Preserving Ambiguities in Generation via Automata Intersection

Abstract

We discuss the problem of generating text that preserves certain ambiguities, a capability that is useful in applications such as machine translation. We show that it is relatively simple to extend a hybrid symbolic /statistical generator to do ambiguity preservation. The paper gives algorithms and examples, and it discusses practical linguistic difficulties that arise in ambiguity preservation. Introduction This paper reports on an aspect of a natural language generation system called Nitrogen (Knight and Hatzivassiloglou, 1995; Langkilde and Knight, 1998ab; Langkilde, 2000), which has been used as part of the Gazelle machine translation (MT) system (Knight et al, 1995). In particular, we show how Nitrogen can generate English text that preserves unresolved ambiguities from non-English documents. One of the first texts translated by Gazelle was a Japanese article with the following noun phrase: 1a. [IC chippu o seizou-suru noni tsukau] dorai-echingu soochi ya suteppa [sil...