Pith. sign in

REVIEW 3 minor 61 references

A pre-trained vision-language alignment model can be directly repurposed for multimodal coreference resolution without target-domain training.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

A pre-trained alignment model is adapted for multimodal coreference resolution via similarity aggregation and evidence theory, reporting gains on the CIN benchmark over prior dedicated methods and VLLMs.

T0 review reviewed 2026-06-27 challenge →

load-bearing objection The paper shows you can repurpose a pre-trained vision-language alignment model for multimodal coreference via similarity aggregation and evidence theory fusion, getting modest gains on CIN without target training.

arxiv 2606.17950 v1 pith:IPUZDGBN submitted 2026-06-16 cs.CV cs.AI

Plug-and-Adapt: Multimodal Coreference Resolution at First Sight with a Pretrained Alignment Model

Jinghan Wu , Jing Li , Ivor W. Tsang , Xuetao Zhang This is my paper
classification cs.CV cs.AI
keywords multimodal coreference resolutionvision-language alignmentplug-and-adaptevidence theory fusionzero-shot adaptationCIN benchmarksimilarity aggregation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

The paper establishes that a carefully pre-trained alignment model between textual and visual information can be adapted on the spot for multimodal coreference tasks. Existing dedicated methods require annotated training on the target dataset, while large vision-language models are too heavy or restricted. The approach uses similarity aggregation of visual and categorical cues combined with evidence theory fusion to boost resolution accuracy. Results on the CIN benchmark show gains over both specialized systems and popular large models, with further tests confirming robustness and generalization.

Core claim

Repurposing a pre-trained fine-grained alignment model via similarity aggregation and evidence theory fusion enables immediate, training-free multimodal coreference resolution that improves CoNLL F1 by 5.31% over state-of-the-art dedicated methods and 2.12% over popular VLLMs on the CIN dataset, while also showing robustness on masked data and generalization on a constructed VCR-MCR dataset.

What carries the argument

The pre-trained alignment model between textual and visual contextual information, repurposed through similarity aggregation fused with evidence theory.

Load-bearing premise

A general pre-trained alignment model between textual and visual information can be directly repurposed to coreference resolution through similarity aggregation and evidence theory fusion without target-domain training.

What would settle it

Failure to improve CoNLL F1 scores on a new multimodal coreference dataset or underperformance compared to fine-tuned baselines when the evidence theory fusion step is removed.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Enables immediate application to new coreference datasets without collecting annotations or running fine-tuning.
  • Reduces dependence on billion-parameter vision-language models that are costly to deploy or access only via APIs.
  • Supports evaluation on masked and constructed datasets to verify robustness and cross-dataset generalization.
  • Demonstrates that fusion of visual and categorical cues can enhance effectiveness beyond either cue alone.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same alignment-plus-fusion pattern could extend to other vision-language reasoning tasks that currently require task-specific training.
  • Smaller alignment models may suffice for many multimodal disambiguation problems if evidence fusion is applied at inference time.
  • Direct plug-and-adapt approaches could reduce the data hunger of multimodal systems in domains where labeled coreference data remains scarce.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

0 major / 3 minor

Summary. The paper proposes a plug-and-adapt method for multimodal coreference resolution (MCR) that pre-trains a fine-grained alignment model on vision-language datasets and repurposes it for MCR via similarity aggregation of visual and categorical cues combined with evidence theory fusion, eliminating the need for target-domain training or large VLLMs. Experiments on the CIN benchmark report 5.31% and 2.12% CoNLL F1 gains over SOTA dedicated methods and popular VLLMs, respectively, with further tests on a masked CIN variant for robustness and a constructed VCR-MCR dataset for generalization.

Significance. If the empirical results hold, the work is significant for offering a lightweight, immediately deployable alternative to both task-specific trained MCR models and inaccessible billion-parameter VLLMs. The approach leverages existing pre-trained alignment models with explicit fusion mechanics, and the manuscript includes ablation tables isolating the fusion contribution plus generalization experiments, which are strengths.

minor comments (3)
  1. Abstract and §4: The reported percentage gains on CIN lack accompanying standard deviations, number of runs, or statistical significance tests; adding these would make the 5.31% and 2.12% improvements easier to interpret.
  2. §3.2: The evidence-theory fusion step is described at a high level; a short worked numerical example showing how similarity scores map to mass functions and are combined would improve clarity without lengthening the section.
  3. §5.3: The construction details of the VCR-MCR dataset (e.g., how coreference pairs were selected or annotated) are only summarized; a brief appendix table with example instances would aid reproducibility.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of our work, the accurate summary of the plug-and-adapt method, and the recommendation for minor revision. We are pleased that the significance for providing a lightweight alternative to task-specific models and large VLLMs is recognized, along with the value of the ablation and generalization experiments.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper pre-trains a fine-grained alignment model on external vision-language alignment datasets, then applies similarity aggregation and evidence-theory fusion to repurpose it for MCR on the CIN benchmark without any target-domain training or fine-tuning. Performance is measured via empirical gains on held-out external benchmarks (CIN, masked CIN, VCR-MCR) against SOTA methods and VLLMs. No equations, fitted parameters, or self-citations are described that would reduce the central claim to its own inputs by construction; the derivation chain remains self-contained against external data and benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are described. The central claim rests on the unstated premise that the pre-trained alignment model already encodes the necessary fine-grained correspondences for coreference.

reviewed 2026-06-27 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Plug-and-Adapt: Multimodal Coreference Resolution at First Sight with a Pretrained Alignment Model." pith.science (2026). https://pith.science/paper/IPUZDGBN

@misc{pith2026260617950,
  author       = {Pith},
  title        = {Pith review of: Plug-and-Adapt: Multimodal Coreference Resolution at First Sight with a Pretrained Alignment Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IPUZDGBN}},
  note         = {Machine review of arXiv:2606.17950}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Visual information helps resolve ambiguity in coreference resolution, leading to notable performance gains. However, existing Multi-modal Coreference Resolution (MCR) methods require training with (partially) annotated data from the target dataset before they can be applied, preventing their direct usability and raising concerns about generalization. While Vision-Language Large Models (VLLMs) with billions of parameters offer promising zero-shot capabilities, they remain largely inaccessible. Their massive size limits deployability, and many are only accessible through paid APIs. In this paper, we propose a plug-and-adapt method that strategically adapts a carefully pre-trained \emph{alignment model} for immediate use in MCR tasks, designed to eliminate the need for training on scarce benchmark datasets or relying on resource-intensive VLLMs. Specifically, we first pre-train a fine-grained alignment model between textual and visual contextual information using vision-language alignment datasets. We then repurpose the alignment model to MCR through similarity aggregation by fusing visual and categorical cues with evidence theory, thereby enhancing effectiveness. Experiments on the Coreference Image Narratives (CIN) benchmark dataset demonstrate the effectiveness of our method, achieving a 5.31\% and 2.12\% improvement in CoNLL F1 over SOTA dedicated methods and popular VLLMs, respectively. We further evaluate our method on a masked CIN dataset for robustness testing and on a specially constructed VCR-MCR dataset for generalization assessment, with results confirming both capabilities.

Figures

Figures reproduced from arXiv: 2606.17950 by Ivor W. Tsang, Jinghan Wu, Jing Li, Xuetao Zhang.

