural Language
Author(s): Karen Sp\"arck Jones
An approach to language understanding most fully developed by Wilks. Preference semantics recognises that, in general, semantic constraints on word combinations cannot be absolute, as this would be incompatible with the creativity of language. The semantic patterns embodied, e.g., in Case Frames, and expressed, e.g., by Semantic Primitives indicate the mutual contextual preferences of words: for example `hit' meaning STRIKE prefers a HUMAN agent. Wilkes' special contribution was a preference metric, minimising information in the interpretation selected. Thus word sense and sentence structure selection in text processing is then determined by maximum Constraint Satisfaction, and does not depend on complete satisfaction. Analogously, where interpretation involves inference, the metric selects the interpretation depending on the shortest inference chain.
- Wilks, Y.A., A preferential, pattern seeking semantics for natural language inference
Artificial Intelligence 6 (1975), 53--74.