Yes, I'm on the Japanese/English beat, especially with emails. Consider this:
Structured communications like emails have a probability-based translation for many of the "frozen" expression. (e.g., This is Name). It is also important for languages that drop the pronouns, translating to languages that have pronouns. Most importantly, you can create a paradigm of detecting a dialogue, and who is speaking to whom, and what social-rank relationship they have. (Master/Servant, what in between.) This can help translate languages that rely highly on context to languages where the context is overtly stated. Such as Japanese to English. And vise-versa. Details on the concept are below. In this snip, you can detect who is the writer and who is the addressee: ○○社長 お疲れ様です。本山です。 ... 明日、奥様に丁重にお詫びします。 大変、申し訳ございませんでした。 本山 Things you can detect 1. Combinatorial patterns. 1a. President blank-blank at the top (likely the addressee) 1b. Next line, "Cheers for hard work" (almost certainly speaking "up" to President) 1c. Repetition of the name 本山 in the first line as well as signature. This tells us the best translation in this first like is "This is Name" (rather than, "Is is Name" or "Name exists"). 1d. Last line before signature, very humble apology, "Terriby sorry for the lack of any excuse" -- now 100% certainty that an employee is writing to their boss. So. Since you are already detecting blocks and patterns, check out patterns among sentences in the communicatins. If you can estabilish with high probability that there is a dialogue happening,and who is talking to whom, then you can surmise what the pronouns are, as well as the overall tone, even when they are not overt. What do you think? -- You received this message because you are subscribed to the Google Groups "General" group. To post to this group, send email to [email protected]. To unsubscribe from this group, send email to [email protected]. For more options, visit this group at http://groups.google.com/group/google-translate-general?hl=en.
