Exploring Sentence Vectors Through Automatic Summarization

Vector semantics, especially sentence vectors, have recently been used successfully in many areas of natural language processing. However, relatively little work has explored the internal structure and properties of spaces of sentence vectors. In this paper, we will explore the properties of sentence vectors by studying a particular real-world application: Automatic Summarization. In particular, we show that cosine similarity between sentence vectors and document vectors is strongly correlated with sentence importance and that vector semantics can identify and correct gaps between the sentences chosen so far and the document. In addition, we identify specific dimensions which are linked to effective summaries. To our knowledge, this is the first time specific dimensions of sentence embeddings have been connected to sentence properties. We also compare the features of different methods of sentence embeddings. Many of these insights have applications in uses of sentence embeddings far beyond summarization.