Speaking of actions: choosing rhetorical status and grammatical form in instructional text generation
Keith Neil Vander Linden · 1993
This thesis addresses a problem fundamental to the development of text generation systems, the problem of managing diversity of expression at the textual level. In any genre, writers employ an extensive set of lexical and grammatical forms of expression, choosing the one they feel will be the most effective given the specific communicative context. The task of the text generation researcher is to identify these forms and the contexts in which they a used, and then to encode this information in the text generation system itself. A temporary solution to these problems, one which is typical of current text generation systems, is for the researcher to choose a small set of lexical and grammatical forms that are sufficient to disambiguate the types of information that the system must express. It is then a simple matter of allowing type of information to determine the appropriate expressional form. These simplifications allow the researcher to address other issues in generation, but do not provide an effective way to manage diversity of expression. The current study specifically addresses this problem. It contends that the only principled way to identify both the relevant expressional forms and the contexts in which they are used, is to perform a detailed linguistic study of a corpus of text from the relevant genre. This thesis proposes the following four step process for identifying both the relevant forms of expression and the contexts in which they are systematically used: (1) Collect a corpus of the relevant text type; (2) Perform a detailed linguistic study of this corpus; (3) Implement the results of this study in a text generation system; (4) Compare the output of the system with the text found in the corpus. This process is performed iteratively until a sufficient match is found between the generator's output and the original corpus text. This approach is demonstrated in the context of instructional text, resulting in the construction of scIMAGENE, an instructional text generation system. scIMAGENE embodies a model of the effective forms of expression consistently used by instructional text writers over a broad range of instruction types.