Abstracts

Wednesday 13th of May

Nicholas Asher (University of Texas at Austin)
Varieties of Discourse Structure in Dialogue
In this paper I examine a variety of discourse structures used in the analysis of dialogue. I propose a way to integrate turn taking structures, question stacks and structures based on rhetorical function, as well as more cognitively based structures like coordination, by distinguishing two tasks, the task of building the discourse structure itself and the task of cognitive modelling.

Robin Cooper (Göteborg University)
Mixing situation theory and type theory to formalize information states in dialogue exchanges
When information is exchanged in dialogue it is relevant to keep track not only of the information content about the domain of discussion but also information about linguistic events which are related to this content. Trying to formalize the situation theoretic approach to this has often involved us in rich and dubious ontologies. Here we blend ideas from situation theory together with work on dependent record types which has been conducted within Martin-Löf's type theory and suggest that this gives us a tractable way of characterizing information states that integrate information about semantic content and the linguistic utterances associated with it.

Wolfgang Heydrich (Universität Hamburg / Universität Bielefeld)
Theory of Mutuality (Syntactic Skeleton)
Lewis (1969) has analyzed common knowledge as mutual reason to believe. I shortly discuss his proposals in order to reconstruct the account in the framework of relevance logic (system R with quantifiers for individuals - persons and situations - and propositions). I formulate three postulates concerning the primitives of Lewis' analysis. A notion of `p's being public information in group G (in symbols: P_Gp ) arises which licenses reason to believe of arbitrary higher (finite and transfinite) degree. Being able to prove that P_Gp implies P_GP_Gp, I wonder which postulates could be added to secure the converse.

Jelle Gerbrandy (ILLC/Universtity of Amsterdam)
Some Remarks on Distributed Knowledge
In this paper, two definitions for the notion of distributed knowledge in possible worlds semantics are discussed and compared. In the first definition distributed knowledge is characterized by intersection of information states, the second defines the concept as `those sentences that are logical consequences of the beliefs of the agents.' I will argue that the effect of the first kind of definition depends on the ontological view one has about possible worlds, and that the second kind of definition depends on the expressive power of the language. I will also show that the logic of the two operators is the same, in the sense that the logics have the same weakly sound and complete axiom system.

Henk Zeevat (ILLC/Universtity of Amsterdam)
Contracts in the Common Ground
The paper explores the introduction of contracts in the common ground by means of requests, promises and questions. Various modal operators are introduced to achieve the modelling of these new additions to the conversational record. The paper extends the treatment of Zeevat (1997) and introduces the beginnings of a theory of action.

Thursday 14th of May

Thomas Clermont, Marc Pomplun, Elke Prestin, Hannes Rieser (Universität Bielefeld)
Eye-movement Research and the Investigation of Dialogue Structure
The paper reports about eye-tracking research as it has been going on in different institutions at Bielefeld University. Initially, a short introduction into eye-tracker technology is given and the two scenarios used in the investigation of agents' behaviour in task-oriented-dialogue are presented: A 2D-setting working with one eye-tracker recording the visual attention of the instructor, and a 3D-setting using eye-tracking for the instructor as well as for the constructor. The 3D-setting is to our knowledge the first one that has been developed for the investigation of agents' actions in dialogue. Subsequently, it is shown that the investigation of situated language behaviour can profit a lot from the application of eye-tracker-technolgy: We present the results of three empirical studies, two 2D-studies and one using the 3Dsetting. They reveal interesting interdependencies between focussing, planning, language production and the organization of larger units in discourse such as sequences of turns.

Wolfgang Heydrich and Peter Kühnlein and Hannes Rieser (Universität Bielefeld)
A DRT-style Modelling of Agents' Mental States in Discourse
Elaboratind and extending the well-known DRT-formalism, Hans Kamp has proposed a format of articulated DRSs in order to represent mental states (Kamp, 1990). Assuming acquaintance with these proposals, we shortly discuss some aspects of their philosophical background and argue that they provide a promising starting-point for the systematic investigation of dialogues. In the main part of this paper, we apply Kamp's method to a sample turn exchange from a construction dialogue (from the Bielefeld SFB setting).

Mieke Rats (Delft University / looking for a job)
Making DRT suitable for the description of information exchange in a dialogue
In my paper, I will take the position that although the construction algoritm of DRT produces intuitionally appealing representations of the meaning of small sequences of sentences, it needs much more refinement for being able to deliver adequate representations of information exchange in dialogue. I will show its inadequacies, present an alternative strategy for dialogues, and point to the consequences for semantic interpretation. The analysis will be based on a profound study of information exchange in two corpora of naturally occurring information dialogues one about flight information and one about train time table information and on experience in building a computer dialogue system for train time table information.

