A Computational Mechanism for Pronominal Reference
35834Authors
nan
Abstract
the syntactically impossible antecedents. This latter This paper describes an implemented mechanism for handling bound anaphora, disjoint reference, and pronominal reference. The algorithm maps over every node in a parse tree in a left-to-right, depth first manner. Forward and backwards coreference, and disjoint reference are assigned during this tree walk. A semantic interpretation procedure is used to deal with multiple antecedents. 1.
Keywords
Full Content
A COMPUTATIONAL MECHANISM FOR PRONOMINAL REFERENCE Robed J. P. Ingria David Stallard BBN Systems and Technologies, Incorporated 10 Mouiton Street Mailstop 009 Cambridge, MA 02238 ABSTRACT the syntactically impossible antecedents. This latter This paper describes an implemented mechanism for handling bound anaphora, disjoint reference, and pronominal reference. The algorithm maps over every node in a parse tree in a left-to-right, depth first manner. Forward and backwards coreference, and disjoint reference are assigned during this tree walk. A semantic interpretation procedure is used to deal with multiple antecedents. 1. INTRODUCTION This paper describes an implemented mechanism for assigning antecedents to bound anaphors and per- sonal pronouns, and for establishing disjoint reference between Noun Phrases. This mechanism is part of the BBN Spoken Language System (Boissn, et al. (1989)). The algorithm used is inspired by the index- ing scheme of Chomsky (1960), augmented by tables analogous to the "Table of Coreference" of Jack- endoff (1972). This mechanism handles only intra- ssntentJal phenomena and only selects the syntac- tically and semantically possible antecedents. Ul- timately, it is meant to be used in conjunction with an extra-sentential reference mechanism like that described in Ayuso (1989) to include antecedents from other utterances and to utilize discourse factors in its final selection of an antecedent. In Section 2 the empirical and theoretical back- ground to this treatment is sketched out. In Section 3, the actual algorithm used is described in detail. In Section 4, the associated semantic interpretation mechanism is presented. In Section 5, we compare the algorithm with related work. Finally, in Section 6, remaining theoretical and implementational issues are discussed. 2. THEORETICAL BACKGROUND While most computational systems are interested in the potential antecedents of pronouns, work in generative grammar by Lasnik (1976) and Reinhart (1976) has led to the conclusion that sentential syntax is responsible for assigning possible antecedents to bound anaphors (reflexives, such as "himself", "herself", "themselves", etc., and the reciprocals "each other" and "one another") but not to personal pronouns ("he", "she", "they", etc). In the case of personal pronouns, sentential syntax only determines procedure is called disjoint reference, since the im- possible antecedents can not even overlap in refer- ence with the pronoun; compare the cases in sen- tences (1) and (2), where the underlined items are non-identical in reference, with those in (3) and (4), where they are non-overlapping in reference. In (1) and (2), "he" and "him" cannot refer to "John" (non- identical reference); while in (3) and (4) "John" cannot be a member of the set referred to by "they" and "them" (non-overlapping or disjoint reference). (1) He likes John. (3) They like John. (2) John likes him. (4) John likes them. Disjoint reference is even more noticeable with first and second person pronouns where it does not merely produce impossible interpretations, but actual ungrammaticality: (5) *1 like me. (7) "We like me. (6) "i like us~ (8) °Yo--'u like yo---u. A crucial notion both for assigining antecedents to bound anaphors and for establishing disjoint refer- ence between …
Content truncated to 500 words
Labeled Text
the syntactically impossible antecedents. This latter This paper describes an implemented mechanism for handling bound anaphora, disjoint reference, and pronominal reference. The algorithm maps over every node in a parse tree in a left-to-right, depth first manner. Forward and backwards coreference, and disjoint reference are assigned during this tree walk. A semantic interpretation procedure is used to deal with multiple antecedents. 1.
Publication Details
- Paper ID:
- 35834
- Year:
- 1989
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan