Abstract
We propose a method to translate Named Entities based on web resources. The method involves automatically learning surface patterns that are composed of English Named Entity, Chinese Translation, and symbols. With identified Named Entity and translation pairs as input, we obtain abstracts from web search engine and the strings composed of the Named Entity, translation and symbols from these abstracts will transform into surface patterns. When translating the Named Entity, we first use a web search engine to collect the abstracts containing the Named Entity and Chinese words. We then could find many probable translations by matching these abstracts with surface patterns. Additionally, we consider Data Redundancy and string length as information to obtain best translation. For evaluation, we collect 3,581 quesiton-answer pairs from the web and select 200 Named Entities from the quesiton-answer pairs as testing data, which are evaluated through two different web search engines. Results show that our methodology has good translation performance and, which is better than the commercial translation software SYSTRAN.