DiaBiz.Kom - towards a Polish Dialogue Act Corpus Based on ISO 24617-2 Standard

Proceedings of the 29th International Conference on Computational Linguistics, pages 3631–3638 October 12–17, 2022. 3631 DiaBiz.Kom – Towards a Polish Dialogue Act Corpus Based on ISO 24617-2 Standard Marcin Oleksy, Jan Wieczorek, Dorota Dru˙zyłowska, Julia Klyus, Aleksandra Domogała, Krzysztof Hwaszcz, Hanna K˛edzierska, Daria Miko´s, Anita Wró˙z Wrocław University of Science and Technology, Wrocław, Poland {marcin.oleksy, jan.wieczorek, dorota.druzylowska}@pwr.edu.pl {julia.klyus, aleksandra.domogala, krzysztof.hwaszcz}@pwr.edu.pl {hanna.kedzierska, daria.dominiak, anita.wroz}@pwr.edu.pl Abstract This article presents the specification and eval- uation of DiaBiz.Kom – the corpus of dialogue texts in Polish. The corpus contains transcrip- tions of telephone conversations conducted ac- cording to a prepared scenario. The transcripts of conversations have been manually annotated with a layer of information concerning com- municative functions. DiaBiz.Kom is the first corpus of this type prepared for the Polish lan- guage and will be used to develop a system of dialogue analysis and modules for creating advanced chatbots. 1 Introduction The rationale of the current research was predomi- nantly connected with the lack of corpora including dialogue texts in Polish which could be used to train artificial intelligence for model creation. In order to bridge this gap, we decided to create a corpus that satisfies our expectations i.e. the one that con- tains dialogue samples from several different fields of business and is annotated for information con- cerning pragmatic functions. At the first stage of our work, we analysed the approaches adopted by other researchers, whose solutions are described in the "Related Works" section. Further, we describe the process of data collection and its manual an- notation. A separate section is dedicated to "Key assumptions and limitations of the guidelines". The paper ends with "Corpus overview" and "Conclu- sions and future works". 2 Related work While creating the Polish corpus of dialogue acts, we analyzed some pre-existing corpora. In theo- retical perspective we refer to the ISO/DIS 24617- 2, standardized annotations and the recommenda- tions from the mentioned document. The ISO standard (Bunt et al., 2010, 2012) is based on particular innovations such as distinguishing be- tween annotations and representations (according to ISO Linguistic Annotation Framework (LAF, ISO 24612:2009) and sets of dialogue participants, dimensions, communicative functions, functional segments and qualifiers (inventory of DiAML). Both manual and automatic annotation of dialogue segments are possible according to the ISO doc- ument and both have been tested in practice and described (Keizer et al., 2011; Petukhova et al., 2014; Bunt et al., 2016; Chowdhury et al., 2016; Ngo et al., 2017; Gilmartin et al., 2018). Dialog- Bank is a language resource containing dialogues with gold standard annotations corresponding with the ISO 24617-2 standard (Bunt et al., 2016). The development of annotation standards for par- ticular corpora can be vividly exemplified by the case of the Switchboard Dialogue Act Corpus (the collection of telephone conversations). Telephone Speech Corpus (LDC97S62) was originally col- lected by Texas Instruments in 1990-1 and consists of approximately 260 hours of speech. The first release of the corpus was published in 1992-31. Ini- tially, the utterances in the corpus were annotated according …

Content truncated to 500 words