This thesis is an inquiry into the nature of the high-level, rhetorical structure of unrestricted natural language texts, computational means to enable its derivation, and two applications (in automatic summarization and natural language generation) that follow from the ability to build such structures automatically. The thesis proposes a first-order formalization of the high-level, rhetorical structure of text. The formalization assumes that text can be sequenced into elementary units; that discourse relations hold between textual units of various sizes; that some textual units are more important to the writer's purpose than others; and that trees are a good approximation of the abstract structure of text. The formalization also introduces a linguistically motivated compositionality criterion, which is shown to hold for the text structures that are valid. The thesis proposes, analyzes theoretically, and compares empirically four algorithms for determining the valid text structures of ...

We describe our experience in developing a discourse-annotated corpus for community-wide use. Working in

... In this paper, we argue that the main theories representing these two approaches, RST (Mann and Thompson 1988) and G&S (Grosz and Sidner 1986), make similar claims about how speakers' intentions determine a structure of their discourse. The similarity occurs because the nucleus-satellite relation among text spans in RST corresponds to the dominance relation among intentions in G&S. Building on this similarity, we sketch a partial mapping between the two theories to show that the main points of the two theories are equivalent. Furthermore, the additional claims found in only RST or only G&S are largely consistent. The issue of what structure is determined by semantic (domain) relations in the discourse and how this structure might be related to the intentional structure is discussed. We suggest the synthesis of the two theories would be useful to researchers in both natural language interpretation and generation.

Graphical presentations can be used to communicate information in relational data sets succinctly and effectively. However, novel graphical presentations about numerous attributes and their relationships are often difficult to understand completely until explained. Automatically generated graphical presentations must therefore either be limited to simple, conventional ones, or risk incomprehensibility. One way of alleviating this problem is to design graphical presentation systems that can work in conjunction with a natural language generator to produce "explanatory captions." This paper presents three strategies for generating explanatory captions to accompany information graphics based on: (1) a representation of the structure of the graphical presentation (2) a framework for identifyingthe perceptual complexity of graphical elements, and (3) the structure of the data expressed in the graphic. We describe an implemented system and illustrate how it is used to generate explanatory cap...

Our goal is to create an “intelligent” 3D agent able to send complex, ‘natural ’ messages to users and, in the future, to converse with them. We look at the relationship between the agent’s communicative intentions and the way that these intentions are expressed into verbal and nonverbal messages. In this paper, we concentrate on the study and generation of coordinated linguistic and gaze communicative acts. In this view we analyse gaze signals according to their functional meaning rather than to their physical actions. We propose a formalism where a communicative act is represented by two elements: a meaning (that corresponds to a set of goals and beliefs that the agent has the purpose to transmit to the interlocutor) and a signal, that is the nonverbal expression of that meaning. We also outline a methodology to generate messages that coordinate verbal with nonverbal signals.

Context appears in Artificial Intelligence (AI) as a challenge for the coming years as shown by the various scientific events focusing on context held since 1995. However, context is already considered in other domains, such as Natural Language Processing, although through few aspects of context. We present in this paper a survey of the literature dealing directly and explicitly with context whatever the domain is. This permits us to have a clear view of the context in AI. One of the conclusions of this survey is to point out the existence of different types of context along knowledge representation, the mechanisms of reasoning on the knowledge, and the interaction of the computer system with humans.

Abstract not found

. CIRCSIM-Tutor v. 3 is a natural-language based ITS for cardiac physiology. In this paper, we describe TIPS, a new text planning engine for CIRCSIM-Tutor based on current research in text generation. Since conversations cannot be completely planned in advance, TIPS plans and executes iteratively. It maintains a goal hierarchy for the tutor while carrying on a conversation with the student. It can handle multi-turn plans on the part of the tutor, and it can back up and replan when the student gives an unexpected answer. In this paper we sketch the design of TIPS using an analysis of human-to-human tutoring transcripts to shape the requirements. 1 Introduction The negative feedback loop which maintains a steady blood pressure in the human body is one of the more difficult topics for first-year medical students to master. CIRCSIM-Tutor v. 3 is the latest in a series of CAI systems intended to help students master the concepts involved. This paper describes TIPS, a new text planner for CI...

One of the key factors for the acceptance of expert systems in real world domains is the capability to explain their reasoning. This paper describes the basic properties that characterize explanation methods and reviews the methods developed up to date for explanation in Bayesian networks.

The linguistic subfields of semantics and pragmatics are both concerned with the study of meaning. We might say that semantics studies what Grice (1967) called the TIMELESS MEANING of a linguistic expression φ —the basic meanings of the words in φ composed as a function of the syntactic structure of φ. Formal semantics, especially since the seminal work of Montague (1973), attempts to develop an empirically adequate theory of semantics for a given language by developing rules that are clear and unambiguous in their application and effect, and in so doing, makes clear predictions about the possible meanings for a given expression. Semanticists assume that words do have basic meanings, and that a given syntactic structure corresponds with a determinate way of composing the meanings of its subparts. 1 Pragmatics, on the other hand, studies utterances of expressions like φ, attempting to explain WHAT SOMEONE MEANT BY SAYING φ ON A PARTICULAR OCCASION. It is clear that there is quite often a difference between the timeless meaning of φ and what someone meant by uttering φ on a given occasion. It is a truism among native speakers that this difference arises because of the way that the context of utterance influences interpretation. We complain if someone quotes what we say out of context because this may well distort our intended meaning. But what is a context of utterance, and how does it influence interpretation? A linguistic interchange is generally called a discourse, hence the problem of understanding