Figure 1
Figure 1. Figure 1: Current MCR solutions: (a) dedicated MCR models, which require [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the adaptation stage of our method. Arrows with different colors in the diagram represent different cues. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: A qualitative example showing coreference resolution result of CLIP, [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Grounding and CR results on the CIN dataset when randomly masking [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Examples of newly constructed data. The mentions in the same color [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of fcor(m, m′ ) for ground truth, visual-cue, category￾cue, and PA-MCR model (left to right). meet the objective of MCR. Some representative examples can be found in [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: CoNLL F1 on the CIN dataset when randomly removing different [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative Error Cases. Each row presents the grounding and [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Grounding and CR results on the CIN dataset when randomly [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: CoNLL F1 on the CIN dataset when randomly removing different [PITH_FULL_IMAGE:figures/full_fig_p010_11.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

61 extracted references · 9 canonical work pages · 4 internal anchors

  1. [1]

    A brief survey on recent advances in coreference resolution,

    R. Liu, R. Mao, A. T. Luu, and E. Cambria, “A brief survey on recent advances in coreference resolution,”Artificial Intelligence Review, vol. 56, no. 12, pp. 14 439–14 481, 2023

    2023

  2. [2]

    Spanbert: Improving pre-training by representing and predicting spans,

    M. Joshi, D. Chen, Y . Liu, D. S. Weld, L. Zettlemoyer, and O. Levy, “Spanbert: Improving pre-training by representing and predicting spans,” Transactions of the Association for Computational Linguistics, vol. 8, pp. 64–77, 2020

    2020

  3. [3]

    Coreference resolution without span representations,

    Y . Kirstain, O. Ram, and O. Levy, “Coreference resolution without span representations,” inProceedings of the Annual Meeting of the Association for Computational Linguistics, 2021, pp. 14–19

    2021

  4. [4]

    Image-based storytelling using deep learning,

    Y . Zhu and W. Q. Yan, “Image-based storytelling using deep learning,” in Proceedings of the International Conference on Control and Computer Vision, 2022, pp. 179–186

    2022

  5. [5]

    Pixels to prose: Understanding the art of image captioning,

    H. Singh, A. Sharma, and M. Pant, “Pixels to prose: Understanding the art of image captioning,”arXiv preprint arXiv:2408.15714, 2024

    work page arXiv 2024

  6. [6]

    Multi-modal self- perception enhanced large language model for 3d region-of-interest captioning with limited data,

    L. Shi, S. Kan, Y . Jin, L. Zhang, and Y . Cen, “Multi-modal self- perception enhanced large language model for 3d region-of-interest captioning with limited data,”IEEE Transactions on Multimedia, 2025

    2025

  7. [7]

    Video storytelling: Textual summaries for events,

    J. Li, Y . Wong, Q. Zhao, and M. S. Kankanhalli, “Video storytelling: Textual summaries for events,”IEEE Transactions on Multimedia, vol. 22, no. 2, pp. 554–565, 2019

    2019

  8. [8]

    What are you talking about? text-to-image coreference,

    C. Kong, D. Lin, M. Bansal, R. Urtasun, and S. Fidler, “What are you talking about? text-to-image coreference,” inProceedings of the Conference on Computer Vision and Pattern Recognition, 2014, pp. 3558–3565

    2014

  9. [9]

    Who’s waldo? linking people across text and images,

    Y . Cui, A. Khandelwal, Y . Artzi, N. Snavely, and H. Averbuch-Elor, “Who’s waldo? linking people across text and images,” inProceedings of the International Conference on Computer Vision, 2021, pp. 1374– 1384

    2021

  10. [10]

    Phrase decoupling cross-modal hierarchical matching and progressive position correction for visual grounding,

    M. Xie, M. Wang, H. Li, Y . Zhang, D. Tao, and Z. Yu, “Phrase decoupling cross-modal hierarchical matching and progressive position correction for visual grounding,”IEEE Transactions on Multimedia, 2025

    2025

  11. [11]

    Gravl-bert: Graphical visual-linguistic representations for multimodal coreference resolution,

    D. Guo, A. Gupta, S. Agarwal, J.-Y . Kao, S. Gao, A. Biswas, C.-W. Lin, T. Chung, and M. Bansal, “Gravl-bert: Graphical visual-linguistic representations for multimodal coreference resolution,” inProceedings of the International Conference on Computational Linguistics, 2022, pp. 285–297

    2022

  12. [12]

