XLE

xle@mailman.uni-konstanz.de

42 discussions

semantic roles for LFG grammar engineering
by Adam Przepiorkowski 03 Feb '14

03 Feb '14

Dear All, [Apologies for cross-posting.] In the context of enriching the Polish LFG grammar with semantic representation, we are looking for a set of semantic roles (Agent, Patient, Beneficiary, etc.) that could be used to mark arguments (and possibly adjuncts) of verbs and other predicates. This set should be exhaustive in the sense that it should be possible to assign – more or less deterministically – a semantic role to any argument (and possibly adjunct) of any predicate. For this reason the standard – in LFG textbooks – sets of some 7 semantic roles do not seem sufficient. Instead, we are looking at larger repertoires proposed in VerbNet, in FrameNet and in John W. Sowa's work on Knowledge Representation. We don't have any strong views about any particular set of semantic roles, as long as it is exhaustive and applicable to real texts (as opposed to being merely theoretically interesting). Has anybody in the LFG community faced a similar task? If so, what set of semantic roles would you recommend? At the moment, we are wavering between VerbNet and Sowa's system, both being more manageable than numerous roles offered in FrameNet, but we are open to other solutions. Many thanks, best regards, Adam P. -- Adam Przepiórkowski ˈadam ˌpʃɛpjurˈkɔfskʲi http://clip.ipipan.waw.pl/ ____ Computational Linguistics in Poland http://jlm.ipipan.waw.pl/ ___________ Journal of Language Modelling http://zil.ipipan.waw.pl/ ____________ Linguistic Engineering Group http://nkjp.pl/ _________________________ National Corpus of Polish

4 3

ParGram sentences for upcoming meeting
by Sebastian Sulger 23 Jan '14

23 Jan '14

Hi, I attach the ParGram sentences for the upcoming ParGram meeting. If you would like to take part in the structure comparison, you should do the following: - translate the sentences to your language; - gloss them and put them in a LaTeX file (use the attached Debrecen handout as a guideline; use the Leipzig Glossing Rules); - parse them; - (opt. disambiguate them); - save the c- and f-structures to eps files; - send the eps files as well as the LaTeX file to me. I volunteer to compile the handout for the meeting. We will not have enough time to look at all 50 sentences in detail, so you are welcome to suggest sentences we can focus on. The plan is to add the sentences to the ParGram Parallel Treebank (accessible on INESS: http://iness.uib.no/iness/main-page) after the meeting, i.e. after we have ensured parallelism. Please let us know Best, Jani -- Sebastian Sulger FB Sprachwissenschaft Universität Konstanz http://ling.uni-konstanz.de/pages/home/sulger

3 3

fs_auntsubattr
by Adam Przepiorkowski 25 Nov '13

25 Nov '13

Dear All, We are currently performing some experiments with training probablistic weights for a Polish LFG grammar using cometc. We are considering possible uses of the fs_auntsubattrs feature template. It is clear that trying all combinations of the grammar's attributes as the feature's arguments makes no sense, so we are trying to come up with some linguistically motivated choices. We thought that it would be useful for us to see how this feature is used in other probabilistic LFG grammars. We would be very grateful for some examples of fs_auntsubattr features which proved efficient in stochastic disambiguation. Many thanks in advance, Katarzyna Krasnowska Adam Przepiórkowski -- Adam Przepiórkowski ˈadam ˌpʃɛpjurˈkɔfskʲi http://clip.ipipan.waw.pl/ ____ Computational Linguistics in Poland http://jlm.ipipan.waw.pl/ ___________ Journal of Language Modelling http://zil.ipipan.waw.pl/ ____________ Linguistic Engineering Group http://nkjp.pl/ _________________________ National Corpus of Polish

1 0

fighting timeouts in XLE grammars?
by Agnieszka Patejuk 09 Oct '13

09 Oct '13

Dear all, We've been developing POLFIE, an LFG grammar of Polish, at ICS PAS for over 3 years. We'd like to concentrate on extending the empirical scope of the grammar, but we're having performance issues which affect the results quite badly. Currently we're testing the grammar on sentences extracted from Składnica, a treebank of Polish. There are 8333 sentences, average sentence length is 10 (all segments are counted, including punctuation). How are these sentences parsed? – an individual dictionary is created for each sentence (so the words are already disambiguated morphosyntactically) – each sentence is parsed on its own, in one XLE run. The following performance variables are used when parsing: – 100 seconds (set timeout 100) – 4096 MB memory (set max_xle_scratch_storage 4096). Current results (out of 8333 sentences): – parsed: 6926 – failed: 154 – out of memory: 11 – timeout: 1228 – unknown error: 14 Almost 15% of sentences are timed out, which is very worrying. The average length of a parsed sentence is almost 9 (8.74), while the average length of a timed out sentence is almost 19 (18.67). Have you had similar problems? Are you parsing real sentences, how long are your sentences? Do you have any suggestions what we could do so as to reduce the number of timed out sentences? Best, Agnieszka

