Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

SAGE: Segment-Aware Gloss-Free Encoding for Token-Efficient Sign Language Translation

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Segmented video tokens enable state-of-the-art gloss-free sign language translation with roughly half the sequence length and 2.67x less memory than prior methods.

desk verdict SAGE is a solid, honestly-written efficiency contribution whose SOTA claim is conditional on disclosing the frozen segmentor's training data and adding error bars. read the letter →

arxiv 2507.09266 v2 pith:EMBLAUIC submitted 2025-07-12 cs.CV

JianHe Low , Ozge Mercanoglu Sincan , Richard Bowden This is my paper · ORCID
classification cs.CV
keywords signlanguagetranslationgloss-freevisualtokenizationsegmentationcontrastivelearningtoken-efficientPHOENIX14T
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that sign language translation can be made both more accurate and much cheaper by first splitting continuous signing video into discrete segments, each corresponding to one sign, and translating each segment as a single visual token. On the PHOENIX14T benchmark, the resulting system, SAGE, reaches a BLEU-4 score of 24.10, beating the previous best gloss-free result of 23.40, while using roughly half the input sequence length and up to 2.67x less peak training memory than prior state-of-the-art. The authors argue that the efficiency gain is not a quality loss: even when previous methods are forced to similar sequence lengths, they fall behind. The paper also introduces a token-level contrastive alignment objective that is claimed to close the visual–language gap without any gloss annotations.

What carries the argument

At the heart of the method is a segment-aware visual tokenizer: a frozen sign segmentation model (the Hands-On model) that predicts boundaries between signs from hand and body pose, followed by a hierarchical encoder that turns each sign segment into a single 1024-dimensional token. This tokenizer is what shrinks a video of hundreds of frames into a sequence of a few dozen tokens (reduction ratio ~0.129), making the Transformer's quadratic self-attention tractable. The alignment machinery is the token-level Cross-Lingual Contrastive Learning (CLCL) loss, applied at two levels of the language encoder—input embeddings and contextualized hidden states—to bind each visual token to a pseudo-gloss word. The final piece is the two-stage protocol: contrastive pretraining of the visual encoder and mapper, then fine-tuning with mBART for translation.

What would settle it

Run SAGE with the Hands-On segmentor replaced by a random segmentation of the same average segment length on the PHOENIX14T dev set; if BLEU-4 does not drop substantially, the specific learned boundaries are not the source of the gain, contradicting the claim that segment-aware tokenization is the key mechanism.

Watch

Extended reading notes

Core claim

SAGE introduces a two-stage gloss-free sign language translation framework. In the first stage, a frozen segmentation model splits continuous sign video into sign units, and a hierarchical visual encoder (ResNet-34 plus a 1D temporal convolution and average pooling) converts each unit into a single 1024-dimensional token; a transformer over these short token sequences captures cross-segment context. The authors align these visual tokens with pseudo-glosses—content words extracted from the spoken sentence via part-of-speech filtering—using a Cross-Lingual Contrastive Learning (CLCL) objective applied both to the mBART input embedding layer and to contextualized hidden states. In the second stage, the pretrained visual encoder and a visual-to-language mapper are attached to the mBART encoder–decoder and fine-tuned for translation. The central discovery is that segment-aware tokenization preserves the linguistic structure of signs while drastically shortening the input sequence, making state-of-the-art translation possible at a memory footprint of roughly 60 GB instead of 80–160 GB.

Load-bearing premise

The load-bearing premise is that the frozen Hands-On segmentation model, without any fine-tuning, produces reliable, semantically meaningful sign boundaries on PHOENIX14T; if those boundaries are noisy or the model's training data overlaps the test split, the token compression loses linguistic content and the efficiency–accuracy trade-off collapses.

