Pith. sign in

REVIEW 4 major objections 6 minor 41 references

From No to Know: Taxonomy, Challenges, and Opportunities for Negation Understanding in Multimodal Foundation Models

T0 review · 4 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read The paper claims that negation, in its many syntactic, morphological, lexical, and prosodic forms, is systematically mishandled by multilingual multimodal foundation models, and that a taxonomy of these forms plus dedicated benchmarks is…

desk verdict A useful agenda-setting perspective on negation in multimodal models, but the taxonomy's categories blur together and the evidence is anecdotal; worth a serious referee with major revisions. read the letter →

arxiv 2502.09645 v1 pith:3HT4RL7P submitted 2025-02-10 cs.CL cs.AI

Mayank Vatsa , Aparna Bharati , Surbhi Mittal , Richa Singh This is my paper · ORCID
classification cs.CLcs.AI
keywords negationmultimodalfoundationmodelsvision-languagetaxonomymultilingualtext-to-imagegenerationbenchmarksaudioandvideounderstanding
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

Negation is everywhere in real instructions — "no ears," "don't use red," "there is no water" — and this paper argues that state-of-the-art multimodal models regularly get it wrong, especially across languages. The authors propose a taxonomy of negation that spans syntactic, morphological, lexical and semantic, and prosodic/paralinguistic/pragmatic forms, with sixteen subtypes, as a shared map for measuring and fixing these failures. They review evidence from text-to-image generation, vision-language retrieval, machine translation, and audio/video AI showing that models often substitute affirmative interpretations or ignore the negating marker entirely. If the taxonomy is adopted, it would let researchers build negation-specific benchmarks, compare models across languages and modalities, and design architectures — language-specific tokenization, fine-grained attention, multimodal fusion — that actually track absence rather than just presence. The contribution is a frame for a research program, not a new model or dataset.

What carries the argument

The carrying mechanism is the proposed taxonomy of negation: four overarching categories (Syntactic, Morphological, Lexical and Semantic, and Prosodic/Paralinguistic/Pragmatic) subdivided into sixteen sub-types, including standard negation, double negation, negative concord, negative copula, tag questions, prohibitives, conditionals, negative existentials, affixal negation, negative prefixes, negative pronouns and determiners, lexical negation, negative polarity items, non-verbal negation, prosodic negation, and dialectal/acoustic variation. It is built on a cross-linguistic typology of negation and is used to organize evidence of model failure, derive research questions, and motivate benchmark design. Its work in the argument is to convert "negation is hard" into a checkable list of phenomena whose presence or absence can be measured in datasets, prompts, and model outputs across languages and modalities.

What would settle it

Run a matched benchmark that presents the same affirmative/negative contrast in each of the sixteen negation subtypes across a dozen languages from different families to current image, text, and video models. If failure rates do not cluster by taxonomy category — for instance, if models fail just as often on simple English "not" as on negative concord or prosodic negation, or if adding negated training samples fixes all categories equally — then the taxonomy's categories do not explain model behavior and the paper's central claim would be weakened. A second falsifier: if a text-only model with no visual or audio input achieves near-perfect negation accuracy on all categories, the paper's implication that multimodal integration is necessary would be undercut.

Watch

Extended reading notes

Core claim

The paper's central claim is that negation is not a single linguistic obstacle but a family of constructs — standard/sentential negation, double negation, negative concord, negative copulas, negation in tag questions, negative imperatives, negation in conditionals, negative existentials, affixal negation, negative prefixes, negative pronouns and determiners, lexical negation, negative polarity items, and non-verbal, prosodic, and dialectal forms — and that each family member stresses multilingual multimodal foundation models differently. The authors assemble these into a four-category taxonomy and argue that current models fail across all of them, with examples such as image generators producing dogs with ears for "a dog with no ears" in English and Hindi alike. They further claim that existing benchmarks under-represent negation, that simply adding negative training samples does not reliably teach semantics, and that progress requires specialized benchmarks plus architectural changes: language-specific tokenization, fine-grained attention to negation scope, and hybrid multimodal fusion that can ingest gestural and prosodic cues. The hoped-for result is models that can distinguish presence from absence and handle "not uncommon," "I ain't got no money," and a head shake in Tamil with equal reliability.

Load-bearing premise

The paper assumes that the taxonomy of negation types taken from one cross-linguistic survey covers the phenomena that actually break multilingual multimodal models, and that the hand-picked examples in its figures are representative of systematic failure rather than cherry-picked outliers.

