Digital Total - Computing & Data Science an der Universität Hamburg und in der Wissenschaftsmetropole Hamburg

Semi-Automatic Subtitle Alignment for Sign Languages

125

Nicht eingeplant

20m

Von-Melle-Park 4

Poster

Beschreibung

To provide subtitles for videos in sign language the translation has to be aligned with the signing.
Doing this manually is time consuming but fully automatic alignment is often not accurate enough. We therefore propose an interactive tool that uses human feedback to optimize the alignment process. For the automatic alignment we use signs detected by a sign spotter and map them to words in the translation. These mappings are used as landmarks for the alignment algorithm. Users can change the alignment directly but also alter, add or delete landmarks to guide the automatic process, so that a correct alignment is reached after a few feedback loops.

Keywords

Sign Languages
Machine Learning
Natural Language Processing
Interface Design
Semi-Automation

Präsentationsmaterialien

SemiautoSubtAlign.pdf