A Characterization of Underspecified Anaphora and its Consequences on the Annotation of Anaphoric Relations

Frédéric Landragin (1)

1 LaTTice - Langues, textes, traitement informatique, cognition

When resolving anaphora, the precise identification of the antecedent is sometimes difficult. In a first set of cases linked to abstract anaphora (“in that context”, “it happens”), the reason lies in the exact delimitation of the antecedent (previous verbal phrase? previous sentence? whole paragraph?). In a second set of cases, several concrete antecedents are possible and the hearer or reader cannot decide between them. Moreover, deciding is not mandatory to the comprehension of the utterance or discourse, and the antecedent can stay underspecified. This phenomenon can be called ‘underspecified anaphora', or anaphora with ‘fuzzy' or ‘vague' antecedent. We illustrate it with French data, and we propose a first classification of types of potential underspecified antecedent. We then deduce principles for their annotation.

Mots clés: Anaphora; antecedent; complex nominal phrases; annotation

Domaines: Linguistique; Sciences de l'information et de la communication; Traitement du texte et du document

Frédéric Landragin. A Characterization of Underspecified Anaphora and its Consequences on the Annotation of Anaphoric Relations. Theoretical and Computational Perspectives on Underspecification (Jahrestagung des SFB 732), Nov 2007, Stuttgart, Germany. ⟨halshs-00186876⟩