A followup to Peter Turney's message from Rada Mihalcea. Sufficiently interesting I thought to merit forwarding to our list!
-- Ted Pedersen http://www.d.umn.edu/~tpederse ---------- Forwarded message ---------- Date: Fri, 16 Jul 2004 13:28:28 -0500 (CDT) From: Rada Mihalcea <[EMAIL PROTECTED]> To: Peter Turney <[EMAIL PROTECTED]> Cc: [EMAIL PROTECTED] Subject: Re: [Senseval-discuss] RE: Re: Senseval-3 Panel on Applications of WSD First, thanks Peter for some most interesting remarks! I believe this is something that many WSD-people have on their mind, given also that so far there was no 'explicit' proof that WSD would help MT. With no intention to "spoil" the panel discussion, let me make a couple of remarks myself (I couldn't refrain ;-) One - although I do not know all the details of the recent NIST/MT evaluations (what track did the two WSD-based teams participate in?, what is the track where SMTs participated? etc.), I do not think that the current test-bed is one that would give much credit to a WSD-based system. Even if we were to have 100%-accurate WSD systems, there isn't much that one could do to "beat" an SMT relying on some 100 mil.words parallel corpora. Perhaps a more realistic test-bed would be to compare various flavors of MT (WSD-for-MT included) on a random genre, on a random pair of languages (a bit like the "surprise language" evaluation from last year) -- which wouldn't give a head-start to SMT. For instance, it took me a few months to collect a 1 mil. words parallel corpus for Romanian-English -- how far could SMT go with such a small corpus? and can we afford spending all this time to put together parallel texts every time we consider a new pair of languages? Second - you made a great comment about Bleu! At the last MT-Summit, somebody (I believe it was Ed Hovy?) pointed out that Bleu is by definition a statistical measure, and thus it favors statistical systems, and discourages other flavors of MT. Which may not be good for the MT world in general. To make an analogy, if you plant several seeds, but you only water one (Bleu watering SMT), you'll never know if one of the other seeds would have grown into a better plant. Maybe WSD-for-MT is one of the other seeds? (or maybe not, but how could we know, if we don't give it some credit) A very interesting topic -- thanks again Peter for bringing it up! Rada _______________________________________________ Senseval-discuss mailing list [EMAIL PROTECTED] http://listserv.hum.gu.se/mailman/listinfo/senseval-discuss ------------------------ Yahoo! Groups Sponsor --------------------~--> Yahoo! Domains - Claim yours for only $14.70 http://us.click.yahoo.com/Z1wmxD/DREIAA/yQLSAA/x3XolB/TM --------------------------------------------------------------------~-> Yahoo! Groups Links <*> To visit your group on the web, go to: http://groups.yahoo.com/group/nlpatumd/ <*> To unsubscribe from this group, send an email to: [EMAIL PROTECTED] <*> Your use of Yahoo! Groups is subject to: http://docs.yahoo.com/info/terms/
