Dynamic Generation of Structured Texts for CALL

Greg Lessard, French Studies

Michael Levison, Computing and Information Science

Queen's University

Keywords: natural language generation, CALL, structured texts, VINCI

The proposed paper will describe, illustrate and evaluate the use of the VINCI natural language generation environment (www.cs.queensu.ca/CompLing/) to dynamically produce multi-paragraph structured texts in French and English for the purposes of CALL. Texts may take a variety of forms, including fictional tales, news events, and biographical accounts, and may be used to test comprehension, grammar, and vocabulary.

Initially, characters and their traits (in the sense of Propp 1968) are specified in the VINCI formalism. By means of a set of guarded syntax rules, these traits trigger the production of a sequence of steps (actions, descriptions) involving the characters. In turn, the steps give rise to paragraph-level rules defined in an RST-like formalism (Mann 2000).

Since VINCI allows for the generation of parallel related trees, the system can produce texts, questions based on the texts, and dynamically-generated analyses of student answers in response to questions.

References

Mann, W. (2000) Rhetorical Structure Theory Web Site. (http://www.sil.org/linguistics/rst/index.htm)

Propp, V. (1968) Morphology of the Folktale. Austin TX: University of Texas Press. [Original text 1928; translation by Laurence Scott; 2nd edition revised and edited with preface by Louis A. Wagner; new introduction by Alan Dundes.]