Pith. sign in

REVIEW 4 major objections 4 minor 46 references

Signformer is all you need: Towards Edge AI for Sign Language

T0 review · 4 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read Signformer, a from-scratch transformer with 3.88 million parameters, reports BLEU-4 23.43 on the PHOENIX14T gloss-free sign language translation benchmark, matching the 7-billion-parameter SignLLM at 23.40.

desk verdict The paper's headline result—23.43 BLEU with 3.88M parameters, '2nd place' and '1807x smaller' than SignLLM—is a test-selected maximum from an unreported configuration sweep, so the central claim is not established, though the from-scratch efficiency angle is worth a serious look. read the letter →

arxiv 2411.12901 v1 pith:PYZCUIT5 submitted 2024-11-19 cs.CL cs.CVcs.CYcs.HCcs.LG

classification cs.CLcs.CVcs.CYcs.HCcs.LG
keywords signlanguagetranslationgloss-freefrom-scratchtransformerparameterefficiencypositionencodingvideo-to-textPHOENIX14TedgeAI
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

Signformer seeks to prove that gloss-free sign language translation does not need billion-parameter pretrained language models. The paper introduces a compact transformer trained from scratch on the PHOENIX14T German corpus and reports that its 3.88-million-parameter Full configuration reaches BLEU-4 23.43, slightly above the 7-billion-parameter SignLLM at 23.40 while being about 1800 times smaller. Its 0.57-million-parameter Feather configuration still scores BLEU-4 20.02, above most existing gloss-free methods. If these numbers hold, the result makes deployable, low-cost sign-to-text translation plausible and reframes the field's recent turn to LLM backbones and extra pretraining as one possible route rather than the necessary one.

What carries the argument

The load-bearing mechanism is the Signformer encoder-decoder stack: a reworked 1D convolution module (pointwise-depthwise-pointwise with a single ReLU6, inside a symmetric layer-normalization flow) that extracts local temporal gesture features; GASLT-style gloss attention that encodes gloss-like units in latent space without explicit gloss labels; absolute position encoding, chosen because German word order is rigid; and contextual position encoding (CoPE) applied in cross-attention to make positions sensitive to local, non-manual cues such as facial expressions. The claim is that this combination, trained with raw spatial and word embeddings, produces the reported parameter-efficiency and BLEU-4 gains.

What would settle it

Recompute BLEU-4 for Signformer-Full and SignLLM with one shared evaluation script (same tokenization, beam size, and detokenization); if the 0.03-point gap reverses or widens materially, the claimed parity and 2nd-place rank collapse. A simpler check is to rerun Signformer-Full several times and see whether 23.43 is the typical score rather than the best of a few runs, since the paper reports only configurations that improve performance.

Watch

Extended reading notes

Core claim

The paper's central claim is that a from-scratch, gloss-free sign language translation architecture can reach near the top of the 2024 gloss-free leaderboard without any pretrained model, extra dataset, gloss annotation, or NLP transfer trick. Signformer-Full reports BLEU-4 23.43 on PHOENIX14T with 3.88M parameters, placing second and edging out SignLLM's 23.40 with 7012M parameters; Signformer-Feather reports BLEU-4 20.02 with 0.57M parameters. The authors attribute this to a direct video-to-text transformer whose convolutional module, gloss attention, and dual position-encoding scheme are designed from an analysis of sign-language structure, and they treat the result as establishing an Edge-AI baseline for sign language translation.

Load-bearing premise

The comparison assumes that the parameter counts and BLEU scores for every competing method, some obtained by email from other authors and some estimated, are measured on the same terms as Signformer's; if they are not, the '2nd place' and '1807x smaller' claims lose their basis.

Editorial extensions

If this is right

  • Gloss-free sign language translation can be tackled as an architecture problem rather than a scaling problem; LLM backbones and CLIP-style pretraining are not required for competitive BLEU.
  • A model under 4 million parameters can sit at or near the top of the gloss-free leaderboard, making CPU-only or mobile deployment realistic.
  • Language-aware choice of position encoding matters: on a rigid-word-order target like German, absolute position encoding outperforms relative and contextual encodings alone.
  • Adding CoPE helps most in small hidden-size configurations and hurts above size 128, so the efficiency gain is specific to compact models.
  • Signformer-Feather at 0.57M parameters gives a concrete lightweight baseline that future small-model research can compare against.

