User modelling in text generation

User modelling in text generation

684 Book reviews I Journal of Pragmatics 22 (1994) 673-686 pragmatists (who are interested in the interface between syntax and pragmatics). There is...

245KB Sizes 0 Downloads 48 Views

684

Book reviews I Journal of Pragmatics 22 (1994) 673-686

pragmatists (who are interested in the interface between syntax and pragmatics). There is much to be learned here, and, more importantly, much to be pondered about what we must still do to attain a better understanding of long-distance reflexivization.

References Battistella, Edwin, 1989. Chinese reflexivization: A movement to INFL approach. Linguistics 27: 987-1012. Bouchard, Denis, 1984. On the content of empty categories. Dordrecht: Foris. Cole, Peter, Gabriella Hermon and Li-May Sung, 1990. Principles and parameters of long-distance reflexives. Linguistic Inquiry 21: l-22. Hestvik, Arild, 1992. LF-movement of pronouns and antisubject orientation. Linguistic Inquiry 23: 557-594. Huang, Yan, 1990. Noam Chomsky, Paul Grice and long-distance reflexivization. Paper presented at the Meeting of the Cambridge Linguistic Society. Huang, Yan, 1991a. A neo-Gricean pragmatic theory of anaphora. Journal of Linguistics 27: 301-335. Huang, Yan, 1991b. Government-Binding theory and long-distance reflexivization. Paper presented at the Spring Meeting of the Linguistics Association of Great Britain. Huang, Yan, 1991~. Long-distance reflexivization revisited: Towards a pragmatic analysis. Paper presented at the Autumn Meeting of the Linguistics Association of Great Britain. Huang, Yan, 1992a. Against Chomsky’s typology of empty categories. Journal of Pragmatics 17: l-29. Huang, Yan, 1992b. Hanyu de kongfanchou [Empty categories in Chinese]. Zhongguo Yuwen 1992 (5): 383-393. Huang, Yan, 1994. The syntax and pragmatics of anaphora. Cambridge: Cambridge University Press. Katada, Fusa, 1991. The LF representation of anaphors. Linguistic Inquiry 22: 287-313. Koster, Jan, 1987. Domains and dynasties: The radical autonomy of syntax. Dordrecht: Foris Kuno, Susumu. 1987. Functional syntax: Anaphora, discourse and empathy. Chicago, IL: University of Chicago Press. Lasnik, Howard, 1989. On the necessity of binding conditions, In: Howard Lasnik ed., 1989, Essays on anaphora, 149-167. Dordrecht: Kluwer. Rappaport, Gilbert, 1986. On anaphor binding in Russian. Natural Language and Linguistic Theory 4: 97-120. Timberlake, Alan, 1979. Reflexivization and cycle in Russian. Linguistic Inquiry 10: 109-141. Xu, Liejiong, 1986. Free empty categories. Linguistic Inquiry 17: 75-93. Zribi-Hertz, Anne, 1989. Anaphor binding and narrative point of view: English reflexive pronouns in sentence and discourse. Language 65: 695-727. SSDI 0378-2166(94)00017-9

CCcile L. Paris, User modelling Publishers,

in text generation.

London

and New York:

Pinter

1993. 205 pp. 539.00.

Reviewed by Jiirg Schtitz, Institute for Applied Luther-Strasse 14, D-661 11 Saarbrticken, Germany.

Information

Sciences,

Martin-

As computer systems become more and more an integral part of a human’s general working environment, they will have to cope with a greater variety of tasks,

