Investigating Dominant Word Order on Universal Dependencies with Graph Rewriting

Proceedings of Recent Advances in Natural Language Processing, pages 281–290 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_033 281 Investigating Dominant Word Order on Universal Dependencies with Graph Rewriting Hee-Soo Choi Sorbonne Universit´e and Universit´e de Lorraine, CNRS, Inria, LORIA, F-54000 Nancy, France hee-soo.choi@loria.fr Bruno Guillaume Universit´e de Lorraine, CNRS, Inria, LORIA, F-54000 Nancy, France Bruno.Guillaume@loria.fr Kar¨en Fort Sorbonne Universit´e / STIH and Universit´e de Lorraine, CNRS, Inria, LORIA, F-54000 Nancy, France karen.fort@loria.fr Guy Perrier Universit´e de Lorraine, CNRS, Inria, LORIA, F-54000 Nancy, France Guy.Perrier@loria.fr Abstract This paper details experiments we performed on the Universal Dependencies 2.7 corpora in order to investigate the dominant word order in the available languages. For this purpose, we used a graph rewriting tool, GREW, which al- lowed us to go beyond the surface annotations and identify the implicit subjects. We first mea- sured the distribution of the six different word orders (SVO, SOV, VSO, VOS, OVS, OSV) in the corpora and investigated when there was a significant difference in the corpora within a given language. Then, we compared the ob- tained results with information provided in the WALS database (Dryer and Haspelmath, 2013) and in ¨Ostling (2015). Finally, we examined the impact of using a graph rewriting tool for this task. The tools and resources used for this research are all freely available. 1 Introduction Language typology has proven to be useful in nat- ural language processing (NLP) (Bender, 2016; O’Horan et al., 2016), for example for improving performance in language transfer (Naseem et al., 2012; Ahmad et al., 2019) and joint learning. As noted by O’Horan et al. (2016) “WALS is currently by far the most commonly-used typolog- ical resource in NLP due to its broad coverage of features and languages”. However, the WALS database (Dryer and Haspelmath, 2013) has been compiled from the work of 55 linguists1 and is not systemically based on a large quantity of data. Moreover, it does not provide all the considered features for all the languages it covers. On the other hand, the Universal Dependencies (UD) framework (Nivre et al., 2016) provides a 1See: https://wals.info/author. large number of corpora annotated in dependency syntax (in version 2.7, there are 183 corpora for 104 languages). We decided to automatically extract from the UD corpora one of the most used features in NLP, the dominant word order, i.e. the way the subject (S), verb (V) and object (O) are ordered in a language (feature 81A in WALS). To do so, we use a freely available graph rewriting tool, which allows us to perform complex searches, to take into account the context of the construction and to add or modify the existing annotations to expose relations which are not directly accessible in the corpora. These experiments led us to define what is a dominant word order, to observe the distribution in word orders within the corpora of a given lan- guage, to determine the frequency of the different word orders in all the considered corpora, and to compare the obtained results with those of existing …

Content truncated to 500 words