Reading between the lines

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

  • A direct corollary the paper does not spell out: if a 3.88M-parameter decoder matches a 7B LLM backbone, the LLM's contribution to this benchmark may be largely replaceable; feeding SignLLM's visual tokens into a small decoder would quantify how much of its BLEU comes from the backbone.
  • The paper's linguistic argument makes a cross-lingual prediction: on a sign language with flexible word order, such as Japanese Sign Language, CoPE should contribute relatively more and APE relatively less; running the same architecture on a second corpus when available would test this.
  • The reported non-monotonicity of CoPE (helpful below hidden size 128, harmful above) suggests a testable general hypothesis about contextual position encodings in video-to-text tasks, not just a tuning detail for this model.
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

4 major / 4 minor

Summary. The paper proposes Signformer, a from-scratch transformer architecture for gloss-free sign language translation, combining a convolutional module, gloss attention, and contextual position encoding (CoPE). It reports that Signformer-Full (3.88M parameters) achieves BLEU-4 23.43 on the PHOENIX14T test set, which it claims is a new 2nd place on the gloss-free leaderboard, approximately matching SignLLM (7B parameters, BLEU-4 23.40) with 1807x fewer parameters. The paper also presents linguistic arguments about German word order to justify the choice of absolute position encoding, and reports a lighter 0.57M-parameter configuration (Signformer-Feather) that outperforms several existing methods.

Significance. If the reported results were robustly established, the paper would make a valuable contribution to efficient sign language translation by demonstrating that a compact from-scratch model can approach the performance of billion-parameter pretrained systems. The emphasis on edge-AI deployment and the analysis of linguistic structure are worthwhile research directions. However, the current evidence is not convincing: the central numbers appear to be selected from an unreported configuration sweep, no error bars or multiple-seed results are provided, the optimizer is changed across configurations, and no code or project link is given. The manuscript therefore does not currently justify its headline claims.

major comments (4)
  1. [§4.1 and Table 4] The paper explicitly states, "We neglect configurations worsening performance as that's trivial to present," and in §4.1 that "we chose to feature configurations offering tangible benefits only." This indicates that the reported Test BLEU-4 of 23.43 for Signformer-Full is the maximum of an unreported configuration sweep, not the performance of a single pre-specified model. No error bars, standard deviations, or multiple-seed results are given, so the 0.03-BLEU margin over SignLLM (23.43 vs. 23.40) is within typical run-to-run noise. Moreover, the Dev BLEU-4 of 21.92 versus Test BLEU-4 of 23.43 for this configuration shows a 1.51-point test-over-dev gap, which is atypical and consistent with selecting the configuration on the test set. The central claims of "2nd place" and "1807x smaller" are therefore not established.
  2. [§4.1 and Tables 1–4] The optimizer is not held fixed: §4.1 states that SophiaG was used for Mid and Full configurations with hidden sizes above 128, while AdamW was adopted for the Feather models with hidden size 64, and that optimizers were chosen to "acquire the best configuration results." This confounds all ablation comparisons: the effects attributed to CoPE, hidden size, or the convolution module could instead be due to the optimizer change. Since no seeds or variance estimates are reported, none of the ablation tables provide a controlled comparison that supports the paper's conclusions about the architecture components.
  3. [§3.3 and Table 2] The position-encoding ablation is not a clean test of the linguistic hypothesis. Table 2's experiments for RPE and CoPE were run with the Multihead Attention module because "Gloss Attention doesn't support RPE," while the final architecture uses Gloss Attention. Thus the comparison conflates the attention mechanism with the position encoding, and the claimed superiority of APE may be an artifact of this mismatch. Additionally, the linguistic rationale (German's strict word order favoring APE) is presented as a post hoc explanation after the empirical results are known; the paper provides no pre-registered or predictive test of this hypothesis, so the "nature analysis" does not independently support the design choice.
  4. [§5.1 and Table 5] The headline comparison against SignLLM rests on a 0.03-BLEU difference (23.43 vs. 23.40), which is smaller than typical run-to-run variance for neural translation models, and the paper provides no confidence intervals or repeated runs. Furthermore, the parameter counts for competing methods are partly estimates or obtained through personal communication, as the table note indicates, and the methods are compared across different training protocols and resource budgets. Without variance estimates and a clearly defined model-selection procedure, the claims of approximate equivalence to SignLLM and of a 1807x parameter reduction are not statistically supported.
minor comments (4)
  1. [Throughout] The manuscript contains numerous typos and grammatical errors, e.g., "sophiscated" (Abstract), "metricuously" (Contributions), "paradign" (§2.3), "applicaiton" (§2.1), "combinded" (§3.2), "syntatic" (§4.2), "leadeboard" (§2.3), and "ture" (§4). A thorough proofreading pass is needed.
  2. [§5.1, Table 5] The table has inconsistent naming: the method is called "STMC-T" in §2.2 but listed as "SMTC-T" in Table 5; the table note also references "SMTC-T" while the text says "STMC-T." The relationship between the confirmed 46.4M parameter count for STMC and the reported 192* for the ensembled STMC-T should be clarified.
  3. [Abstract and §1] The phrase "We release official code and project page available here" is not accompanied by an actual link or repository identifier anywhere in the manuscript. For a reproducibility-focused claim, a working URL or a note explaining the lack of release is necessary.
  4. [§4 and §5] The paper states that CSL-Daily was not included due to access restrictions, which is an understandable limitation, but the claim that "we have ensured both datasets are representative and consistent" is unsupported because no experiments on CSL-Daily are presented. The generalizability claim should be toned down or substantiated with results on at least one additional dataset.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: Signformer's reported BLEU scores are external benchmark measurements, and the architectural ablations are empirical comparisons rather than derivations whose conclusions are assumed by their inputs.

full rationale

The paper's central claims—Signformer-Full scoring BLEU-4 23.43 on PHOENIX14T and being 1807x smaller than SignLLM—are empirical measurements against an external benchmark, not derivations from assumptions that entail the results. The ablations in Tables 2–4 compare alternative position encodings, convolution modules, and hidden sizes on the same benchmark; the position-encoding discussion in Section 3.3 is a linguistic motivation, not a definitional equivalence between the analysis and the outcome. There are no self-citations or imported uniqueness theorems: the cited prior work (SL-Transformer, GASLT, Conformer, CoPE) is external, and the paper explicitly reports scratch training without pretrained weights. The selection note in Table 4 ('We neglect configurations worsening performance as that's trivial to present') and the absence of error bars are legitimate methodological limitations that affect reliability of the 2nd-place and margin claims, but they do not make the predictions equivalent to the model's inputs by construction. The position-encoding rationale may be post hoc, but a post hoc explanation is not circularity under the stated criteria because it does not reduce the experimental result to the theory's own assumptions. Accordingly, no circular step is identified.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claim rests on benchmark comparisons and architectural choices that are not independently verified. The configuration selection is effectively fitting to the test set, and the parameter counts of competitors are obtained informally.

free parameters (1)
  • Configuration selection = hidden size 64 to 256, CoPE on/off, optimizer choice
    The paper reports only configurations with tangible gains, effectively selecting settings based on test BLEU, which constitutes model fitting to the test set.
assumptions (4)
  • domain assumption PHOENIX14T is a valid benchmark for gloss-free SLT and the reported parameter counts of other methods are accurate as obtained via email.
    Section 5.1 states parameter info is acquired from authors through emails, which is not independently verified.
  • domain assumption GASLT's deformable attention mechanism is an effective component for gloss-free SLT.
    Section 3.1 relies on GASLT's attention without independent justification.
  • ad hoc to paper German's strict word order makes APE preferable to RPE, and CoPE adds useful locality.
    Section 3.3 presents a qualitative linguistic argument without quantitative evidence; it is used to justify post hoc choices based on empirical results.
  • domain assumption The reported BLEU scores are computed with the same evaluation protocol across all compared methods.
    The paper compares against published numbers but doesn't verify the evaluation setup.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Signformer is all you need: Towards Edge AI for Sign Language." pith.science (2026). https://pith.science/paper/PYZCUIT5

@misc{pith2026241112901,
  author       = {Pith},
  title        = {Pith review of: Signformer is all you need: Towards Edge AI for Sign Language},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PYZCUIT5}},
  note         = {Machine review of arXiv:2411.12901}
}
read the original abstract

Sign language translation, especially in gloss-free paradigm, is confronting a dilemma of impracticality and unsustainability due to growing resource-intensive methodologies. Contemporary state-of-the-arts (SOTAs) have significantly hinged on pretrained sophiscated backbones such as Large Language Models (LLMs), embedding sources, or extensive datasets, inducing considerable parametric and computational inefficiency for sustainable use in real-world scenario. Despite their success, following this research direction undermines the overarching mission of this domain to create substantial value to bridge hard-hearing and common populations. Committing to the prevailing trend of LLM and Natural Language Processing (NLP) studies, we pursue a profound essential change in architecture to achieve ground-up improvements without external aid from pretrained models, prior knowledge transfer, or any NLP strategies considered not-from-scratch. Introducing Signformer, a from-scratch Feather-Giant transforming the area towards Edge AI that redefines extremities of performance and efficiency with LLM-competence and edgy-deployable compactness. In this paper, we present nature analysis of sign languages to inform our algorithmic design and deliver a scalable transformer pipeline with convolution and attention novelty. We achieve new 2nd place on leaderboard with a parametric reduction of 467-1807x against the finests as of 2024 and outcompete almost every other methods in a lighter configuration of 0.57 million parameters.

