Scholarlink

<p>Sign languages pose unique computational challenges because meaning is distributed across handshapes, motion, facial expression, and body posture. Focusing on Turkish Sign Language (Türk İşaret Dili, TİD), this Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)-guided review screens eight databases and brings together 86 peer-reviewed studies published from 2003 to 2026. We trace the journey from marker-based rigs to today’s multimodal deep-learning and graph architectures, showing how benchmark corpora such as BosphorusSign, BosphorusSign22k, and AUTSL triggered step-wise accuracy jumps. Many advances coincide with the incorporation of linguistically motivated cues (e.g., non-manuals, spatial constructions, dialect tags) into data and models; however, reported gains may also reflect confounds such as larger datasets, stronger backbones, or augmentation, so we interpret causal attributions cautiously. In controlled isolated-sign benchmarks, current systems can exceed 95 percent top-1 accuracy and early continuous-sign translators now exist; however, deployment still wrestles with limited dialect coverage, sparse facial annotation, evaluation-protocol variability, and latency/robustness when processing new signers. To close these gaps we highlight three priorities. First, build larger and more demographically balanced corpora that synchronize RGB, depth, and skeleton streams. Second, adopt linguist-led annotation schemes that record clause structure, prosody, and regional variants. Third, develop domain-adaptation pipelines and user studies in direct partnership with Deaf communities. Concrete prototypes already point the way: classroom tools that turn spoken Turkish into avatar-based TİD and voice back students’ signed questions, hospital kiosks that mediate triage dialogs, and broadcast overlays that render live news in sign form. By weaving linguistic theory into machine-learning practice, TİD research is emerging as a versatile testbed for sign-language technology and a model for human-centered AI that advances accessibility in education, healthcare, and public media.</p>

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 insights into Turkish sign language recognition, generation, and translation: a comprehensive systematic review

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月9日

查看全文
  • Fesih Keskin,
  • Gültekin Işık

摘要

Sign languages pose unique computational challenges because meaning is distributed across handshapes, motion, facial expression, and body posture. Focusing on Turkish Sign Language (Türk İşaret Dili, TİD), this Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)-guided review screens eight databases and brings together 86 peer-reviewed studies published from 2003 to 2026. We trace the journey from marker-based rigs to today’s multimodal deep-learning and graph architectures, showing how benchmark corpora such as BosphorusSign, BosphorusSign22k, and AUTSL triggered step-wise accuracy jumps. Many advances coincide with the incorporation of linguistically motivated cues (e.g., non-manuals, spatial constructions, dialect tags) into data and models; however, reported gains may also reflect confounds such as larger datasets, stronger backbones, or augmentation, so we interpret causal attributions cautiously. In controlled isolated-sign benchmarks, current systems can exceed 95 percent top-1 accuracy and early continuous-sign translators now exist; however, deployment still wrestles with limited dialect coverage, sparse facial annotation, evaluation-protocol variability, and latency/robustness when processing new signers. To close these gaps we highlight three priorities. First, build larger and more demographically balanced corpora that synchronize RGB, depth, and skeleton streams. Second, adopt linguist-led annotation schemes that record clause structure, prosody, and regional variants. Third, develop domain-adaptation pipelines and user studies in direct partnership with Deaf communities. Concrete prototypes already point the way: classroom tools that turn spoken Turkish into avatar-based TİD and voice back students’ signed questions, hospital kiosks that mediate triage dialogs, and broadcast overlays that render live news in sign form. By weaving linguistic theory into machine-learning practice, TİD research is emerging as a versatile testbed for sign-language technology and a model for human-centered AI that advances accessibility in education, healthcare, and public media.