| 摘要 |
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for translating text using a bridge language. One of the methods includes receiving original text in a source language and data identifying a target language; obtaining bridge language translations for the original text into a bridge language, wherein the bridge language translations comprise one or more dictionary translations; for each bridge language translation, obtaining one or more candidate translations for the bridge language translation into the target language, wherein the candidate translations comprise one or more dictionary translations; determining that a candidate translation is a dictionary translation obtained for a particular bridge language translation and that the particular bridge language translation is a dictionary translation obtained for the original text; determining that the candidate translation is obtained for two or more distinct bridge language translations; and selecting the candidate translation as a preferred translation of the original text. |
| 主权项 |
1. A computer-implemented method comprising:
receiving, at a machine translation system, original text in a source language and data identifying a target language; determining, at the machine translation system, that no translations from the source language to the target language are included in a translation data set associated with the machine translation system; obtaining, at the machine translation system, a plurality of bridge language translations for the original text from the source language into a bridge language, wherein the plurality of bridge language translations comprises one or more dictionary translations obtained from a dictionary associated with the machine translation system that maps each of a plurality of terms in the source language to respective alternative terms in the bridge language and one or more phrase table translations obtained from a phrase table associated with the machine translation system and generated by a statistical machine translation system that maps source language terms to bridge language terms; for each bridge language translation, obtaining, at the machine translation system, one or more candidate translations for the bridge language translation into the target language, wherein the candidate translations comprise one or more dictionary translations obtained from a dictionary associated with the machine translation system that maps each of a plurality of terms in the bridge language to respective alternative terms in the target language and one or more phrase table translations associated with the machine translation system and obtained from a phrase table generated by the statistical machine translation system that maps bridge language terms to target language terms; determining, at the machine translation system, that a first candidate translation is a dictionary translation obtained for a first bridge language translation and that the first bridge language translation is a dictionary translation obtained for the original text; determining, at the machine translation system, that the first candidate translation is obtained for two or more first distinct bridge language translations, wherein the two or more first distinct bridge language translations comprise the first bridge language translation and one or more dictionary translations for the original text that are distinct from the first bridge language translation; determining, at the machine translation system, that a second candidate translation is a dictionary translation obtained for a second bridge language translation and that the second bridge language translation is a dictionary translation obtained for the original text; determining, at the machine translation system, that the second candidate translation is obtained for two or more second distinct bridge language translations, wherein the two or more second distinct bridge language translations comprise the second bridge language translation and one or more phrase table translations that are distinct from the second bridge language translation; determining, at the machine translation system, that respective parts of speech of the first candidate translation and the second candidate translation match a part of speech of the original text; selecting, at the machine translation system, the first candidate translation and the second candidate translation as preferred translations of the original text from the source language to the target language for the machine translation system; and generating, at the machine translation system, a respective score for each preferred translation based at least in part on a combination of a first score obtained for a bridge language translation for which the preferred translation is obtained and a second score obtained for the candidate translation to which the preferred translation corresponds and on a number of distinct bridge language translations for which the preferred translation is obtained. |
