Generating Natural Language Descriptions with Integrated Text and Examples

Generating Natural Language Descriptions with Integrated Text and Examples

Generating Natural Language Descriptions with Integrated Text and Examples

Generating Natural Language Descriptions with Integrated Text and Examples

Synopsis

This book discusses issues in generating coherent, effective natural language descriptions with integrated text and examples. This is done in the context of a system for generating documentation dynamically from the underlying software representations.

Good documentation is critical for user acceptance of any complex system. Advances in areas such as knowledge-based systems, natural language, and multimedia generation now make it possible to investigate the automatic generation of documentation from the underlying knowledge bases. This has several important benefits: it is always accessible; it is always current, because the documentation reflects the underlying representation; and, it can take the communication context, such as the user, into account.

The work described in this book compiles results from cognitive psychology and education on effective presentation of examples, as well as work on computational generation of examples from intelligent tutoring systems. It also takes into account computational learning from examples, and a characterization of good examples for just this purpose. Issues arising from these research areas--as well as issues coming from the author's own corpus analysis of instructional and explanatory texts--are discussed in the context of generating natural language descriptions of software constructs.

A text planner is used for a hierarchy of communicative goals. Examples are treated as an integral part of the planning process and their interaction with text is represented at all stages. The strengths and limitations of this approach are also discussed. Although the focus of this book is the generation of natural language descriptions, a similar set of issues need to be addressed in the generation of multimedia descriptions.

This book will be of interest to all researchers working in the areas of natural language interfaces, intelligent tutoring systems, documentation and technical writing, and educational psychology.

Excerpt

Most of this research was done as part of my doctoral work at the Information Sciences Institute of the University of Southern California. I was a research assistant in the Explainable Expert Systems (EES) Project, where the goal was to design an expert system shell that would be capable of justifying its decisions and explaining its execution trace in natural language. Procedural knowledge in EES, in the form of plans, was specified in a language called INTEND. INTEND was designed to satisfy the dual goals of being powerful enough to allow the specification of complex problem solving actions and plans, while also facilitating natural language paraphrases of both its syntactic constructs as well as its execution structure. However, initially there was little documentation on INTEND, its grammar, language actions, or their exact semantics. Some of the original developers of the language felt, like all true hackers at heart, that (a) the language could hardly be clearer, with all of its constructs (and the associated syntax) being completely obvious, and (b) for the few exceptional cases that were not the interpreter source code was always available for perusal. While this user model was usually quite accurate after one had been working on the project for some time, new arrivals to the project often found themselves at sea for a while before they became familiar with INTEND.

Having made the requisite number of mistakes initially in writing INTEND code that refused to work as I thought it should, I decided to work on a dynamic documentation system for INTEND. The goal was that when a user made a mistake, the system would generate appropriate documentation of the relevant construct from the underlying specification. Since the documentation was being generated dynamically, it was guaranteed to be accurate, and furthermore, could be varied to . . .

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.