    Reclip: A strong zero-shot baseline for referring ex- pression comprehension,

    S. Subramanian, W. Merrill, T. Darrell, M. Gardner, S. Singh, and A. Rohrbach, “Reclip: A strong zero-shot baseline for referring ex- pression comprehension,” inProceedings of the Annual Meeting of the Association for Computational Linguistics, 2022, pp. 5198–5215

    2022

  13. [13]

    A dual reinforcement learning framework for weakly supervised phrase grounding,

    Z. Wang, C. Yang, B. Jiang, and J. Yuan, “A dual reinforcement learning framework for weakly supervised phrase grounding,”IEEE Transactions on Multimedia, vol. 26, pp. 394–405, 2023

    2023

  14. [14]

    Simmc 2.0: A task-oriented dialog dataset for immersive multimodal conversations,

    S. Kottur, S. Moon, A. Geramifard, and B. Damavandi, “Simmc 2.0: A task-oriented dialog dataset for immersive multimodal conversations,” arXiv preprint arXiv:2104.08667, 2021

    work page arXiv 2021

  15. [15]

    Who are you referring to? coreference resolution in image narrations,

    A. Goel, B. Fernando, F. Keller, and H. Bilen, “Who are you referring to? coreference resolution in image narrations,” inProceedings of the International Conference on Computer Vision, 2023, pp. 15 247–15 258

    2023

  16. [16]

    Semi-supervised multimodal coreference resolution in image narrations,

    ——, “Semi-supervised multimodal coreference resolution in image narrations,” inProceedings of the Conference on Empirical Methods in Natural Language Processing, 2023, pp. 11 067–11 081

    2023

  17. [17]

    Self-adaptive fine-grained multi-modal data augmentation for semi- supervised multi-modal coreference resolution,

    L. Zheng, B. Chen, H. Fei, F. Li, S. Wu, L. Liao, D. Ji, and C. Teng, “Self-adaptive fine-grained multi-modal data augmentation for semi- supervised multi-modal coreference resolution,” inProceedings of the International Conference on Multimedia, 2024, pp. 8576–8585. IEEE TRANSACTIONS ON MULTIMEDIA 12

    2024

  18. [18]

    Connecting vision and language with localized narratives,

    J. Pont-Tuset, J. Uijlings, S. Changpinyo, R. Soricut, and V . Ferrari, “Connecting vision and language with localized narratives,” inProceed- ings of the European Conference on Computer Vision. Springer, 2020, pp. 647–664

    2020

  19. [19]

    Revisiting multi-modal llm evaluation,

    J. Lu, S. Srivastava, J. Chen, R. Shrestha, M. Acharya, K. Kafle, and C. Kanan, “Revisiting multi-modal llm evaluation,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 555–564

    2025

  20. [20]

    Knowledge en- hanced vision and language model for multi-modal fake news detection,

    X. Gao, X. Wang, Z. Chen, W. Zhou, and S. C. Hoi, “Knowledge en- hanced vision and language model for multi-modal fake news detection,” IEEE Transactions on Multimedia, vol. 26, pp. 8312–8322, 2024

    2024

  21. [21]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inProceedings of the International Conference on Machine Learning. PmLR, 2021, pp. 8748–8763

    2021

  22. [22]

    Combination of evidence in dempster-shafer theory,

    K. Sentz and S. Ferson, “Combination of evidence in dempster-shafer theory,” 2002

    2002

  23. [23]

    Jøsang,Subjective logic

    A. Jøsang,Subjective logic. Springer, 2016, vol. 3

    2016

  24. [24]

    VisualBERT: A Simple and Performant Baseline for Vision and Language

    L. H. Li, M. Yatskar, D. Yin, C. Hsieh, and K. Chang, “Visualbert: A simple and performant baseline for vision and language,”arXiv preprint arXiv:1908.03557, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1908

  25. [25]

    Uniter: Universal image-text representation learning,

    Y .-C. Chen, L. Li, L. Yu, A. El Kholy, F. Ahmed, Z. Gan, Y . Cheng, and J. Liu, “Uniter: Universal image-text representation learning,” in Proceedings of the European Conference on Computer Vision. Springer, 2020, pp. 104–120