Editorial extensions

If this is right

  • A single ~60 GB GPU budget suffices for state-of-the-art gloss-free training, roughly 2.67x less memory than Sign2GPT and 1.33x less than GFSLT.
  • At reduction ratio 0.129, the token sequence is about half the length used by comparable methods, cutting the Transformer's quadratic attention cost proportionally.
  • Token-level CLCL alignment improves BLEU-1 through BLEU-3 more than global CLIP-style alignment, indicating finer-grained visual-to-word correspondence.
  • Under matched half-length inputs, SAGE's BLEU-4 of 24.10 exceeds GFSLT's 19.25 and FLa-LLM's 20.02, showing that semantic tokenization retains quality where uniform downsampling degrades it.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A learned, end-to-end boundary predictor optimized jointly with translation could outperform the frozen segmentor; the paper's own ablation discards the pretraining Transformer, suggesting the visual encoder itself may be simplified further.
  • Pseudo-gloss extraction via POS filtering may be replaced by a learnable text tokenizer or by mBART's own subword units, which could better handle multi-word signs and reduce the mismatches shown in the similarity matrix.
  • The 0.129 reduction ratio is not a hard floor: co-articulated signs with redundant frames could be merged, potentially halving token count again without loss.
  • If SAGE transfers to continuous sign language recognition, segment tokens could replace frame-wise CTC alignment, since gloss boundaries are typically weak supervision in that task.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper introduces SAGE, a two-stage gloss-free sign language translation (SLT) framework. In stage 1, a frozen segmentor (Hands-On) partitions continuous signing video into sign segments, each encoded via a ResNet-34 and a lightweight temporal encoder followed by average pooling to produce one visual token per segment. A token-to-token contrastive loss (CLCL) aligns these visual tokens with pseudo-glosses extracted from spoken sentences using POS tagging, with supervision applied at both the mBART token-embedding level and the hidden-state level. In stage 2, the pretrained visual encoder and mapper are attached to an mBART-large-50 encoder-decoder and fine-tuned for translation. On PHOENIX14T, the method reports a BLEU-4 of 24.10, a token reduction ratio of 0.129, and peak training memory of roughly 60 GB, claiming state-of-the-art gloss-free performance and improved scalability. Ablations evaluate loss choices, the weighting hyperparameter \beta, and pretraining configurations on the development set.

Significance. If the claims hold, SAGE would be the first gloss-free SLT method to use sign segmentation as a visual tokenizer, achieving a meaningful reduction in input sequence length (about half of prior 0.25 ratios) while matching or exceeding current gloss-free BLEU scores. The idea of segment-aware tokenization with token-level contrastive alignment is a plausible and potentially scalable direction for SLT, and the paper is honest about some limitations, such as pseudo-gloss mismatches in the qualitative analysis. Strengths include the released code, clearly described architecture, internally consistent ablations, and explicit acknowledgment in Section 4.5 that spoken-language-derived pseudo-glosses do not always match signed content. However, the central empirical claims rest on an unverified frozen segmentor and on small performance differences without statistical support, so the significance is conditional on those points being resolved.

major comments (3)
  1. [§3.1.1, §4.3, Table 1] The frozen Hands-On segmentor [25] is load-bearing for both the claimed token compression (ratio 0.129 in Table 2) and the BLEU-4 improvement (24.10 vs. 23.40 in Table 1), but the paper does not report Hands-On's training corpus, nor any boundary-quality measure on PHOENIX14T. If Hands-On was trained on PHOENIX14T or its near-duplicates, the gloss-free framing and the comparison to prior work are not a fair test; if its boundaries are noisy, average pooling within segments can discard linguistic content. The manuscript defers these details to 'supplementary materials', but no supplementary file is present in the arXiv v2 submission. Please disclose the segmentor's training data, report segment-boundary quality (e.g., F1 against gloss-aligned boundaries or per-segment token statistics) on the PHOENIX14T test split, and, ideally, compare against a segmentor trained on a disjoint corpus.
  2. [§4.3, Table 1] The central SOTA claim rests on a +0.7 BLEU-4 difference (24.10 vs. 23.40) over SignLLM, with no error bars, number of seeds, or statistical significance test. For sequence-to-sequence models on this benchmark, such a margin is within typical run-to-run variance; please provide mean and standard deviation over at least three seeds for SAGE and, if feasible, for the closest baselines, or a paired bootstrap test over test sentences.
  3. [§4.3.2, Table 3] The memory comparison reports '∼60 GB' and '∼3 × RTX 3090' as approximate values without specifying how peak VRAM was measured (e.g., per-component versus whole-training peak, batch size, sequence-length distribution). Because each baseline uses different hardware and batch sizes, the claimed 2.67× reduction is not on equal footing. Please report measured peak memory under matched conditions (same GPU, same batch size, comparable sequence lengths) or provide the full training configurations so the comparison can be reproduced.
minor comments (4)
  1. [§3.1.2, Eq. (1)] The notation t_i is used both for visual segment tokens in Section 3.1.1 and for subword tokens in Eq. (1); please use distinct symbols (e.g., s_i or u_i) to avoid ambiguity.
  2. [§3.1.3] The output dimension of the temporal encoder is written as R^{(n−4)×1024} because of the kernel size k=5, but the text does not explain that no padding is applied; please clarify this dimension change.
  3. [§4.4, Table 4] The table captions do not state that the ablations are on the development set, although the text in Section 4.4 does; please make this explicit in the captions.
  4. [General] The paper refers to 'supplementary materials' in Sections 3.1.1, 4.2, and 4.5, but the arXiv v2 submission does not include a supplementary file; the final version must include the referenced material or clearly indicate where it is available.

