MarsaGram: an excursion in the forests of parsing trees

MarsaGram: an Excursion in the Forests of Parsing Trees Philippe Blache, St´ephane Rauzy, Gr´egoire Montcheuil Laboratoire Parole et Langage - UMR 7309 5 avenue Pasteur, 13100 Aix-en-Provence, France firstname.lastname@lpl-aix.fr Abstract The question of how to compare languages and more generally the domain of linguistic typology, relies on the study of different linguistic properties or phenomena. Classically, such a comparison is done semi-manually, for example by extracting information from databases such as the WALS. However, it remains difficult to identify precisely regular parameters, available for different languages, that can be used as a basis towards modeling. We propose in this paper, focusing on the question of syntactic typology, a method for automatically extracting such parameters from treebanks, bringing them into a typology perspective. We present the method and the tools for inferring such information and navigating through the treebanks. The approach has been applied to 10 languages of the Universal Dependencies Treebank. We approach is evaluated by showing how automatic classification correlates with language families. Keywords: Typology, Treebank, Syntactic properties, Grammar extraction 1. Introduction Treebanks becoming available for many different languages (see for example (Nivre et al., 2015)), they are now impor- tant resources especially in a typology perspective. Asso- ciating syntactic information to large scale natural data of- fers both the extraction of regularities and the description of specific realizations of syntactic constructions. Further- more, thanks to the use of a precise annotation guide, it becomes possible to extract automatically information and apply machine learning techniques to study the distribution of different phenomena. However, the type of information classically encoded into a treebank remains at a high level of generality, that more- over very often remain implicit. For example, constituency- based treebanks encode an implicit phrase-structure gram- mar, that can be sometimes enriched with the annotation of the main syntactic relations. In this case, in order to ren- der such information explicit, it is possible to automatically extract the grammar from the treebank and identify all the realizations of its rules. On top of this, it is also necessary to describe finer-grained information, such as government phenomena, linear order, cooccurrence, etc., which char- acterize better a language than the strict set of grammar’s rules. The same kind of questions arises in a typology perspec- tive: comparing the grammars extracted from treebanks of different languages is not meaningful per se. On the other hand, typology very often focus only on specific informa- tion such as verb/arguments relation or head/modifiers or- ders. We propose in this article an approach based on a specific representation of the syntactic information, in the perspec- tive of characterizing languages in a typology point of view. This approach mainly relies on a new tool, MarsaGram1, making it possible to automatically infer the typological in- formation, independently of the formalism (constituency or dependency). We present in the following the type informa- 1Available on the Ortolang platform, hdl:11041/ortolang- 000917 tion that can be extracted and illustrate the approach with different treebanks. 2. Methods 2.1. Infering the context-free grammar of …

Content truncated to 500 words