CISC 883: Natural Language Generation

Catalog Description:
Current research directions in generation, models of generation, differences between generation and understanding, text structure and coherence, grammars for generation.

Current Text: Selected papers from journals and conference proceedings.

Goals:
This course is an in-depth look at the area of natural language generation. Most of the work in natural language processing has concentrated on understanding text. Instead, we look at the problems involved in generating text. Generation brings up issues not apparent in understanding. The job of the understander is to recognize which choice has been taken. The generator, on the other hand, must decide why to make one choice over another. Generation, then, may force the researcher to come to terms with issues concerning what kind of information must be available to the linguistic component and where it must be gotten (among others). Such questions can be ignored in doing research on understanding.

The course will take the tact of studying ongoing research projects involving generation. We will look at a number of generation methodologies and discuss how they handle various aspects of the generation problem. We will try to tease out some of the advantages and disadvantages of these approaches and get a feel for the questions currently being looked at by the generation community.

Content:

• Models of the generation process
• Deciding what to say:
  • discourse strategies
  • focus constraints
  • rhetorical structure theory
• Sentence Generators
  • general strategies for sentence generation
  • functional unification grammars
  • systemic grammars
  • tree adjoining grammars
• Planning formalisms in generation
• Explanation in expert systems
• Generation in intelligent tutoring systems
• Discourse and User Models
• Planning large texts

Typical Course Requirements: Critical analysis of research papers (with both written evidence and active discussion during class), class presentation of research paper(s), project/term paper (with class presentation).

Required Background: CISC 681: Artificial Intelligence, CISC 882: Natural Language Processing.