Transition network grammars for natural language analysis
Communications of the ACM
| Hi-index | 0.00 |
Case grammars are useful formalisms in natural language processing. Their usefulness in realistic implementation projects have been demonstrated in the form of working systems by a number of researchers. However a close inspection, made during an attempt to use case grammars in generating user tailored error messages, has revealed serious limitations: There are so many anomalies and inconsistencies in earlier models that a completely new approach to the characterization of case grammars has been found necessary. After reviewing representative work about case grammars the present state of crisis is explained. To solve these problems a new model is proposed, which is based on dividing sentences into a small number of categories depending on similarity in their meanings. For each meaning category, a unique paradigm is defined to formally characterize the semantic relations between sentence components. Together with the use of well defined combination rules, the proposed set of paradigms can represent sentences with arbitrary complexity.