Computational Linguistics Research Research Articles

Control Issues in Anaphor Resolution

Much research in computational linguistics has concentrated on treating the individual phenomena of natural language rather than on how these treatments can be made to work together. This paper discusses four respects in which processes embodying individual treatments must be made to co-operate with one another if anaphoric expressions are to be interpreted correctly. These are the co-ordination of (1) treatments of anaphoric and non-anaphoric ambiguity in a sentence; (2) treatments of different kinds of knowledge relevant to interpretation; (3) the resolution of several anaphors in one sentence; and (4) candidate referents arising from several sources. It is argued that a flexible control structure based on numerical scoring allows the required co-operation to take place, whereas a more limited depth-first architecture seems not to allow this. The discussion is grounded in comparisons between two very different implemented systems that resolve anaphors. The first, SPAR, is an experimental system embodying the control structure argued for. The second, the SRI Core Language Engine, is a wide-coverage system intended for practical applications. Its control structure is, at the time of writing, depth-first, but planned enhancements will allow the problems discussed in the paper to be tackled, resulting in the development of a powerful and widely applicable natural processor.

Research in computational linguistics

The Linguistic String Project and Computer Science Department of New York University are engaged in a long-term program of basic and applied research in computational linguistics, aimed at the automated retrieval and processing of information in natural language scientific and technical texts. Over the past fifteen years we have developed a large array of resources, including a broad-coverage English grammar, dictionaries, a program for parsing and transformational analysis, and various utilities for managing grammar, dictionary, and text files. The grammar and parsing program have been incorporated into two main application systems, one for the information formatting of natural language medical texts, the other for natural language data base retrieval.

Research at Stanford University

Support and Personnel: Research in Computational Linguistics at Stanford is being carried out in the context of the Knowledge Base Management Systems (KBMS) project (Defense Advanced Research Projects Agency contracts MDA903-77-C-0322 and N000-39-80G-0132 Gio Wiederhold, Principal Investigator). The KBMS project is concerned with the application of Artificial Intelligence techniques to the design, performance, and use of database systems. Additional funding is provided by NSF Grant ISTRO-23889. Primary contributors are Jerry Kaplan (Research Associate), and students Jim Davidson, Francisco Corella, and Lena Yesil. Participants on a sub-contract to SRI International are Barbara Grosz, Norman Haas, and Daniel Sagalowicz.

Dependencies of discourse structure on the modality of communication

With the genesis of speech understanding systems, computational linguistics research is becoming less wedded to teletype-based interaction. It would therefore be wise to look for systematic ways in which the discourse structure of voice interaction differs from teletype interaction. Such discoveries could lead to substantially different discourse components for speech understanding systems.Rubin (1980) points out that language experiences should not simply be characterized as oral or written. Rather, there is a set of dimensions along which language experiences such as having a conversation and writing/reading a letter might differ, including: the ability to interact, the sharing of space and time between speaker and hearer, the concreteness of referents, and the use of voice or print.Following Rubin's taxonomy, and influenced by Chapanis et al.'s [1977] communication mode study and Grosz' [1977] task-oriented dialogue work, we conducted a study to explore how the structure of an instruction-giving discourse depends on the communication situation in which it takes place. Twenty-five subjects were videotaped as they instructed twenty-five others in assembling a toy water pump. Five "dialogues" each took place face-to-face, via telephone, teletype, audiotape, and written text. We chose to analyze telephone and teletype dialogues first since results would have direct implications for the design of speech understanding and production systems.Preliminary results indicate that the structure of telephone dialogues is markedly different from that of teletype dialogues. In telephone mode, speakers frequently, explicitly, and often indirectly, request hearers to identify the referents of noun phrases. For example, utterances used indirectly to perform such requests include "there is a NP" and "the NP?". In contrast, teletype "speakers" rarely accomplish the goal of referring in a separate step. Instead, the goals of referring and requesting an assembly action are achieved with one utterance, usually an imperative such as "Insert the green plunger into the large tube with threads on one end".As for computational implications, we are led to conclude that, within the framework of a plan-based theory of speech acts (Perrault and Allen [1980]), referent identification should be treated as a planned action by language production and comprehension systems. That is, by planning to facilitate the healer's plan, producers should design their noun phrases so that hearers can identify the referents. Conversely, comprehenders should reason about what the producer intended to be done with the uttered NP -- find its referent, supply a co-referring NP. etc. Our goal, then is to develop production and comprehension systems capable of reasoning about reference.

The 1969 international conference on computational linguistics: A progress report

T 1969 International Conference on Computational Linguistics, held at S~nga-Siiby, Sweden, September 1-4, attracted participants from some 20 countries. The conference is noteworthy as evidence of a new trend in computational linguistics research, meriting the attention of scholars in the humanities. In recent years, linguistic theory has been characterized by a syntax-based orientation, which was at once an advance over previous linguistic theory and a misunderstanding of the more basic role of semantics. The centrality of syntax in linguistic theory greatly influenced developments in computational linguistics, where research efforts were concentrated on parsing strategies and the aptly named "algebraic linguistics," with semantics playing a very subsidiary role. Such a bias does not reflect a proper assessment of the role of semantics in the analysis of natural-language text, whether by computer or by humans. Viewed as a problem in artificial intelligence, automatic analysis and representation of the content of natural-language text can be defined as a process by which the computer "understands" the input data. The process of understanding involves identification of the concepts contained in the text as well as the relations obtaining between these concepts; this information must then be explicitly represented in some standard form. The formal representation serves as the basis for further operations on the text-for example, automatic abstracting, questionanswering, translation, stylistic analysis, etc. Because of the bias toward syntax, the attention of most investigators in computational linguistics throughout its brief history has been focused on a single dimension of the relations between concepts. The other aspects of the problem-namely, explication of the concepts themselves and encoding of this information in some formal languagehave been largely ignored. Quite recently, there has been increasing concern with the role of semantics and with the problem of specifying a formalism for semantic representation; this concern is reflected in the contributions to the 1969 computational linguistics conference. Of the 71 papers accepted for presentation, almost a third deal with some aspect of the much-neglected semantic dimension of language. Since it is this topic which is likely to be