Leiden University Student Repository

Biesheuvel, Merel Levina (2024)

Predicting Implicitness: Alignment of Evoked Questions and Discourse Relations in TED Talks

Bachelor thesis | Taalwetenschap (BA)

When reading, people often form questions to understand how ideas in a text are connected. This study examined whether the alignment between these questions and the text’s actual connections – where alignment means that the expected connection matches the one in the text – can explain when and why some connections are stated explicitly, while others are left implicit. We analysed TED Talk transcripts using two datasets: TED-Q, which captures questions that were evoked by readers, and the TED Multilingual Discourse Bank (TED-MDB), which annotated the discourse relations. A new method was developed to link the questions from TED-Q to the annotations from TED-MDB, which proved to be reliable and scalable. The results showed that aligned connections were nearly three times more likely to be implicit. Alignment was also the strongest predictor of implicitness. These findings highlight alignment’s utility for understanding implicit discourse and its potential for advancing discourse...Show moreWhen reading, people often form questions to understand how ideas in a text are connected. This study examined whether the alignment between these questions and the text’s actual connections – where alignment means that the expected connection matches the one in the text – can explain when and why some connections are stated explicitly, while others are left implicit. We analysed TED Talk transcripts using two datasets: TED-Q, which captures questions that were evoked by readers, and the TED Multilingual Discourse Bank (TED-MDB), which annotated the discourse relations. A new method was developed to link the questions from TED-Q to the annotations from TED-MDB, which proved to be reliable and scalable. The results showed that aligned connections were nearly three times more likely to be implicit. Alignment was also the strongest predictor of implicitness. These findings highlight alignment’s utility for understanding implicit discourse and its potential for advancing discourse analysis.Show less

Discourse analysis, implicit discourse, discourse relations, discourse annotation, TED Talks

Author

Biesheuvel, Merel Levina

Faculty

Faculty of Humanities

Specialisation

Language & Cognition

Supervisors

Westera, M.

ECTS Credits

10

Language

en