6 8

ParGram/XLE Redmine
by Sebastian Sulger 15 Aug '13

15 Aug '13

Hi all, In an effort to be a little more sophisticated about websites/wikis/forums, I have created a Redmine (http://www.redmine.org/) site for collaborative work on all things related to ParGram/XLE. The site is at: http://typo.uni-konstanz.de/redmine The platform has two projects: ParGram and XLE. - ParGram project: wiki, calendar, documents modules. This wiki replaces the old one (http://typo.uni-konstanz.de/pargramwiki is obsolete, so please update your bookmarks). - XLE project: documents, support forum. The support forum is for XLE related questions. The idea here is that we can give some support to people new to XLE (this was a suggestion that came up during the Debrecen meeting). Only people that are registered with the Redmine system are allowed to make changes (create/update wikipages, post forum messages). You can create your account, I will then activate it. To simply view the pages, one does not need an account. Feel free to distribute the above address to people interested in ParGram/XLE matters. ---- On a related note, the ParGram workspace with all the files from past meetings is here: http://ling.uni-konstanz.de/pages/home/pargram_urdu/pargram To access that page, you (still) need a user account, which you can get by contacting Anja Leiderer (anja.leiderer(a)hotmail.de) with your desired username/password combination. Please contact me with any questions that may come up. Enjoy your holidays! Best, Jani -- Sebastian Sulger FB Sprachwissenschaft Universität Konstanz http://ling.uni-konstanz.de/pages/home/sulger

1 0

REMINDER: Extension of ParGramBank
by Sebastian Sulger 06 Jun '13

06 Jun '13

Hi everyone, this email is a reminder for those grammar projects that take part in ParGramBank. We had decided we should extend the treebank *by June 30th* to include all 50 of last year's test sentences (I attach them here for your convenience). Extending the treebank by the end of June will enable us to discuss issues/exchange ideas concerning the treebank (e.g., wrt. alignment) at this year's ParGram meeting in Debrecen. It would be great if as many of you as possible could participate in making this treebank a success! If your grammar project is not a part of ParGramBank as of yet, please send me an email to find out how to contribute in the future. Best, Jani (from the Urdu grammar team at Konstanz) -- Sebastian Sulger FB Sprachwissenschaft Universitaet Konstanz http://ling.uni-konstanz.de/pages/home/sulger

1 0

Re: [XLE] ParGram Parallel Treebank
by Sebastian Sulger 28 Nov '12

28 Nov '12

Hello ParGrammers, I am emailing you again to ask you for your participation in building a parallel treebank for the ParGram project. We have already received contributions from the Wolof, Norwegian and Georgian grammars (and thanks a LOT for that!). There are many more languages represented in the ParGram project, though, and the more participate in this task, the better! I am attaching the file with the sentences as well as a readme on how to submit your parses to this email. So far, we have decided to submit the first 16 sentences. Ideally, you would submit your parses already disambiguated (as you do anyway for the ParGram meetings) - but if you can't manage to disambiguate them, you can always to that at a later stage. We would like to extend the deadline once more: please submit your structures until the end of the year. Also, if you have questions about how to submit or problems with the Subversion repository, please let me know. Best wishes, Jani (from the Urdu team in Konstanz) Am 25.09.12 16:52, schrieb Sebastian Sulger: > Dear ParGrammers, > > At the last ParGram meeting, we discussed possible ways of consolidating > our accumulated pargram knowledge and making it more widely available. > > One idea was to begin creating a parallel treebank based on the > f-structure comparison exercise we do routinely at ParGram. For the Bali > meeting, we had circulated examples of constructions (intransitive, > transitive, passives, relative clauses, questions, ... - see attached > file sentences_2012.txt) We managed to do the first 20 or so of these > and would like to use these as a basis for the parallel treebank. > > It would be great if as many of you as possible could participate! > > We had decided on an October 31st deadline for submitting the first 16 > sentences to the repository, but would now extend this deadline to > November 30th. > > To facilitate the development of this resource, we have created a > Subversion repository here at Konstanz for storing c- and f-structures > as well as sentence glosses. > > A readme file with detailed instructions on how to proceed with > uploading your structures to the repository is also attached to this > email. You can check out the repository like this: > > svn checkout > svn+ssh://pargram@cald.sprachwiss.uni-konstanz.de/Users/pargram/bali-meeting-structures/repos > > > The password is 'pargramrocks'. > > If you need help with Subversion (http://subversion.apache.org/), please > let me know. > > Cheers, > Jani & the Konstanz team > -- Sebastian Sulger FB Sprachwissenschaft Universitaet Konstanz http://ling.uni-konstanz.de/pages/home/sulger

