Silent Anaphora

Charles J. Fillmore
International Computer Science Institute
Berkeley, USA
e-mail: fillmore@icsi.berkeley.edu

One part of the research effort of the "FrameNet" project involves collecting attestations (from the British National Corpus) of the combinatorial properties of the predicates we examine. At first we thought such documentation could be provided in the form of sentences whose surface syntactic constituents were annotated for the semantic and syntactic complements (in a broad sense of "complement") of the words we have targeted for analysis, but it soon became obvious that we need to include, in each such collection, examples of sentences -- and the means of annotating them -- in which certain elements are represented as "understood but missing". In the need to be clear about that, we have come to distinguish Constructionally Licensed Null Instantiation, Indefinite Null Instantiation, and Definite Null Instantiation, as separate ways of cataloguing the "missing" elements.

Of particular interest for the study of text is the registration of instances of the last-mentioned - omissions that require resolution in the current or preceding linguistic context. (Discussed in Fillmore 1984 "Pragmatically controlled zero anaphora" - BLS 12 - but treated elsewhere by J. Panevova and identified in various valence dictionaries and other lexical resources.) Part of the FrameNet database, then, will be a registry of the words that permit definite null instantiation of particular semantic roles. The paper will discuss the value of having such a database and the challenges created by such phenomena for the semantic annotation of text and problems of the theory of translation.

See previous, next abstract.