[ Skip to the content ]

Institute of Formal and Applied Linguistics

at Faculty of Mathematics and Physics, Charles University, Prague, Czech Republic


[ Back to the navigation ]

Publication


Year 2016
Type in proceedings
Status published
Language English
Author(s) Kríž, Vincent Hladká, Barbora
Title Improving Dependency Parsing Using Sentence Clause Charts
Czech title Zlepšení závislostního parsingu pomocí klauzálních grafů
Proceedings 2016: Stroudsburg, PA, USA: ACL 2016: Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics – Student Research Workshop
Pages range 86-92
How published online
URL http://acl2016.org/files/acl-srw-2016.pdf
Supported by 2016 SVV 260 333 (Teoretické základy informatiky a výpočetní lingvistiky) 2012-2016 PRVOUK P46 (Informatika)
Czech abstract Navrhujeme metodu, která zlepšuje závislostní parsing složených vět. Metoda předpokládá segmentaci věty do klauzí a nevyžaduje přetrénování parseru. Klauzální strukturu věty reprezentujeme pomocí klauzálních grafů, které poskytují informaci o vnoření každé klauze. Navrhujeme postup, ve kterém parsujeme klauze nezávisle a vzniklé závislostní stromy vkládáme jako podstromy do finálního stromu pro celou větu. Metodu aplikujeme na češtinu a experimentujeme s MST parserem natrénovaným na PDT 2.0. Dosahujeme zvýšení UAS o 0.97%.
English abstract We propose a method for improving the dependency parsing of complex sentences. This method assumes segmentation of input sentences into clauses and does not require to re-train a parser of one's choice. We represent a sentence clause structure using clause charts that provide a layer of embedding for each clause in the sentence. Then we formulate a parsing strategy as a two-stage process where (i) coordinaed and subordinated clauses of the sentence are parsed separately with respect to the sentence clause chart and (ii) their dependency trees become subtrees of the final tree of the sentence. The object language is Czech and the parser used is a maximum spanning tree parser trained on the Prague Dependency Treebank. We have achieved an average 0.97% improvement in the unlabeled attachment score. Although the method has been designed for the dependency parsing of Czech, it is useful for other parsing techniques and languages.
Specialization linguistics ("jazykověda")
Confidentiality default – not confidential
Open access no
ISBN* 978-1-945626-02-9
Address* Stroudsburg, PA, USA
Month* August
Venue* Humboldt University
Publisher* Association for Computational Linguistics
Institution* Association for Computational Linguistics
Creator: Common Account
Created: 8/8/16 4:38 PM
Modifier: Almighty Admin
Modified: 2/25/17 10:07 PM
***

Content, Design & Functionality: ÚFAL, 2006–2016. Page generated: Fri Nov 24 05:05:01 CET 2017

[ Back to the navigation ] [ Back to the content ]

100% OpenAIRE compliant