From Factorial to Quadratic Time Complexity for Sentence Realization Using Nearest Neighbour Algorithm

Abstract

{karthikg@students,sriram@research,taraka@students}.iiit.ac.in Abstract. Sentence Realization is the task of generating a well-formed sentence from a bag of words. Sentence Realization is a major step in many Natural Language Processing applications like Machine Translation (MT), Summariza- tion and Dialogue Systems. In this paper, we explore a graph based Nearest Neighbour Algorithm for the task of Sentence Realization. Sentence Realization is a major step in many Natural Language Processing applications like Machine Translation (MT), Summarization and Dialogue Systems. The task of Sen- tence Realization involves formation of a well-formed sentence from a bag of lexical items. These lexical items may be attached syntactically with one another. The level of syntactic information varies from application to application. Our aim consists of achiev- ing quality sentence realiser using as much as minimum syntactic information and of minimal computational complexity. As such our experiments assume only basic syntactic information, such as unlabeled dependency relationships between the lexical items. Graph based algorithms for Natural Language applications such as Pars- ing (McDonald et al. 2005), Summarization (Mihalcea and Tarau 2005) and Word sense disambiguation (Mihalcea 2005) have been well explored. For the task of Sentence Re- alization, graph based algorithms have yet to be explored. This paper is a novel effort in that direction.