Mark T. Maybury

Advances in Automatic Text Summarization

It has been said for decades (if not centuries) that more and more information is becoming available and that tools are needed to handle it. Only recently, however, does it seem that a sufficient quantity of this information is electronically available to produce a widespread need for automatic summarization. Consequently, this research area has enjoyed a resurgence of interest in the past few years, as illustrated by a 1997 ACL Workshop, a 1998 AAAI Spring Symposium and in the same year SUMMAC: a TREC-like TIPSTER-funded summarization evaluation conference. Not unexpectedly, there is now a book to add to this list: Advances in Automatic Summarization, a collection of papers edited by Inderjeet Mani and Mark T. Maybury and published by The MIT Press. Half of it is a historical record: thirteen previously published papers, including classics such as Luhns 1958 word-counting sentence-extraction paper, Edmundsons 1969 use of cue words and phrases, and Kupiec, Pedersen, and Chens 1995 trained summarizer. The other half of the book holds new papers, which attempt to cover current issues and point to future trends. It starts with a paper by Karen Sp~irck Jones, which acts as an overall introduction. In it, the summarization process and the uses of summaries are broken down into their constituent parts and each of these is discussed (it reminded me of a much earlier Sp~rck Jones paper on categorization [1970]). Despite its comprehensiveness and authority, I must confess to finding this opener heavy going at times. The rest of the papers are grouped into six sections, each of which is prefaced with two or three well-written pages from the editors. These introductions contain valuable commentary on the coming papers--even pointing out a possible flaw in the evaluation part of one. The opening section holds three papers on so-called classical approaches. Here one finds the oft-cited papers of Luhn, Edmundson, and Pollock and Zamora. As a package, these papers provide a novice with a good idea of how basic summarization works. My only quibble was in their reproduction. In Luhns paper, an article from Scientific American is summarized and it would have been beneficial to have this included in the book as well. Some of the figures in another paper contained very small fonts and were hard to read; fixing this for a future print run is probably worth thinking about. The next section holds papers on corpus-based approaches to summarization, starting with Kupiec et al.s paper about a summarizer trained on an existing corpus of manually abstracted documents. Two new papers building upon the Kupiec et al. work follow this. Exploiting the discourse structure of a document is the topic of the next section. Of the five papers here, I thought Daniel Marcus was the best, nicely describing summarization work so far and then clearly explaining his system, which is based on Rhetorical Structure Theory. The following section on knowledge-rich approaches to summarization covers such things as Wendy Lehnerts work on breaking

New Directions in Question Answering

Question answering systems, which provide natural language responses to natural language queries, are the subject of rapidly advancing research encompassing both academic study and commercial applications, the most well-known of which is the search engine Ask Jeeves. Question answering draws on different fields and technologies, including natural language processing, information retrieval, explanation generation, and human computer interaction. Question answering creates an important new method of information access and can be seen as the natural step beyond such standard Web search methods as keyword query and document retrieval. This collection charts significant new directions in the field, including temporal, spatial, definitional, biographical, multimedia, and multilingual question answering.After an introduction that defines essential terminology and provides a roadmap to future trends, the book covers key areas of research and development. These include current methods, architecture requirements, and the history of question answering on the Web; the development of systems to address new types of questions; interactivity, which is often required for clarification of questions or answers; reuse of answers; advanced methods; and knowledge representation and reasoning used to support question answering. Each section contains an introduction that summarizes the chapters included and places them in context, relating them to the other chapters in the book as well as to the existing literature in the field and assessing the problems and challenges that remain.

Communicative acts for explanation generation

Abstract Knowledge-based systems that interact with humans often need to define their terminology, elucidate their behavior or support their recommendations or conclusions. In general, they need to explain themselves. Unfortunately, current computer systems, if they can explain themselves at all, often generate explanations that are unnatural, ill-connected or simply incoherent. They typically have only one method of explanation which does not allow them to recover from failed communication. At a minimum, this can irritate an end-user and potentially decrease their productivity. More dangerous, poorly conveyed information may result in misconceptions on the part of the user which can lead to bad decisions or invalid conclusions, which may have costly or even dangerous implications. To address this problem, we analyse human-produced explanations with the aim of transferring explanation expertise to machines. Guided by this analysis, we present a classification of explanatory utterances based on their content and communicative function. We then use these utterance classes and additional text analysis to construct a taxonomy of text types. This text taxonomy characterizes multisentence explanations according to the content they convey, the communicative acts they perform, and their intended effect on the addressees knowledge, beliefs, goals and plans. We then argue that the act of explanation presentation is an action-based endeavor and introduce and define an integrated theory of communicative acts (rhetorical, illocutionary, and locutionary acts). To illustrate this theory we formalize several of these communicative acts as plan operators and then show their use by a hierarchical text planner (TEXPLAN—Textual EXplanation PLANner) that composes natural language explanations. Finally, we classify a range of reactions readers may have to explanations and illustrate how a system can respond to these given a plan-based approach. Our research thus contributes (1) a domain-independent taxonomy of abstract explanatory utterances, (2) a taxonomy of multisentence explanations based on these utterance classes and (3) a classification of reactions readers may have to explanations as well as (4) an illustration of how these classifications can be applied computationally.

