Speakers
Hendrik Kröger
Kamil Piwowar
(Goethe Universität)
Description
Using large language models, which represent language elements in multidimensional semantic space, we can track how speech unfolds over time. For structured narratives, this approach allows us to examine how semantic content evolves from the opening sentence to the climax by dividing the story into schema-based events. We present a computational pipeline that traces the progression of spoken summaries of Hänsel and Gretel. The spoken input is transcribed into utterances, which are then embedded with a language model and aligned to a predefined set of major narrative events. A cognitive-map-like representation of the narrative is generated by projecting the utterances into two- and three-dimensional space.
Authors
Hendrik Kröger
Kamil Piwowar
(Goethe Universität)
Co-author
Oliver Grimm
(Goethe-Universität)