    2020

  26. [26]

    Vinvl: Revisiting visual representations in vision-language models,

    P. Zhang, X. Li, X. Hu, J. Yang, L. Zhang, L. Wang, Y . Choi, and J. Gao, “Vinvl: Revisiting visual representations in vision-language models,” inProceedings of the Conference on Computer Vision and Pattern Recognition, 2021, pp. 5579–5588

    2021

  27. [27]

    Zero-shot referring expression comprehension via structural similarity between images and captions,

    Z. Han, F. Zhu, Q. Lao, and H. Jiang, “Zero-shot referring expression comprehension via structural similarity between images and captions,” inProceedings of the Conference on Computer Vision and Pattern Recognition, 2024, pp. 14 364–14 374

    2024

  28. [28]

    Models overview - anthropic,

    Anthropic, “Models overview - anthropic,” https://docs.anthropic.com/ en/docs/models-overview., 2024, accessed 17-07-2025

    2024

  29. [29]

    LLaVA-OneVision: Easy Visual Task Transfer

    B. Li, Y . Zhang, D. Guo, R. Zhang, F. Li, H. Zhang, K. Zhang, P. Zhang, Y . Li, Z. Liuet al., “Llava-onevision: Easy visual task transfer,”arXiv preprint arXiv:2408.03326, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  30. [30]

    Qwen2.5-VL Technical Report

    S. Bai, K. Chen, X. Liu, J. Wang, W. Ge, S. Song, K. Dang, P. Wang, S. Wang, J. Tanget al., “Qwen2.5-vl technical report,”arXiv preprint arXiv:2502.13923, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  31. [31]

    Can gpt-4v (ision) serve medical applications? case studies on gpt-4v for multimodal medical diagnosis,

    C. Wu, J. Lei, Q. Zheng, W. Zhao, W. Lin, X. Zhang, X. Zhou, Z. Zhao, Y . Zhang, Y . Wanget al., “Can gpt-4v (ision) serve medical applications? case studies on gpt-4v for multimodal medical diagnosis,”arXiv preprint arXiv:2310.09909, 2023

    work page arXiv 2023

  32. [32]

    Gpt-4 in a cancer center—institute-wide deployment challenges and lessons learned,

    R. Umeton, A. Kwok, R. Maurya, D. Leco, N. Lenane, J. Willcox, G. A. Abel, M. Tolikas, and J. M. Johnson, “Gpt-4 in a cancer center—institute-wide deployment challenges and lessons learned,”The New England Journal of Medicine Artificial Intelligence, vol. 1, no. 4, p. AIcs2300191, 2024

    2024

  33. [33]

    3ur-llm: An end- to-end multimodal large language model for 3d scene understanding,

    H. Xiong, Y . Zhuge, J. Zhu, L. Zhang, and H. Lu, “3ur-llm: An end- to-end multimodal large language model for 3d scene understanding,” IEEE Transactions on Multimedia, 2025

    2025

  34. [34]

    Hico: A benchmark for recognizing human-object interactions in images,

    Y .-W. Chao, Z. Wang, Y . He, J. Wang, and J. Deng, “Hico: A benchmark for recognizing human-object interactions in images,” inProceedings of the IEEE international conference on computer vision, 2015, pp. 1017– 1025

    2015

  35. [35]

    Grounded situation recognition,

    S. Pratt, M. Yatskar, L. Weihs, A. Farhadi, and A. Kembhavi, “Grounded situation recognition,” inProceedings of the European Conference on Computer Vision. Springer, 2020, pp. 314–332

    2020

  36. [36]

    Visual genome: Connecting language and vision using crowdsourced dense image annotations,

    R. Krishna, Y . Zhu, O. Groth, J. Johnson, K. Hata, J. Kravitz, S. Chen, Y . Kalantidis, L.-J. Li, D. A. Shammaet al., “Visual genome: Connecting language and vision using crowdsourced dense image annotations,” International Journal of Computer Vision, vol. 123, pp. 32–73, 2017

