A data-driven study of temporal adverbials as discourse segmentation markers

International audienceMany linguistic and psycholinguistic studies present sentence-initial temporal adverbials as "good" markers of discourse segmentation. This paper proposes a corpus-based evaluation of their potential to signal discontinuity such as a discourse break or shift. We look in particular at how temporal adverbials interact with other features, such as position in the text (in relation to document structure), and type of referring expression as grammatical subject in the host sentence. Our methodology calls upon a large diversified tagged corpus, and combines quantitative and qualitative approaches in order to systematically explore these configurations and their relation to text-type. Temporal adverbials are shown in this corpus to signal discontinuity only when in certain configurations, i.e., they cannot be considered as segmentation markers in their own right