Multimedia summaries of broadcast news

Increasing amounts of text, audio, and video content has fueled efforts to provide direct, content based access to these materials. Summaries are often necessary to enable timely relevancy assessments, information extraction, or information analysis from source material. Whereas text summarization research is receiving increasing attention, comparatively few investigators have examined video summarization. We report on the extension of a broadcast news access system to provide multimedia summaries. We briefly overview our system for video analysis, focusing on our novel integration of image, speech and language processing techniques to support automated video summarization. We outline algorithms for proper name and keyphrase extraction, story segmentation, and key frame extraction which together underpin our current ability to automatically summarize video. We describe the systems ability to generate multimedia video summaries tailored to a user query. We discuss evaluation metrics for measuring the (quality) value of these summary artifacts.

Intelligent user interfaces: an introduction

Intelligent user interfaces promise to improve the interaction for all. Drawing upon material from the recently completed Readings in Intelligent User Interfaces (IUI) (Maybury and Wahlster, 1998), this tutorial will define terms, outline the history, describe key subfields, and exemplify and demonstrate intelligent user interfaces in action.

Personalized multimedia information access

Ask questions, get personalized answers.

Personalized Digital Television

This chapter presents the recommendation techniques applied in Personal Program Guide (PPG). This is a system generating personalized Electronic Program Guides for Digital TV. The PPGmanages a user model that stores the estimates of the individual user’s preferences for TV program categories. This model results from the integration of di¡erent preference acquisitionmodules that handle explicit user preferences, stereotypical information about TV viewers, and information about the user’s viewing behavior. The observation of the individual viewing behavior is particularly easy because the PPG runs on the set-top box and is deeply integrated with the TV playing and the video recording services o¡ered by that type of device.

Planning multisentential English text using communicative acts

Abstract : This technical report contains a dissertation submitted for the degree of Doctor of Philosophy in the University of Cambridge. This dissertation discusses the investigation of the hypothesis that the generation of multiple utterances, as with planning single utterances (Appelt, 1985), is a plan-based activity that is based on communicative acts. This requires the analysis of text in search for underlying communicative acts that achieve distinct effects of the hearers knowledge, beliefs, and desires. Thus this dissertation investigates the communicative structure and communicative function of a range of explanations for knowledge based systems. Related to this is the issue of how focus constrains generation (Sidner, 1979; McKeown, 1982) and so a second theoretical aim is to investigate how attentional constraints relate to text planning and linguistic realization.

Research in multimedia and multimodal parsing and generation

This overview introduces the emerging set of techniques for parsing and generating multiple media (e.g., text, graphics, maps, gestures) using multiple sensory modalities (e.g., auditory, visual, tactile). We first briefly introduce and motivate the value of such techniques. Next we describe various computational methods for parsing input from heterogeneous media and modalities (e.g., natural language, gesture, gaze). We subsequently overview complementary techniques for generating coordinated multimedia and multimodal output. Finally, we discuss systems that have integrated both parsing and generation to enable multimedia dialogue in the context of intelligent interfaces. The article concludes by outlining fundamental problems which require further research.

Knowledge-based multimedia: The future of expert systems and multimedia

Abstract Recent advances have fueled investigation into the integration of knowledge-based systems and multimedia technology. These intelligent or knowledge-based multimedia systems go beyond traditional hypertext and hypermedia and promise to provide capabilities such as knoledge-based search and retrieval from multimedia data bases, interfaces that intelligently communicate with users via multiple media, and multimedia training systems that tailor interaction to individual student characteristics and needs. After some definitions, we illustrate a number of expert multimedia tools and applications that are or will emerge from this integration. We then provide details on how this integration of knowledge and media will support a new generation of intelligent multimedia interfaces. In doing so, we describe what additional knowledge and processes need to be formalized to support this integration and describe a promising new approach that views multimedia communication as an action-based endeavor. We conclude by briefly recommending an agenda for future research.