Labeled Morphological Segmentation with Semi-Markov Models

Proceedings of the 19th Conference on Computational Language Learning, pages 164–174, Beijing, China, July 30-31, 2015. c⃝2015 Association for Computational Linguistics Labeled Morphological Segmentation with Semi-Markov Models Ryan Cotterell1,2 Department of Computer Science1 Johns Hopkins University, USA ryan.cotterell@jhu.edu Thomas Müller2 Alexander Fraser2 Center for Information and Language Processing2 University of Munich, Germany muellets@cis.lmu.de Hinrich Schütze2 Abstract We present labeled morphological segmentation—an alternative view of morphological processing that unifies sev- eral tasks. We introduce a new hierarchy of morphotactic tagsets and CHIPMUNK, a discriminative morphological segmen- tation system that, contrary to previous work, explicitly models morphotactics. We show improved performance on three tasks for all six languages: (i) morpho- logical segmentation, (ii) stemming and (iii) morphological tag classification. For morphological segmentation our method shows absolute improvements of 2-6 points F1 over a strong baseline. 1 Introduction Morphological processing is often an overlooked problem since many well-studied languages (e.g., Chinese and English) are morphologically impov- erished. But for languages with complex mor- phology (e.g., Finnish and Turkish) morphological processing is essential. A specific form of mor- phological processing, morphological segmenta- tion, has shown its utility for machine translation (Dyer et al., 2008), sentiment analysis (Abdul- Mageed et al., 2014), bilingual word alignment (Eyigöz et al., 2013), speech processing (Creutz et al., 2007b) and keyword spotting (Narasimhan et al., 2014), inter alia. We advance the state-of-the- art in supervised morphological segmentation by describing a high-performance, data-driven tool for handling complex morphology, even in low- resource settings. In this work, we make the distinction between unlabeled morphological segmentation (UMS ) (often just called “morphological segmentation”) and labeled morphological segmentation (LMS). The labels in our supervised discriminative model for LMS capture the distinctions between different types of morphemes and directly model the mor- photactics. We further create a hierarchical uni- versal tagset for labeling morphemes, with differ- ent levels appropriate for different tasks. Our hi- erarchical tagset was designed by creating a stan- dard representation from heterogeneous resources for six languages. This allows us to use a single unified framework to obtain strong performance on three common morphological tasks that have typically been viewed as separate problems and addressed using different methods. We give an overview of the tasks addressed in this paper in Figure 1. The figure shows the expected output for the Turkish word gençle¸smelerin ‘of rejuvenat- ings’. In particular, it shows the full labeled mor- phological segmentation, from which three repre- sentations can be directly derived: the unlabeled morphological segmentation, the stem/root 1 and the structured morphological tag containing POS and inflectional features. We model these tasks with CHIPMUNK, a semi- Markov conditional random field (semi-CRF) (Sarawagi and Cohen, 2004), a model that is well- suited for morphology. We provide a robust eval- uation and analysis on six languages and CHIP- MUNK yields strong results on all three tasks, in- cluding state-of-the-art accuracy on morphologi- cal segmentation. Section 2 presents our LMS framework and the morphotactic tagsets we use, i.e., the labels of the sequence prediction task CHIPMUNK solves. Sec- tion 3 …

Content truncated to 500 words