Sentence generation from a bag of words using N-gram model

We are presenting in this paper, a method of sentence generation from a given bag of words. The task of sentence generation has its usage in text summarization, question answering system etc. The focus of our task is to generate all possible correct sentences from a given bag of words. The technique that we have applied is N-gram language model. The N-gram model is trained by a text corpus to generate only candidate sequences from a given bag of words. For N input words, instead of considering all possible N! permuted orders as candidate sequence, we have generated only candidate sequences less then N! by applying DFS (Depth First Search) filtering technique at run time. We have two corpora namely text corpus and annotated corpus of POS tags. We have extracted all valid POS trigram tags from the annotated corpus. Each of the generated candidate sequence has a probability score. The candidate sequences were ranked by matching it with valid trigram POS tag signature and probability score. Preliminary experimental work carried out in this direction by using the above mentioned model shows promising results.