Search

In this paper, we address the problem, "What makes an answer appropriate?". We do so by investigating indirect answers to questions in task-oriented dialogues. Three cases are distinguished: (1) The response, though indirect, answers the question asked; (2) the response denies a presupposition of the question; and (3) the response answers to higher goals the questioner was trying to achieve. Detailed analysis shows the need for knowledge about the task, the role of the participants, and communication goals, in the construction of appropriate answers. We conclude with a preliminary formulation of the appropriateness of an answer in terms of the goals of the questioner and the knowledge of the respondent., Sponsored in part by DARPA.

This paper describes a tuneable performance grammar currently being developed for speech understanding. It shows how attributes of words are defined and propagated to successively larger phrases, how other attributes are acquired, how factors reference them to help the parser choose among competing definitions to interpret the utterance correctly, and how these factors can easily be changed to adapt the grammar to other discourses and contexts. Factors that might be classified as "syntactic" are emphasized, but the attributes they reference need not be, and seldom are, purely syntactic., Technical Note No. 112. SRI Project 3804. Presented at the Annual Meeting of the Association for Computational Linguistics (13th) held in Boston, MA, on 30 Oct-1 Nov 1975. Sponsored in part by the Defense Advanced Research Projects Agency (DARPA).

Evidence is offered to support the view that linguistic competence cannot in principle be divorced from linguistic performance in order to abstract universal properties of grammars, that rules of grammar inevitably incorporate perceptual strategies and constraints, and that grammaticality and acceptability are related to predictability. The theory of systematic variation affords better direction for gathering data on rule-governed language use and a means for representing the results in formal grammars that predict speech behavior. Some of the strategies and constraints operating in performance and the rule- governed regularities they produce are demonstrated in the analysis of seven tape- recorded task-oriented dialogs.

A set of grammar codes and rules for analyzing, or parsing, English sentences automatically on a digital computer is presented. The introduction briefly discusses the nature of parsing and the general model of English structure on which this approach is based, and the codes and rules are discussed in detail. Although the computer program is not explained at length, the logic underlying it is illustrated in the discussion of the rules. A complete list of the rules written so far and a sample output of parsed sentences are appended. The linguist and the programmer will find the presentation sufficiently detailed so that they can add to, refine, or modify the codes and rules, and design programs for applying them to text.

A major problem is to provide for automatic recognition of sameness of meaning in spite of structural differences so that when a request for factual information is processed, any of the possible paraphrases of an answer will be retrieved from a natural language text. The RAND parsing grammar has been extended and refined so that identical labels are attached to comparable elements of structurally different paraphrases, such as the active-passive pair, 'X commands the third fleet' and 'The third fleet is commanded by X.' The sentences are first parsed automatically; then the output of the PARSE routine is fed into a subroutine called FLAG, which attaches labels that make the sameness of meaning explicit. The labels are attached by using coded information derived from the rules applied in the course of parsing the sentences. A sample of the output from the PARSE and FLAG routines shows how the differences in structure are resolved automatically and how the labels are attached that make possible the automatic retrieval of both sentences in response to a single relevant request. It is shown in detail how interrogative sentences also may be recognized as paraphrases of a single request and how the output from their processing can be used as input to a retrieval search. (Author)