Morphological Inflection Generation Using Character Sequence to Sequence Learning

Proceedings of NAACL-HLT 2016, pages 634–643, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics Morphological Inflection Generation Using Character Sequence to Sequence Learning Manaal Faruqui1 Yulia Tsvetkov1 Graham Neubig2 Chris Dyer1 1Language Technologies Institute, Carnegie Mellon University, USA 2Graduate School of Information Science, Nara Institute of Science and Technology, Japan {mfaruqui,ytsvetko,cdyer}@cs.cmu.edu neubig@is.naist.jp Abstract Morphological inflection generation is the task of generating the inflected form of a given lemma corresponding to a particular linguis- tic transformation. We model the problem of inflection generation as a character sequence to sequence learning problem and present a variant of the neural encoder-decoder model for solving it. Our model is language inde- pendent and can be trained in both supervised and semi-supervised settings. We evaluate our system on seven datasets of morphologically rich languages and achieve either better or comparable results to existing state-of-the-art models of inflection generation. 1 Introduction Inflection is the word-formation mechanism to ex- press different grammatical categories such as tense, mood, voice, aspect, person, gender, number and case. Inflectional morphology is often realized by the concatenation of bound morphemes (prefixes and suffixes) to a root form or stem, but noncon- catenative processes such as ablaut and infixation are found in many languages as well. Table 1 shows the possible inflected forms of the German stem Kalb (calf) when it is used in different cases and numbers. The inflected forms are the result of both ablaut (e.g., a→¨a) and suffixation (e.g., +ern). Inflection generation is useful for reducing data sparsity in morphologically complex languages. For example, statistical machine translation suffers from data sparsity when translating morphologically-rich languages, since every surface form is considered an singular plural nominative Kalb K¨alber accusative Kalb K¨alber dative Kalb K¨albern genitive Kalbes K¨alber Table 1: An example of an inflection table from the German noun dataset for the word Kalb (calf). independent entity. Translating into lemmas in the target language, and then applying inflection gener- ation as a post-processing step, has been shown to alleviate the sparsity problem (Minkov et al., 2007; Toutanova et al., 2008; Clifton and Sarkar, 2011; Fraser et al., 2012; Chahuneau et al., 2013a). Mod- eling inflection generation has also been used to im- prove language modeling (Chahuneau et al., 2013b), identification of multi-word expressions (Oflazer et al., 2004), among other applications. The traditional approach to modeling inflec- tion relies on hand-crafted finite state transducers and lexicography, e.g., using two-level morphology (Koskenniemi, 1983; Kaplan and Kay, 1994). Such systems are appealing since they correspond to lin- guistic theories, but they are expensive to create, they can be fragile (Oflazer, 1996), and the com- posed transducers can be impractically large. As an alternative, machine learning models have been proposed to generate inflections from root forms as string transduction (Yarowsky and Wicentowski, 2000; Wicentowski, 2004; Dreyer and Eisner, 2011; Durrett and DeNero, 2013; Ahlberg et al., 2014; Hulden, 2014; Ahlberg et al., 2015; Nicolai et al., 2015). However, these impose either assumptions about the set of possible morphological processes 634 inflection generation …

Content truncated to 500 words