A Study on Flexibility in Natural Language Generation through a Statistical Approach to Story Generation

This paper presents a novel statistical Natural Language Generation (NLG) approach relying on language models (Positional and Factored Language Models). To prove and validate our approach, we carried out a series of experiments in the scenario of story generation. Through the different configurations tested, our NLG approach is able to produce either a regeneration in the form of a summary of the original story, or a recreation of one story, i.e., a new story based on the entities and actions that the original narration conveys, showing its flexibility to produce different types of stories. The results obtained and the subsequent analysis of the generated stories shows that the macroplanning addressed in this manner is a key step in the process of NLG, improving the quality of the story generated, and decreasing the error rate with respect to not including this stage.