A Probabilistic Morphological Analyzer for Syriac

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 810–820, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics A Probabilistic Morphological Analyzer for Syriac Peter McClanahan, George Busby, Robbie Haertel, Kristian Heal †, Deryle Lonsdale‡, Kevin Seppi, Eric Ringger Department of Computer Science, ‡Department of Linguistics, †Center for the Preservation of Ancient Religious Texts (CPART) Brigham Young University Provo, Utah 84604 USA http://nlp.cs.byu.edu/ Abstract We define a probabilistic morphological ana- lyzer using a data-driven approach for Syriac in order to facilitate the creation of an annotated corpus. Syriac is an under-resourced Semitic language for which there are no available lan- guage tools such as morphological analyzers. We introduce novel probabilistic models for segmentation, dictionary linkage, and morpho- logical tagging and connect them in a pipeline to create a probabilistic morphological analyzer requiring only labeled data. We explore the per- formance of models with varying amounts of training data and find that with about 34,500 labeled tokens, we can outperform a reason- able baseline trained on over 99,000 tokens and achieve an accuracy of just over 80%. When trained on all available training data, our joint model achieves 86.47% accuracy, a 29.7% re- duction in error rate over the baseline. 1 Introduction Our objective is to facilitate the annotation of a large corpus of classical Syriac (referred to simply as “Syr- iac” throughout the remainder of this work). Syr- iac is an under-resourced Western Semitic language of the Christian Near East and a dialect of Aramaic. It is currently employed almost entirely as a liturgi- cal language but was a true spoken language up un- til the eighth century, during which time many pro- lific authors wrote in Syriac. Even today there are texts still being composed in or translated into Syr- iac. By automatically annotating these texts with lin- guistically useful information, we will facilitate sys- tematic study by scholars of Syriac, the Near East, and Eastern Christianity. Furthermore, languages that are linguistically similar to Syriac (e.g., Arabic and Hebrew) may benefit from the methodology pre- sented here. Our desired annotations include morphological segmentation, links to dictionary entries, and mor- phological attributes. Typically, annotations of this kind are made with the assistance of language tools, such as morphological analyzers, segmenters, or part-of-speech (POS) taggers. Such tools do not exist for Syriac, but some labeled data does exist: Kiraz (1994) compiled an annotated version of the Peshitta New Testament (1920) and a concordance thereof. We aim to replicate this kind of annota- tion on a much larger scale with more modern tools, building up from the labeled New Testament data, our only resource. Motivated by this state of affairs, our learning and annotation framework requires only labeled data. We approach the problem of Syriac morphological annotation by creating five probabilistic sub-models that can be trained in a supervised fashion and com- bined in a joint model of morphological annota- tion. We introduce novel algorithms for segmenta- tion, dictionary linkage, and morphological tagging. We then …

Content truncated to 500 words