May 15, 2003

Meaningful machines?  

Metafilter links to this article about improving the quality of machine translations. It seems like a smart enough approach, but it's not going to solve all the translation problems out there. Specifically, I don't think there's a wide enough pool of translated discourse for a computer to be able to solve larger contextual problems.

Let me put it this way: Is the context of this sentence the paragraph in which it appears, the post in which it appears, this whole blog, the blogosphere itself? The problem isn't just that we need to look at context when we translate or understand something. We actually have to be able to figure out which context is important. This kind of thing is going to be much more difficult for a computer to handle - and if the context turns out to have a very wide scope, it may render statistical techniques like this one useless.

I do think this is a step in the right direction - given the computational strength we have at our disposal, this kind of narrow contextual analysis seems like it will clean things up considerably. But we have a long way to go before we can handle more serious translation problems like metaphor or irony.


Post a comment

Remember personal