Rosetta is a language generation system originally developed for the CMU Communicator system by Kevin Lenzo. Rosetta is an active template system written in Perl; it separates generic processing (such as interfaces to the dialog system) from domain-specific template processing.
Requests for generation specify a form and sub-form, as well as attribute values that are to be substituted into the template. Templates are filled by arbitrary Perl code and can thus generate variations predicated on input values and context.
An alternate approach to surface-form generation was introduced by Oh and Rudnicky (2000) and implemented stochastic generation processes. The latter approach uses a corpus of prompts spoken by human operators to create language models corresponding to different prompt categories. These categories can correspond to the form/sub-form classification used in Rosetta and allow stochastic generation to use the same interfaces as the existing Rosetta system.
- Oh, A. H. & Rudnicky, A. 2000. "Stocastic language generation for spoken dialogue systems". Proceedings of the ANLP/NAACL workshop on conversational systems, 27-32.