Dependency Parsing with Reference to Slovene, Spanish and Swedish

Proceedings of the 10th Conference on Computational Natural Language Learning (CoNLL-X), pages 196–200, New York City, June 2006. c⃝2006 Association for Computational Linguistics Dependency Parsing with Reference to Slovene, Spanish and Swedish Simon Corston-Oliver Natural Language Processing Microsoft Research One Microsoft Way Redmond WA 98052 simonco@microsoft.com Anthony Aue Natural Language Processing Microsoft Research One Microsoft Way Redmond WA 98052 anthaue@microsoft.com Abstract We describe a parser used in the CoNLL 2006 Shared Task, “Multingual Depen- dency Parsing.” The parser first identi- fies syntactic dependencies and then labels those dependencies using a maximum en- tropy classifier. We consider the impact of feature engineering and the choice of ma- chine learning algorithm, with particular focus on Slovene, Spanish and Swedish. 1 Introduction The system that we submitted for the CoNLL 2006 Shared Task, “Multingual Dependency Parsing,” (Buchholz et al., 2006) is a two stage pipeline. The first stage identifies unlabeled directed dependen- cies using an extension of the parser described in (Corston-Oliver et al., 2006). The second stage is a maximum entropy classifier that labels the directed dependencies. The system was trained on the twelve obligatory languages, as well as the optional lan- guage, Bulgarian (Hajiˇc et al., 2004; Simov et al., 2005; Simov and Osenova, 2003; Chen et al., 2003; B¨ohmov´a et al., 2003; Kromann, 2003; van der Beek et al., 2002; Brants et al., 2002; Kawata and Bar- tels, 2000; Afonso et al., 2002; Dˇzeroski et al., 2006; Civit Torruella and Mart´ı Anton´ın, 2002; Nilsson et al., 2005; Oflazer et al., 2003; Atalay et al., 2003). Table 1 presents the results of the system de- scribed in the current paper on the CoNLL shared task, including the optional evaluation on Bulgar- ian. For Slovene, we ranked second with a labeled Language Unlabeled Labeled Attachment Attachment Arabic 78.40 63.53 Bulgarian 90.09 83.36 Chinese 90.00 79.92 Czech 83.02 74.48 Danish 87.94 81.74 Dutch 74.83 71.43 German 87.20 83.47 Japanese 92.84 89.95 Portugese 88.96 84.59 Slovene 81.77 72.42 Spanish 84.87 80.36 Swedish 89.54 79.69 Turkish 73.11 61.74 Table 1: Results on CoNLL 2006 shared task. dependency accuracy of 72.42%. This was not sta- tistically significantly different from the top-ranked score of 73.44%. For Spanish, our labeled depen- dency accuracy of 80.36% is within 0.1% of the third-ranked score of 80.46%. Our unlabeled de- pendency accuracy for Swedish was the best of all the systems at 89.54%. Our labeled accuracy for Swedish, however, at 79.69%, fell far short of the third-best score of 82.31%. We therefore focus on Swedish when considering the impact of our choice of learning algorithm on our label accuracy. 2 Data We divided the shared data into training and devel- opment test sets, using larger development test sets 196 for the languages supplied with more data. The de- velopment test set consisted of 250 sentences for Arabic, Slovene, Spanish and Turkish, 500 sen- tences for Danish and Portuguese, and 1,000 sen- tences for the other languages. 3 The Parser The baseline parser predicts unlabeled directed de- pendencies. As described in (Corston-Oliver …

Content truncated to 500 words