A Deep Learning Framework for Real-Time Word-Level Translation of American and Turkish Sign Languages
Bibliographic record
Abstract
Hearing loss or deafness is a problem faced by an increased number of persons worldwide. Deafness is generally not attributed to age or gender, and it is possible to find young persons suffering from deafness. It is a problem which disconnects the suffering persons from properly communicating with the community around them and at large. This may bring up a number of personal and psychological complications on both the person suffering from deafness and the close family. Thanks to the sign language which somehow broke the barriers and highly helped persons with deafness by integrating them in the society. However, the English language is dominating the existing tools targeting this group of people. Particularly, the American Sign Language (ASL) is rendering the existing tools non-generalizable. This paper describes a real-time system sign language system to help in the communication with persons with hearing loss or deafness. The target is to bypass the dominance of the ASL by covering another language, namely the Turkish Sign Language (TSL). The system may be considered as a major step to bridge the gap by providing word-level interpretation for both ASL and TSL. The method described in this paper uses a video- keypoint computer vision pipeline: MediaPipe and a long short- term memory (LSTM) network to interpret the temporal flow of each sign. To provide a wide-ranging and culturally relevant vocabulary, the developed system was trained on two large datasets — Word-Level American Sign Language (WLASL) for ASL and Ankara University Turkish Sign Language (AUTSL) for TSL. The integrated system is supported by a user-friendly interface for real-time practical use by naive users. The models attain impressive scores, 90 % for WLASL and 91 % for AUTSL in accuracy, with F1-scores over 90 %. These findings demonstrate how the developed system forms a promising step forward in the direction towards an accessible communication tool with comprehensive coverage of all the available sign languages for different cultures and ethnicities.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.006 | 0.005 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".