Growing research in sign language recognition, generation, and translation AI has been accompanied by calls for ethical development of such technologies. While these works are crucial to helping individual researchers do better, there is a notable lack of discussion of systemic biases or analysis of rhetoric that shape the research questions and methods in the field, especially as it remains dominated by hearing non-signing researchers. Therefore, we conduct a systematic review of 101 recent papers in sign language AI. Our analysis identifies significant biases in the current state of sign language AI research, including an overfocus on addressing perceived communication barriers, a lack of use of representative datasets, use of annotations lacking linguistic foundations, and development of methods that build on flawed models. We take the position that the field lacks meaningful input from Deaf stakeholders, and is instead driven by what decisions are the most convenient or perceived as important to hearing researchers. We end with a call to action: the field must make space for Deaf researchers to lead the conversation in sign language AI.
LINK
Posterpresentatie op Conferentie. Introduction: Classifiers are handshapes (sometimes combined with a specific orientation) that, when combined with the other parameters of movement and location form a ‘verb of motion or location’. There is a limited body of research available on the acquisition of classifiers by children. The available studies have focused on deaf children of deaf (DOD) parents, who are native signers. Results show that classifiers emerge at 3 years and approach an adult like level at the age of 9 (Beal Alvarez & Easterbrooks, 2013). This small study was set out to investigate the production of classifiers in DOH children who acquire Sign Language of the Netherlands. Our expectation was that DOH children produce classifiers, but fail to use them correctly in all instances due to lack of pragmatic control (Slobin et al., 2003). Method: Four children (two girls, two boys) were recruited at a school for the Deaf in The Netherlands (5;10 – 6;8 years). All children were deaf or severely hearing-impaired from birth. Children used (sign supported) Dutch at home and sign language at school and had approximately three years of exposure to sign language. Narratives (Frog-story) were recorded. The recordings were transcribed and analyzed using ELAN-software. Analysis focused on type of classifier (entity and handling) and accuracy in production. Results: The children produced 22 classifiers in total, 20 entity classifiers and 2 handling classifiers. Ten percent of the entity classifiers was incorrect; the handshape to express the entity did not match the handshape frequently selected for that entity. Conclusion: DOH children produce classifiers after three years of exposure to sign language. Errors in classifier production involved errors in handshape selection. This compares to type of errors frequently found for DOD children. Results will be discussed in relation to the iconic and gestural properties of classifiers (Cormier et al., 2012). References: Beal-Alvarez, J.S. & Easterbrooks, S.R. (2013). Increasing children’s ASL classifier production: A multicomponent intervention. American Annals of the Deaf, 158, 311 – 333. Cormier, K., Quinto-Pozos, D., Sevcikova, Z., Schembri, A. (2012). Lexicalisation and de-lexicalisation processes in sign languages: Comparing depicting constructions and viewpoint gestures. Language & Communication, 32, 329 – 348. Slobin, D., Hoiting, N., Kuntze, K., Lindert, R., Weinberg, A. Pyers, J., Anthony, M., Biederman, Y., Thumann, H. (2003). A cognitive/functional perspective on the acquisition of ‘classifiers’. In: Emmorey, K. (Ed.). Perspectives on classifier constructions in sign languages. Lawrence Erlbaum Associates, Mahwah, NJ. Pp 297 – 310.
We present a number of methodological recommendations concerning the online evaluation of avatars for text-to-sign translation, focusing on the structure, format and length of the questionnaire, as well as methods for eliciting and faithfully transcribing responses.
LINK