Editorial extensions

If this is right

  • Negation needs its own benchmarks: generic NLI or captioning sets undercount negated constructs, so a model's reported competence can hide systematic failure on "no," "not," "without," "never," and language-specific markers.
  • Failure modes should be predictable by taxonomy category: double negation and negative concord in Romance and Slavic languages will trip models more than English sentential "not," and prosodic or gestural negation will be nearly invisible to text-only training.
  • Architectural fixes are on the table: language-specific tokenization, fine-grained attention to negation scope, and multimodal fusion of audio and visual cues should improve negation handling more than simply adding negative examples to training data.
  • Real-world systems such as translation, image generation, medical and legal text processing, and conversational AI will keep producing confidently wrong outputs on negated instructions until these benchmarks and architectures are adopted.
  • The taxonomy frames the question of whether a universal negation-handling framework is viable or whether strategies must be tailored to individual languages and modalities.

Reading between the lines

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

  • [Editorial inference] If the taxonomy's categories track genuine model difficulty, then a single evaluation suite could rank languages and modalities by negation robustness, turning the taxonomy into a diagnostic instrument rather than just a description.
  • [Editorial inference] The paper's examples suggest models may be using a heuristic of ignoring negators and generating the most probable scene; if so, adversarial tests that swap negators ("with ears" vs. "without ears") would expose shortcut learning more cleanly than the hand-picked images do.
  • [Editorial inference] Because the taxonomy includes gestural and prosodic negation, it implies that text-only language models cannot fully solve negation understanding no matter how much text they see — a claim that could be tested by comparing text-only versus audio-visual models on matched items.
  • [Editorial inference] A quantitative version of the argument would score models on matched affirmative/negative pairs across the sixteen subtypes; the paper motivates but does not run such a benchmark.
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 / 6 minor

Summary. The manuscript is a perspective paper arguing that negation is a persistent failure mode for multilingual multimodal foundation models. It surveys psycholinguistic, NLP, vision-language, and audio literature, proposes a taxonomy of negation with four macro-categories and sixteen subtypes, presents illustrative text-to-image failure examples in English, Hindi, Spanish, Bangla, and Japanese, poses eleven research questions, and advocates for specialized benchmarks, language-specific tokenization, fine-grained attention mechanisms, and multimodal fusion architectures. The central load-bearing claim is that the proposed taxonomy is comprehensive and usable for building negation-specific benchmarks and model improvements.

Significance. If the taxonomy were sound, the paper would provide a useful shared vocabulary and research agenda for an under-explored problem. It consolidates scattered evidence from psycholinguistics, LLM benchmarks, VLM compositionality, and audio foundation models, and it raises concrete research questions that could guide future work. The figure captions are honest in acknowledging that the examples are not comprehensive and may reflect model biases. However, the contribution is entirely conceptual: there is no systematic evaluation, and the taxonomy's internal consistency must be established before benchmarks can be reliably built on it.

