Arabic Syntactic Trees: from Constituency to Dependency

Arabic Syntactic Trees: from Constituency to Dependency Zdeneli ‘iabokrtskST and Otakar Smri Center for Computational Linguistics Faculty of Mathematics and Physics Charles University in Prague {zabokrtsky,smrz}@ckl.mff.cuni.cz Abstract This research note reports on the work in progress which regards automatic transformation of phrase-structure syn- tactic trees of Arabic into dependency- -driven analytical ones. Guidelines for these descriptions have been developed at the Linguistic Data Consortium, Uni- versity of Pennsylvania, and at the Fac- ulty of Mathematics and Physics and the Faculty of Arts, Charles University in Prague, respectively. The transformation consists of (i) a re- cursive function translating the topology of a phrase tree into a corresponding de- pendency tree, and (ii) a procedure as- signing analytical functions to the nodes of the dependency tree. Apart from an outline of the annota- tion schemes and a deeper insight into these procedures, model application of the transformation is given herein. 1 Introduction Exploring the relationship between constituency and dependency sentence representations is not a new issue—the first studies go back to the 60's (Gaifman (1965); for more references, see e.g. Schneider (1998)). Still, some theoretical find- ings had not been applicable until the first de- pendency treebanks with well-defined annotation schemes came into existence just in the very last years (Haji6 et al., 2001). The need to convert Arabic treebank data of different descriptions arises from a co-operation between the Linguistic Data Consortium (LDC), University of Pennsylvania, and three concerned institutions of Charles University in Prague, namely the Center for Computational Linguistics, the Institute of Formal and Applied Linguistics, and the Institute of Comparative Linguistics. The two parties intend to share the resources they create. Prior to this exchange, 10,000 words from the LDC Arabic Newswire A Corpus were manually annotated in both syntactic styles as a step to ensure that the annotations are re-usable and their concepts mutually compatible. Here we attempt the constituency–dependency direction of the transfer. 1.1 Phrase-structure trees The input data come from the LDC team (Maamouri et al., 2003). The annotation scheme is based on constituent-syntax bracketing style used at the University of Pennsylvania (Maamouri and Cieri, 2002). The trees include nodes for surface text tokens as well as non-terminal nodes follow- ing from the descriptive grammar. Not only syn- tactic elements, but also several kinds of structural reconstructions (traces) are captured here. 1.2 Analytical trees Under the analytical tree structure we understand a representation of the surface sentence in form of a dependency tree. The node set consists of all the tokens determined after morphological analysis of the text, and the sentence root node. The descrip- tion recovers the relation between a governor and a node dependent on it. The nature of the govern- ment is expressed by the analytical functions of the nodes being linked 183 0 VERB ya+kun (it) was *TRACE NP-SBJ 0 PREP min from 0 NEG_PART -lam not NP-P D 0 NOUN muwAjah+ap confrontation NP CONJ ma- and 0 CONJ wa- and 0 CONJ wa- and 0 PERIOD 0� 0 PRON VERB …

Content truncated to 500 words