Besides, with projects such as Google books and simply by indexing the web, Google has acquired corpuses of text that are biased towards English. Presumably Google mostly gets requests of translations to and from English. Besides, people seldom need to translate things from Swahili to Yiddish. That's a lot of data, and many of these bilingual corpuses might be so small the quality of the translation would be pretty bad anyway. This means that Google would need 71*70 = 4970 statistical models from each language to the other, generated from 71*70/2 = 2485 bilingual corpuses, since presumably a single language corpus can be used to generate translation models in both directions. Google Translate handles 71 languages if I counted correctly. Statistical Machine Translation can only get you directly from one language to another. I hope you realize by now how this relates to Google Translate.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |