Variation in Coreference Strategies across Genres and Production Media

Proceedings of the 28th International Conference on Computational Linguistics, pages 5774–5785 Barcelona, Spain (Online), December 8-13, 2020 5774 Variation in Coreference Strategies across Genres and Production Media Berfin Aktas¸ SFB1287 Research Focus Cognitive Sciences University of Potsdam, Germany berfinaktas@uni-potsdam.de Manfred Stede SFB1287 Research Focus Cognitive Sciences University of Potsdam, Germany stede@uni-potsdam.de Abstract In response to (i) inconclusive results in the literature as to the properties of coreference chains in written versus spoken language, and (ii) a general lack of work on automatic coreference resolution on both spoken language and social media, we undertake a corpus study involving the various genre sections of Ontonotes, the Switchboard corpus, and a corpus of Twitter conversations. Using a set of measures that previously have been applied individually to different data sets, we find fairly clear patterns of ”behavior” for the different genres/media. Besides their role for psycholinguistic investigation (why do we employ different coreference strategies when we write or speak) and for the placement of Twitter in the spoken–written continuum, we see our results as a contribution to approaching genre-/media-specific coreference resolution. 1 Introduction Research on strategies for producing referring expressions has often investigated the differences (if any) between spoken and written language, but as we will show in Section 2, results have been inconclusive. Sometimes, claims are simply contradictory, but the more important problem is that usually, the exact ways of measuring the properties of coreference chains are not being made transparent. In addition, the data that has been used can vary considerably, and it is not always clear how studies can be compared. Our primary goal here is to shed light on coreference with respect to the spoken/written distinction, by undertaking a careful comparative corpus analysis and explicitly stating our methods of measurement. In an earlier study, we presented a quantitative study on different genre sections of the OntoNotes corpus (Aktas¸ et al., 2019). We extend that study in two directions: We augment the rather small proportion of spoken data in OntoNotes with the Switchboard corpus (Godfrey et al., 1992), and we introduce new quantitative features for the comparative analysis. The secondary goal is to explore how the medium microblog, specifically Twitter, relates to the spoken- written spectrum for coreference strategies. In a recent study, we investigated this research question empirically through computational experiments on a state-of-the-art ”standard” coreference resolution system (Aktas¸ et al., 2020). We showed that the choice of genre and the medium (spoken vs. written) in training data can make a bigger difference than the bare amount of data. In the current paper, we apply corpus-based empirical methods to situate the coreference strategies found in Twitter conversations in the medium/genre spectrum. For this purpose, we finally extend the (Aktas¸ et al., 2019) study by adding the comparison with Twitter texts, thus achieving a wider range of production media. Although Switchboard and OntoNotes have previously been used for investigating coreference, to our knowledge they have not yet been systematically compared. Accordingly, an important part of our work is in harmonizing …

Content truncated to 500 words