Language and accessibility technology

British Sign Language translation

Work on AI methods for British Sign Language translation, including ways to represent and assess visual, temporal and linguistic information.

Current position

At a glance

Area
Research
Category
Language and accessibility technology
Current status
In development
Stage
Research
Back to research

The research question

What this is—and why we are working on it.

This programme investigates AI methods for understanding and translating British Sign Language. BSL is visual and spatial: hand shape, movement, body position, facial expression, timing and the relationships between them can all carry meaning.

That makes the problem different from translating one written language into another. A useful system must represent information across video and time, learn from carefully described data and be evaluated with the people and contexts it is intended to support.

42able is interested in the problem because it brings language, computer vision and accessibility together around a clear need: better ways for services and information to support BSL users.

What we’re investigating

The work behind the question.

The work brings video representation, sequence modelling, language data and human-centred evaluation into one research programme.

  • Video representations that retain movement, timing and spatial relationships
  • Methods for combining manual signs with non-manual information such as facial expression and body movement
  • Experiments with sequence models, intermediate representations and translation outputs
  • Evaluation approaches that examine meaning, failure modes and multiple measures of translation quality

Possible applications

Where this work could matter.

The research could support accessibility and communication tools developed with Deaf BSL users and relevant language professionals.

Communication and accessibility tools

Support new interfaces that help Deaf BSL users access or exchange information across signed and written formats.

Translation assistance

Give translators, interpreters or content teams assistive components for preparing and reviewing BSL material.

Public-facing information

Explore ways to connect BSL video with information published by public-facing organisations.

BSL-accessible services

Investigate components for services that need better BSL accessibility while keeping people and language expertise in control.

Current status

Where the work stands.

The programme is in development. Current work focuses on technical experiments, dataset requirements and evaluation methods across different users and settings.

Open questions

What we’re exploring next.

  • How can a model retain manual, facial, bodily, spatial and temporal information without flattening the language into an inadequate representation?
  • Which datasets and annotation practices cover enough signers, settings and language variation for meaningful evaluation?
  • How should translation quality be assessed when several outputs may be linguistically acceptable?
  • Where can an assistive tool add value while keeping Deaf users and language professionals in control?

Questions and collaboration

Explore the next question with 42able.

We collaborate where a research problem benefits from shared domain knowledge, relevant data and rigorous technical work.

Start a conversation