nadiatims wrote:

what happens if you ask one of the AIs "When did WW2 start?" or something anyone with access to wikipedia can easily check?
Can you really answer questions like that just with statistical text analysis?

Yep, it's been done. https://www.google.co.uk/search?q=when+did+wwii+start

Well, if giving the answer in the form of a webpage strikes you as a copout, I think Google is working on their own sort of knowledge database that's probably a bit closer to what you're thinking of.

dizmox wrote:

I guess this is straying away from the topic of machine translation, I don't think anything nearly as powerful as an AI that could beat the Turing test would be needed for 99% accurate machine translation.

Back to the Turing test though: feed all the communication history of mankind into a machine with the the most sophisticated of text analysis algorithms conceivable if you want - it might be a fantastic oracle, but it will still be easily outed as a machine by someone asking the right questions - ones that rely on its lack of understanding of the real world (and more generally, anything) and its inability to deal with novel scenarios.

Maybe it could you fool into thinking it's an extremely dull witted human who didn't have an intelligent opinion on any new situation described to it mid-conversation (and thus wouldn't be able to just mindlessly search its data banks for one), but I'd argue that's not really in the spirit of the test.

The Turing test is a weird thing though.  In reality, almost any crappy chatbot will pass the Turing test if the person talking to it has no reason to believe they are talking to a computer.  So if you don't tell people that they might be talking to a computer, even simple programs can pass the turing test.

Likewise, if you do a regular Turing test (i.e., people know they might be talking to computers) using people who don't know much about computers and AI, a fairly good program will pass the Turing test.

But if you use people knowledgable about the Turing test, as well as computers and AI, almost any program will fail the Turing test.

My favorite is the backwards Turing test.  I read a blog many years ago at the end of the chat era, and the blogger said that someone had posted his Yahoo chat name as being an interesting bot.  Of course, it was really him, a human.  He got a lot of messages from people who thought he was a bot, and try as he might he was unable to convince most of them that he was in fact a human.

Anyway, I don't know why you think a computer that can pass the Turing test needs to be more powerful than a computer that can do extremely accurate machine translation.

Last edited by Tzadeck (2012 July 22, 8:07 pm)

turvy wrote:

What about Chinese <-> English?

I can't test this myself, but from what I can gather with my risible knowledge of Mandarin and common sense, I think it translates much better than whatever you can get from Japanese.

Splatted wrote:

... use of a search engine may allow the computer to find several possible ways of translating a word or phrase, but what criteria does it use to evaluate these choices? If this was a game of chess the criteria would be obvious, whether it leads to a win, a loss or a draw, but no such criteria exist for translation, so I fail to see how the computer could make a reliably good decision.

What if the program searched for corresponding context for all of the possible translations and then selected the translation that occurs most often in similar contexts to the phrase that needs to be translated?

For example, we want to translate 「空気が悪い」. Probably the two most common meanings of this phrase are "the air is stale" (i.e. literally uncomfortable to breathe) or "there is tension in the air". A good MT program would look at the context of 「空気が悪い」 and the compare it to all other contexts in which 「空気が悪い」 is used, and then select the translation that most often occurs in similar contexts (of course, this presumes that the input field doesn't include just the phrase 「空気が悪い」 by itself). For example, if the preceding context mentions a fight or dire situation, then MT program will choose "there is tension in the air". However, if the context refers to dust, difficulty breathing, or a need for fresh air, then the program will choose "the air is stale".

I'd imagine such a program would need to somehow search its database for similar contexts and determine the probability that a certain translation is the most appropriate one given its context. However, it doesn't seem as insurmountable of a problem as it initially may seem.

For a Turing test, if I ask it three obscure questions and it gets them right, does it fail the test?  Unless it is a Jeopardy winner, I wouldn't expect a real person to get the answers right.