Coordination Structures in Dependency Treebanks

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 517–527, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Coordination Structures in Dependency Treebanks Martin Popel, David Mareˇcek, Jan ˇStˇep´anek, Daniel Zeman, Zdenˇek ˇZabokrtsk´y Charles University in Prague, Faculty of Mathematics and Physics Institute of Formal and Applied Linguistics ( ´UFAL) Malostransk´e n´amˇest´ı 25, CZ-11800 Praha, Czechia {popel|marecek|stepanek|zeman|zabokrtsky}@ufal.mff.cuni.cz Abstract Paratactic syntactic structures are noto- riously difficult to represent in depen- dency formalisms. This has painful con- sequences such as high frequency of pars- ing errors related to coordination. In other words, coordination is a pending prob- lem in dependency analysis of natural lan- guages. This paper tries to shed some light on this area by bringing a system- atizing view of various formal means de- veloped for encoding coordination struc- tures. We introduce a novel taxonomy of such approaches and apply it to treebanks across a typologically diverse range of 26 languages. In addition, empirical obser- vations on convertibility between selected styles of representations are shown too. 1 Introduction In the last decade, dependency parsing has grad- ually been receiving visible attention. One of the reasons is the increased availability of depen- dency treebanks, be they results of genuine depen- dency annotation projects or converted automat- ically from previously existing phrase-structure treebanks. In both cases, a number of decisions have to be made during the construction or conversion of a dependency treebank. The traditional notion of dependency does not always provide unambiguous solutions, e.g. when it comes to attaching func- tional words. Worse, dependency representation is at a loss when it comes to representing paratactic linguistic phenomena such as coordination, whose nature is symmetric (two or more conjuncts play the same role), as opposed to the head-modifier asymmetry of dependencies.1 1We use the term modifier (or child) for all types of de- pendent nodes including arguments. The dominating solution in treebank design is to introduce artificial rules for the encoding of coor- dination structures within dependency trees using the same means that express dependencies, i.e., by using edges and by labeling of nodes or edges. Ob- viously, any tree-shaped representation of a coor- dination structure (CS) must be perceived only as a “shortcut” since relations present in coordination structures form an undirected cycle, as illustrated already by Tesni`ere (1959). For example, if a noun is modified by two coordinated adjectives, there is a (symmetric) coordination relation between the two conjuncts and two (asymmetric) dependency relations between the conjuncts and the noun. However, as there is no obvious linguistic in- tuition telling us which tree-shaped CS encoding is better and since the degree of freedom has sev- eral dimensions, one can find a number of distinct conventions introduced in particular dependency treebanks. Variations exist both in topology (tree shape) and labeling. The main goal of this pa- per is to give a systematic survey of the solutions adopted in these treebanks. Naturally, the interplay of dependency and co- ordination links in a single tree leads to …

Content truncated to 500 words