ParGram March 2014

pargram@mailman.uni-konstanz.de

2 participants
3 discussions

Spring ParGram Meeting 2014
by Sebastian Sulger 03 Aug '15

03 Aug '15

Hi, At this year's ParGram meeting, we decided we should have 2 meetings in 2014 (spring and fall). Regarding the spring meeting, Ron Kaplan has volunteered to host it at Nuance in California. To find a suitable date, I created a Doodle poll which you can find here: http://doodle.com/niwz4hgptcn3rpsr Please participate in the poll so that we can find a date that suits most of us. By the way: This will be a special occasion since we will be celebrating 20 years of ParGram! Cheers, Jani -- Sebastian Sulger FB Sprachwissenschaft Universität Konstanz http://ling.uni-konstanz.de/pages/home/sulger

3 3

ParGram publications
by Sebastian Sulger 21 Mar '14

21 Mar '14

Hi, I am updating the ParGram website at http://pargram.b.uib.no. If you have any publications related to ParGram which are not listed there under "References", please let me know their details. Best, Jani -- Sebastian Sulger FB Sprachwissenschaft Universität Konstanz http://ling.uni-konstanz.de/pages/home/sulger

1 0

[XLE] Problem with print-feature-forest?
by Katarzyna Krasnowska 03 Mar '14

03 Mar '14

Dear All, I have a question concerning the print-feature-forest command of XLE and would appreciate any help from someone who has some experience using this command. I am working on training a probabilistic variant of the Polish LFG grammar using XLE's functionalities. When preparing the feature forests for cometc using the print-feature-forest command, I came across a problem: it seems that print-feature-forest sometimes gets the feature counts wrong. I ran the print-feature-forest command on the obj-ex.pl file attached to this e-mail, with a features file containing only one line, "fs_attrs OBJ" (so the only feature counted here is the number of OBJ's in the f-structure). obj-ex.pl is a packed parse representation produced by the Polish LFG grammar, with some attributes non-relevant to the problem, such as morphosyntactic ones, removed. The sentence translates as "Look ahead of you and read in the beard". It contains 2 verbs and 2 adjunct prepositional phrases and has 3 possible parses; in two parses, each verb has one of the PP's attached to it; in one parse, the second PP is shared between the verbs. Since a PP which is an adjunct is represented as an f-structure headed by the preposition whose OBJ is the noun's f-structure, it seems that the first two parses contain 2 OBJ's and the third one contains 3 of them since the shared PP appears twice in the f-structure. However, the output feature-forest is: (7 1:2 {1 (1 {2 (3 ) (4 1:2 ) } ) (2 ) } ) which corresponds to two parses with 2 OBJ's and one with 4 of them. Knowing from my previous experience that subsumption can be tricky in XLE, I replaced the two subsume statements in obj-ex.pl with equalities (in the case of this sentence, there actually is an equality), and the feature forest produced for the modified f-structure seems correct: (7 1:1 {1 (1 {2 (3 1:1 ) (4 1:2 ) } ) (2 1:1 ) } ) Perhaps someone knows why it works like this? Am I wrong in expecting what the feature forest should look like? Or maybe it is an known issue with print-feature-forest? I would be very grateful for any advice, since data preparation for my experiments relies heavily on the print-feature-forestcommand . Many thanks in advance, Katarzyna Krasnowska _______________________________________________ XLE mailing list XLE(a)mailman.uni-konstanz.de https://mailman.uni-konstanz.de/mailman/listinfo/xle

1 0

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

ParGram March 2014