[09:57:31] *** Joins: vinit-ivar (~vinit@122.169.33.35) [09:58:49] *** Quits: vin-ivar (~vinit@122.179.140.213) (Ping timeout: 250 seconds) [10:39:29] *** vinit-ivar is now known as vin-ivar [12:34:14] *** Quits: vin-ivar (~vinit@122.169.33.35) (Ping timeout: 250 seconds) [12:36:01] *** Joins: vinit-ivar (~vinit@122.170.42.13) [12:49:48] *** Quits: vinit-ivar (~vinit@122.170.42.13) (Ping timeout: 244 seconds) [12:56:25] *** Joins: vinit-ivar (~vinit@182.70.11.35) [15:24:35] *** vinit-ivar is now known as vin-ivar [15:43:10] *** Joins: jmvanel (~jmvanel@78.193.21.40) [16:08:37] inariksit: [16:30:46] yes? [16:43:02] *** Quits: jmvanel (~jmvanel@78.193.21.40) (Remote host closed the connection) [17:02:07] whoops [17:02:07] i could have sworn i'd sent the next message :s [17:02:07] do you have a moment? had some ideas [17:02:23] I don't have time now, going to choir [17:02:45] cool [17:02:51] ping when you do :) [17:02:55] ok! [21:52:10] vin-ivar: I'm back [21:58:09] hey [21:58:20] so [21:59:22] i've realised that there isn't a one-to-one tag mapping from GF to apertium, there's sometimes missing tags in one of them, etc [21:59:24] however [21:59:52] I *think* it should be possible to deduce the apertium based on the particular combination of GF tags [21:59:53] and maybe context [22:00:40] (i hope i'm not annoying you guys too much with all of this) [22:01:11] anyway, what i'd thought was supervised [22:03:29] no worries, it's good that you ask when you have questions! [22:04:13] and yeah, that would be interesting to see if the information is easily reconstructed [22:04:53] lexical info helps too, like if your GF tags say Det Sg Masc and the word is "el", then what's missing from apertium is just the [22:09:02] cool [22:09:05] yeah [22:09:17] i'd thought something similar [22:10:35] i think i'll give the learning bit a shot (i've written a script to get neat dictionaries out of the stream) [22:10:35] tomorrow [22:10:35] what I'm thinking is [22:11:11] generate a bunch of random strings -> apertium parse, GF parse -> use the apertium parse as a feature [22:11:12] train [22:11:12] et al [22:44:51] sounds like a comp ling conference paper ^_^