Circularity Check

0 steps flagged · score 1.0 of 10

No demonstrated circularity; the central claim is an empirical benchmark result, and the only concern is an undisclosed training-set overlap for the self-cited Hands-On segmentor.

full rationale

The paper's derivation chain is empirical, not formal: a two-stage pretraining-and-fine-tuning pipeline is evaluated on PHOENIX14T, and there is no equation-level reduction in which an output is defined in terms of an input or a fitted parameter is renamed as a prediction. The visual tokenizer is a frozen model, Hands-On [25], from the same research group; the method and its efficiency gains (0.129 token ratio; ~60 GB VRAM) depend on this segmentor. The paper does not disclose Hands-On's training corpus or measure segment-boundary quality on PHOENIX14T (Section 3.1.1 only refers to supplementary materials), so if [25] had been trained on PHOENIX14T or near-duplicates, the benchmark comparison would not be independent. However, the manuscript offers no evidence of such overlap, and the central BLEU results are self-contained comparisons against published numbers from external systems (GFSLT, FLa-LLM, SignLLM, etc.). The pseudo-gloss extraction 'Following [37]' is a POS-filtering step reused from the authors' Sign2GPT, but it is not a claimed derivation and does not force the translation result. The CLCL loss is externally published (Cico [9]) and is applied here at token level, a new combination rather than a circular import. The absent supplementary runtime/segmentation details are a reproducibility gap, not demonstrated circularity. Thus the circularity score is low.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim pulls most of its machinery from prior work: Hands-On for segmentation, Sign2GPT for pseudo-gloss extraction, Cico for CLCL, and mBART. The hyperparameters beta and alpha control the objective, and L controls pretraining depth. No new physical or conceptual entities are proposed.

free parameters (3)
  • beta (loss weighting in Eq. 5) = 0.6
    Balances Cross-Embedding and Cross-Hidden-State losses; selected by dev-set ablation in Table 4b, where BLEU-4 ranges from 22.78 to 23.81.
  • alpha (loss weighting in Eq. 4) = not reported
    Balances video-to-text and text-to-video CLCL losses; the paper does not state its value or ablate it.
  • pretraining transformer depth L = 3
    Chosen over 12 layers based on Table 5; the 3-layer setting gives BLEU-4 23.81 versus 23.43 for 12 layers.
assumptions (4)
  • domain assumption Hands-On segmentation generalizes to PHOENIX14T without fine-tuning.
    Invoked in Section 3.1.1 as a frozen tokenizer; no validation on PHOENIX14T is shown and its training data is not disclosed.
  • domain assumption POS-filtered pseudo-glosses approximate the semantic content of signed video.
    Section 3.1.2 follows Sign2GPT; Section 4.5 admits mismatches, including words that are not signed and signs that are missing from pseudo-glosses.
  • domain assumption mBART-large-50 encodes German words well enough to serve as alignment targets.
    Section 3.1.2 uses mBART input embeddings and hidden states as text targets without checking DGS-specific coverage.
  • domain assumption CLCL transfers from per-frame I3D features to segment-level token embeddings.
    Section 3.1.5 adapts Cico's CLCL [9], originally designed for frame features, to segment tokens; the transfer is assumed rather than justified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SAGE: Segment-Aware Gloss-Free Encoding for Token-Efficient Sign Language Translation." pith.science (2026). https://pith.science/paper/EMBLAUIC

@misc{pith2026250709266,
  author       = {Pith},
  title        = {Pith review of: SAGE: Segment-Aware Gloss-Free Encoding for Token-Efficient Sign Language Translation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EMBLAUIC}},
  note         = {Machine review of arXiv:2507.09266}
}
read the original abstract

