Evaluating Morphological Resources: a Task-Based Study for French Question Answering
Abstract
Morphology is a key component for many Natural Language Processing applications. In this article, we focus on one prototypical application, namely Question Answering (QA). In QA, morphological relations, especially those relying on the derivation and compounding processes, are often addressed in a superficial manner. Considering that some resources are able to provide deep and precise knowledge about a large spectrum of morphological processes, the issue lies first in determining the morphological phenomena which are most relevant for QA systems and second in evaluating the coverage of existing resources in this respect. To this aim, we describe a manual annotation and analysis of French question-answer pairs, which was performed in order to produce a unique and well-characterised reference dataset. Based on this study, we evaluate five different morphological resources for French and show that some dedicated resources are still lacking, which would cover phenomena such as denominal adjectives and agent deverbal nouns.
Origin : Files produced by the author(s)
Loading...