PISA - A procedure for analyzing the structure of explanatory texts.

Abstract

Linguistic analyses of text corpora have contributed to the understanding of natural language processing in both reading and writing. However, the impact of text analysis in psycho-linguistic research has been limited, mainly because the analyses hardly ever concern text structure. Existing models for text structure analysis tend to rely heavily on analysts intuitions and world knowledge, and they are hardly formulated explicitly enough to be applied in an objective and reliable way.This paper discusses a procedure that aims at solving this in determinacy for the domain of explanatory texts. We have formulated out intuitions about text structures in ordered sets of production rules that constitute an analytic procedure. This procedure automatically builds structure representations in an incremental way, that is, clause-by-clause in a left-to-rightmanner. It makes well-defined use of linguistic knowledge and hardly relies on world knowledge. Linguistic cues that play an important role in the linking of text Segments are, among others, referential continuity (anaphors, lexical reiteration), verb semantics, markers of coherence relations (connectives, temporal adverbs), sentence form (ellision, subordination) and position in the sentence frame (initial, final).To illustrate the workings of the procedure and the relevance of its products, results are presented of procedural analyses of a corpus of explanatory texts. The texts were first draft versions of essays written by 12-year-old children. The Output of the procedure (especially the hierarchical structure assigned to a text), appears to lead to fruitful insights into the cognitive representation of language users and has implications for research on writing proficiency, text quality, and text processing.