Institute of Formal and Applied Linguistics

Traditionally, task-oriented dialogue (TOD) systems require experts' extensive involvement to design annotation schemes and prepare the collected dataset. We investigate the problem from different perspectives. In our work, we propose several mechanisms to mitigate the need for structure annotations for TOD modeling. First, we explore the possibilities of automatic data annotation with clustering algorithms. Next, we introduce an unsupervised task-oriented dialogue model with learned latent representations of dialogue actions. Finally, we discuss applications of large pre-trained language models (LLMs) to task-oriented dialogue modeling and share experiment to asses LLM's performance concerning dialogue success rate.

*** The talk will be delivered in person (MFF UK, Malostranské nám. 25, 4th floor, room S1) and will be streamed via Zoom. For details how to join the Zoom meeting, please write to sevcikova et ufal.mff.cuni.cz ***