Authors:
Alicia Sagae
and
Scott E. Fahlman
Affiliation:
Carnegie Mellon University, United States
Keyword(s):
Image Retrieval, Textual Similarity, Textual Inference.
Related
Ontology
Subjects/Areas/Topics:
Applications
;
Artificial Intelligence
;
Data Engineering
;
Domain Analysis and Modeling
;
Enterprise Information Systems
;
Information Systems Analysis and Specification
;
Knowledge Engineering and Ontology Development
;
Knowledge Representation
;
Knowledge-Based Systems
;
Natural Language Processing
;
Ontologies and the Semantic Web
;
Ontology Engineering
;
Pattern Recognition
;
Symbolic Systems
Abstract:
Background knowledge resources contribute to the performance of many current systems for textual inference tasks (QA, textual entailment, summarization, retrieval, and others). However, it can be difficult to assess how additions to such a knowledge base will impact a system that relies on it. This paper describes the incremental, task-driven development of an ontology that provides features to a system that retrieves images based on their textual descriptions. We perform error analysis on a baseline system that uses lexical features only, then focus ontology development on reducing these errors against a development set. The resulting ontology contributes more to performance than domain-general resources like WordNet, even on a test set of previously unseen examples.