    2017

  37. [37]

    A. S. Asratian, T. M. Denley, and R. H ¨aggkvist,Bipartite graphs and their applications. Cambridge university press, 1998, vol. 131

    1998

  38. [38]

    Classification-then- grounding: Reformulating video scene graphs as temporal bipartite graphs,

    K. Gao, L. Chen, Y . Niu, J. Shao, and J. Xiao, “Classification-then- grounding: Reformulating video scene graphs as temporal bipartite graphs,” inProceedings of the Conference on Computer Vision and Pattern Recognition, 2022, pp. 19 497–19 506

    2022

  39. [39]

    Llm meets scene graph: Can large language models understand and generate scene graphs? a benchmark and empirical study,

    D. Yang, M. Kim, S. Mac Kim, B.-w. Kwak, M. Park, J. Hong, W. Woo, and J. Yeo, “Llm meets scene graph: Can large language models understand and generate scene graphs? a benchmark and empirical study,” inACL, 2025, pp. 21 335–21 360

    2025

  40. [40]

    Information extraction,

    R. Grishman, “Information extraction,”IEEE Intelligent Systems, vol. 30, no. 5, pp. 8–15, 2015

    2015

  41. [41]

    Open information extraction: A review of baseline techniques, approaches, and applications,

    S. Kamp, M. Fayazi, Z. Benameur-El, S. Yu, and R. Dreslinski, “Open information extraction: A review of baseline techniques, approaches, and applications,”arXiv preprint arXiv:2310.11644, 2023

    work page arXiv 2023

  42. [42]

    Faster r-cnn: Towards real-time object detection with region proposal networks,

    S. Ren, K. He, R. Girshick, and J. Sun, “Faster r-cnn: Towards real-time object detection with region proposal networks,”Advances in Neural Information Processing Systems, vol. 28, 2015

    2015

  43. [43]

    Generalization of dempster–shafer theory: A complex mass function,

    F. Xiao, “Generalization of dempster–shafer theory: A complex mass function,”Applied Intelligence, vol. 50, pp. 3266–3275, 2020

    2020

  44. [44]

    Trusted multi-view classi- fication with dynamic evidential fusion,

    Z. Han, C. Zhang, H. Fu, and J. T. Zhou, “Trusted multi-view classi- fication with dynamic evidential fusion,”IEEE transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 2, pp. 2551–2566, 2022

    2022

  45. [45]

    Two effects, one trigger: On the modality gap, object bias, and informa- tion imbalance in contrastive vision-language models,

    S. Schrodi, D. T. Hoffmann, M. Argus, V . Fischer, and T. Brox, “Two effects, one trigger: On the modality gap, object bias, and informa- tion imbalance in contrastive vision-language models,”arXiv preprint arXiv:2404.07983, 2024

    work page arXiv 2024

  46. [46]

    Bridge the modality and capability gaps in vision-language model selection,

    C. Yi, Y . He, D.-C. Zhan, and H.-J. Ye, “Bridge the modality and capability gaps in vision-language model selection,”Advances in Neural Information Processing Systems, vol. 37, pp. 34 429–34 452, 2024

    2024

  47. [47]

    Towards under- standing the modality gap in clip,

    P. Shi, M. C. Welle, M. Bj ¨orkman, and D. Kragic, “Towards under- standing the modality gap in clip,” inProceedings of the International Conference on Learning Representations Workshop on Multimodal Rep- resentation Learning, 2023

    2023

  48. [48]

    Confidence- aware contrastive learning for selective classification,

    Y .-C. Wu, S.-H. Lyu, H. Shang, X. Wang, and C. Qian, “Confidence- aware contrastive learning for selective classification,” inICML, 2024, pp. 53 706–53 729

    2024

  49. [49]