Figures

Figures reproduced from arXiv: 2411.12901 by the authors.

Figure 1
Figure 1. (TOP): Gloss-Free TOP5 Leaderboard 2024; (Down): Sign Language Translation (Gloss-Based & Gloss-Free) TOP5 Leaderboard 2024. Our Signformers, using 3M parameters, achieve the 2nd place on against finests ranged from 1-7B parame￾ters, and exhibit approximate performance to most SLT gloss-based approaches, while considerably advancing in efficiency, Informa￾tion Density, and NetScore [39] 1 arXiv:2411.12901v1 [cs.CL] … view at source ↗
Figure 2
Figure 2. Architecture of Signformer, composed of a convolutional module, CoPE-Gloss Attention, and CoPE-Cross Attention. Model is built and trained from-scratch without external embedding or pretrained source, utilizing raw spatial and word embedding layers. 2.3. Gloss-Free Sign Language Translation Gloss-Free paradign has appeared the most important shift as ultimate video2text modelling enables more authentic linguistic an… view at source ↗
Figure 3
Figure 3. Convolution Module, composed of a stacked Pointwise-Depthwise-Pointwise 1D convolution encapsulated within a symmetric LayerNormalization flow, designed to incorporate with gloss attention to extract higher-level visual features. layer is used. Although it is acknowledged that CLIP, VGG, and other large, pretrained multimodal ViT-based [10] models offer superior embedding performance, we insist on inheriting origina… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 37 canonical work pages

  1. [1]

    Word error rate estimation for speech recognition: e-WER

    Ahmed Ali and Steve Renals. Word error rate estimation for speech recognition: e-WER. In Proceedings of the 56th Annual Meeting of the Association for Computational Lin- guistics (Volume 2: Short Papers), pages 20–24, Melbourne, Australia, 2018. Association for Computational Linguistics. 2

  2. [2]

    A colloquial corpus of Japanese Sign Language: Linguistic resources for observing sign language conversa- tions

    Mayumi Bono, Kouhei Kikuchi, Paul Cibulka, and Yutaka Osugi. A colloquial corpus of Japanese Sign Language: Linguistic resources for observing sign language conversa- tions. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14) , pages 1898–1904, Reykjavik, Iceland, 2014. European Language Resources Associatio...

  3. [3]

    The clausal syntax of German Sign Language: A cartographic approach (Volume 5)

    Fabian Bross. The clausal syntax of German Sign Language: A cartographic approach (Volume 5). Language Science Press,

  4. [4]

    Neural sign language trans- lation

    Necati Cihan Camgoz, Simon Hadfield, Oscar Koller, Her- mann Ney, and Richard Bowden. Neural sign language trans- lation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7784–7793, 2018. 2, 7

  5. [5]

    Sign language transformers: Joint end-to- end sign language recognition and translation, 2020

    Necati Cihan Camgoz, Oscar Koller, Simon Hadfield, and Richard Bowden. Sign language transformers: Joint end-to- end sign language recognition and translation, 2020. 2, 3, 7

  6. [6]

    A simple multi-modality transfer learning base- line for sign language translation, 2023

    Yutong Chen, Fangyun Wei, Xiao Sun, Zhirong Wu, and Stephen Lin. A simple multi-modality transfer learning base- line for sign language translation, 2023. 2, 7

  7. [7]

    Fully Convolutional Networks for Continuous Sign Language Recognition, page 697–714

    Ka Leong Cheng, Zhaoyang Yang, Qifeng Chen, and Yu- Wing Tai. Fully Convolutional Networks for Continuous Sign Language Recognition, page 697–714. Springer International Publishing, 2020. 2

  8. [8]

    Acquiring a first language in adolescence: The case of basic word order in american sign language

    Qi Cheng and Rachel I Mayberry. Acquiring a first language in adolescence: The case of basic word order in american sign language. Journal of child language, 46(2):214–240, 2019. 5

Show all 46 references
  1. [9]

    Le, and Ruslan Salakhutdinov

    Zihang Dai, Zhilin Yang, Yiming Yang, Jaime Carbonell, Quoc V . Le, and Ruslan Salakhutdinov. Transformer-xl: At- tentive language models beyond a fixed-length context, 2019. 5

  2. [10]

    An image is worth 16x16 words: Transformers for image recognition at scale, 2021

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

  3. [11]

    Matthew S. Dryer. Order of subject, object and verb (v2020.3). In The World Atlas of Language Structures Online. Zenodo,

  4. [12]

    Piater, and Hermann Ney

    Jens Forster, Christoph Andreas Schmidt, Thomas Hoyoux, Oscar Koller, Uwe Zelle, Justus H. Piater, and Hermann Ney. Rwth-phoenix-weather: A large vocabulary sign language recognition and translation corpus. In International Confer- ence on Language Resources and Evaluation, 2012. 5, 6

  5. [13]

    Contextual position encoding: Learning to count what’s important, 2024

    Olga Golovneva, Tianlu Wang, Jason Weston, and Sainbayar Sukhbaatar. Contextual position encoding: Learning to count what’s important, 2024. 5

  6. [14]

    Llms are good sign language translators, 2024

    Jia Gong, Lin Geng Foo, Yixuan He, Hossein Rahmani, and Jun Liu. Llms are good sign language translators, 2024. 2, 3, 7

  7. [15]

    Multi-stream keypoint attention network for sign language recognition and translation

    Mo Guan, Yan Wang, Guangkun Ma, Jiarui Liu, and Mingzu Sun. Multi-stream keypoint attention network for sign language recognition and translation. arXiv preprint arXiv:2405.05672, 2024. 7

  8. [16]

    Conformer: Convolution-augmented transformer for speech recognition,

    Anmol Gulati, James Qin, Chung-Cheng Chiu, Niki Par- mar, Yu Zhang, Jiahui Yu, Wei Han, Shibo Wang, Zheng- dong Zhang, Yonghui Wu, and Ruoming Pang. Conformer: Convolution-augmented transformer for speech recognition,

  9. [17]

    Bpemb: Tokenization-free pre-trained subword embeddings in 275 languages, 2017

    Benjamin Heinzerling and Michael Strube. Bpemb: Tokenization-free pre-trained subword embeddings in 275 languages, 2017. 4

  10. [18]

    The kinetics human action video dataset, 2017

    Will Kay, Joao Carreira, Karen Simonyan, Brian Zhang, Chloe Hillier, Sudheendra Vijayanarasimhan, Fabio Viola, Tim Green, Trevor Back, Paul Natsev, Mustafa Suleyman, and Andrew Zisserman. The kinetics human action video dataset, 2017. 2

  11. [19]

    Word-level deep sign language recognition from video: A new large-scale dataset and methods comparison

    Dongxu Li, Cristian Rodriguez, Xin Yu, and Hongdong Li. Word-level deep sign language recognition from video: A new large-scale dataset and methods comparison. In The IEEE Winter Conference on Applications of Computer Vision, pages 1459–1469, 2020. 2

  12. [20]

    Tspnet: Hierarchical fea- ture learning via temporal semantic pyramid for sign language translation, 2020

    Dongxu Li, Chenchen Xu, Xin Yu, Kaihao Zhang, Ben Swift, Hanna Suominen, and Hongdong Li. Tspnet: Hierarchical fea- ture learning via temporal semantic pyramid for sign language translation, 2020. 7, 8

  13. [21]

    How language-neutral is multilingual bert?, 2019

    Jindˇrich Libovick´y, Rudolf Rosa, and Alexander Fraser. How language-neutral is multilingual bert?, 2019. 2

  14. [22]

    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

  15. [23]

    Few-shot learning with multi- lingual language models, 2022

    Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O’Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mon...

