Determining Case in Arabic: Learning Complex Linguistic Behavior Requires Complex Linguistic Features

Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pp. 1084–1092, Prague, June 2007. c⃝2007 Association for Computational Linguistics Determining Case in Arabic: Learning Complex Linguistic Behavior Requires Complex Linguistic Features Nizar Habash†, Ryan Gabbard‡, Owen Rambow†, Seth Kulick‡ and Mitch Marcus‡ †Center for Computational Learning Systems, Columbia University New York, NY, USA {habash,rambow}@cs.columbia.edu ‡Department of Computer and Information Science,University of Pennsylvania Philadelphia, PA, USA {gabbard,skulick,mitch}@cis.upenn.edu Abstract This paper discusses automatic determina- tion of case in Arabic. This task is a ma- jor source of errors in full diacritization of Arabic. We use a gold-standard syntac- tic tree, and obtain an error rate of about 4.2%, with a machine learning based system outperforming a system using hand-written rules. A careful error analysis suggests that when we account for annotation errors in the gold standard, the error rate drops to 0.8%, with the hand-written rules outperforming the machine learning-based system. 1 Introduction In Modern Standard Arabic (MSA), all nouns and adjectives have one of three cases: nominative (NOM), accusative (ACC), or genitive (GEN). What sets case in MSA apart from case in other languages is most saliently the fact that it is usually not marked in the orthography, as it is written using diacrit- ics which are normally omitted. In fact, in a re- cent paper on diacritization, Habash and Rambow (2007) report that word error rate drops 9.4% ab- solute (to 5.5%) if the word-final diacritics (which include case) need not be predicted. Similar drops have been observed by other researchers (Nelken and Shieber, 2005; Zitouni et al., 2006). Thus, we can deduce that tagging-based approaches to case identification are limited in their usefulness, and if we need full diacritization for subsequent process- ing in a natural language processing (NLP) applica- tion (say, language modeling for automatic speech recognition (Vergyri and Kirchhoff, 2004)), we need to perform more complex syntactic processing to re- store case diacritics. Options include using the out- put of a parser in determining case. An additional motivation for investigating case in Arabic comes from treebanking. Native speakers of Arabic in fact are native speakers of one of the Arabic dialects, all of which have lost case (Holes, 2004). They learn MSA in school, and have no native-speaker intuition about case. Thus, determin- ing case in MSA is a hard problem for everyone, including treebank annotators. A tool to catch case- related errors in treebanking would be useful. In this paper, we investigate the problem of de- termining case of nouns and adjectives in syntactic trees. We use gold standard trees from the Arabic Treebank (ATB). We see our work using gold stan- dard trees as a first step towards developing a sys- tem for restoring case to the output of a parser. The complexity of the task justifies an initial investiga- tion based on gold standard trees. And of course, the use of gold standard trees is justified for our other objective, helping quality control for treebanking. …

Content truncated to 500 words