A Nearest-Neighbor Approach to the Automatic Analysis of Ancient Greek Morphology

CoNLL 2008: Proceedings of the 12th Conference on Computational Natural Language Learning, pages 127–134 Manchester, August 2008 A Nearest-Neighbor Approach to the Automatic Analysis of Ancient Greek Morphology John Lee Spoken Language Systems MIT Computer Science and Artificial Intelligence Laboratory Cambridge, MA 02139, USA jsylee@csail.mit.edu Abstract We propose a data-driven method for au- tomatically analyzing the morphology of ancient Greek. This method improves on existing ancient Greek analyzers in two ways. First, through the use of a nearest- neighbor machine learning framework, the analyzer requires no hand-crafted rules. Second, it is able to predict novel roots, and to rerank its predictions by exploiting a large, unlabelled corpus of ancient Greek. 1 Introduction The civilization of ancient Greece, from which the Western world has received much of its heritage, has justly received a significant amount of schol- arly attention. To gain a deeper understanding of the civilization, access to the essays, poems, and other Greek documents in the original language is indispensable. Ancient Greek is a highly inflected Indo- European language1. A verb, for example, is in- flected according to its person, number, voice, tense/aspect and mood. According to (Crane, 1991), “a single verb could have roughly 1,000 forms, and, if we consider that any verb may be preceded by up to three distinct prefixes, the num- ber of forms explodes to roughly 5,000,000.” The inflections are realized by prefixes and suffixes to c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. 1All Greek words are transcribed into the Roman alpha- bet in this paper. The acute, grave and circumflex accents are represented by diacritics, as in ´o, `o and ˜o, respectively. Smooth breathing marks are omitted; rough breathing marks are signalled by h. Underbars used in e and o represent eta and omega. the stem, and sometimes spelling changes within the stem. These numerous forms can be further complicated by accents, and by additional spelling changes at morpheme boundaries for phonological reasons. The overall effect can yield an inflected form in which the root2 is barely recognizable. Indeed, a staple exercise for students of ancient Greek is to identify the root form of an inflected verb. This skill is essential; without knowing the root form, one cannot understand the meaning of the word, or even look it up in a dictionary. For Classics scholars, these myriad forms also pose formidable challenges. In order to search for occurrences of a word in a corpus, all of its forms must be enumerated, since words do not frequently appear in their root forms. This procedure be- comes extremely labor-intensive for small words that overlap with other common words (Crane, 1991). Automatic morphological analysis of ancient Greek would be useful for both educational and research purposes. In fact, one of the first analyz- ers was developed as a pedagogical tool (Packard, 1973). Today, a widely used analyzer is embed- ded in the Perseus Digital Library (Crane, 1996), an internet resource utilized by both students and researchers. …

Content truncated to 500 words