Gloss-free Sign Language Translation (SLT) has advanced rapidly, achieving strong performances without relying on gloss annotations. However, these gains have often come with increased model complexity and high computational demands, raising concerns about scalability, especially as large-scale sign language datasets become more common. We propose a segment-aware visual tokenization framework that leverages sign segmentation to convert continuous video into discrete, sign-informed visual tokens. This reduces input sequence length by up to 50% compared to prior methods, resulting in up to 2.67x lower memory usage and better scalability on larger datasets. To bridge the visual and linguistic modalities, we introduce a token-to-token contrastive alignment objective, along with a dual-level supervision that aligns both language embeddings and intermediate hidden states. This improves fine-grained cross-modal alignment without relying on gloss-level supervision. Our approach notably exceeds the performance of state-of-the-art methods on the PHOENIX14T benchmark, while significantly reducing sequence length. Further experiments also demonstrate our improved performance over prior work under comparable sequence-lengths, validating the potential of our tokenization and alignment strategies.

Figures

Figures reproduced from arXiv: 2507.09266 by the authors.

Figure 1
Figure 1. Method Comparison. Conventional approaches rely on temporal downsampling to reduce the number of video frames; however, this often still results in lengthy and redundant sequences being passed to the language encoder. In contrast, our method in￾troduces a visual tokenization strategy that groups frames into dis￾crete visual tokens, significantly reducing input sequence length. across space and time through coordinat… view at source ↗
Figure 2
Figure 2. Architecture Overview. Our method consists of two stages. (i) In the first stage, a sign segmentor tokenizes continuous sign videos, and a visual encoder maps each segment into an embedding. We then apply contrastive pretraining to align the visual representations with the textual embeddings from a language encoder, encouraging semantically rich and discriminative features. (ii) In the second stage, the visual token… view at source ↗
Figure 3
Figure 3. Contrastive Learning Comparison. Prior pretraining methods adopt CLIP-style objectives to align video and text glob￾ally via summarized [CLS] tokens. In contrast, our approach per￾forms fine-grained cross-lingual alignment by computing token￾level similarities across all the video-text pairs within a batch. ordering. As our pseudo-glosses follow spoken language order rather than the temporal progression of the video… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: shows the similarity matrix between segmented vi￾sual tokens and pseudo-gloss candidates via dot-product of their embeddings. The clear diagonal patterns show that the CLCL objective effectively enforces fine-grained, token￾level alignment with semantically relevant ps…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Sign Spotting Disambiguation using Large Language Models

    cs.CV 2025-07 conditional novelty 5.0 of 10

    LLM-based beam search disambiguation improves dictionary sign spotting WER from 47.2% to 44.4% on an internal BSL dataset.

Reference graph

Works this paper leans on

48 extracted references · 44 canonical work pages · cited by 1 Pith paper

  1. [25]

    Hands-on: Segmenting individual signs from continuous sequences

    Jianhe Low, Harry Walsh, Ozge Mercanoglu Sincan, and Richard Bowden. Hands-on: Segmenting individual signs from continuous sequences. In FG, 2025. 2, 3

    work page 2025

  2. [1]

    Bsl-1k: Scaling up co-articulated sign language recognition using mouthing cues

    Samuel Albanie, G ¨ul Varol, Liliane Momeni, Triantafyllos Afouras, Joon Son Chung, Neil Fox, and Andrew Zisserman. Bsl-1k: Scaling up co-articulated sign language recognition using mouthing cues. In ECCV, pages 35–53. Springer,

    work page

  3. [2]

    Distri- bution and duration of signs and parts of speech in swedish sign language

    Carl B ¨orstell, Thomas H ¨orberg, and Robert ¨Ostling. Distri- bution and duration of signs and parts of speech in swedish sign language. Sign Language & Linguistics, 19(2):143–196,

    work page

  4. [3]

    Neural sign language trans- lation

    Necati Cihan Camgoz, Simon Hadfield, Oscar Koller, Her- mann Ney, and Richard Bowden. Neural sign language trans- lation. In CVPR, pages 7784–7793, 2018. 1, 2, 6

    work page 2018

  5. [4]

    Sign language transformers: Joint end- to-end sign language recognition and translation

    Necati Cihan Camgoz, Oscar Koller, Simon Hadfield, and Richard Bowden. Sign language transformers: Joint end- to-end sign language recognition and translation. In CVPR, pages 10023–10033, 2020. 1, 2, 6

    work page 2020

  6. [5]