major comments (4)
  1. [Section 2, 'Double Negation' and 'Negative Concord'] The definitions of Double Negation and Negative Concord are not mutually exclusive, and several examples are assigned to the wrong category. The manuscript states that double negation 'reinforces the negative meaning' and that negative concord occurs when 'multiple negative elements... collectively express a single negation'; under these definitions, Spanish 'No vi a nadie' satisfies both. More seriously, French 'Je ne fais rien', Russian 'Ya ne nikogda ne skazal', and AAVE 'I ain't got no money' are standard textbook examples of negative concord, not of a separate reinforcing double negation, yet the first is filed under Double Negation and the latter two are also described there. Because a benchmark item's subtype is underdetermined by the definitions, the taxonomy cannot support the precise subtype-level analyses promised in Section 4.
  2. [Section 2, 'Syntactic Negations'] Several subtypes are contexts or speech acts rather than negation constructions. Negative Imperatives ('Don't go'), Negation in Conditionals ('If you don't come...'), Negative Existentials ('There is no water'), and Negation in Tag Questions are all environments in which standard (sentential) negation can appear; they are not disjoint from Standard Negation. For example, 'Don't go' is standard negation in an imperative, and 'There is no water' is standard negation of an existential construction. A taxonomy that mixes construction types with contexts will not yield stable, non-overlapping labels, so benchmarks built on it will inherit label noise.
  3. [Section 2, opening; Abstract] The 'comprehensive taxonomy' claim is not supported by the evidence offered. The taxonomy is built solely on Miestamo [19] and is not validated against a broad sample of languages, modalities, or existing negation benchmarks. Since the paper's main contribution is the taxonomy, the authors should either add a validation study (e.g., annotating a sample of negated prompts from SugarCrepe, CC-Neg, or This-is-not-a-Dataset and reporting inter-annotator agreement and subtype coverage) or explicitly reframe the taxonomy as a preliminary framework whose coverage is an open question.
  4. [Figures 1 and 3] The empirical claim that models 'consistently produced images of dogs with ears' and 'fail while creating images using negative prompts in different languages' is presented from hand-picked examples without an experimental protocol. No model versions, seeds, number of generations, or quantitative success rates are reported, and the captions themselves concede that images 'may inadvertently reflect biases' and 'do not capture the full complexity.' As a perspective, illustrative examples are acceptable, but the text should not overstate them as systematic evidence; either soften the wording or provide a small controlled evaluation.
minor comments (6)
  1. [Section 2, 'Double Negation'] The text writes 'AA VE' where 'AAVE' (African American Vernacular English) is meant; please fix the spacing.
  2. [Section 2, 'Negative Prefixes'] The German example 'unm¨oglich' appears with a broken diacritic encoding; it should be 'unmöglich'.
  3. [Figure 1 caption and Section 1] Model naming is inconsistent: the figure uses 'Dalle3' while the text uses 'DALL-E 3'; please standardize.
  4. [Section 3, RQ4] RQ4 cites 'not uncommon' as an example of double or nested negation, but under the paper's own taxonomy this is better characterized as lexical negation or litotes; the example should be aligned with the taxonomy.
  5. [Section 2, 'Prosodic, Paralinguistic, and Pragmatic Negation'] The boundary between Non-verbal Negation and Prosodic Negation is unclear, since examples such as Tamil intonational negation involve both prosody and non-verbal cues; the subtype definitions should be clarified.
  6. [General] The manuscript would benefit from a brief limitations paragraph stating that the taxonomy is untested and that the figures are illustrative; this would better align the framing with the evidence provided.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the proposed taxonomy is transparently grounded in an external typological survey and the paper makes no fitted or predicted quantities that reduce to its own inputs.

full rationale

The paper is a perspective piece whose central artifact, the taxonomy in Section 2, is explicitly 'Building upon the cross-linguistic typology of negation [19]' (Miestamo 2007), an external source; the taxonomy is not derived from the paper's own conclusions. There are no equations, fitted parameters, or benchmark results in the paper, so no prediction is statistically forced by construction. The only self-citations are [8] (CC-Neg) and [38] (bias in Indic text-to-image generation), cited in Section 4 as examples of 'initial efforts' in negation research; they are background references and do not carry the argument, close off alternatives, or serve as a uniqueness theorem. Figures 1 and 3 are illustrative demonstrations of model failures rather than predictions generated from the taxonomy. The potential overlap between the Double Negation and Negative Concord subtypes is a taxonomy-precision concern, not circularity, because the subtypes are defined independently and neither is defined in terms of the other to produce a result. In short, no claim in the paper reduces by construction or by self-citation to its own inputs.

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

The central claims rest on the assumed completeness of an external typology, the presumed generality of model failures, and the representativeness of selected examples. No free parameters or invented entities are introduced.

assumptions (3)
  • domain assumption The cross-linguistic typology of negation (Miestamo, 2007) is comprehensive and accurate.
    The paper's taxonomy is built directly on this source (Section 2, 'Building upon the cross-linguistic typology of negation [19]') without independent validation of its coverage.
  • domain assumption Multimodal foundation models exhibit generalized difficulty with negation beyond the specific examples shown.
    The paper treats model failures as a systemic issue, citing prior benchmarks and its own figures, but does not measure failure rates or control for prompt/model variation.
  • domain assumption Negation is under-represented in training corpora.
    This is cited from literature [13,21] and used to motivate the problem, but the paper adds no new data confirming corpus statistics.

how reviews work

0 comments
Cite this review

Pith. "Pith review of From No to Know: Taxonomy, Challenges, and Opportunities for Negation Understanding in Multimodal Foundation Models." pith.science (2026). https://pith.science/paper/3HT4RL7P

