Improving Document Summarization Using Crosslingual Word Embeddings

In recent years, due to the increase of information available online in multiple languages and the inability of a user to examine it manually, several text summarization techniques have been developed. This thesis proposes a new methodology to extract significant sentences from a collection of textual documents written in multiple languages. Specifically, it aims at extracting a summary in any of the source languages by exploiting also the semantic relationships between cross-lingual content. To this purpose, it exploits aligned word embedding models to extract cross-lingual relationships and a graph-based approach to pick the most significant sentences. The results demonstrate that using cross-lingual text correlations improves summarizer performance.