A Brief Survey of Textual Dialogue Corpora

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 1264–1274 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 1264 A Brief Survey of Textual Dialogue Corpora Hugo Gonc¸alo Oliveira1,2, Patr´ıcia Ferreira1,3, Daniel Martins3, Catarina Silva1,2, Ana Alves1,3 1CISUC, Universidade de Coimbra, Portugal 2DEI, FCTUC, Universidade de Coimbra, Portugal 3ISEC, Instituto Polit´ecnico de Coimbra, Portugal hroliv@dei.uc.pt, patriciaf@dei.uc.pt, daniel.martins2997@gmail.com, catarina@dei.uc.pt, ana@dei.uc.pt Abstract Several dialogue corpora are currently available for research purposes, but they still fall short for the growing interest in the development of dialogue systems with their own specific requirements. In order to help those requiring such a corpus, this paper surveys a range of available options, in terms of aspects like speakers, size, languages, collection, annotations, and domains. Some trends are identified and possible approaches for the creation of new corpora are also discussed. Keywords: Dialogue Corpora, Dialogue Analysis, Dialogue Systems 1. Introduction Driven by the growing adoption of dialogue systems and conversational agents in recent years (Dale, 2016), the need for conversational data to study specific di- alogue phenomena and train the aforementioned sys- tems has also increased. However, despite the broad range of dialogue corpora currently available, many re- leased in the last years, this number is still far from cov- ering all the necessities of real-world applications of di- alogue systems. These include all possibly envisioned dialogue analysis tasks, domains, and their specificity, among others, such as languages. Additionally, the ac- cess to dialogue data and its usage are typically very restricted, due to the logistics involved in their collec- tion and, especially, privacy concerns. For instance, a common application of dialogue systems is automated customer-support, e.g., through call-centers. However, real conversations of this type typically include a large amount of personal data, not always easy to anonymise, as well as sensitive information such as unsatisfied cus- tomers expressing their opinion. Hence, in most cases, there will not be a perfect cor- pus for a new project involving dialogue. This means that researchers will often have to choose between: (i) going through all the work involved in the collection (and annotation) of data for creating a new corpus from scratch; or (ii) looking at publicly available corpora for selecting the most suitable for their task, possibly after some adaptations. This survey targets both previous scenarios. It is not as extensive as Serban et al. (2018), but it gives a more practical perspective, and is more up-to-date, with a fo- cus on corpora of human-human dialogue, available in a textual format, not exclusively in English. More pre- cisely, we compile a list of corpora of such dialogues, and put forward a brief analysis, covering the following aspects: number and type of speakers, size, languages, annotations, data collection, and contents in general. Interested researchers may use this survey as a refer- ence for selecting appropriate corpora for their projects, but also for deciding on suitable approaches for creat- ing new dialogue corpora. With this analysis, we can identify trends, which …

Content truncated to 500 words