1 0

ParGram material available
by Sebastian Sulger 02 Nov '12

02 Nov '12

APOOLOGIES FOR CROSS-POSTING Dear All, since the Bergen BSCW workspace for storing ParGram files has been closed, we have been working on an alternative solution at Konstanz. Anja Leiderer, one of our students, has now implemented a website using PHP for you to browse ParGram related files (meeting discussions, presentations, etc). The website is here: http://ling.uni-konstanz.de/pages/home/pargram_urdu/pargram/index.php To access the directory, you need to have a username and a password. Those of you who wish to have access can send email to Anja and me with their choice of a username/password combination, and we will give you access: anja.leiderer(a)hotmail.com sebastian.sulger(a)uni-konstanz.de People can send future meeting files to us for upload. The search does not work as of yet --- we are working on debugging and improving it at the moment. You can browse and download files, though. Suggestions are welcome, of course! Best wishes, Jani & the Urdu team at Konstanz -- Sebastian Sulger FB Sprachwissenschaft Universitaet Konstanz http://ling.uni-konstanz.de/pages/home/sulger

2 2

subsumption and constraining equations
by Dag Haug 17 May '12

17 May '12

Dear all, I have a problem understanding how subsumption and constraining equations work together in the XLE. As far as I can see, XLE's behavior is different from what the documentation says, but I am not sure I understand. The basic problem is this: I want to allow nominalizations of S's that have a CASE value but not other S's. The rule I use is this: NP --> { N' | S: ! << ^ (! CASE) } So an NP can expand to S, in which case the S's f-structure should subsume that of the NP, so all the information gets passed up from the f-structure of the S to the f-structure of the NP. Then there is a constraining/existential equation requiring the S's f-structure to have CASE - or so I thought. It wouldn't be enough that the NP gets assigned CASE from above, so to speak. If I read it correctly, he documentation of explicit subsumption says that it behaves differently from implicit subsumption (in sets) in that constraining equations don't get distributed. But as it happens, XLE accepts a solution where S in itself has no case, but the NP gets case from the verb. So even though it displays two different f-structures for the NP and the S (since they differ in CASE), it allows for the existential constraint to be satisfied in the "wrong" f-structure. I can achieve the effect I want by adding @(COMPLETE(! CASE)), forcing local satisfaction of the existential constraint. In that case, there is no need to use subsumption at all, of course. So implementation-wise I am ok, but I am interested in subsumption from a theoretical point of view as well, so if anyone can help explain XLE's behavior and whether there is a particular theoretical reason for it, I'd be grateful. I have a minimal implementation which recreates the problem, if anyone is interested. Best, Dag Haug

4 7

Question: generating from s-structures
by Francois Lareau 16 Apr '12

16 Apr '12

Dear XLE-ists, We are using XLE in a bilingual text generation project (http://web.science.mq.edu.au/~ayeye), and we would like to be able to generate sentences from s-structures rather than f-structures. It seems that projections other than c- and f-structures get encoded in the exported prolog files as attributes of the f-structure. This means that generating from an s-structure technically boils down to generating from an underspecified f-structure (where everything is missing except the "s::" attributes). But even when declaring all attributes of f-structures addable, we still cannot generate from s-structures. I think this is because the PRED attribute cannot be added to an underspecified f-structure, due of its central role in ensuring completeness and coherence. Is there a way to work around this limitation, short of using transfer rules? Has anyone ever tried to generate from other projections than the f-structure? Any help would be much appreciated. Kind regards, François Lareau Post-doc fellow Centre for Language Technology Macquarie University

2 1

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

XLE