Jean-louis Dessalles (ENST, Paris)
The interplay of desire and necessity in dialogue
The purpose of this paper is to suggest that many argumentative moves in casual dialogues can be explained in terms of conflicting desires and conflicting beliefs, in such a way that these moves may be predicted. Participants approase the different outcomes of the conflicting situation and try to find, together, through dialogue, a solution that they consider as acceptable. We show how realistic dialogues can emerge through a simple recursive process from an initial cognitive conflict. This model is implemented in our program PARADISE which can reconstruct the argumentative moves of some real conversations.

Steve Pulman (Cambridge SRI)
The TRINDI Project: Some Preliminary Themes
The TRINDI project (Task oRiented INstructional DIalogue) is a recently started EU project involving the Universities of Gothenberg, Edinburgh, and Saarbrücken, as well as SRI Cambridge and Xerox Research Centre Europe (http://www.ling.gu.se/~cooper/trindi/ ). This presentation describes the aims of the project and discusses some preliminary themes that have emerged as central to the achievement of those aims.

Wieland Eckert (AT&T)
Automatic Evaluation of Dialogue Systems
We advocate an objective evaluation methodology for the automated evaluation of spoken dialogue systems that eliminates manual interaction and reduces annotation errors and personal bias. The evaluation is performed by observing interactions between the system and a simulated user. We argue that user simulation is an inexpensive and feasible method for optimizing a dialogue system in the lab. Using a simulated user we can conduct dialogues until the performance measure reaches a predetermined confidence level. A simulated user not only exercises the dialogue system and points out defects, it also helps predict the success of a modified dialogue strategy.

Ian Lewin (SRI International)
Formal Design, Verification and Simulation of Multi-Modal Dialogues
We have designed and implemented a dialogue management design tool for use in dialogue design as a component of user-interface design in multi-modal applications. The tool provides: a formal language (typed feature structures) for describing states and events; a simple rule formalism for specifying dialogues; an automatic dialogue-property checking module; a dialogue-simulator for interactive testing of designs. It is designed to be compatible with GUIDE (Mellor 1995) - a hardware/software environment permitting arbitrary coupling of input/output devices to facilitate experimentation in multi-modal design - but this does not restrict its generality.

Marc Blasband (Nederlandse Spoorwegen)
A Simple Semantic Model
Dialogue and natural language applications that are recently developed are based on a pragmatic semantic model, that has been developed in parallel by a number of different organisations: Philips, LIMSI, CSELT. The model is used by NS/OVR, KPN and Philips when developing VIOS (a train schedule information system using natural language). The same model seems to be used by the Technical University of Twente for a reservation application for the theatre and also by NUANCE, British Telecom and Dragon. Here the model will be explained.

Friday 15th of May

Stefan van Oord and Rieks op den Akker (University of Twente)
Fuzzy Natural Language Dialogue Systems
In natural language processing in general and particularly in dialogue management systems there are many sources of uncertainty. Current natural language dialogue systems lack the ability to handle this uncertainty. In this paper we investigate these sources of uncertainty. We explain why fuzzy set theory and possibility theory are suitable to deal with them. Furthermore, we unfold a model with rational agents that can have uncertain knowledge. We describe how they can reason with their knowledge and how knowledge can be revised while paying respect to its uncertainty.

John Barnden (University of Birmingham)
Uncertain Reasoning About Agents' Beliefs and Reasoning, with special attention to Metaphorical Mental State Reports
In handling dialogue, or natural language discourse more generally, it is important for a system to reason uncertainly about the beliefs and the reasoning of agents. An implemented system is described that accomplishes this, thoroughly integrating an uncertain form of simulative reasoning into a general uncertain-reasoning framework. The system can also reason (uncertainly) on the basis of metaphorical utterances about mental states. Such utterances are common in mundane discourse.

Nicolas Maudet and Fabrice Evrard (IRIT-ENSEEIHT, Toulouse)
A Generic framework for dialogue game implementation
Many authors have proposed to use {\em dialogue games} as a structure to model conventions of language use that exist in dialogue interactions. Dialogue games allow two speakers to dialogue in accordance with rules. The main idea is to prohibit some sequences of 'moves' given previous moves (in the manner of {\it dialogue grammar}) but also given the players' mental states.
Unfortunately, each model is built upon a formalism stricly designed for a specific type of interaction (e.g. it is only concerned with local coherence). It is our purpose in this paper to determine common requirements for those models and to propose a generic framework for dialogue game implementation, allowing games to be part of a larger structure reflecting the global coherence of dialogue.

