| 主权项 |
1. A system comprising:
at least one memory, in a networked language translation system, to store linguistic resources for one or more translation models comprising a translation memory database, at least one morphological dictionary, and at least one glossary, wherein the translation memory database stores a plurality of parallel phrases in at least one first language and at least one second language, wherein the glossary stores glossary terms that each comprise one or more words in the first language, wherein the morphological dictionary stores possible forms for each of a plurality of words in the first language, and wherein each of the possible forms has a basic form; an interface, in the networked language translation system, to receive a source document in the first language to be translated to the second language; at least one hardware processing unit, in the networked language translation system, configured to:
segment the source document into a plurality of initial segments in the first language;match one or more of the initial segments in the first language with one or more of the parallel phrases in the first language from the translation memory database to identify one or more of the parallel phrases in the second language from the translation memory database;generate at least one machine translation of the initial segments based on the translation models and the identified ones of the parallel phrases in the second language, wherein the machine translation comprises a plurality of pre-translation segments corresponding to ones of the initial segments;generate similarity scores for the pre-translation segments, wherein each similarity score in the similarity scores corresponds to a pre-translation segment in the pre-translation segments and measures a similarity between a corresponding one of the initial segments in the first language and a corresponding one of the parallel phrases in the first language;determine for each pre-translation segment in the pre-translation segments whether human translation is required for the pre-translation segment based on the similarity score of the pre-translation segment;in response to a determination that a set of one or more of the pre-translation segments require human translation, select multiple human translators from a plurality of human translators based on preference scores for the plurality of human translators, wherein each preference score among the preference scores is based on qualities of translations for documents previously translated by a human translator among the plurality of human translators that are similar to the source document;provide a set of one or more of the initial segments in the first language and the set of the pre-translation segments in the second language to translator systems over a network for translation of the set of the initial segments by the multiple human translators;receive one or more translated segments in the second language from the translator systems over the network, wherein the translated segments correspond to the set of the pre-translation segments;assemble the translated segments and ones of the pre-translation segments to generate a translation of the source document in the second language; andretrain one or more of the language models based on the translated segments. |
