Mining Unseen Name Translations Via Detecting Comparable News

摘要

We develop a framework for mining unseen name translations from daily multilingual news stories. Multilingual news articles from various sources are automatically downloaded from the Web. Comparable news in different languages are discovered via a gloss translation and an un-supervised learning algorithm. Multilingual name cognates are extracted from each comparable news cluster and matched by a phonetic matching model. Experiments have been conducted on the daily online news and the results show that unseen multilingual name translations can be successfully discovered by our framework.