Shallow Discourse Annotation for Chinese TED Talks

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 99–103 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 99 Exploiting Cross-Lingual Hints to Discover Event Pronouns Sharid Lo´aiciga1∗, Christian Hardmeier2 & Asad Sayeed3 1 CoLab Potsdam, Department of Linguistics, University of Potsdam 2 Department of Linguistics and Philology, Uppsala University 3 CLASP, Department of Philosophy, Linguistics and Theory of Science, University of Gothenburg sharid.loaiciga@gmail.com, christian.hardmeier@lingfil.uu.se, asad.sayeed@gu.se Abstract Non-nominal co-reference is much less studied than nominal coreference, partly because of the lack of annotated corpora. We explore the possibility of exploiting parallel multilingual corpora as a means of cheap supervision for the classification of three different readings of the English pronoun it: entity, event or pleonastic, from their translation in several languages. We found that the ‘event’ reading is not very frequent, but can be easily predicted provided that the construction used to translate the it example is a pronoun as well. These cases, nevertheless, are not enough to generalize to other types of non-nominal reference. Keywords: it, reference, Europarl corpus 1. Introduction Nominal coreference has been studied extensively, but work on the automatic recognition of non-nominal anaphora is scarce, as are annotated data sets. Among the challenges of non-nominal anaphora is the difficulty of characterizing the large variance of antecedent types, which often include clauses, sentences, and even paragraphs. Here we focus on the English pronoun it and its capacity to func- tion as anaphor for nominal entity and non-nominal event antecedents, and as a pleonastic token. Examples 1 to 3 below illustrate these different readings using English pas- sages from the Europarl corpus and their French parallel translations. In this paper, we evaluate the potential of multilingual par- allel data as a source of indirect supervision for the anno- tation and classification of different readings of the English pronoun it. We explore the hypothesis that languages have different strategies and preferences to encode referential re- lationships, and that these differences surface as systematic patterns in multilingual parallel data. Therefore, the com- peting readings of the pronoun it correspond to different patterns of translation across languages. We present a method for creating artificially labeled data for the classification of three different readings of it: en- tity, event or pleonastic, from their translation in several languages. We found that the ‘event’ reading is not very frequent, but can be easily predicted provided that the con- struction used to translate the it example is a pronoun as well. These cases, nevertheless, are not enough to general- ize to other types of non-nominal reference. 1. ENTITY READING Madam President, I have been deluged with messages from growers from all over the south-east of England who regard this proposal as near catastrophic. It will result, they tell me, in smaller crops and in higher prices. ∗Work completed while the first author was affiliated with the Department of Philosophy, Linguistics and Theory of Science, University of Gothenburg. Madame la Pr´esidente, j’ai ´et´e assailli de messages de …

Content truncated to 500 words