HITS' Monolingual and Cross-lingual Entity Linking System at TAC 2012: A Joint Approach

This paper presents HITS’ system for monolingual and cross-lingual entity linking at TAC 2012. We propose a joint system for entity disambiguation, recognition of NILs and clustering using Markov Logic. The proposed model (1) is global, i.e. a group of mentions in a text is disambiguated in one single step combining various global and local features, and (2) performs disambiguation, unknown entity detection and clustering jointly. The model for all languages is exclusively trained on English Wikipedia articles. The results achieved in the TAC monolingual and cross-lingual entity linking tasks show that our approach is competitive: our best English run achieves 8.5 percent points above median, while we outperformed all other participating systems in the Chinese cross-lingual subtask. The results for the Spanish subtask are lower due to a bug. Our unofficial Spanish results (after fixing the bug) are close to the ones of the best system.