Paper: Searching For Grammaticality: Propagating Dependencies In The Viterbi Algorithm

ACL ID W05-1628
Title Searching For Grammaticality: Propagating Dependencies In The Viterbi Algorithm
Venue ENLG
Session Main Conference
Year 2005
Authors

In many text-to-text generation scenarios (for in- stance, summarisation), we encounter human- authored sentences that could be composed by re- cycling portions of related sentences to form new sentences. In this paper, we couch the generation of such sentences as a search problem. We in- vestigate a statistical sentence generation method which recombines words to form new sentences. We propose an extension to the Viterbi algorithm designed to improve the grammaticality of gener- ated sentences. Within a statistical framework, the extension favours those partially generated strings with a probable dependency tree structure. Our preliminary evaluations show that our approach generates less fragmented text than a bigram base- line.