Learning Class-to-Class Selectional Preferences

Title: Learning Class-to-Class Selectional Preferences

Abstract: This research paper presents a model that learns selectional preferences for classes of verbs. The motivation behind this is that different senses of a verb may have different preferences, and some classes of verbs can share preferences. The model is tested on a word sense disambiguation task, using subject-verb and object-verb associations extracted from a small sense-disambiguated corpus.

Main Research Question: Can we develop a model that learns selectional preferences for classes of verbs, improving upon previous statistical models that focus on word-to-class relations?

Methodology: The paper uses a corpus-based approach to extract subject-verb and object-verb relations from Semcor, a corpus tagged with WordNet word-senses. It employs the Minipar parser to extract syntactic relations. The paper defines a word sense disambiguation exercise to evaluate the extracted preferences, using a sample of words and documents from Semcor.

Results: The results show that the model successfully learns selectional preferences for classes of verbs, improving upon previous statistical models. The paper provides examples of verb senses with related selectional preferences, demonstrating that the model can generalize and say that a class of verbs has a particular selectional preference.

Implications: The research has implications for the field of natural language processing, as it extends previous statistical models to class-to-class preferences, providing a more comprehensive understanding of selectional preferences in verbs. This can lead to improved performance in tasks such as word sense disambiguation, syntactic parsing, and semantic role labeling.

Link to Article: https://arxiv.org/abs/0109029v1 Authors: arXiv ID: 0109029v1