Book reviews I Journal of Pragmatics 22 (1994) 673-686

685

including natural language processing (NLP) support. Therefore, an NLP module cannot be viewed as a stand-alone product, but as one component in, say, a hightechnology office environment. In such an environment NLP can be integrated within a documentation and document production environment sharing knowledge resources with other user-supporting utilities, such as grammar and style checkers, indexing and retrieval facilities, explanation modules and other tools. Obviously, such systems have to be able to adapt their behaviour to different kinds of users with whom they interact. In this book CCcile Paris discusses and validates an approach for the generation of texts for the purpose of information explanation that allows the system to tailor the text according to the user’s level of expertise. Her approach is not based on the common assumption that an expert of a domain under consideration needs less information than, for instance, a novice user, but that different kinds of users need different sorts of information to achieve the goal of effective communication. Her hypothesis is grounded on an analysis of a variety of texts, and is made to operate in the text generation system TAILOR that is able to vary its output for a wide,range of users. The book starts with an introduction to the overall field of text generation and the need for (general) user modelling as one of the central focal points for effective and efficient communication. To achieve this goal, one also has to identify the different sorts of information and knowledge that will affect the generation process. This process is facilitated by an intensive analysis of different text types including encyclopedias, high school text books and car manuals for novice and expert readers in the respective domains. Chapter 2 addresses the specific aspects and requirements of user modelling in the TAILOR system by taking into account previous approaches in this field, and focussing on the inferential potential over several knowledge sources, i.e. background knowledge of and about the user, user questions and given discourse information, that have to be employed. Chapter 3 then presents an intensive and careful study of different text sorts which leads to the identification of the strategies used in real life texts to explain and describe devices to readers with different levels of expertise. These strategies, the two main ones being the constituency schema, which makes use of rhetorical predicates as defined by McKeown, and the process trace, the newly identified strategy, which makes use of directives and is organised around functional information, form the basis for the process of forrnalising and implementing the generation capabilities of the TAILOR system. The text analyses are based on standard techniques used in computational linguistics, such as McKeown’s decomposition of texts by rhetorical predicates, Hobb’s coherence relations and the rhetorical structure theory of Mann and Thompson. The two main identified strategies for the organisation of texts cover the extreme ends of the user spectrum of the system. A combination of these strategies shall permit the covering of intermediate levels of expertise, thus providing the users with an appropriate mixture of structural and functional information. For the formalisation process, the author uses the well-elaborated theory of augmented transition networks (ATN) of Woods. At first sight, this might look old-fash-

686

Book reviews I Journal of Pragmatics 22 (1994) 673-686

ioned, especially when taking into account recent developments in kowledge-based approaches to NLP. However, since the aim of the research was primarily to amend and extend the strategy of discourse constituency, which has already been formalised by means of ATNs, the adopted approach provides “convenient and efficient solutions to generate coherent multi-sentential text”, and therefore, “it was important to use the same formalism for both strategies, in order to be able to combine them easily” (p. 75). In chapters 4 and 5 this formalisation is presented in detail, and exemplified by various text examples. Chapter 6 describes the actual implementation of the TAILOR system, which includes the specification of the various interfaces of the system. The knowledge base used in TAILOR is that of RESEARCHER, a system developed at Columbia University to read, remember and generalise from patent abstracts. This knowledge is accessed according to the user model and the strategy that is appropriate for the generation of the output text for the different kinds of users. The information types of the knowledge base comprise structural, attributive and functional knowledge, which is organised in a frame-based representation. The user model, which triggers the access, consists of two information slots which characterise the kowledge level of the user. The first slot contains information about the local expertise of the user and the second slot lists the basic concepts about which the user has a (basic) understanding. These kinds of knowledge are essential for finding the appropriate paths through the ATN representations that are responsible for triggering the text generation process. The surface string generator, i.e. the actual output text, is that of the TEXT system and is based on a functional unification grammar. One essential limitation of the TAILOR system is that it cannot be used as an interactive system, which would certainly ease its integration into a computational office environment (as mentioned above) for various documentation tasks. Chapter 7 presents related work in the field of user modelling and text generation, and shows again how the role of a user’s domain knowledge influences the tailoring of user-specific texts. In the last chapter the author discusses various application areas for such a system, ranging form dialogue and phrasal structuring capabilities to the controlling of the generation process with register information and other adaptive capabilities. This book addresses some of the essential aspects which are necessary when designing and implementing a system that takes into account the level of expertise of the user of the system. The adapted approach is well elaborated and is based on thorough empirical text analyses. The whole readability and understandability is supported by appropriate examples. I really enjoyed reading this book, as McKeown promised in her foreword. SSDZ 0378-2166(94)00006-Z