    On calibration of modern neural networks,

    C. Guo, G. Pleiss, Y . Sun, and K. Q. Weinberger, “On calibration of modern neural networks,” inICML. PMLR, 2017, pp. 1321–1330

    2017

  50. [50]

    Overview of results of the muc-6 evaluation,

    B. M. Sundheim, “Overview of results of the muc-6 evaluation,” in Proceedings of the Message Understanding Conference, 1995

    1995

  51. [51]

    Algorithms for scoring coreference chains,

    A. Bagga and B. Baldwin, “Algorithms for scoring coreference chains,” inProceedings of the International Conference on Language Resources and Evaluation Workshop, vol. 1. Citeseer, 1998, pp. 563–566

    1998

  52. [52]

    On coreference resolution performance metrics,

    X. Luo, “On coreference resolution performance metrics,” inProceed- ings of the Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing, 2005, pp. 25–32

    2005

  53. [53]

    Conll- 2012 shared task: Modeling multilingual unrestricted coreference in ontonotes,

    S. Pradhan, A. Moschitti, N. Xue, O. Uryupina, and Y . Zhang, “Conll- 2012 shared task: Modeling multilingual unrestricted coreference in ontonotes,” inProceedings of the Joint Conference on EMNLP and CoNLL-shared task, 2012, pp. 1–40

    2012

  54. [54]

    Stanford’s multi-pass sieve coreference resolution system at the conll-2011 shared task,

    H. Lee, Y . Peirsman, A. Chang, N. Chambers, M. Surdeanu, and D. Jurafsky, “Stanford’s multi-pass sieve coreference resolution system at the conll-2011 shared task,” inProceedings of the Conference on Computational Natural Language Learning, 2011, pp. 28–34

    2011

  55. [55]

    End-to-end neural coreference resolution,

    K. Lee, L. He, M. Lewis, and L. Zettlemoyer, “End-to-end neural coreference resolution,” inProceedings of the Conference on Empirical Methods in Natural Language Processing, 2017, pp. 188–197

    2017

  56. [56]

    On gen- eralization in coreference resolution,

    S. Toshniwal, P. Xia, S. Wiseman, K. Livescu, and K. Gimpel, “On gen- eralization in coreference resolution,” inProceedings of the Workshop on Computational Models of Reference, Anaphora and Coreference, 2021, pp. 111–120

    2021

  57. [57]

    Qwen2.5 Technical Report

    Qwen, A. Yang, B. Yang, B. Zhang, B. Hui, B. Zheng, B. Yu, C. Li, D. Liu, and F. Huang, “Qwen2.5 technical report,”arXiv preprint arXiv:2412.15115, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  58. [58]

    Maf: Multimodal alignment framework for weakly-supervised phrase grounding,

    Q. Wang, H. Tan, S. Shen, M. Mahoney, and Z. Yao, “Maf: Multimodal alignment framework for weakly-supervised phrase grounding,” inPro- ceedings of the Conference on Empirical Methods in Natural Language Processing, 2020, pp. 2030–2038

    2020

  59. [59]

    Are language models robust coreference resolvers?

    N. T. Le and A. Ritter, “Are language models robust coreference resolvers?” inProceedings of the Conference on Language Modeling, 2024

    2024

  60. [60]

    Open information extraction via chunks,

    K. Dong, A. Sun, J.-J. Kim, and X. Li, “Open information extraction via chunks,”Proceedings of the Conference on Empirical Methods in Natural Language Processing, 2023

    2023

  61. [61]

    From recognition to cogni- tion: Visual commonsense reasoning,

    R. Zellers, Y . Bisk, A. Farhadi, and Y . Choi, “From recognition to cogni- tion: Visual commonsense reasoning,” inProceedings of the Conference on Computer Vision and Pattern Recognition, 2019, pp. 6720–6731. IEEE TRANSACTIONS ON MULTIMEDIA 13 Jinghan Wureceived the B.S. degree in automation science and technology in Xi’an Jiaotong University, Shaanxi, C...

    2019

This paper was first reviewed by grok-4.3 on June 27, 2026.