Named entities (NE), the noun or noun phrases referring to persons, locations and organizations, are among the most information-bearing linguistic structures. Extracting and translating named entities benefits many natural language processing problems such as cross-lingual information retrieval, cross-lingual question answering and machine translation.; In this theisis we propose an efficient and effective framework to extract and translate NEs from text and speech. We adopt the hidden Markov model (HMM) as a baseline NE extraction system, and investigate its performance in multiple language pairs with varying amounts of training data. We expand the baseline text NE tagger with a context-based NE extraction model, which aims to detect and correct NE recognition errors from automatic speech recognition hypotheses. We also adapt the broadcast stews trained NE tagger for meeting transcripts.; We develop several language-independent features to capture phonetic and semantic similarity measures between source and target NE pairs. We incorporate these features to solve various NE translation problems presented in different language pairs (Chinese to English, Arabic to English and Hindi to English), with varying resources (parallel and non-parallel corpora as well as the World Wide Web) and different input data streams (text and speech).; We also propose a cluster-specific name transliteration framework. By grouping names from similar origins into one cluster and training cluster-specific transliteration and language models, we manage to dramatically reduce the name transliteration error rates.
展开▼
