Identifying Negation in the DGS Corpus

Zusammenfassung

Negation is a semantic phenomenon that indicates that a presupposed fact or event does not hold (Polanyi and Zaenen, 2006). In spoken languages, this can be expressed for example through negation words (no, not, without), content words (abandon, alleviated, destruction), connectives (however, but) and modal operators (if, would).

Sign languages have an equally rich set of devices to express negation (Quer, 2012). These include negation particles, manual negation morphemes as well as headshake and facial expression.

We seek to provide corpus evidence of negation devices in German sign language (DGS) by analysing the DGS Corpus. In order to provide such evidence it is a necessary first step to identify all units of signing in the DGS Corpus that elicit negation in a phrasal expression. As negation can be caused by a variety of devices, manual annotation of all occurrences is time consuming. To reduce the annotation workload, we introduce automatic measures to identify negation in the corpus.

Only a select few parts of the DGS Corpus have undergone detailed analysis regarding negation. For the majority of the corpus, only basic levels of annotation are available. This basic annotation includes lemmatisation (type-token matching), mouthing, and translation into German. Annotators were encouraged to go beyond this level where possible by providing comments on significant nonmanual behaviour. For example, in over five thousand instances the comment “Kopfschütteln” (headshake) was provided. This information further qualifies the annotated token, as do qualifiers (Konrad et al., 2012) indicating that alpha negation was applied to its type. Such interactions between modalities can be identified using the DGS Lexical Database (Langer et al., 2016), which lists types that can take single-token headshake as well as types being blends from negation particles.

In order to identify negation, we combine information gained from the annotation and the German translation with an automatic analysis of body movements. The whole corpus has undergone pose estimation by using OpenPose (Cao et al., 2018), resulting in joint position timeseries in 2D coordinates. We use the pose information to identify nonmanual behaviour, such as headshakes, which is then compared to and combined with the available annotations. The resulting instances of potential negation in DGS are then contrasted to occurrences of negation in the German translation. For this we use lexical resources such as the lexicon of negation words by Wilson et al. (2005) and the list of negating content words by Schulder et al. (2018). These resources may also be used in conjunction with meanings attributed to DGS lexical entities to identify additional instances of negation.

To evaluate our approach, we apply it to those parts of the corpus that have undergone detailed analysis regarding the occurrence of negation. The resulting automatic negation detection system can be used for automatic classification, as assistance feature for human annotation and to detect annotation mistakes.

Referenzen

Zhe Cao, Gines Hidalgo, Tomas Simon, Shih-En Wei, and Yaser Sheikh. Open-Pose: Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields. arXiv:1812.08008 [cs], 2018.

Reiner Konrad, Thomas Hanke, Susanne König, Gabriele Langer, Silke Matthes, Rie Nishio, and Anja Regen. From form to function. A database approach to handle lexicon building and spotting token forms in sign languages. In Proceedings of the Workshop on the Representation and Processing of Sign Languages: Interactions between Corpus and Lexicon. Language Resources and Evaluation Conference (LREC), pages 87–94, Istanbul, Turkey, 2012.

Gabriele Langer, Thomas Troelsgård, Jette Kristoffersen, Reiner Konrad, Thomas Hanke, and Susanne König. Designing a Lexical Database for a Combined Use of Corpus Annotation and Dictionary Editing. In Workshop on the Representation and Processing of Sign Languages: Corpus Mining. Language Resources and Evaluation Conference (LREC), pages 143–152, 2016.

Livia Polanyi and Annie Zaenen. Contextual Valence Shifters. In James G. Shanahan, Yan Qu, and Janyce Wiebe, editors, Computing Attitude and Affect in Text: Theory and Applications, volume 20 of The Information Retrieval Series, pages 1–10. Springer Netherlands, Dordrecht, Netherlands, 2006. ISBN 978-1-4020-4102-0.

Josep Quer. Negation. In Roland Pfau, Markus Steinbach, and Bencie Woll, editors, Sign Language: An International Handbook, pages 316–339. De Gruyter Mouton, 2012. ISBN 978-3-11-026132-5.

Marc Schulder, Michael Wiegand, and Josef Ruppenhofer. Automatically Creating a Lexicon of Verbal Polarity Shifters: Mono- and Cross-lingual Methods for German. In Proceedings of the International Conference on Computational Linguistics (COLING), pages 2516–2528, Santa Fe, New Mexico, USA, August 2018. International Committee on Computational Linguistics.

Theresa Wilson, Janyce Wiebe, and Paul Hoffmann. Recognizing Contextual Polarity in Phrase-level Sentiment Analysis. In Proceedings of the Joint Conferences on Human Language Technology and on Empirical Methods in Natural Language Processing (HLT/EMNLP), pages 347–354, Vancouver, British Columbia, Canada, 2005. Association for Computational Linguistics.