Scholarlink

DEEP is a novel bidirectional Italian sign language (LIS)-to-Italian automatic translation system. Sign languages are notoriously hard to recognize by a computer vision system, and it is also quite difficult to generate signs in a smooth and comprehensible way. DEEP focuses on two case study scenarios: pharmacy and passport/identity office, providing a vertical, high-quality dataset, and implementing two pipelines that permit to translate LIS to Italian (with a BLEURT of 0.81), and Italian to LIS (generating a very good avatar, as reported by several deaf subjects and LIS interpreters). Whereas the LIS-to-Italian pipeline uses a generic translator, the Italian-to-LIS pipeline is more constrained, relying on a predefined list of known sentences. Despite this limitation, we argue the whole system could be a pragmatic and useful choice for several scenarios.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

DEEP: An Automatic Bidirectional Translator for Italian Sign Language

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Nicolas Tagliabue,
  • Elisa Colletti,
  • Roberto Tedesco,
  • Francesco R. Dani,
  • Sonia Cenceschi,
  • Alessandro Trivilini

摘要

DEEP is a novel bidirectional Italian sign language (LIS)-to-Italian automatic translation system. Sign languages are notoriously hard to recognize by a computer vision system, and it is also quite difficult to generate signs in a smooth and comprehensible way. DEEP focuses on two case study scenarios: pharmacy and passport/identity office, providing a vertical, high-quality dataset, and implementing two pipelines that permit to translate LIS to Italian (with a BLEURT of 0.81), and Italian to LIS (generating a very good avatar, as reported by several deaf subjects and LIS interpreters). Whereas the LIS-to-Italian pipeline uses a generic translator, the Italian-to-LIS pipeline is more constrained, relying on a predefined list of known sentences. Despite this limitation, we argue the whole system could be a pragmatic and useful choice for several scenarios.