Robust pronoun resolution with limited knowledge

Abstract

Most traditional approaches to anaphora resolution rely heavily on linguistic and domain knowledge. One of the disadvantages of developing a knowledgebased system, however, is that it is a very labourintensive and time-consuming task. This paper presents a robust, knowledge-poor approach to resolving pronouns in technical manuals, which operates on texts pre-processed by a part-of-speech tagger. Input is checked against agreement and for a number of antecedent indicators. Candidates are assigned scores by each indicator and the candidate with the highest score is returned as the antecedent. Evaluation reports a success rate of 89.7% which is better than the success rates of the approaches selected for comparison and tested on the same data. In addition, preliminary experiments show that the approach can be successfully adapted for other languages with minimum modifications.

Annotations en chaînes de coréférences et anaphores dans un corpus de discours spontané en français

“…L'accord en genre et en nombre est une contrainte très commune, toujours considérée par les processus de résolution de l'anaphore. Elle exprime une contrainte obligatoire pour les correcteurs symboliques (Lappin et Leas, 1994), alors qu'elle joue le plus souvent un rôle de préférence pour les approches heuristiques (Mitkov, 1998). Le genre et le nombre sont, enfin, des propriétés toujours considérées par les techniques basées sur un apprentissage automatique sur des corpus annotés (Recasens, 2010).…”

Extracting and Visualizing Quotations from News Wires

“…Automatic extraction of quotations is not among the most explored fields in NLP applications, despite the interest it raises for domain related tools such as news exploring or sentiment detection.Nevertheless, we can mention several applications developed in this perspective: Google InQuotes 2 , News Explorer 3 , Excom 4 . Like SAPIENS, News Explorer aims at the detection of quotations in news -although SAPIENS is specialized for press agency news wires -and their author; it is based on word lists and pattern matching.…”

“…Pleonastic il pronouns have been filtered out by the ILIMP system of [1] which has been integrated within SXPipe. Following [3] or [4], we model resolution as a two-step process, whereby: (1) we apply a series of hard constraints which has the effect of eliminating candidates semantically incompatible, and (2) we rank of the remaining candidates on the basis of salience preferences.…”

An Incremental Discourse Parser Architecture

“…By restraining the domains the antecedents of anaphors are to be looked for, VT makes the existing anaphora resolution techniques [10] more effective and more efficient. The model we have studied for experimenting VT-driven anaphora resolution is based on a three tiered representation and a two steps process.…”