Mental models as representations of discourse and text By Alan Garnham (review)

192 LANGUAGE, VOLUME 66, NUMBER 1 (1990) heretofore largely ignored in the literature (e.g. mágnanákaw 'robber' versus magnanákaw 'will rob'). F argues that these facts fall naturally out of a general metrical account of Tagalog stress, lending support for the analysis. Both grid and tree incarnations of Metrical Phonology are employed. Ch. 4 (97-110) presents an interesting discussion of Baligtad Tagalog, an extremely complicated form of speech disguise involving a large number of interacting processes (e.g. ditoh 'hexe' —» dótih via metathesis; hindP'no' —» dimihinin via syllable reversal and complete infixation). F argues that the Baligtad Tagalog data support several of her previous analyses, particularly with respect to infixation and stress. Since the data are presented solely as supporting evidence and nothing in the author's previous analyses hinges on these facts, the discussion is quite intriguing. Ch. 5 (1 13-17) provides abriefconclusion and summary; additionally, there are two appendices describing Tagalog segments and clusters and a very complete bibliography. Whether or not one agrees with the author's largely syllablebased approach, the primary strength of the book remains its clear, concise, and complete presentation of both the facts of, and the theoretical issues raised by, the major morphophonological processes of Tagalog. This presentation is strengthened by F's responsible use of current theoretical mechanisms to describe the phenomena. Consequently, this work should be of interest both to Malayo-Polynesianists interested in current theoretical issues and to theoreticians who want a rapid introduction to a language in which many morphological and phonological processes interact in complex ways. [Bradley L. Pritchett, Northwestern University.] Mental models as representations of discourse and text. By Alan Garnham . Chichester: Ellis Horwood. 1987. Pp. 198. How do people understand natural languages ? Garnham's eventual goal is to find an answer to this question from the perspective of a cognitive psychologist. This work does not provide such a comprehensive psychological theory, but reports a number of experiments that establish certain properties of the mental representations of text that are constructed during language comprehension. The main body ofthe book is a revised version of G's 1981 D.Phil thesis (University of Sussex). The claim underlying this investigation is that previous psychological accounts of language representation, based too heavily on linguistic theories, are inadequate. A survey chapter on linguistic theories of semantics and pragmatics, and on Artificial Intelligence approaches to text understanding, provides some background in these fields, as well as allowing the author to point out where he perceives these theories to fall short of psychological adequacy. According to G, what is represented in mental models is at least partly independent of how it is represented. Thus, the author's central claim relates to the content of mental models, and not to how they are constructed by hearers, how they might be represented, or what types of mental processes operate on them. G's conclusion is that mental models are made up of 'people, objects, and events, rather than linguistic structures'. That is, not only do people forget the syntactic structure of sentences (this had generally been acknowledged by psycholinguists ), but it is also wrong to characterize mental models as semantic representations. Assuming that mental models are indeed constructed during text comprehension, this hypothesis leads to a number of empirically testable predictions. If mental models contain tokens representing discourse referents, rather than semantic expressions denoting such referents , then (other things being equal) which linguistic expression is used to refer to such a token should not matter. This prediction is validated by two experiments that show that people confuse coreferential descriptions, and that confusion of descriptions is correlated with coreferentiality. Since mental models represent situations in the world, G assumes that they should be derived from the meaning of the text taken together with general world knowledge. If this is the case, hearers should be expected to confuse information derived from linguistic input with information derived from world knowledge. Using experimental materials with 'highly probable instruments', such as a key used to unlock a door, G finds that hearers often are unsure about the source of their knowledge. Similarly, semantically different sentences that describe very similar situations are confused more often...