Ask HN: Automatic Translation, why not? | Hacker News Reader

Ask HN: Automatic Translation, why not?

1 pointmfalcon9 comments

I'm currently working in a project which was the creation of a series of tools to improve language learning in an original way. I've recently incorpored in the project which was being executed a while ago.

The thing is, some of the people here are convinced that we could make a translator better than any of the translators in the market(google, babylon, etc...).

In my opinion the group doesn't have possibilities of succeding in the topic because of the lack of experience in the field and of course, because it's a really hard problem.

I would like to find material to show them that we're going to lose time and money trying to make something we can't. The approach to the problem is the sentence analysis, identify every element of the sentence being aware of the context in order to choose the correct tranlation of a word.

I've been reading the Stanford course of NLP but it doesn't especify why the approach mentioned above is discarded.

Consider: "I'll set this here."

How many different meanings can you find for that? What about the classic: "Time flies like an arrow, but fruit flies like a banana."

Sentence parsing is a huge field with a lot of very good work. If you only have "grammatically correct" (whatever that means) sentences, then the "Link Grammar" approach is very good at parsing in context.

Natural sounding language production is extremely difficult. Almost everything machine produced sounds stilted and non-natural.

On the other hand, everyone has to start somewhere, and a truly new approach may yield better results than anyone else's. Unless you take the time, effort, money and risk, how will you know?

Thanks for mentioning "Link Grammar", I googled it and I find very interesting information.

What is their background? If they have none as you say, it is going to be unlikely.

You mentioned reading the stanford course, there are also video lectures:

http://see.stanford.edu/see/lecturelist.aspx?coll=63480b48-8...

If they decide to proceed, two good NLP books:

1.Speech and Language Processing

2.Foundations of Statistical Natural Language Processing

The group is composed by: a linguist, an "experienced programmer", me, two business guys and a translator.

The "experienced programmer" is what they said he is, but after a week I realised that he's not. I didn't know how to make an HTML parser, so I have to do it for him(I begin programming this year).

Thanks you for the books, I'll give them a try.

Think about how often idiomatic phrases come up in English. Then try to translate those phrases into, say, Chinese. And vice versa. Good luck with that.

"Steel rice bowl" makes no sense in English, and "apple of your eye" makes no sense in Chinese.

They say we could have a table with a series of idiomatic phrases. Every sentence has to be compared with that table in order to know if its an idiomatic phrase or not. I think that's not a good idea.

That's really, really not a good idea. You need to get them to talk to someone who knows for real how hard this problem is.

I think that's what everyone tries to do. What your group needs to do is talk to some of the people who have already worked on this.

Then again, there's no reason, a priori, that your group can't build the better mousetrap if they've got the requisite skills and resources.

The problem is that we are from Argentina, sorry not to mention that. As far as I know, nobody in my country had a similar project.

Reply on news.ycombinator.com