[01:34:31] *** Joins: spectre (~fran@124.115.202.84.customer.cdi.no) [02:13:57] *** Quits: Masxmasx (~masx@unaffiliated/masxmasx) (*.net *.split) [02:16:52] *** Quits: spectre (~fran@124.115.202.84.customer.cdi.no) (Ping timeout: 264 seconds) [02:24:57] *** Joins: Masxmasx (~masx@unaffiliated/masxmasx) [02:37:06] *** Quits: Masxmasx (~masx@unaffiliated/masxmasx) (*.net *.split) [02:37:06] *** Quits: jmvanel (~jmvanel@78.193.21.40) (*.net *.split) [02:37:08] *** Quits: Flammie (~flammie@80.111.32.7) (*.net *.split) [02:37:09] *** Quits: evariste (~evariste@dhcp2-225004.cs.chalmers.se) (*.net *.split) [02:37:14] *** Quits: vlopez (~vlopez@schumpeter.ekb.sgsnet.se) (*.net *.split) [02:37:15] *** Quits: vin-ivar (~vinit@122.170.55.215) (*.net *.split) [02:37:17] *** Quits: esg (~emil@esg.xen.prgmr.com) (*.net *.split) [02:52:09] *** Joins: Masxmasx (~masx@unaffiliated/masxmasx) [02:52:09] *** Joins: vin-ivar (~vinit@122.170.55.215) [02:52:09] *** Joins: jmvanel (~jmvanel@78.193.21.40) [02:52:09] *** Joins: Flammie (~flammie@80.111.32.7) [02:52:09] *** Joins: vlopez (~vlopez@schumpeter.ekb.sgsnet.se) [02:52:09] *** Joins: evariste (~evariste@dhcp2-225004.cs.chalmers.se) [02:52:09] *** Joins: esg (~emil@esg.xen.prgmr.com) [07:03:58] was asleep [07:04:00] anyway [07:04:04] about the PGF library [07:04:30] parse p l t "this fish" essentially returns a list of parse trees for the sentence? [07:27:06] i think i'm finding it hard to see how morph disambiguation really works. the probs file is essentially just unigram probabilities for every form of the word? [07:27:17] ie, non-contextual [09:58:09] *** Quits: vin-ivar (~vinit@122.170.55.215) (Read error: No route to host) [10:01:56] *** Joins: vin-ivar (~vinit@122.170.55.215) [10:05:12] *** Joins: spectre (~fran@124.115.202.84.customer.cdi.no) [11:15:40] vin-ivar: the probs file has probabilities for all functions, not just lexical [11:15:45] and there's nothing about the word forms [11:16:08] for instance, question clauses have really low probability because the probs are from wall street journal [11:16:52] there's not really morphological disambiguation; if you do a morpho lookup for a single word, you get all analyses that the word on its own could be [11:17:12] if you parse it as a part of a sentence, some options are discarded just because they are not grammatical [11:17:25] ah, that's what I meant [11:17:34] so what part of the whole chain does the discarding? [11:17:39] no wait [11:17:42] that's silly of me [11:17:48] e.g. w="can", you get verb and a beer can; "I can do it", you get only the verb reading [11:17:54] the parse trees that don't form are obviously rejected [11:18:02] yeah :) [11:18:43] and then you have ambiguities that are still not solved, then you get the most probable (according to wall street journal) trees first [11:19:01] there should be a function that outputs the probability as well [11:19:23] that's in the C runtime, there's a haskell FFI [11:19:30] it's called PGF2 [11:20:19] iin src/runtime/haskell-bind [11:20:51] damn, that looks nice, I'll check it out [11:20:52] (but if you're happy using the haskell version for now, you don't need to change :) there are small differences in some of the functions, mainly just what you need to give as argument) [11:21:01] inariksit: you're familiar with Haskell, yeah? [11:21:05] yeah [11:21:29] would you mind taking a look at my code? i've never contributed to a significant haskell project so I'm not sure how, well, 'clean' it is [11:21:33] i'll send it over tonight [11:21:38] okay, can do! [11:21:43] ^^ [11:21:59] thanks :) [11:30:28] okay, so if I'm not reading this wrong [11:30:37] porting the morph. disambiguation [11:30:55] initially, just build the morph analysis, then build parse trees, reject the ones that don't show up in any tree [11:30:56] spectre^ [11:31:09] ooh this is interesting functionally because no state, hehe [12:15:47] *** Quits: spectre (~fran@124.115.202.84.customer.cdi.no) (Ping timeout: 246 seconds) [12:42:20] *** Joins: spectre (~fran@dhcp49-ans.wifi.uit.no) [16:23:59] *** Quits: spectre (~fran@dhcp49-ans.wifi.uit.no) (Ping timeout: 246 seconds) [16:56:02] *** Joins: spectre (~fran@115.114.202.84.customer.cdi.no) [18:20:33] *** Quits: jmvanel (~jmvanel@78.193.21.40) (Quit: Quitte) [21:04:22] *** Quits: Flammie (~flammie@80.111.32.7) (Read error: Connection reset by peer) [21:09:43] *** Joins: Flammie (~flammie@80.111.32.7) [21:15:57] *** Quits: Flammie (~flammie@80.111.32.7) (Read error: Connection reset by peer) [21:18:17] *** Joins: Flammie (~flammie@80.111.32.7)