A Joint Model of Orthography and Morphological Segmentation

Proceedings of NAACL-HLT 2016, pages 664–669, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics A Joint Model of Orthography and Morphological Segmentation Ryan Cotterell Tim Vieira Department of Computer Science Johns Hopkins University, USA {ryan.cotterell,tim.f.vieira}@gmail.com Hinrich Sch¨utze Center for Information and Language Processing LMU Munich, Germany inquiries@cis.lmu.org Abstract We present a model of morphological seg- mentation that jointly learns to segment and restore orthographic changes, e.g., funniest �→ fun-y-est. We term this form of analysis canon- ical segmentation and contrast it with the tra- ditional surface segmentation, which segments a surface form into a sequence of substrings, e.g., funniest �→ funn-i-est. We derive an im- portance sampling algorithm for approximate inference in the model and report experimental results on English, German and Indonesian. 1 Introduction Morphological segmentation is useful for NLP appli- cations, such as, automatic speech recognition (Afify et al., 2006), keyword spotting (Narasimhan et al., 2014), machine translation (Clifton and Sarkar, 2011) and parsing (Seeker and C¸ etino˘glu, 2015). Prior work cast the problem as surface segmentation: a word form w is segmented into a sequence of substrings whose concatenation is w. In this paper, we introduce the problem of canonical segmentation: w is ana- lyzed as a sequence of canonical morphemes, based on a set of word forms that have been “canonically” annotated for supervised learning. Each canonical morpheme c corresponds to a surface morph s, de- fined as its orthographic manifestation, i.e., as the substring of w that is generated by applying editing operations like insertion and deletion. Consider the following example: funniest has a canonical segmen- tation fun-y-est with three morphs funn-i-est. Arriv- ing at the canonical analysis requires two edit op- erations: delete n in funn and replace i with y in funniest funyest fun y est memikulmu menpikulmu men pikul mu Zulassung zulassenung zu lassen ung Orthography Underlying Form Segmentation Figure 1: Examples of canonical segmentation for English (top), Indonesian (middle) and German (bottom). i. Figure 1 gives examples of orthography (i.e., the concatentation of surface morphs), underlying form (i.e., the concatentation of canonical morphemes) and canonical segmentation in three languages. Canonical segmentation is motivated in the fol- lowing three ways: (i) Computational morphology is the study of how words and their meanings are composed from smaller units. This goal is better supported by canonical morphemes than by surface morphemes because the smaller units are more ac- curately modeled. For funniest, composition can reason with canonical morphemes fun and y, whereas surface segmentation must work with funn and i. (ii) Morphological analysis is typically done with attribute-value pairs (AVP), e.g., [lemma=FUNNY, degree=SUPER]. While AVP is a good represen- tation for inflectional morphology, it is not pow- erful enough for derivational morphology. If we represent the derivation of funnier as [lemma=FUN, deriv-suffix=-Y, degree=SUPER], then it is no longer clear in this fixed representation whether 664 degree = SUPER applies to fun or fun+y.1 Canon- ical segmentation is more flexible—allowing us to express derivational relations without committing to a fixed attribute-value structure, …

Content truncated to 500 words