Tuesday, May 24, 2011

Birthing an unprofessional society

I found something interesting a couple days ago.

Computational linguists/NLP researchers/implementers are really bad at communicating.

Obviously, the problem does not lie in their language skills.

Now, here is the issue. The ACL (Association of Computational Linguistics) is more or less the only association that binds together people interested in CL/NLP. And well, they are an academic association, interested in things like membership dues, trade shows, focus groups, papers, publishing, etc.

Nobody cares about the ACL, except professors trying to get/keep tenure, researchers looking to steal/borrow ideas, and engineers who have no clue where to start. Start with the research, that makes sense. Or does it?

I looked really hard in and around the ACL, and couldn't even find a forum or a chatroom.

No forum. No chatroom.

To me, this is the internet equivalent of a city with no phone lines. There is no way to ask questions, show off stuff, or do jack, really.

So, there's a vaccum. Time to fill the vaccum. The thoughts that I have had for CL/NLP community are as follows:

1) Must have a community website. A good example of a thriving community is TeamLiquid.net -- a Starcraft 2 community.

2) Must have examples. LOTS of examples of functioning, attachable code. If I want to know how to perform the TFIDF method, I should be able to see that and work through it, step by step. To this end, a wiki needs to be attached. This is the groundwork necessary to perform automated ranking and testing in a non-piecewise way.

3) May be sponsored, never affiliated. Anything left here is free for the taking. This would represent published, open techniques that are plentiful. This is very different from a professional society, because it taps the sea of amateurs. It places demands on the CL/NLP community to have their algorithms available as web services, as applications that can be interfaced with everything and anything else.

4) Communication must be informal, instantaneous, and recordable. Forums and chatrooms are perfect for this.

5) Quality must be ever-improving, but with a low bar of entry. Quality takes time. Hacks take less time. Both show off a concept, and one is simply more attractive than the other.

Work may begin on this soon, but so far this is the whole of the problem that I see in getting the Semantic Web established and integrated with CL/NLP. They need each other, and one currently has no way to speak ot the other. So... here we go.

No comments: