American Sign Language Recognition Using a Multimodal Transformer Network
Bibliographic record
Abstract
The North American Deaf community, who commonly communicates using American Sign Language (ASL), faces obstacles regarding education, employment, and access to mental health services [1], [2], [3]. The invention of sign language translation services could assist in reducing the cultural disconnect felt by the Deaf community and encourage hearing individuals to learn sign language. Sign language translation is performed using a camera or video feed to capture a person’s hand signs while performing ASL. The visual data are analyzed using computer vision technologies and passed through neural networks for recognition prediction. This paper proposes a scalable proof-of-concept deep learning solution that can recognize hand signs when trained on limited datasets using a unique Multimodal Transformer Network (MTN) approach. This architecture improves the learning and recognition process by using combined and partitioned skeletal landmarks simultaneously, while further expanding context with Convolutional Neural Network (CNN) features to include relevant information not found within the skeletal data. MediaPipe Holistic [4] and the ResNet50 ImageNet [5] model extract skeletal landmarks and CNN feature data from video feeds, respectively. These data are used as input channels spread among five modalities, achieving an accuracy of 94% with 22 classifications when trained on a dataset with little variety.
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 | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.005 | 0.001 |
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".