Discovery of Unknown Events From Multi-lingual News

We have proposed a new approach to detect topically-related events from multi-lingual news sources. In particular, we are interested in Chinese and English on-line newswire stories. Three categories of named entities terms, namely, people names, geographical location names, and organization names, together with the story content terms constitute the basis for story representation. The named entities are extracted by transformation-based linguistic taggers. One for Chinese and one for English. For Chinese stories, we tackle the unknown word problems by means of a hybrid solution of rule-based and statistical-based methods. To conduct event detection in multi-lingual settings, we conduct gross translation on Chinese story representation into English. One gross translation approach is the basis translation method using only a bilingual dictionary. The second approach makes use of a parallel corpus as an additional resource. Unsupervised learning technique is employed to discover new events.