Abstract

A considerable amount of research has lately been conducted on developing neural architectures for learning vector representations of word forms (i.e., word embeddings). However, they have clear limitations when it comes to deep language understanding as they conflate different meanings of a word into a single representation and consequently are unable to accurately model semantics of individual word senses. A field of research has tried to address this issue with word representations by breaking them into those of their individual meanings. In this presentation I will give an overview of current representation techniques with a special emphasis on knowledge-based representations and NASARI (http://lcl.uniroma1.it/nasari/), our recently developed multilingual representation of concepts and entities. Finally, I will briefly present some of its most successful applications to date, namely semantic similarity, word and named entity disambiguation, sense clustering, domain labeling and text classification.

Bio: Jose Camacho Collados is a Google Doctoral Fellow and PhD student at Sapienza University of Rome (http://wwwusers.di.uniroma1.it/~collados). His research focuses on Natural Language Processing and on the area of lexical and distributional semantics in particular. Jose co-organized a tutorial on “Semantic Representations of Word Senses and Concepts” at ACL 2016 (http://acl2016.org/index.php?article_id=58) and an EACL 2017 workshop on “Sense, Concept and Entity Representations and their Applications” (https://sites.google.com/site/senseworkshop2017/). He is additionally co-organizing a SemEval shared task on multilingual and cross-lingual semantic similarity (http://alt.qcri.org/semeval2017/task2/). His background education includes an Erasmus Mundus Master in Natural Language Processing and Human Language Technology and a 5-year BSc degree in Mathematics.