Annotating the meaning of discourse connectives by looking at their translation: The translation-spotting technique

Abstract

The various meanings of discourse connectives like while and however are difficult to identify and annotate, even for trained human annotators. This problem is all the more important that connectives are salient textual markers of cohesion and need to be correctly interpreted for many NLP applications. In this paper, we suggest an alternative route to reach a reliable annotation of connectives, by making use of the information provided by their translation in large parallel corpora. This method thus replaces the difficult explicit reasoning involved in traditional sense annotation by an empirical clustering of the senses emerging from the translations. We argue that this method has the advantage of providing more reliable reference data than traditional sense annotation. In addition, its simplicity allows for the rapid constitution of large annotated datasets.

Disambiguating Discourse Connectives for Statistical Machine Translation

“…Therefore, we have annotated from the beginning of Europarl about 2,200 instances of seven highly ambiguous English discourse connectives: although, however, meanwhile, since, (even) though, while, and yet [14]. They were annotated using translation spotting [15], i.e. indicating the French translation, and then clustering and mapping them to a set of seven sense labels: CONTRAST, CONCESSION, TEMPORAL, CAUSAL, AD-VERB, TEMPORAL/CONTRAST, and TEMPORAL/CAUSAL.…”

Causality and Subjectivity in Spanish Connectives: Exploring the Use of Automatic Subjectivity Analyses in Various Text Types

“…The proliferation of studies focusing on the diffi culties related to annotation and segmentation procedures (Bayerl & Paul, 2011;Cartoni et al, 2013;Hoek et al, 2017;, among others) is not a coincidence. In fact, diff erent theoretical approaches to the analysis of coherence relations such as the "Penn Discourse Treebank" (PDTB) (Prasad et al, 2008), the "Rhetorical Structure Theory Treebank" (Carlson et al, 2003) and "Segmented Discourse Representation Theory" (Asher & Lascarides, 2003) have provided extensive resources that explain the diff erent proposed sets of coherence relations and describe the processes of annotation and segmentation in detail (see the manuals developed by the PDTB Research Group [2007] and Carlson & Marcu [2001]).…”

Towards a discourse relation-aware approach for Chinese-English machine translation

“…Recent works focus on the translation of ambiguous DCs, such as 'since' in the temporal sense vs. 'since' in the reason sense. This is achieved by annotating the DCs in the training data by 'translation spotting', which is to manually align the DCs of the source text to their translation in the target text, either occurring as DCs or other expressions (Meyer et al, 2011;Meyer and Polakova, 2013;Cartoni et al, 2013). Experiments of these works have been conducted in English-to-French, Czech and German translation and only explicit DCs were considered.…”

“…Therefore, the PDTB convention is adopted for the annotation of connectives on both sides of the parallel corpus. Instead of sense annotation, the DCs are aligned in similar manner as the 'translation spotting' approach (Meyer et al, 2011;Cartoni et al, 2013). In other words, the 'senses' are disambiguated by the translation of the DCs.…”