Task-Oriented Dialogue (TOD) systems provide interactive assistance to a user in order to accomplish a specific task such as making a reservation at a restaurant or booking a room in a hotel. Speech presents itself as a natural interface for TOD systems. A typical approach to implement them is to use a modular architecture (Gao et al., 2018). A core component of such dialogue systems is Spoken Language Understanding (SLU) whose goal is to extract the relevant information from the user’s utterances. While spoken dialogue was the focus of earlier work (Williams et al., 2013; Henderson et al., 2014), recent work has focused on text inputs with no regard for the specificities of spoken language (Wu et al., 2019; Heck et al., 2020; Feng et al., 2021). However, this approach fails to account for the differences between written and spoken language (Faruqui and Hakkani-Tür, 2022) such as disfluencies. My research focuses on Spoken Language Understanding in the context of Task-Oriented Dialogue. More specifically I am interested in the two following research directions: • Annotation schema for spoken TODs, • Integration of dialogue history for contextually coherent predictions.

Though Dialogue State Tracking (DST) is a core component of spoken dialogue systems, recent work on this task mostly deals with chat corpora, disregarding the discrepancies between spoken and written language. In this paper, we propose OLISIA, a cascade system which integrates an Automatic Speech Recognition (ASR) model and a DST model. We introduce several adaptations in the ASR and DST modules to improve integration and robustness to spoken conversations. With these adaptations, our system ranked first in DSTC11 Track 3, a benchmark to evaluate spoken DST. We conduct an in-depth analysis of the results and find that normalizing the ASR outputs and adapting the DST inputs through data augmentation, along with increasing the pre-trained models size all play an important role in reducing the performance discrepancy between written and spoken conversations.