@misc{pith2026250209645,
  author       = {Pith},
  title        = {Pith review of: From No to Know: Taxonomy, Challenges, and Opportunities for Negation Understanding in Multimodal Foundation Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3HT4RL7P}},
  note         = {Machine review of arXiv:2502.09645}
}
read the original abstract

Negation, a linguistic construct conveying absence, denial, or contradiction, poses significant challenges for multilingual multimodal foundation models. These models excel in tasks like machine translation, text-guided generation, image captioning, audio interactions, and video processing but often struggle to accurately interpret negation across diverse languages and cultural contexts. In this perspective paper, we propose a comprehensive taxonomy of negation constructs, illustrating how structural, semantic, and cultural factors influence multimodal foundation models. We present open research questions and highlight key challenges, emphasizing the importance of addressing these issues to achieve robust negation handling. Finally, we advocate for specialized benchmarks, language-specific tokenization, fine-grained attention mechanisms, and advanced multimodal architectures. These strategies can foster more adaptable and semantically precise multimodal foundation models, better equipped to navigate and accurately interpret the complexities of negation in multilingual, multimodal environments.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 33 canonical work pages

  1. [19]

    Language and linguistics compass 1(5), 552–570 (2007) 15

    Miestamo, M.: Negation–an overview of typological research. Language and linguistics compass 1(5), 552–570 (2007) 15

    work page 2007

  2. [1]

    In: International Conference on Machine Learning, pp

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from natural language supervision. In: International Conference on Machine Learning, pp. 8748–8763 (2021)

    2021

  3. [2]

    In: International Conference on Machine Learning, pp

    Li, J., Li, D., Xiong, C., Hoi, S.: Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In: International Conference on Machine Learning, pp. 12888–12900 (2022). PMLR

    2022

  4. [3]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Zhai, X., Wang, X., Mustafa, B., Steiner, A., Keysers, D., Kolesnikov, A., Beyer, L.: Lit: Zero-shot transfer with locked-image text tuning. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 18123– 18133 (2022)

    work page 2022

  5. [4]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Zhang, P., Li, X., Hu, X., Yang, J., Zhang, L., Wang, L., Choi, Y., Gao, J.: Vinvl: Revisiting visual representations in vision-language models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 5579–5588 (2021)

    work page 2021

  6. [5]

    Acta Psychologica, 1–7 (2014)

    Khemlani, S., Orenes, I., Johnson-Laird, P.N.: The negations of conjunctions, conditionals, and disjunctions. Acta Psychologica, 1–7 (2014)

    work page 2014

  7. [6]

    In: Proceed- ings of the 12th Joint Conference on Lexical and Computational Semantics, pp

    Truong, T.H., Baldwin, T., Verspoor, K., Cohn, T.: Language models are not naysayers: an analysis of language models on negation benchmarks. In: Proceed- ings of the 12th Joint Conference on Lexical and Computational Semantics, pp. 101–114 (2023)

    work page 2023

  8. [7]

    In: Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track (2023)

    Hsieh, C.-Y., Zhang, J., Ma, Z., Kembhavi, A., Krishna, R.: Sugarcrepe: Fix- ing hackable benchmarks for vision-language compositionality. In: Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track (2023). https://openreview.net/forum?id=Jsc7WSCZd4

    work page 2023

Show all 41 references
  1. [8]

    no” to say “yes

    Singh, J., Shrivastava, I., Vatsa, M., Singh, R., Bharati, A.: Learn “no” to say “yes” better: Improving vision-language models via negations. arXiv preprint arXiv:2403.20312 (2024) 14

    2024 arXiv

  2. [9]

    In: The Thirty-eighth Annual Conference on Neural Informa- tion Processing Systems (2024)

    Patel, M., kusumba, N.S.A., Cheng, S., Kim, C., Gokhale, T., Baral, C., Yang, Y.: TripletCLIP: Improving compositional reasoning of CLIP via synthetic vision- language negatives. In: The Thirty-eighth Annual Conference on Neural Informa- tion Processing Systems (2024). https:/...