Jonathan Ginzburg (Hebrew University, Jerusalem)
Clarifying Utterances
The paper argues for the importance of utterances as basic units in dialogue, as opposed to "sentences in context". The main case study is clarification: a characterization is provided within KOS, a recently developed synthesis of situation semantics and dialogue games, of the clarifications that can follow up on an utterance. This involves outlining: (a) a theory of understanding for dialogue, (b) a semantics for clarification acts, and (c) an account of how utterances update contexts. I will show that a relational view of meaning, whose original motivation was purely logical, can provide an important component for a theory of utterance understanding and utterance updates.

Jeroen Groenendijk (ILLC/Universtity of Amsterdam)
Questions in Update Semantics
The paper presents a static and dynamic version of an update semantics for a language of predicate logic enriched with simplex interrogatives. The static version gives rise to the same logic as the denotational semantics given for the same language in Groenendijk & Stokhof (1997). The dynamic semantics has some additional features, which concern anaphoric relations, and present us with a notion of existential quantification which seems to correspond to the specific use of indefinites. At meta-logical level an absolute notion of discourse coherence is defined which covers the usual notion of answerhood as a special case. More fine-grained notions of relevance and coherence are given as well, and are applied to simple interrogative dialogues.

Robert van Rooy (ILLC/University of Amsterdam)
Modal subordination in Questions
In this paper it is discussed how questions should change information states in dynamic semantics. The main claim is that the meaning of following utterances can be dependent on questions in a similar way as these meanings can be dependent on quantificational sentences, or more broadly, that this dependence is one of `modal subordination'. The most important contribution of this paper, however, is to show how this dependence between questions and later utterances can be accounted for in a general and systematic way.

Stafan Larsson (Göteborg University)
Questions Under Discussion and Dialogue Moves
The QUD provides a partially ordered representation of several types of questions which can be addressed at a certain stage of a dialogue. A question can, of course, be seen a kind speech act or dialogue move. Dialogue moves are often defined in terms of the BDI model, which provides e.g. mechanisms for interpretation of indirect questions. However, the generality of the BDI model makes it rather complex, and it is not specifically designed for managing dialogue in the way that the QUD model is. By combining both representations and relate them, a model with the advantages of both models may hopefully be constructed. Dialogue moves can be formalized in terms of the BDI model and still interact with the QUD.

Massimo Poesio (University of Edinburgh) and David Traum (UMIACS, University of Maryland)
Towards an Axiomatization of Dialogue Acts
Conversations involve all sorts of verbal activities beyond those strictly related to the performance of the task at hand. Among other things, the participants in a conversation have to make sure they both understand what's going on, to manage turn taking, and to keep each other informed about their progress in achieving their task. The participants share information about the status of all of these processes; this suggests that the view of the conversational score they share is rather more complex than assumed in previous accounts. We proposed a preliminary formalization of this more complex view of the conversational score in previous work; in this paper we revise that earlier model, and use our theory of the conversational score to give a partial specification of the effect of the dialogue acts included in the DRI classification.

And also ...

Adam Zachary Wyner (Bar Ilan University)
Adverbs and Anaphora
In the paper, similarities are demonstrated between sentences with manner and factive adverbs and discourses with correlated adjectives which predicate of a pronoun which refers to an antecedent event or fact. It is argued that the intrasentential adverbial cases should be analyzed in terms of the intersentential adjectival cases. An account, the Discourse Theory of adverbial modification, is provided for the intersentential cases using a version of DRT (Asher, 1993), then extended to the intrasentential cases. It is argued that this theory is superior to the current and widespread "Association Theory" of adverbial modification, in which particular sorts of adverbs are adjoined to particular adjunction sites. Furthermore, it is argued that adverbs are not be combinators with respect to events or facts.

Soo-Jun Park, Keon-Hoe Cha, Won-Kyung Sung, Do Gyu Song, Hyun-A Lee, Jay Duke Park, Dong-In Park (SERI, Korea), Jörg Höhle (GMD/FIT Birlinghoven)
MALBOT: An Intelligent Dialogue Model using User Modeling
In this paper we present a plan-based Korean dialogue system as a natural language user interface for a database. MALBOT supports mixed initiative to give users more control over dialogues. Mixed initiative, user modeling, alternative solution suggestion and handling circumlocution are few of MALBOT's characteristics. By employing user modeling, MALBOT reflects the user's preference in dialogues. The user modeling shell system BGP-MS is adapted to suit the requirements of MALBOT's domain. User modeling provides more intelligent system response by reflecting user's preference. The idea of user modeling in dialogue systems has found a good soil in this study.