A Single Generative Model for Joint Morphological Segmentation and Syntactic Parsing

Proceedings of ACL-08: HLT, pages 371–379, Columbus, Ohio, USA, June 2008. c⃝2008 Association for Computational Linguistics A Single Generative Model for Joint Morphological Segmentation and Syntactic Parsing Yoav Goldberg Ben Gurion University of the Negev Department of Computer Science POB 653 Be’er Sheva, 84105, Israel yoavg@cs.bgu.ac.il Reut Tsarfaty Institute for Logic Language and Computation University of Amsterdam Plantage Muidergracht 24, Amsterdam, NL rtsarfat@science.uva.nl Abstract Morphologicalprocesses in Semitic languages deliver space-delimited words which intro- duce multiple, distinct, syntactic units into the structure of the input sentence. These words are in turn highly ambiguous, breaking the assumption underlying most parsers that the yield of a tree for a given sentence is known in advance. Here we propose a single joint model for performing both morphological segmenta- tion and syntactic disambiguation which by- passes the associated circularity. Using a tree- bank grammar, a data-driven lexicon, and a linguistically motivated unknown-tokens han- dling technique our model outperforms previ- ous pipelined, integrated or factorized systems for Hebrew morphological and syntactic pro- cessing, yielding an error reduction of 12% over the best published results so far. 1 Introduction Current state-of-the-art broad-coverage parsers as- sume a direct correspondence between the lexical items ingrained in the proposed syntactic analyses (the yields of syntactic parse-trees) and the space- delimited tokens (henceforth, ‘tokens’) that consti- tute the unanalyzed surface forms (utterances). In Semitic languages the situation is very different. In Modern Hebrew (Hebrew), a Semitic language with very rich morphology, particles marking con- junctions, prepositions, complementizers and rela- tivizers are bound elements prefixed to the word (Glinert, 1989). The Hebrew token ‘bcl’1, for ex- ample, stands for the complete prepositional phrase 1We adopt here the transliteration of (Sima’an et al., 2001). “in the shadow”. This token may further embed into a larger utterance, e.g., ‘bcl hneim’ (literally “in-the-shadow the-pleasant”, meaning roughly “in the pleasant shadow”) in which the dominated Noun is modified by a proceeding space-delimited adjec- tive. It should be clear from the onset that the parti- cle b (“in”) in ‘bcl’ may then attach higher than the bare noun cl (“shadow”). This leads to word- and constituent-boundaries discrepancy, which breaks the assumptions underlying current state-of-the-art statistical parsers. One way to approach this discrepancy is to as- sume a preceding phase of morphological segmen- tation for extracting the different lexical items that exist at the token level (as is done, to the best of our knowledge, in all parsing related work on Arabic and its dialects (Chiang et al., 2006)). The input for the segmentation task is however highly ambiguous for Semitic languages, and surface forms (tokens) may admit multiple possible analyses as in (Bar- Haim et al., 2007; Adler and Elhadad, 2006). The aforementioned surface form bcl, for example, may also stand for the lexical item “onion”, a Noun. The implication of this ambiguity for a parser is that the yield of syntactic trees no longer consists of space- delimited tokens, and the expected number of leaves in the syntactic analysis in not known in advance. Tsarfaty (2006) …

Content truncated to 500 words