Research map / Video / Video-language understanding
Aligning Subtitles in Sign Language Videos
Hannah Bull, Triantafyllos Afouras, Gül Varol, Samuel Albanie, Liliane Momeni, Andrew Zisserman
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| feed the audio-aligned subtitle timing as a prior (ablation) | SAT · BSL-1K aligned | F1@.50: 30.66 -> 45.10 | Improved |
| shift the audio prior by 3.2 seconds (ablation) | SAT · BSL-1K aligned | F1@.50: 45.10 -> 48.15 | Improved |
| randomise the prior duration by up to 2s (ablation) | SAT · BSL-1K aligned | F1@.50: 48.15 -> 46.72 | Hurt |
| use the subtitle text instead of random text during training (ablation) | SAT · BSL-1K aligned | F1@.50: 40.04 -> 48.15 | Improved |
| add DTW global alignment to resolve overlaps (ablation) | SAT · BSL-1K aligned | F1@.50: 41.27 -> 48.15 | Improved |
| enlarge the search window from 8 to 20 seconds (ablation) | SAT · BSL-1K aligned | F1@.50: 44.13 -> 48.15 | Improved |
| add text augmentation (shuffle and add/delete words) (ablation) | SAT · BSL-1K aligned | F1@.50: 45.31 -> 48.15 | Improved |
| add positional encodings to BERT text features (ablation) | SAT · BSL-1K aligned | F1@.50: 48.15 -> 46.36 | Hurt |
| use token embeddings instead of a single sentence embedding (ablation) | SAT · BSL-1K aligned | F1@.50: 41.71 -> 48.15 | Improved |
| add word-level sign localisation pretraining (ablation) | SAT · BSL-1K aligned | F1@.50: 42.47 -> 48.15 | Improved |
| feed video to the encoder with frame-wise binary output instead of DETR-style regression (ablation) | SAT · BSL-1K aligned | F1@.50: 34.33 -> 48.15 | Improved |
| use BERT instead of word2vec text encoding (ablation) | SAT · BSL-1K aligned | F1@.50: 42.06 -> 48.15 | Improved |
Closest papers by meaning
- Lost in Translation, Found in Embeddings: Sign Language Translation and Alignment (NeurIPS 2026 · similarity 0.94)
- Scaling Sign Language Translation (NeurIPS 2024 · similarity 0.90)
- Lost in Translation, Found in Context: Sign Language Translation with Contextual Cues (CVPR 2025 · similarity 0.90)
- Conditional Variational Autoencoder for Sign Language Translation with Cross-Modal Alignment (AAAI 2024 · similarity 0.89)
- LLMs are Good Sign Language Translators (CVPR 2024 · similarity 0.89)
- Bridging Sign and Spoken Languages: Pseudo Gloss Generation for Sign Language Translation (NeurIPS 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
