HITS' Cross-lingual Entity Linking System at TAC 2011: One Model for All Languages
Angela Fahrni, Michael Strube · Theory and applications of categories · 2011
This paper presents HITS’ system for crosslingual entity linking at TAC 2011. We approach the task in three stages: (1) context disambiguation to obtain a language-independent representation, (2) entity disambiguation, (3) clustering of the queries that have not been linked in the second step. For each of these steps one single model is trained and applied to both languages, i.e. English and Chinese. A multilingual knowledge base derived from Wikipedia is at the core of the system. The results achieved in the TAC cross-lingual entity task support our one-model-for-all-languages approach: the F1 scores of all three runs we submitted exceed the median value by 4.8 to 5.5 percent points.