    2024

  3. [10]

    In: Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pp

    Hossain, M.M., Anastasopoulos, A., Blanco, E., Palmer, A.: It’s not a non-issue: Negation as a source of error in machine translation. In: Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pp. 3869–3885 (2020)

    2020

  4. [11]

    In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp

    Kassner, N., Sch¨ utze, H.: Negated and misprimed probes for pretrained language models: Birds can talk, but cannot fly. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp. 7811–7818 (2020)

    2020

  5. [12]

    guesswhat?! dialogues as a case study

    Testoni, A., Greco, C., Bernardi, R.: Artificial intelligence models do not ground negation, humans do. guesswhat?! dialogues as a case study. Frontiers in big Data, 736709 (2022)

    2022

  6. [13]

    In: Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp

    Safavi, T., Zhu, J., Koutra, D.: Negater: Unsupervised discovery of negatives in commonsense knowledge bases. In: Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp. 5633–5646 (2021)

    2021

  7. [14]

    In: Ger- man Conference on Artificial Intelligence (K¨ unstliche Intelligenz), pp

    Schon, C., Siebert, S., Stolzenburg, F.: Negation in cognitive reasoning. In: Ger- man Conference on Artificial Intelligence (K¨ unstliche Intelligenz), pp. 217–232 (2021). Springer

    2021

  8. [15]

    2022 ieee

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. 2022 ieee. In: CVF Conference on Computer Vision and Pattern Recognition, vol. 1 (2021)

    2021

  9. [16]

    https://arxiv.org/abs/2406.05494

    Varshney, N., Raj, S., Mishra, V., Chatterjee, A., Sarkar, R., Saeidi, A., Baral, C.: Investigating and Addressing Hallucinations of LLMs in Tasks Involving Negation (2024). https://arxiv.org/abs/2406.05494

    2024 arXiv

  10. [17]

    In: Pro- ceedings of the Joint International Conference on Computational Linguistics, Language Resources and Evaluation, pp

    Christen, R., Shaitarova, A., St¨ urmer, M., Niklaus, J.: Resolving legalese: A multilingual exploration of negation scope resolution in legal documents. In: Pro- ceedings of the Joint International Conference on Computational Linguistics, Language Resources and Evaluation, pp...

    2024

  11. [18]

    JMIR Med Inform 2023;11:e46348 (2023)

    Weng, K., Liu, C., Chen, C.: Deep learning approach for negation and speculation detection for automated important finding flagging and extraction in radiology report: Internal validation and technique comparison study. JMIR Med Inform 2023;11:e46348 (2023)

    2023

  12. [20]

    In: Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pp

    Hossain, M.M., Kovatchev, V., Dutta, P., Kao, T., Wei, E., Blanco, E.: An anal- ysis of natural language inference benchmarks through the lens of negation. In: Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pp. 9106–9118 (2020)

    2020

  13. [21]

    Transactions of the Association for Computational Linguistics 8, 34–48 (2020)

    Ettinger, A.: What bert is not: Lessons from a new suite of psycholinguistic diag- nostics for language models. Transactions of the Association for Computational Linguistics 8, 34–48 (2020)

    2020

  14. [22]

    IEEE transactions on affective computing 10(3), 374–393 (2017)

    Alarcao, S.M., Fonseca, M.J.: Emotions recognition using eeg signals: A survey. IEEE transactions on affective computing 10(3), 374–393 (2017)

    2017

  15. [23]

    PhD thesis, University of Oxford (1978)

    Pea, R.: The development of negation in early child language. PhD thesis, University of Oxford (1978)

    1978

  16. [24]

    Natural Language & Linguistic Theory, 775–825 (2020)

    Greco, M.: On the syntax of surprise negation sentences: A case study on expletive negation. Natural Language & Linguistic Theory, 775–825 (2020)

    2020

  17. [25]

    looking at

    Orenes, I.: “looking at” negation: Faster processing for symbolic rather than iconic representations. Journal of Psycholinguistic Research (6), 1417–1436 (2021)

    2021

  18. [26]

    Psychological bulletin 138(3), 427 (2012)

    Khemlani, S., Johnson-Laird, P.N.: Theories of the syllogism: A meta-analysis. Psychological bulletin 138(3), 427 (2012)

    2012

  19. [27]

    Journal of Psycholinguistic Research 50(6), 1511–1534 (2021)

    Wang, S., Sun, C., Tian, Y., Breheny, R.: Verifying negative sentences. Journal of Psycholinguistic Research 50(6), 1511–1534 (2021)

    2021

  20. [28]

    Advances in neural information processing systems 35, 24824–24837 (2022)

    Wei, J., Wang, X., Schuurmans, D., Bosma, M., Xia, F., Chi, E., Le, Q.V., Zhou, D., et al.: Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems 35, 24824–24837 (2022)

