Empirical Studies of Discourse Representations for Natural Language Interfaces

EMPIRICAL STUDIES OF DISCOURSE REPRESENTATIONS FOR NATURAL LANGUAGE INTERFACES Ntis Dahlb~ick Ame JOnsson Natural Language Processing Laboratory Department of Computer and Information Science LinkOping University, S-581 83 LINKOPING, SWEDEN Intemet: NDA@LIUIDA.SE, ARJ@LIUIDA.SE Phone +46 13281644, +46 13281717 ABSTRACT We present the results from a series of ex- periments aimed at uncovering the discourse structure of man-machine communication in nat- ural language (Wizard of Oz experiments). The results suggest the existence of different classes of dialogue situations, requiring computational discourse representations of various complexity. Important factors seem to be the number of dif- ferent permissible tasks in the system and to what extent the system takes initiative in the dialogue. We also analyse indexical expressions and espe- cially the use of pronouns, and suggest a psy- chological explanation of their restricted oc- currence in these types of dialogues. INTRODUCTION Natural Language interfaces will in the fore- seeable future only be able to handle a subset of natural language. The usability of this type of in- terfaces is therefore dependent on finding subsets of natural language that can be used without the user experiencing inexplicable "holes" in the sys- tem performance, i.e. finding subsets for which we can computationally handle complete linguistic and conceptual coverage. This points to the need for theories of the 'sublanguage' or 'sublanguages' used when communicating with computers (Kittredge and Lehrberger, 1982). But unfortunately: "we have no well-developed lin- guistics of natural-language man-machine com- munication." (von Hahn, 1986 p. 523) One way of tackling this problem is to sim- ulate the man-machine dialogue by letting users communicate with a background system through an interface which they have been told is a natural language interface, but which in reality is a per- son simulating such a device (sometimes called a Wizard of Oz experiment, see Guindon, Shuld- berg, and Conner, 1987). While not being a new technique, early examples are Malhotra (1975, 1977), Thomas (1976), and Tennant (1979, 1981), only a limited number of studies have been conducted so far. A considerably larger number of similar studies have been conducted where the users knew that they were communi- cating with a person. This is unfortunate, since those researchers who have considered the issue have noted that the language used when commu- nicating with a real or simulated natural language interface has differed from the language used in teletyped dialogues between humans, although it has been difficult to the exact nature of these dif- ferences. The language used has been described as 'formal' (Grosz, 1977), 'telegraphic' (Guindon et al, 1987), or 'computerese' (Reilly, 1987). Only a few Wizard of Oz studies have been run, using different background systems and dif- feting in questions asked and methods of analysis used. It is therefore premature to draw any far- reaching conclusions. With some caution, bow- ever, perhaps the following can be accepted as a summary of the pattem of results obtained so far: The syntactic structure is not too complex (Guindon et al, 1987, Reilly, 1987), and presum- ably within the capacity …

Content truncated to 500 words