Abstracts of HCRC Research papers, 1995

Amy Isard and Jean Carletta
Transaction and Action Coding in the Map Task Corpus
Task-oriented dialogues can normally be divided into subdialogues, each of which reflects collaboration on a particular substep of the task, and which we call `transactions'. We have devised a way of identifying transactions and their associated actions for HCRC Map Task dialogues, and we have tested the replicability of our coding scheme using naive subjects. In this report we describe the coding and the replication study.
(March 1995; 27 pages)
Ref. No. HCRC/RP-65 Price: UKL 1.20

Jasper Taylor, Jean Carletta and Chris Mellish
Requirements for Belief Models in Cooperative Dialogue
Models of rationality typically rely on underlying logics that allow simulated agents to entertain beliefs about one another to any depth of nesting. We argue that representations of individual deeply nested beliefs are in principle unnecessary for any cooperative dialogue. We examine some existing dialogue systems and conclude that where deeply-nested beliefs are supported, they are used only in situations where cooperation is not assumed.
Use of deeply-nested beliefs is associated with nested reasoning (i.e., reasoning about other agents' reasoning), which is limited by the architecture in most systems. Such systems cope well with cooperative dialogues because these do not require such reasoning. We show that summarizing representations of beliefs by conjoining all levels of nesting beyond the second allows agents capable of simple planning and plan recognition to cooperate in situations where agents equipped with unrestricted belief models would need to embody unrealistic assumptions.
(March 1995; 41 pages)
Ref. No. HCRC/RP-66 Price: UKL 1.60

Massimo Poesio
Semantic Ambiguity and Perceived Ambiguity
I explore some of the issues that arise when trying to establish a connection between the underspecification hypothesis pursued in the NLP literature and work on ambiguity in semantics and in the psychological literature. A theory of underspecification is developed `from the first principles', i.e., starting from a definition of what it means for a sentence to be semantically ambiguous and from what we know about the way humans deal with ambiguity. An underspecified language is specified as the translation language of a grammar covering sentences that display three classes of semantic ambiguity: lexical ambiguity, scopal ambiguity, and referential ambiguity. The expressions of this language denote sets of senses. A formalization of defeasible reasoning with underspecified representations is presented, based on Default Logic. Some issues to be confronted by such a formalization are discussed.

(May 1995; 45 pages)
Ref. No. HCRC/RP-68 Price: UKL 1.60

Andrei Mikheev
Domain Knowledge for Natural Language Processing
In this paper we describe the organization and contents of the knowledge base (KB) developed for the processing of patient discharge summaries (PDSs)---letters sent by a hospital consultant to a patient's own doctor. This KB is the major component in the processing of natural language by our system and includes both the conceptual and lexical knowledge represented in a uniform way. Conceptual knowledge in the KB is built and organized in a top-down fashion and explicitly targeted to represent information which was identified by domain experts to be important. Linguistic knowledge is built around the conceptual structures in a bottom-up and empirically driven way and encode the actual textual representations of particular classes of conceptual structures. The fundamental idea behind this approach is grouping of lexical knowledge around conceptual structures by means of lexico-semantic constructions (LSCs), which cover entire classes of NL expressions and are prepacked with rules for deterministic interpretation of these expressions into conceptual schemata of the domain. Although this KB is explicitly targeted to the medical domain and its sublanguage we argue that the approach advocated in this paper can be applied to general i.e. multi-domain NLP. This requires the parallel development of lexico-semantic constructions for capturing information from the text and conceptual structures to accommodate this information. In support of this claim we present strategies and tools for semi-automatic knowledge acquisition from the underlying corpus. On the basis of some example texts and system output, we will illustrate how acquired and represented background knowledge is used in processing to yield the desired interpretations.

(July 1995; 31 pages)
Ref. No. HCRC/RP-70 Price: UKL 1.30

Keith Stenning and Richard Tobin
Assigning information to modalities: comparing graphical treatments of the syllogism
Our long term goal is a cognitive theory of the effects of assigning information to different media and modalities. Our theory is based on the observation that graphical representations are limited in their logical expressiveness and that this leads to tractable inference. Our theory predicts that usability should be a function of logical expressiveness.
We begin by clarifying the intended senses of media and modalities. We then suggest a method for studying the cognitive effects of modality assignment. Comparing alternative assignments of the same information to different modalities allows many unknown factors affecting usability to be minimised. We choose categorial syllogisms as an example domain and describe three methods of varying expressive power for implementing syllogistic reasoning.
The expressive power of the three systems is then considered and predictions of usability made. Although empirical evidence about the usability of the three systems is at present limited, their analysis is sufficient to raise issues about what is required to connect empirical observations to the kind of theory sketched here. Strong intuitions about relative usability of these well-specified systems suggests that the theory may be developing along promising lines. The intuitively most usable system is the one whose expressive power is closely matched to the abstractive requirements of the inferential task at hand.

(September 1995; 23 pages)
Ref. No. HCRC/RP-71 Price: UKL 1.10

Jonathan Ginzburg, Zurab Khasidashvili and Enric Vallduvi, organizers
Tbilisi Symposium on Language, Logic, and Computation
The Tbilisi Symposium on Language, Logic, and Computation, the first of a regular series, is scheduled to take place in the Republic of Georgia on October 19-22, 1995. It will be a four-day conference with seven keynote speakers. 51 papers were submitted (length ca. 10 pages) from which 26 were selected for presentation. A selection of papers presented in the conference will be published by CSLI, Stanford.
Contributed papers cover topics such as natural language syntax, formal semantics, dynamic logic, quantified extensions of modal systems and intermediate logics, statistics and language processing, automated deduction and logic programming, process algebras, and lambda and combinatory calculi. Since work in the fields of Logic, Language and Computation is currently at a stage where cross--fertilisation is of increasing importance, it is expected that the proceedings will be of interest to a wide audience from among researchers in computer science, artificial intelligence, logic, linguistics and philosophy. In addition, the proceedings will provide an opportunity for high quality research from Georgia, Armenia and other FSU countries to reach the international LLC community.

(September 1995; 67 pages)
Ref. No. HCRC/RP-72 Price: UKL 2.40

Antonio Moreno
Dynamic Belief Analysis
The process of {\em rational inquiry} can be defined as the evolution of the beliefs of a rational agent as a consequence of its internal inference procedures or its interaction with the environment. These beliefs can be modelled in a formal way using {\em doxastic logics}. The {\em (possible worlds model} and its associated {\em Kripkean semantics} provide an intuitive semantics for these logics, but they seem to model agents which are {\em logically omniscient} (and {\em perfect reasoners.} These problems can be avoided with a syntactic view of possible worlds, defining them as arbitrary sets of sentences in a propositional doxastic logic. This approach does not account for any kind of evolution in the set of beliefs, though. In this work this syntactic view of possible worlds is taken, and a three-dimensional dynamic analysis of the beliefs of the agent is suggested in order to model the process of {\em rational inquiry} in which the agent is permanently engaged. The agent can make a {\em logical} analysis of the beliefs (using a modified version of the{\em analytic tableaux} method); the results of this analysis guide the {\em experimental} analysis, that allows the agent to perform certain tests in the environment, that corroborate or refute the results of the logical analysis. In a third ({\em axiomatic}) kind of analysis, the agent can transform its set of beliefs into an axiomatic system (a set of axioms and a set of inference rules.)
(September 1995; 29 pages)
Ref. No. HCRC/RP-73 Price: UKL 1.20