  16. [24]

    Sophia: A scalable stochastic second-order optimizer for language model pre-training, 2024

    Hong Liu, Zhiyuan Li, David Hall, Percy Liang, and Tengyu Ma. Sophia: A scalable stochastic second-order optimizer for language model pre-training, 2024. 6

  17. [25]

    Multilingual denoising pre-training for neural machine translation, 2020

    Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, and Luke Zettle- moyer. Multilingual denoising pre-training for neural machine translation, 2020. 2

  18. [26]

    Decoupled weight decay regularization, 2019

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization, 2019. 6

  19. [27]

    Shortgpt: Layers in large language models are more redun- dant than you expect, 2024

    Xin Men, Mingyu Xu, Qingyu Zhang, Bingning Wang, Hongyu Lin, Yaojie Lu, Xianpei Han, and Weipeng Chen. Shortgpt: Layers in large language models are more redun- dant than you expect, 2024. 4

  20. [28]

    Neural sign language trans- lation by learning tokenization

    Alptekin Orbay and Lale Akarun. Neural sign language trans- lation by learning tokenization. In 2020 15th IEEE interna- tional conference on automatic face and gesture recognition (FG 2020), pages 222–228. IEEE, 2020. 7 9

  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]

    Learning transferable visual models from natural language supervision, 2021

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision, 2021. 2

  23. [31]

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified text-to-text transformer, 2023. 2

  24. [32]

    Glu variants improve transformer, 2020

    Noam Shazeer. Glu variants improve transformer, 2020. 4

  25. [33]

    Efficient attention: Attention with linear com- plexities, 2024

    Zhuoran Shen, Mingyuan Zhang, Haiyu Zhao, Shuai Yi, and Hongsheng Li. Efficient attention: Attention with linear com- plexities, 2024. 4

  26. [34]

    Very deep con- volutional networks for large-scale image recognition, 2015

    Karen Simonyan and Andrew Zisserman. Very deep con- volutional networks for large-scale image recognition, 2015. 2

  27. [35]

    Grammaticalization of auxiliaries in sign languages

    Markus Steinbach, Roland Pfau, et al. Grammaticalization of auxiliaries in sign languages. TRENDS IN LINGUISTICS STUDIES AND MONOGRAPHS, 188:303, 2007. 5

  28. [36]

    Llama: Open and efficient foundation language mod- els, 2023

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Mar- tinet, Marie-Anne Lachaux, Timoth ´ee Lacroix, Baptiste Rozi`ere, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. Llama: Open and efficient foundation lan...

  29. [37]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need, 2023. 5

  30. [38]

    Stochas- tic transformer networks with linear competing units: Ap- plication to end-to-end sl translation

    Andreas V oskou, Konstantinos P Panousis, Dimitrios Kos- mopoulos, Dimitris N Metaxas, and Sotirios Chatzis. Stochas- tic transformer networks with linear competing units: Ap- plication to end-to-end sl translation. In Proceedings of the IEEE/CVF International Conference on Co...

  31. [39]

    Netscore: Towards universal metrics for large-scale performance analysis of deep neural networks for practical on-device edge usage, 2018

    Alexander Wong. Netscore: Towards universal metrics for large-scale performance analysis of deep neural networks for practical on-device edge usage, 2018. 1, 8

  32. [40]

    Sign2gpt: Leveraging large language models for gloss-free sign language translation, 2024

    Ryan Wong, Necati Cihan Camgoz, and Richard Bowden. Sign2gpt: Leveraging large language models for gloss-free sign language translation, 2024. 2, 7

  33. [41]

    Gloss attention for gloss-free sign language translation, 2023

    Aoxiong Yin, Tianyun Zhong, Li Tang, Weike Jin, Tao Jin, and Zhou Zhao. Gloss attention for gloss-free sign language translation, 2023. 3, 4, 7

  34. [42]

    Better sign language translation with stmc-transformer, 2020

    Kayo Yin and Jesse Read. Better sign language translation with stmc-transformer, 2020. 2, 7

  35. [43]

    Gloss-free sign language translation: Improving from visual-language pretraining, 2023

    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, 2023. 3, 7

  36. [44]

    Spatial-temporal multi-cue network for continuous sign lan- guage recognition, 2020

    Hao Zhou, Wengang Zhou, Yun Zhou, and Houqiang Li. Spatial-temporal multi-cue network for continuous sign lan- guage recognition, 2020. 2

  37. [45]

    Improving sign language translation with mono- lingual data by sign back-translation, 2021

    Hao Zhou, Wengang Zhou, Weizhen Qi, Junfu Pu, and Houqiang Li. Improving sign language translation with mono- lingual data by sign back-translation, 2021. 7

  38. [46]

    A survey on efficient training of transformers, 2023

    Bohan Zhuang, Jing Liu, Zizheng Pan, Haoyu He, Yuetian Weng, and Chunhua Shen. A survey on efficient training of transformers, 2023. 4 10

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.