@inproceedings{miyazaki:24004:sign-lang:lrec,
  author    = {Miyazaki, Taro and Tan, Sihan and Uchida, Tsubasa and Kaneko, Hiroyuki},
  title     = {Sign Language Translation with Gloss Pair Encoding},
  pages     = {32--38},
  editor    = {Efthimiou, Eleni and Fotinea, Stavroula-Evita and Hanke, Thomas and Hochgesang, Julie A. and Mesch, Johanna and Schulder, Marc},
  booktitle = {Proceedings of the {LREC-COLING} 2024 11th Workshop on the Representation and Processing of Sign Languages: Evaluation of Sign Language Resources},
  maintitle = {2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation ({LREC-COLING} 2024)},
  publisher = {{ELRA Language Resources Association (ELRA) and the International Committee on Computational Linguistics (ICCL)}},
  address   = {Torino, Italy},
  day       = {25},
  month     = may,
  year      = {2024},
  isbn      = {978-2-493814-30-2},
  language  = {english},
  url       = {https://www.sign-lang.uni-hamburg.de/lrec/pub/24004.html},
  abstract  = {Because sign languages are the first language for those who are born deaf or who lost their hearing in early childhood, it is better to use sign languages rather than transcribed spoken language to provide important information to these people. We have been developing a sign language computer graphics generation system to provide information to deaf people, and in this paper, we present a translation method from spoken language to sign language that can be used in the system. In general, since the number of glosses used when transcribing sign language is limited, a single meaning is often expressed by a combination of multiple sign words, i.e., the word "library" is expressed in Japanese Sign Language with two words: "book" and "building." To merge these expressions into one token, we propose gloss pair encoding (GPE), which is inspired by bite pair encoding (BPE). This technique is expected to enable more accurate handling of expressions that have a single meaning in multiple sign words. We also show that it is effective as data augmentation on the sign language side in sign language translation, which has not been done much so far.}
}

@inproceedings{uchida:24011:sign-lang:lrec,
  author    = {Uchida, Tsubasa and Miyazaki, Taro and Kaneko, Hiroyuki},
  title     = {{HamNoSys-based} Motion Editing Method for Sign Language},
  pages     = {90--99},
  editor    = {Efthimiou, Eleni and Fotinea, Stavroula-Evita and Hanke, Thomas and Hochgesang, Julie A. and Mesch, Johanna and Schulder, Marc},
  booktitle = {Proceedings of the {LREC-COLING} 2024 11th Workshop on the Representation and Processing of Sign Languages: Evaluation of Sign Language Resources},
  maintitle = {2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation ({LREC-COLING} 2024)},
  publisher = {{ELRA Language Resources Association (ELRA) and the International Committee on Computational Linguistics (ICCL)}},
  address   = {Torino, Italy},
  day       = {25},
  month     = may,
  year      = {2024},
  isbn      = {978-2-493814-30-2},
  language  = {english},
  url       = {https://www.sign-lang.uni-hamburg.de/lrec/pub/24011.html},
  abstract  = {We have developed a Japanese-to-Japanese Sign Language (JSL) translation system to expand sign language services for the Deaf. Although recording the motion data of isolated JSL by motion capture (MoCap) and avatar animation driven by MoCap data is effective for capturing the more natural movements of sign language, the disadvantage is that they lack the flexibility to reproduce the contextual modification of signs. We therefore propose a sign language motion data editing method based on the Hamburg Notation System for Sign Languages (HamNoSys) for use in a hybrid system that combines a MoCap data-driven technique and a phonological generation technique. The proposed method enables the editing of handshape, hand orientation, and location of the motion data based on HamNoSys components to generate contextual modifications for motion-captured citation form signs in translated gloss sequences. Experimental results demonstrate that our method achieves the flexibility to generate contextual modifications and new movements while preserving natural human-like movements without the need for additional MoCap processes.}
}