DiaBiz – an Annotated Corpus of Polish Call Center Dialogs
Proceedings of the Thirteenth International Conference on Language Resources and Evaluation (LREC 2022)
Abstract
This paper introduces DiaBiz, a large, annotated, multimodal corpus of Polish telephone conversations conducted in varied business settings, comprising 4036 call centre interactions from nine different domains, i.e. banking, energy services, telecommunications, insurance, medical care, debt collection, tourism, retail and car rental. The corpus was developed to boost the development of third-party speech recognition engines, dialog systems and conversational intelligence tools for Polish. Its current size amounts to nearly 410 hours of recordings and over 3 million words of transcribed speech. We present the structure of the corpus, data collection and transcription procedures, challenges of punctuating and truecasing speech transcripts, dialog structure annotation and discuss some of the ecological validity considerations involved in the development of such resources.