Connecting prosody and syntax: Building a multilayer workflow for spoken Russian
Connecting prosody and syntax: Building a multilayer workflow for spoken Russian

Summary/Abstract: This article presents a pipeline developed to transform a collection of audio recordings into a structured linguistic corpus, searchable for both sound- and text-related phenomena, as well as combinations thereof. The workflow builds on the dataset of telephone conversations created by Jones et al. (2016) and proceeds in three main stages. First, the recordings are automatically transcribed with WhisperX (Bain et al. 2023). Second, pitch levels and pitch movements of the speech signal are automatically labeled via Prosogram (Mertens 2022). Third, Universal Dependencies–style morphosyntactic annotation via Stanza (Qi et al. 2020) is performed on the transcription text. The output of the automatic tools is evaluated in terms of how suitable it is for linguistic research, illustrated on the use case of Russian yes/no-questions.

• Details
• Contents