Part-of-Speech Tagging for Middle English through Alignment and Projection of Parallel Diachronic Texts

Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pp. 390–399, Prague, June 2007. c⃝2007 Association for Computational Linguistics Part-of-speech Tagging for Middle English through Alignment and Projection of Parallel Diachronic Texts Taesun Moon and Jason Baldridge Department of Linguistics University of Texas at Austin 1 University Station B5100 Austin, TX 78712-0198 USA tsmoon, jbaldrid@mail.utexas.edu Abstract We demonstrate an approach for inducing a tagger for historical languages based on ex- isting resources for their modern varieties. Tags from Present Day English source text are projected to Middle English text using alignments on parallel Biblical text. We explore the use of multiple alignment ap- proaches and a bigram tagger to reduce the noise in the projected tags. Finally, we train a maximum entropy tagger on the output of the bigram tagger on the target Biblical text and test it on tagged Middle English text. This leads to tagging accuracy in the low 80’s on Biblical test material and in the 60’s on other Middle English material. Our re- sults suggest that our bootstrapping meth- ods have considerable potential, and could be used to semi-automate an approach based on incremental manual annotation. 1 Introduction Annotated corpora of historical texts provide an im- portant resource for studies of syntactic variation and change in diachronic linguistics. For example, the Penn-Helsinki Parsed Corpus of Middle English (PPCME) (Kroch and Taylor, 2000) has been used to show the existence of syntactic dialectal differ- ences between northern and southern Middle En- glish (Kroch et al., 2000) and to examine the syn- tactic evolution of the English imperative construc- tion (Han, 2000). However, their utility rests on their having coverage of a significant amount of annotated material from which to draw patterns for such stud- ies, and creating resources such as the PPCME re- quire significant time and cost to produce. Corpus linguists interested in diachronic language studies thus need efficient ways to produce such resources. One approach to get around the annotation bottle- neck is to use semi-automation. For example, when producing part-of-speech tags for the Tycho Brahe corpus of Historical Portuguese (Britto et al., 2002), a set of seed sentences was manually tagged, and the Brill tagger (Brill, 1995) was then trained on those and consequently used to tag other sentences. The output was inspected for errors, the tagger was re- trained and used again to tag new sentences, for sev- eral iterations. We also seek to reduce the human effort involved in producing part-of-speech tags for historical cor- pora. However, our approach does so by leveraging existing resources for a language’s modern varieties along with parallel diachronic texts to produce accu- rate taggers. This general technique has worked well for bilingual bootstrapping of language processing resources for one language based on already avail- able resources from the other. The first to explore the idea were Yarowsky and Ngai (2001), who in- duced a part-of-speech tagger for French and base noun phrase detectors for French and Chinese …

Content truncated to 500 words