A Topic-Aligned Multilingual Corpus of Wikipedia Articles for Studying Information Asymmetry in Low Resource Languages

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 220–228 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 220 Developing a Corpus of Indirect Speech Act Schemas Antonio Roque, Alexander Tsuetaki, Vasanth Sarathy, Matthias Scheutz Tufts University Medford, MA, USA {Firstname.Lastname}@tufts.edu Abstract Resolving Indirect Speech Acts (ISAs), in which the intended meaning of an utterance is not identical to its literal meaning, is essential to enabling the participation of intelligent systems in peoples’ everyday lives. Especially challenging are those cases in which the interpretation of such ISAs depends on context. To test a system’s ability to perform ISA resolution we need a corpus, but developing such a corpus is difficult, especialy given the contex-dependent requirement. This paper addresses the difficult problems of constructing a corpus of ISAs, taking inspiration from relevant work in using corpora for reasoning tasks. We present a formal representation of ISA Schemas required for such testing, including a measure of the difficulty of a particular schema. We develop an approach to authoring these schemas using corpus analysis and crowdsourcing, to maximize realism and minimize the amount of expert authoring needed. Finally, we describe several characteristics of collected data, and potential future work. Keywords: corpus methodologies, corpus construction, crowdsourcing 1. Introduction Indirect speech acts (ISAs) involve utterances whose lit- eral meanings are not identical to their intended meanings (Searle, 1975). For example, the utterance “can you open the door?” has the surface form of a question and a literal meaning of an elicitation of information from the hearer re- garding the hearer’s ability to open the door. In some cases, the speaker’s intent may in fact be to have the hearer reply with the answer to that question. However, if the intended meaning is a request that the hearer perform the action of opening the door, then an ISA has been performed. Over the years, researchers have developed inferential (Per- rault and Allen, 1980), idiomatic (Wilske and Kruijff, 2006), and hybrid (Briggs and Scheutz, 2013) implemen- tations of ISA analysis. Studies have shown that peo- ple commonly use ISAs in everyday conversations as well as in task-based dialogues with robots (Williams et al., 2018), and that context is important for understanding ISAs (Gibbs Jr, 1979). In fact, we argue that the detec- tion and interpretation of ISAs is modulated by numerous contextual factors, such as location, task, interactant roles, co-present objects, and recent dialogue history. Even slight changes in this context can change whether an utterance’s intended meaning is different from its literal meaning. In- telligent systems that interact with people in real-world en- vironments need to be able to use context to determine whether an utterance should be interpreted literally, or as part of an ISA. An important part of developing such a ca- pability is testing it. But doing so requires answering sev- eral difficult questions regarding how to collect and repre- sent such content. In this paper we make the following contributions. We de- scribe …

Content truncated to 500 words