IPaGe - An Incremental Parallel Generator for Natural Language

Matthias Rehm

Linguistics on the way into the third millenium, Seiten 777 - 785

Verlag: Peter Lang


It is generally assumed that the production of natural language utterances can be split into two parts, the so-called what-to-say and how-to-say parts. If such a modular building is embraces, interesting questions arise concerning the take over of processing between these parts. One crucial aspect often stressed is the necessity of an incremental processing behaviour. In this paper a system is described that realizes the how-to-say part of the generation process. The notion of flexibility will be used as a key to an incremental and parallel processing behaviour. It will be shown that splitting an input structure in different increments can yield the possibility to implement different levels of parallel processing. Thereby the role of the input structures to the generation process will be stressed. These constitute the interface between a possible application or - in production terms - the what-to-say part of the generation process. The different levels of distributing the processing along with the characterisitc features of the input structures used support a certain system architecture. Such an architecture is described here that has been implemented as IPaGe (Incremental Parallel Generator).


  • BibTeX  -  (BibTeX.txt, 0 KB)