Vocabulary Tailored Summary Generation

Proceedings of the 27th International Conference on Computational Linguistics, pages 795–805 Santa Fe, New Mexico, USA, August 20-26, 2018. 795 Vocabulary Tailored Summary Generation Kundan Krishna Adobe Research kunkrish@adobe.com Aniket Murhekar IIT Bombay aniket1602@gmail.com Saumitra Sharma IIT Guwahati sharmasaumitra15@gmail.com Balaji Vasan Srinivasan Adobe Research balsrini@adobe.com Abstract Neural sequence-to-sequence models have been successfully extended for summary generation. However, existing frameworks generate a single summary for a given input and do not tune the summaries towards any additional constraints/preferences. Such a tunable framework is desirable to account for linguistic preferences of the specific audience who will consume the summary. In this paper, we propose a neural framework to generate summaries constrained to vocabulary- defined linguistic preferences of a target audience. The proposed method accounts for the gen- eration context by tuning the summary words at the time of generation. Our evaluations indicate that the proposed approach tunes summaries to the target vocabulary while still maintaining a superior summary quality against a state-of-the-art word embedding based lexical substitution algorithm, suggesting the feasibility of the proposed approach. We demonstrate two applica- tions of the proposed approach - to generate understandable summaries with simpler words, and readable summaries with shorter words. 1 Introduction Automatic text summarization (Nenkova and McKeown, 2011) is the task of generating summaries of an input document while retaining the important points. These summaries are used for presenting the most relevant and important information in a long text in a succinct form. They are useful in places where a quick consumption of the information in a long article is preferred. Earlier works in summarization select sentences/textual units from the input article and put them together into an “extractive” summary. How- ever, humans summarize an article by understanding the content and paraphrasing the understood content into the desired summary. Therefore, extractive summarization is unable to produce “human-like” sum- maries. This has led to efforts towards “abstractive” summarization which paraphrases summaries the input article. Several models have been proposed, with the most recent ones based on neural networks. Often, it is desirable to tune the summaries to the linguistic preferences of the readers. For exam- ple, a medical report may contain a lot of technical jargon beyond the understanding of the common population. When such a report is consumed by a patient, it makes sense to use lesser jargon to suit a patient’s knowledge level. Similarly, while reading articles, teenagers would prefer more informal and trendy words, while older people might like a more formal vocabulary. Summaries which are tuned to such “linguistic” preferences of the target population segment are likely to appeal better and catch their attention. A standard approach to incorporate vocabulary tuning would be to post-process a generated summary to achieve the desired goal by replacing a few words (e.g. replacing words with their simpler alternatives). However, such an approach might not preserve the context and hence can result in a complete change in the meaning of the content. Consider the sentence “The baseball pitcher was seen with a pitcher of …

Content truncated to 500 words