Towards Universal Multilingual Knowledge Bases

Lexical, ontological, as well as encyclopedic knowledge is increasingly being encoded in machine-readable form. This paper deals with knowledge representation in multilingual settings. It begins by proposing a generic graph-based knowledge base framework, and then, in three case studies, explains how preexisting knowledge can be cast into this framework. The first case study involves enriching WordNet with information about human languages and their relationships. The second study shows how machine learning techniques can be used to bootstrap a large-scale multilingual version of WordNet where semantic relationships between terms in many languages are captured. The final study examines how information can be extracted from Wiktionary to produce a lexical network of etymological and derivational relationships between words.