TED-Q: TED Talks and the Questions they Evoke

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1118–1127 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1118 TED-Q: TED Talks and the Questions they Evoke Matthijs Westera, Laia Mayol, Hannah Rohde Universitat Pompeu Fabra (×2), University of Edinburgh Barcelona (Spain), Edinburgh (Scotland) {matthijs.westera, laia.mayol}@upf.edu, hannah.rohde@ed.ac.uk Abstract We present a new dataset of TED-talks annotated with the questions they evoke and, where available, the answers to these questions. Evoked questions represent a hitherto mostly unexplored type of linguistic data, which promises to open up important new lines of research, especially related to the Question Under Discussion (QUD)-based approach to discourse structure. In this paper we introduce the method and open the first installment of our data to the public. We summarize and explore the current dataset, illustrate its potential by providing new evidence for the relation between predictability and implicitness – capitalizing on the already existing PDTB-style annotations for the texts we use – and outline its potential for future research. The dataset should be of interest, at its current scale, to researchers on formal and experimental pragmatics, discourse coherence, information structure, discourse expectations and processing. Our data-gathering procedure is designed to scale up, relying on crowdsourcing by non-expert annotators, with its utility for Natural Language Processing in mind (e.g., dialogue systems, conversational question answering). Keywords: Discourse structure, discourse relation, evoked question, question under discussion, TED-talks, crowdsourcing, im- plicit connective 1. Introduction Discourse structure is relevant for a variety of semantic and pragmatic phenomena and is increasingly important for a number of language technologies. It is integrated into theoretical and psycholinguistic models of a range of context-driven effects (Cummins and Rohde, 2015), in- cluding those in coreference (Kehler and Rohde, 2013; Polanyi, 1988), presupposition (Kim et al., 2015), impli- cature (Beaver and Clark, 2008), discourse particles and cue phrases (Hirschberg and Litman, 1993), among others. Within computational systems, multiple domains rely on semantic resources to support the derivation of meaning in text processing and to produce natural sounding language in generation tasks. Discourse structure informs applications such as anaphora resolution (Voita et al., 2018), argument mining (Hewett et al., 2019), machine translation (Xiong et al., 2019), and text simplification (Siddharthan, 2003). One way of articulating the structure of a text is to iden- tify the questions and subquestions that are raised and an- swered by subsequent spans of text. Models of Questions Under Discussion (QUDs) posit underlying structures that are built around a sequence of discourse moves consisting of questions and their answers (Carlson, 1983; Ginzburg, 1994; Ginzburg and Sag, 2000; van Kuppevelt, 1995; Lars- son, 1998; Roberts, 1996). These questions and answers can be understood in terms of their use in moving a dis- course forward to achieve communicative goals and sub- goals. QUDs influence both the surface form of the an- swer and the meaning derived from that answer. But not all QUDs are explicit, in fact most are not, particularly in natural discourse. Recovering implicit QUDs is …

Content truncated to 500 words