AI Use Policy
The videos and original captions on this site are entirely human-produced.
AI has been used in a supporting role both to curate the collection and to generate short summaries and descriptions that help learners navigate the material. AI-generated content is clearly marked with an AI badge where it appears.
What is human-produced
- Videos — all recordings are from the BSL Corpus Project,
filmed by UCL researchers with 249 deaf signers across the UK.
- BSL gloss annotations — the sign-level captions used to generate subtitles were produced by trained human annotators working from the original recordings, with some processing described below.
- Free English translations — the English renderings of
each story were written by human translators, time-aligned at sentence
level.
What AI was used for in creating this site
Background curation
Exploration — Claude (Anthropic) used the English translation text and full BSL gloss (EAF) to identify, suggest and locate relevant content and steer the direction of this site.
Converting EAF gloss to BSL captions To make the original expert-labelled sign glosses useful to novice learners, we used Claude (Anthropic) under close human supervision to detect typos or errors, and convert from technical EAF format to readable BSL captions. For example,
"DSH(SPHERE)-MOVE/G:CA:PLACES-BABY-DOWN" -> "[places baby down]", and
"PT:POSS3SG" -> "his" or "her" depending on context.
Coding — Claude Code was used to write much of the website and scripts used to generate content.
Clip selection — representative sentence clips for each story, topic or grammatical feature, generated by Claude (Anthropic) from the English translation text.
User-facing text
- Story summaries — two-sentence overviews of each story, generated by Claude (Anthropic) from the English translation text.
- Sign notes — the one-sentence explanations of why a
particular sign or moment is interesting to a BSL learner are AI-generated. This is the highest risk use of AI and we are open to feedback on how to minimise or validate this content.
AI-generated content is clearly marked with an AI badge where it appears.
Accuracy and limitations
AI-generated summaries and notes are generally reliable but may occasionally be inaccurate, oversimplified, or miss important nuance — particularly for linguistically complex moments. They are intended as orientation aids, not authoritative descriptions.
The underlying human annotations are the authoritative source. If you spot
a significant error in the AI-generated content, please let us know.