    How2sign: A large-scale multimodal dataset for continuous american sign language

    Amanda Cardoso Duarte, Shruti Palaskar, Lucas Ven- tura Ripol, Deepti Ghadiyaram, Kenneth DeHaan, Flo- rian Metze, Jordi Torres Vi ˜nals, and Xavier Gir ´o Nieto. How2sign: A large-scale multimodal dataset for continuous american sign language. In CVPR, pages 2734–2743. Insti- tute of Electrical and Electronics Engineers (IEEE), 2021. 1

    work page 2021

  7. [6]

    A simple multi-modality transfer learning base- line for sign language translation

    Yutong Chen, Fangyun Wei, Xiao Sun, Zhirong Wu, and Stephen Lin. A simple multi-modality transfer learning base- line for sign language translation. In CVPR, pages 5120– 5130, 2022. 1, 2, 6

    work page 2022

  8. [7]

    Two-stream network for sign language recognition and translation

    Yutong Chen, Ronglai Zuo, Fangyun Wei, Yu Wu, Shu- jie Liu, and Brian Mak. Two-stream network for sign language recognition and translation. NeurIPS, 35:17043– 17056, 2022. 2, 6

    work page 2022

Show all 48 references
  1. [8]

    Factorized learn- ing assisted with large language model for gloss-free sign language translation

    Zhigang Chen, Benjia Zhou, Jun Li, Jun Wan, Zhen Lei, Ning Jiang, Quan Lu, and Guoqing Zhao. Factorized learn- ing assisted with large language model for gloss-free sign language translation. In Proceedings of the 2024 Joint In- ternational Conference on Computational Linguist...

    2024

  2. [9]

    Cico: Domain-aware sign language re- trieval via cross-lingual contrastive learning

    Yiting Cheng, Fangyun Wei, Jianmin Bao, Dong Chen, and Wenqiang Zhang. Cico: Domain-aware sign language re- trieval via cross-lingual contrastive learning. InCVPR, pages 19016–19026, 2023. 5

    2023

  3. [10]

    Bert: Pre-training of deep bidirectional trans- formers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional trans- formers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the As- sociation for Computational Linguistics: Human L...

    2019

  4. [11]

    Geo-sign: Hyperbolic contrastive regularisation for geometrically aware sign lan- guage translation

    Edward Fish and Richard Bowden. Geo-sign: Hyperbolic contrastive regularisation for geometrically aware sign lan- guage translation. arXiv preprint arXiv:2506.00129 , 2025. 2

    2025

  5. [12]

    Llms are good sign language translators

    Jia Gong, Lin Geng Foo, Yixuan He, Hossein Rahmani, and Jun Liu. Llms are good sign language translators. In CVPR, pages 18362–18372, 2024. 1, 2, 3, 6, 7

    2024

  6. [13]

    Noise-contrastive estimation: A new estimation principle for unnormalized statistical models

    Michael Gutmann and Aapo Hyv ¨arinen. Noise-contrastive estimation: A new estimation principle for unnormalized statistical models. In Proceedings of the Thirteenth Inter- national Conference on Artificial Intelligence and Statistics, pages 297–304. JMLR Workshop and Conferenc...

    2010

  7. [14]

    Hand-model- aware sign language recognition

    Hezhen Hu, Wengang Zhou, and Houqiang Li. Hand-model- aware sign language recognition. InAAAI, pages 1558–1566,

  8. [15]

    Contin- uous sign language recognition with correlation network

    Lianyu Hu, Liqing Gao, Zekang Liu, and Wei Feng. Contin- uous sign language recognition with correlation network. In CVPR, pages 2529–2539, 2023. 2

    2023

  9. [16]

    Improving 3d pose estimation for sign language

    Maksym Ivashechkin, Oscar Mendez, and Richard Bowden. Improving 3d pose estimation for sign language. In ICASSP, pages 1–5. IEEE, 2023. 3

    2023

  10. [17]

    Skeleton aware multi-modal sign language recognition

    Songyao Jiang, Bin Sun, Lichen Wang, Yue Bai, Kunpeng Li, and Yun Fu. Skeleton aware multi-modal sign language recognition. In CVPR, pages 3413–3423, 2021. 2

    2021

  11. [18]

    Oscar Koller, Jens Forster, and Hermann Ney. Continuous sign language recognition: Towards large vocabulary statisti- cal recognition systems handling multiple signers.Computer Vision and Image Understanding, 141:108–125, 2015. 2

    2015

  12. [19]

    Weakly supervised learning with multi- stream cnn-lstm-hmms to discover sequential parallelism in sign language videos

    Oscar Koller, Necati Cihan Camgoz, Hermann Ney, and Richard Bowden. Weakly supervised learning with multi- stream cnn-lstm-hmms to discover sequential parallelism in sign language videos. IEEE TPAMI, 42(9):2306–2320, 2019. 1

    2019

  13. [20]

    Tspnet: Hier- archical feature learning via temporal semantic pyramid for sign language translation

    Dongxu Li, Chenchen Xu, Xin Yu, Kaihao Zhang, Benjamin Swift, Hanna Suominen, and Hongdong Li. Tspnet: Hier- archical feature learning via temporal semantic pyramid for sign language translation. NeurIPS, 33:12034–12045, 2020. 6

    2020

  14. [21]

    Transferring cross-domain knowledge for video sign language recognition

    Dongxu Li, Xin Yu, Chenchen Xu, Lars Petersson, and Hongdong Li. Transferring cross-domain knowledge for video sign language recognition. In CVPR, pages 6205– 6214, 2020. 2

    2020

  15. [22]

    Uni-sign: Toward uni- fied sign language understanding at scale

    Zecheng Li, Wengang Zhou, Weichao Zhao, Kepeng Wu, Hezhen Hu, and Houqiang Li. Uni-sign: Toward uni- fied sign language understanding at scale. arXiv preprint arXiv:2501.15187, 2025. 2, 3

    2025 arXiv

  16. [23]

    One grammar or two? sign languages and the nature of human language

    Diane C Lillo-Martin and Jon Gajewski. One grammar or two? sign languages and the nature of human language. Wiley Interdisciplinary Reviews. Cognitive Science , 5(4): 387–401, 2014. 1

    2014

  17. [24]

    Rouge: A package for automatic evaluation of summaries

    Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. In Text Summarization Branches Out, pages 74–81, 2004. 6

    2004

  18. [26]

    Soft: Softmax-free transformer with linear complexity

    Jiachen Lu, Jinghan Yao, Junge Zhang, Xiatian Zhu, Hang Xu, Weiguo Gao, Chunjing Xu, Tao Xiang, and Li Zhang. Soft: Softmax-free transformer with linear complexity. NeurIPS, 34:21297–21309, 2021. 2

    2021

  19. [27]

    Visual alignment constraint for continuous sign language recognition

    Yuecong Min, Aiming Hao, Xiujuan Chai, and Xilin Chen. Visual alignment constraint for continuous sign language recognition. In ICCV, pages 11542–11551, 2021. 2

    2021

  20. [28]

    Stochastic fine-grained labeling of multi-state sign glosses for continuous sign language recog- nition

    Zhe Niu and Brian Mak. Stochastic fine-grained labeling of multi-state sign glosses for continuous sign language recog- nition. In ECCV, pages 172–186. Springer, 2020. 2

    2020

  21. [29]

    Bleu: A method for automatic evaluation of machine translation

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: A method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics , pages 311–318,

  22. [30]

    Reconstruct- ing hands in 3d with transformers

    Georgios Pavlakos, Dandan Shan, Ilija Radosavovic, Angjoo Kanazawa, David Fouhey, and Jitendra Malik. Reconstruct- ing hands in 3d with transformers. In CVPR, pages 9826– 9836, 2024. 3

    2024

  23. [31]

    Elements of the language of signs

    Harvey P Peet. Elements of the language of signs. American Annals of the Deaf and Dumb, 5(2):83–95, 1853. 1

  24. [32]

    Can’t you see the difference? sources of variation in sign language structure

    Pamela Perniss, Roland Pfau, and Markus Steinbach. Can’t you see the difference? sources of variation in sign language structure. Emotion, pages 1–34, 2007. 1

    2007

  25. [33]

    Iterative align- ment network for continuous sign language recognition

    Junfu Pu, Wengang Zhou, and Houqiang Li. Iterative align- ment network for continuous sign language recognition. In CVPR, pages 4165–4174, 2019. 2

    2019

  26. [34]