    2022

  21. [29]

    In: The Twelfth International Conference on Learning Representations (2024)

    Goyal, S., Ji, Z., Rawat, A.S., Menon, A.K., Kumar, S., Nagarajan, V.: Think before you speak: Training language models with pause tokens. In: The Twelfth International Conference on Learning Representations (2024). https://openreview.net/forum?id=ph04CRkPdC

    2024

  22. [30]

    In: Proceedings of the Conference on Empirical Methods in Natural Language Processing, pp

    Garc ´ ıa-Ferrero, I., Altuna, B., Alvez, J., Gonzalez-Dios, I., Rigau, G.: This is not a dataset: A large negation benchmark to challenge large language models. In: Proceedings of the Conference on Empirical Methods in Natural Language Processing, pp. 8596–8615 (2023)

    2023

  23. [31]

    Yuksekgonul, M., Bianchi, F., Kalluri, P., Jurafsky, D., Zou, J.: When and why vision-language models behave like bags-of-words, and what to do about it? In: International Conference on Learning Representations (2022)

    2022

  24. [32]

    In: Proceedings of the ACM International Conference on Multimedia, 16 pp

    Wang, Z., Chen, A., Hu, F., Li, X.: Learn to understand negation in video retrieval. In: Proceedings of the ACM International Conference on Multimedia, 16 pp. 434–443 (2022)

    2022

  25. [33]

    https://arxiv.org/abs/2305.11834

    Deshmukh, S., Elizalde, B., Singh, R., Wang, H.: Pengi: An Audio Language Model for Audio Tasks (2024). https://arxiv.org/abs/2305.11834

    2024 arXiv

  26. [34]

    https: //arxiv.org/abs/2406.11768

    Ghosh, S., Kumar, S., Seth, A., Evuru, C.K.R., Tyagi, U., Sakshi, S., Nieto, O., Duraiswami, R., Manocha, D.: GAMA: A Large Audio-Language Model with Advanced Audio Understanding and Complex Reasoning Abilities (2024). https: //arxiv.org/abs/2406.11768

    2024 arXiv

  27. [35]

    In: Findings of the Association for Computational Linguistics, pp

    Caffagni, D., Cocchi, F., Barsellotti, L., Moratelli, N., Sarto, S., Baraldi, L., Cornia, M., Cucchiara, R.: The revolution of multimodal large language mod- els: A survey. In: Findings of the Association for Computational Linguistics, pp. 13590–13618 (2024)

    2024

  28. [36]

    In: Findings of the Association for Computational Linguistics, pp

    Zhang, D., Li, S., Zhang, X., Zhan, J., Wang, P., Zhou, Y., Qiu, X.: Speechgpt: Empowering large language models with intrinsic cross-modal conversational abilities. In: Findings of the Association for Computational Linguistics, pp. 15757–15773 (2023)

    2023

  29. [37]

    In: Findings of the Association for Computational Linguistics, pp

    Zhang, Y., Yasunaga, M., Zhou, Z., HaoChen, J.Z., Zou, J., Liang, P., Yeung, S.: Beyond positive scaling: How negation impacts scaling trends of language models. In: Findings of the Association for Computational Linguistics, pp. 7479–7498 (2023)

    2023

  30. [38]

    ECCV (2024)

    Mittal, S., Sudan, A., Vatsa, M., Singh, R., Glaser, T., Hassner, T.: Navigating text-to-image generative bias across indic languages. ECCV (2024)

    2024

  31. [39]

    In: Findings of the Association for Computational Linguistics, pp

    Limisiewicz, T., Balhar, J., Mareˇ cek, D.: Tokenization impacts multilingual lan- guage modeling: Assessing vocabulary allocation and overlap across languages. In: Findings of the Association for Computational Linguistics, pp. 5661–5681 (2023)

    2023

  32. [40]

    arXiv preprint arXiv:2410.12989 (2024)

    Chelombitko, I., Safronov, E., Komissarov, A.: Qtok: A comprehensive frame- work for evaluating multilingual tokenizer quality in large language models. arXiv preprint arXiv:2410.12989 (2024)

    2024 arXiv

  33. [41]

    Neurocomputing 284, 171–176 (2018) 17

    Choi, H., Cho, K., Bengio, Y.: Fine-grained attention mechanism for neural machine translation. Neurocomputing 284, 171–176 (2018) 17

    2018

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.