From: Jon Blossom Subject: Re: APPS: I want my VR-TV NOW!! (Lingo) Date: Thu, 3 Jun 93 10:20:12 PDT This whole thread started with a post on the question: would it be possible to create a multi-national virtual world in which everyone spoke their native language, and everyone's local machine translated everything for them? People speaking Balinese could talk freely with people speaking German if they had a "driver" for it. Well, somehow we got off on a tangent about AI and language understanding, and I hope I can it to rest with this post. Sorry to waste the bandwidth for those who aren't interested in AI. Yes, such a virtual world would be great, but believe me, it's nowhere in the near future... The problem of computerized language translation are not at all trivial, and most AI researchers agree with me. I know - after years of playing text adventures I was enamored with Natural Language Processing, and I explored it to fairly lengthy extents. The biggest problem with machine translation is in assuming that there is a one-to-one mapping of words in one language to words in another. Sure, you can say "chair" in English and translate it to "siege" in French or "silla" in Spanish. But if you've learned more than a couple of languages, you'll realize that most words have to be translated based on their contextual meaning. Which means that *the computer has to understand the precise meaning of every word in the target phrase in context*! For instance, there's a word in Arabic that means something like "my father's sister's daughter." We'd call it "cousin" with a great lack of information. If we were translating from English to Arabic and were talking specifically about our father's sister's daughter, we'd have to use their word for it in the translation. It's not just a matter of the complexity of the English language. Yes, ours is a very twisted grammar. But Japanese? German? Those languages have a more hellish grammar than English. And in ANY language, you can express the same thought in a myriad of ways. Believe me, translation is not a simple task. Haven't you ever been involved in a translation and encountered a word where you say "Hmm... there's no word for this in English!" Try religious terms. Buddha. Karma. Tao. You know why we use those words in English? Because there's just no way to communicate their contextual richness directly in the English language. The Japanese have hundreds of words describing the taste of tuna. English supports very few. "Tasty. Salty. Sweet." They just don't cut it for people who have lived their lives on a big island and are descended from fishermen. >fsmar@acad3.alaska.edu writes: >first there are speech to text coversion programs availible now (i Speech-to-text is just a question of Speech Recognition. Sure, the computer can hear certain morphemes and perform complicated searches on an internal database. But does it understand what it's typing? Could it render your sentence into another language? >....also there are language traslators availible now to...you type in >what you want to say and it spits it out to you in whatever language >the translator is made for...there is spanish, and i belive You're a victim of marketing if you believe that these things will translate whatever you want to say. Those little things are electronic dictionaries that do basic, one-word, one-to-one translations of words like "bathroom" from one language to another. (Incidentally, if you're asking someone in France where the bathroom is, would you ask for "le cabinet", "la salle de bains", "la toilette", or something else? Depends on what you mean by bathroom!) These things are great for travelers, horrible for translators. Don't you think the UN would use them instead of live real-time translators if they could? In short, the problem of machine translation is NOT one that has been solved, and it is probably not one that will be solved any time soon. Sure, we might be able to translate some very specific subset with a very specific grammer ("GET THE MAGIC LAMP AND RUB IT"), and that might be enough for some basic interaction. But full-text recognition and understanding is something for Star Trek and beyond.