Integrating Selectional Preferences in WordNet

Title: Integrating Selectional Preferences in WordNet

Research Question: How can we improve the integration of selectional preferences in WordNet, a large database of English words and their relationships?

Methodology: The researchers proposed a new approach to learn selectional preferences for classes of verbs. They trained a model on subject-verb and object-verb relationships extracted from a sense-disambiguated corpus (Semcor), which is tagged with WordNet word-senses. They used a syntactic parser to extract these relationships. Their approach differed from previous work by focusing on classes of verbs instead of individual verbs, allowing them to capture the relationships between classes in a hierarchy.

Results: The researchers provided examples illustrating the validity of their approach and showing that it was feasible. They also presented experimental results on a word sense disambiguation task, which demonstrated the effectiveness of their method.

Implications: This new approach to learning selectional preferences for classes of verbs has several implications. First, it provides a better formalization than previous verb-to-class models. Second, it allows for the integration of selectional preferences in WordNet, improving the accuracy and comprehensiveness of the database. Lastly, it can be applied to larger, non-disambiguated corpora, making it a scalable and practical solution for improving natural language processing systems.

Link to Article: https://arxiv.org/abs/0204027v1 Authors: arXiv ID: 0204027v1