    Learn- ing transferable visual models from natural language super- vision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. In ICML, pages 8748–8763. PMLR, 2021. 2, 3, 5

    2021

  27. [35]

    Mul- tilingual translation with extensible multilingual pretraining and finetuning

    Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, and Angela Fan. Mul- tilingual translation with extensible multilingual pretraining and finetuning. 2020. 3

    2020

  28. [36]

    Youtube-sl-25: A large- scale, open-domain multilingual sign language parallel cor- pus

    Garrett Tanzer and Biao Zhang. Youtube-sl-25: A large- scale, open-domain multilingual sign language parallel cor- pus. arXiv preprint arXiv:2407.11144, 2024. 3

    2024 arXiv

  29. [37]

    Sign2gpt: Leveraging large language models for gloss-free sign language translation

    Ryan Wong, Necati Cihan Camgoz, and Richard Bowden. Sign2gpt: Leveraging large language models for gloss-free sign language translation. In ICLR. 1, 2, 3, 6, 7

  30. [38]

    Learnt contrastive concept embeddings for sign recognition

    Ryan Wong, Necati Cihan Camgoz, and Richard Bowden. Learnt contrastive concept embeddings for sign recognition. In ICCV, pages 1945–1954, 2023. 2

    1945

  31. [39]

    Improving gloss-free sign language translation by reducing representation density

    Jinhui Ye, Xing Wang, Wenxiang Jiao, Junwei Liang, and Hui Xiong. Improving gloss-free sign language translation by reducing representation density. In NeurIPS, 2024. 2, 3

    2024

  32. [40]

    Gloss attention for gloss-free sign language translation

    Aoxiong Yin, Tianyun Zhong, Li Tang, Weike Jin, Tao Jin, and Zhou Zhao. Gloss attention for gloss-free sign language translation. In CVPR, pages 2551–2562, 2023. 3, 6

    2023

  33. [41]

    Better sign language translation with STMC-transformer

    Kayo Yin and Jesse Read. Better sign language translation with STMC-transformer. In Proceedings of the 28th Inter- national Conference on Computational Linguistics , pages 5975–5989, Barcelona, Spain (Online), 2020. International Committee on Computational Linguistics. 1

    2020

  34. [42]

    Sltunet: A simple unified model for sign language translation

    Biao Zhang, Mathias M ¨uller, and Rico Sennrich. Sltunet: A simple unified model for sign language translation. arXiv preprint arXiv:2305.01778, 2023. 1, 6

    2023 arXiv

  35. [43]

    Conditional sentence generation and cross-modal reranking for sign language translation

    Jian Zhao, Weizhen Qi, Wengang Zhou, Nan Duan, Ming Zhou, and Houqiang Li. Conditional sentence generation and cross-modal reranking for sign language translation. IEEE TMM, 24:2662–2672, 2022. 6

    2022

  36. [44]

    Conditional variational autoencoder for sign language translation with cross-modal alignment

    Rui Zhao, Liang Zhang, Biao Fu, Cong Hu, Jinsong Su, and Yidong Chen. Conditional variational autoencoder for sign language translation with cross-modal alignment. In AAAI, pages 19643–19651, 2024. 2, 6

    2024

  37. [45]

    Gloss-free sign language translation: Improving from visual- language pretraining

    Benjia Zhou, Zhigang Chen, Albert Clap ´es, Jun Wan, Yanyan Liang, Sergio Escalera, Zhen Lei, and Du Zhang. Gloss-free sign language translation: Improving from visual- language pretraining. In ICCV, pages 20871–20881, 2023. 1, 2, 3, 5, 6, 7

    2023

  38. [46]

    Improving sign language translation with monolingual data by sign back-translation

    Hao Zhou, Wengang Zhou, Weizhen Qi, Junfu Pu, and Houqiang Li. Improving sign language translation with monolingual data by sign back-translation. In CVPR, pages 1316–1325, 2021. 1, 2, 6

    2021

  39. [47]

    C2slr: Consistency-enhanced continuous sign language recognition

    Ronglai Zuo and Brian Mak. C2slr: Consistency-enhanced continuous sign language recognition. In CVPR, pages 5131–5140, 2022. 2

    2022

  40. [48]

    Natural language-assisted sign language recognition

    Ronglai Zuo, Fangyun Wei, and Brian Mak. Natural language-assisted sign language recognition. In CVPR, pages 14890–14900, 2023. 2

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.