NEWS · RESEARCH · #255
DiscoSign: Discourse-aware text-to-sign-language-gloss translation using LLMs
This paper introduces DiscoSign, a modular LLM-based framework for discourse-aware translation from written text to sign language glosses. The approach is grounded in linguistic research and explicitly addresses discourse phenomena such as (i) spatial coreference resolution and (ii) Question-Answer Clauses (QACs); the excerpt indicates a third phenomenon is also handled but does not specify it.
KEY POINTS
- This paper introduces DiscoSign, a modular LLM-based framework for discourse-aware translation from written text to sign language glosses.
- The approach is grounded in linguistic research and explicitly addresses discourse phenomena such as (i) spatial coreference resolution and (ii) Question-Answer Clauses (QACs); the excerpt indicates a third phenomenon is also handled but does not specify it.
- Handling discourse-level phenomena (e.g., spatial coreference and QACs) can make text→sign-gloss translation more coherent and linguistically appropriate, which matters for more natural sign language generation and downstream signer comprehension.
WHY IT MATTERS
Handling discourse-level phenomena (e.g., spatial coreference and QACs) can make text→sign-gloss translation more coherent and linguistically appropriate, which matters for more natural sign language generation and downstream signer comprehension.