Distinguishing Literal and Non-Literal Usage of German Particle Verbs

Proceedings of NAACL-HLT 2016, pages 353–362, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics Distinguishing Literal and Non-Literal Usage of German Particle Verbs Maximilian Köper Sabine Schulte im Walde Institut für Maschinelle Sprachverarbeitung Universität Stuttgart, Germany {maximilian.koeper,schulte}@ims.uni-stuttgart.de Abstract This paper provides a binary, token-based classification of German particle verbs (PVs) into literal vs. non-literal usage. A random forest improving standard features (e.g., bag- of-words; affective ratings) with PV-specific information and abstraction over common nouns significantly outperforms the major- ity baseline. In addition, PV-specific classifi- cation experiments demonstrate the role of shared particle semantics and semantically related base verbs in PV meaning shifts. 1 Introduction Automatic detection of non-literal expressions (in- cluding metaphors and idioms) is critical for many natural language processing (NLP) tasks such as information extraction, machine translation, and sentiment analysis. For this reason, the last decade has seen an increase in research on identifying literal vs. non-literal meaning (Birke and Sarkar, 2006; Birke and Sarkar, 2007; Li and Sporleder, 2009; Sporleder and Li, 2009; Turney et al., 2011; Shutova et al., 2013; Tsvetkov et al., 2014), as well as the establishment of workshops on metaphori- cal language in NLP.1 In this paper, we explore the prediction of lit- eral vs. non-literal language usage of a compu- tationally challenging class of multiword expres- sions: German particle verbs (PVs) such as an- lachen (laugh at) are compositions of a base verb 1sites.google.com/site/metaphorinnlp2016/ (BV) such as lachen (smile/laugh) and a verb par- ticle such as an. German PVs are highly produc- tive (Springorum et al., 2013b; Springorum et al., 2013a), and the particles are notoriously ambigu- ous (Lechler and Roßdeutscher, 2009; Haselbach, 2011; Springorum, 2011). Furthermore, the par- ticles often trigger (regular) meaning shifts when they combine with base verbs (Springorum et al., 2013b), so the resulting PVs represent frequent cases of non-literal meaning. The contributions of this paper are as follows: 1. We present a random forest classifier that cor- rectly identifies 86.8% of literal vs. non-literal language usage within a novel dataset of 6436 annotated sentences, in comparison to a ma- jority baseline of 64.9%. 2. We successfully incorporate salient PV- specific features and noun clusters in addition to standard bag-of-words features and affective ratings. 3. We demonstrate that PVs with semantically similar particles and semantically similar base verbs can predict each others’ literal vs. non-literal language usage. 4. We illustrate the potential and the limits of the most salient classification features in pre- dicting PV non-literal language usage. In the remainder of this paper we describe pre- vious work on non-literal language identification and computational models of German particle verbs (Section 2), before we introduce our dataset 353 on German particle verbs (Section 3), the particle verb features (Section 4), and the experiments, re- sults and analyses (Section 5). 2 Related Work Previous work relevant to this paper includes re- search on identifying non-literal language usage, and computational work on (German) particle verb meaning. Identification of non-literal language usage: Birke and Sarkar (2006), Birke and …

Content truncated to 500 words