JOURNAL TRANSCRIPT
RACHEL
A TEXT-BASED
GIORA
ANALYSIS OF NON-NARRATIVE TEXTS*
Most linguistic theories are concerned with the rendering explicit of what ordinary speakers find intuitively acceptable. Specifically, a linguistic theor~ is to provide rules for sentence well-formedness. A step forward is taken by recent researchers into text well-formedness. The conditions for text well-formedness can be viewed as requirements on either the surface structure or the semantic organization of the text. This study is a research into the constraints on the semantic structure of the text. In this study text well-formedness is d,fined in terms of the Relevance Requirement. On this view, a text is well-formed if all its main assertion propositions are relevant to a Topic of Discourse (DT). Thus, Relevance is viewed as a relation between a proposition or a set of propositions and a DT. To be able to account for text well-formedness, then, it is necessary to make explicit the notions of Relevance and DT. My suggestion here is to explicate the above notions in cognitive terms, Specifically, I propose here the application of categorical organization (in the sense established by Rosch) to nonnarrative texts. Such texts, I argue, get organized in the way categorial concepts are formed.
1.
Introduction
The main thrust of psychological research into the structure of texts is concerned with the schematic organization of the events and scenes that make up narratives. A scheme (or frame, to use Minski's 1975 term) isa cognitive structure - an organized representation of a body of knowledge which is spa.tially and/or temporally organized. Most recent research concerning text processing attests to the fact that narrative texts are processed in terms of schema (Rumelhart 1975, for instance). On the basis of Mandler's (1979, 1984) distinction between schematic and categorical organization, I propose here the application of categorical organization to non-narrative/ *
I am obliged to Mira Ariel, Simon Garrod, Yeshayahu Shen and Sidney Strauss for their insightful comments on an earlier. draft.
116
Rache1 Giora A text-based analysis of non-narrative texts
expository texts. For the purpose of this paper, expository jinformative texts are those that convey maximum information in the most economical and efficient way, in terms of processing. Such texts, I will argue, get arganized in the way categorical concepts are formed. Like schematic organization, categorical organization refers to hierarchically arranged cognitive structures, which govern our understanding of the relationships among superordinate, subordinate and coordinate classes. Since I propose to study text organization in categorical terms, the main concern of this paper lies in the rules that govern categorical classifications (see Section 2.3.1.2). However, before attempting the projection of categorical organization on the structure of non-schematic texts, let me first propose a processing discourse model. The following (Section 2) therefore, is a brief discussion of my previous attempts (Giora 1985a, 1985b, 1985c) to draw a context set (discourse model) for the text.
2.
A Dis-course
Topic
Oriented
Context
Set for the Text
Giora (1985a) models a context set for the text along the lines suggested by Reinhart (1981) for the' construction of the context set for the sentence. Reinhart views the context set for the sentence as organized under local entries in the form of Sentence Topics (STs). The context set for the text, however, is viewed as organized under global entries in the form of Discourse Topics (DTs). The DT of a given text segment is defined as the element relative to which the whole set of propositions (of that segment) is taken to be "about". Given this view, the discussion of discourse organization, then, has to cope with the notion of DT. The treatment of DT discussed here is handled within the framework of a theory of Relevance.
2.1.
On the Definition of Relevance
Given the assumption that informaivejexpository texts convey information in the most economical and efficient way, the Relevance Requirement is viewed as a necessary condition for text well-formedness. Thus in a previous work of mine Giora (1985b) I show that a text or a text segment is well-formed i. e., coherent· if all its main assertion propositios are either relevant to a DT or marked either as digr~ssion or as subordinate to the immediately preceding main assertion proposition. Thus Relevance is viewed as a relation between a proposition or a set of propositions and a DT.
117
It means that for a set of propositions to be relevant they must be interpretable as being "about" that DT (for a different view see Appendix). Clearly, an explicit formulation of the Relevance Requirement stipulated above necessitates both the explication of the notion of DT and that of the "aboutness" relation. I will start by specifying the notion of pragmatic "aboutness" .
2.2.
Pragmaticaboutness
Rather than treating the relation of "being about" from a philosophical perspective (Goodman 1972, for instance), the formulation of the relation of "being about" is viewed here as derived from what we know about text processing. The assumption is that the way readers process a text, that is, the procedures a reader puts into effect while interpreting what the text is about, are conducive to the formulation of the aboutness relation. Following Strawson (1964) and Stalnaker (1978) Reinhart (1981) views the processing model for the sentence as reflecting the way our knowledge expands - the way we add the propositional content of a sentence or an utterance to our knowledge store. The process of expanding our knowledge according to Reinhart is twofold: we first assess the truth value of a proposition about to be added to our knowledge store. Having found no reason to reject it as false, we then add it to the set of propositions already in our context set. Going beyond Strawson and Stalnaker, Reinhart claims that these two procedures are Topic-oriented. Thus, to assess ~hetruth value of 'all crows are black', for instance, we check the members of the set of crows to see if any of them is not black, rather than checking the non-black things to see if any of them is a crow. Though in principle both of these strategies are feasible, in practice we follow the first because we view the formulation that 'all crows are black' as classified under crows, and it is our knowledge of crows that we search in order to assess this proposition. The strategy of assessing the truth value of a given message is thus Topic-oriented, in the sense that it is our knowledge of the Topic of a given proposition that affects the process of verification. The addition of the propositional content of a statement is similarly relative to a Topic: It is implausible to assume, for reasons of cognitive economy, that we store lists of propositions (the range of recall being seven items or so), and Reinhart thus suggests that the storage procedure, too, be construed as Topic-oriented. She argues this by reference to the catalogue
118
119
Raehel Giora
A text-based analysis of non-narrative texts
metaphor, where Topic functions as an entry under which we classify and store information. Topics, then, instruct the reader of the construction of a context set: having assessed a proposition relative to its Topic constituent, the reader stores it under an entry corresponding to that Topic. A Topic, then, is attributed the cognitivejunction of an entry which is interpreted as an instruction from a writer to a reader on how to construct a discourse model. The account of sentence processing is thus viewed as centering around the notion of Sentence Topic: for a sentence to be "about" a Topic is to be assessed and stored relative to that Topic.
blond" as a primitive discourse unit in terms of the Semantic Values