Pith. sign in

REVIEW 4 major objections 3 minor 47 references

PanDent: Toward Comprehensive Tooth-Level Structure-Language Consistency in Dental Radiology

T0 review · 4 major / 3 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read Current multimodal models can write fluent dental reports yet fail to tie findings to the correct teeth, and fine-tuning on the new PanDent benchmark substantially improves that structure-language consistency.

desk verdict A genuinely useful dental OPG dataset with tooth-level structured annotations, but the benchmark test split and normalization pipeline need major repair before the headline claims hold. read the letter →

arxiv 2607.27378 v1 pith:YAUFULS4 submitted 2026-07-29 cs.CV cs.MM

Xiaohan Li , Xinyu Liu , Chang Liu , Sum Wing Au Yeung , Jun Liu , Yixuan Yuan , Hui Chen This is my paper · ORCID
classification cs.CVcs.MM
keywords dentalpanoramicradiographyorthopantomogrammultimodallargelanguagemodelsstructure-languageconsistencytooth-leveldiagnosisradiologyreportgenerationclinicalbenchmarkfine-grainedlocalization
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

The paper seeks to establish that multimodal large language models—models that read dental panoramic X-rays and produce free-text reports—can sound fluent while making clinically consequential mistakes: they miss teeth that are absent, attribute findings to the wrong tooth numbers, and overlook restorations or bone loss. To make this failure measurable, it introduces PanDent, a set of 9,524 panoramic radiographs with expert-validated tooth-level annotations and template-generated reports that tie each sentence to specific findings. On a held-out 500-case set, the paper shows that fine-tuning general-purpose models on PanDent roughly doubles their average exact-match clinical accuracy, which still leaves tooth-index grounding as the weakest skill. A sympathetic reader would care because structure-language consistency is the missing evaluation axis for medical image-text models: fluency scores do not tell whether a model actually sees the disease at the correct anatomical location. The paper itself concedes that rare findings are long-tailed and unevenly covered.

What carries the argument

The load-bearing object is the PanDent annotation-report pair, built from 21 normalized clinical findings, tooth-index-aware sets under FDI numbering, and binary regional findings. A clinician-defined template bank converts these structured findings into free-text reports with deterministic structure-language alignment; an LLM-assisted verifier and expert audit check consistency. Evaluation proceeds by prompting a model separately for each highlighted attribute and exact-matching its normalized answer against expert labels, which isolates visual localization and diagnostic correctness from language style.

What would settle it

Use the released code to reproduce the exact-match numbers on the 500-case split, then run the same zero-shot versus fine-tuned models on a fresh OPG cohort whose missing-tooth, bone-loss, and crown prevalences are matched to the training split; if the fine-tuning advantage collapses or reverses, the claimed clinical gain is a distribution artifact rather than genuine tooth-level grounding.

Watch

Extended reading notes

Core claim

The central claim is that current MLLMs fail to produce clinically consistent tooth-level descriptions and that PanDent's structured annotations provide supervision that substantially fixes this. The authors argue that the gap is hidden by conventional report-level metrics: BLEU/ROUGE scores make models look adequate, while exact-match comparisons at the level of individual FDI tooth numbers reveal large errors. Their experiments on proprietary, open-source, and medical-domain models show the same pattern—fluency without grounding—and their fine-tuned models achieve the largest gains on the benchmark's clinical accuracy track. The paper also claims that this evaluation protocol, combining re

Load-bearing premise

The load-bearing premise is that the 500-case held-out test set fairly represents the same clinical distribution as the 9,024 training images; the reported prevalence shifts (missing teeth 88.4% vs 68.7%; maxillary bone loss 54.2% vs 30.8%) raise the possibility that fine-tuning gains come from domain shift rather than tooth-level understanding.

Editorial extensions

If this is right

  • If PanDent is right, fine-tuning general-domain multimodal models on it yields substantial gains in both report-level similarity and exact-match clinical accuracy.
  • Fluency metrics alone are not a reliable proxy for clinical quality; attribute-level exact match is the limiting evaluation.
  • Medical-domain models trained on other modalities (for example, chest X-ray) do not transfer well to dental panoramic imaging, performing worse than general-purpose models.
  • Tooth-index grounding remains the hardest failure mode even after fine-tuning, so progress will require domain-specific visual grounding methods.
  • The dual-track evaluation (linguistic coherence plus structural factuality) can be replicated for other anatomy where findings must be localized to specific structures.

Reading between the lines

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

  • The template-generated reports may make fine-tuning gains look larger than they would on free-form clinical dictations; a clinician-reading study on unstructured reports would test whether the improvement is about radiological content or about mimicking template style.
  • The large prevalence shifts between train and test (for example, missing teeth in 88.4% of test cases versus 68.7% of training cases) suggest that part of the measured gain could be a domain-shift artifact; evaluating on prevalence-matched cohorts would separate benchmark-specific overfitting from genuine tooth-level understanding.
  • A natural next step would be to add location-aware supervision, such as detection heads or grounding tokens at the tooth level, since even fine-tuned models still fail at exact FDI indexing.
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 / 3 minor

Summary. The paper introduces PanDent, a dental panoramic radiography (OPG) benchmark of 9,524 cases with expert-validated tooth-level structured annotations, from which free-text radiology reports are generated via a clinician-designed template bank. The authors evaluate a wide range of general-domain and medical-domain MLLMs on report-level NLG metrics and attribute-level exact-match accuracy, report that current models produce fluent but clinically inconsistent reports, and show that fine-tuning Qwen-based models on PanDent substantially improves both linguistic and clinical accuracy. The central claim is that PanDent provides a rigorous benchmark and training resource for tooth-level structure-language consistency.

Significance. If the evaluation is valid, PanDent would be a valuable contribution: the dataset is large, expert-validated, publicly released with code, and the attribute-level exact-match protocol is a principled way to test fine-grained clinical grounding. The deterministic template-based report generation and explicit mapping between structured findings and text are also useful design elements. However, the empirical conclusions rest on several untested assumptions about test-set representativeness, the normalization pipeline, and the clinical meaningfulness of the report-level metrics. These issues are load-bearing because the paper's headline claims — that MLLMs fail at tooth-level consistency and that fine-tuning on PanDent fixes this — are only as strong as the benchmark's validity.

major comments (4)
  1. [Sec. 3, Table 4] The test-set composition is almost certainly confounded by source. The dataset contains 9,019 public and 505 in-house cases, and the split is 9,024 train / 500 test, so the test set is effectively the entire in-house batch. Table 4 confirms large prevalence shifts between train and test (e.g., missing teeth 68.7%→88.4%, maxillary GCBL 30.8%→54.2%, crowns 17.7%→32.8%, mandibular angular bone loss 1.9%→8.0%, with rare findings at 0% in test). Thus zero-shot failures may reflect out-of-distribution acquisition/annotation style, and fine-tuning gains may reflect adaptation to a single clinic distribution and template format rather than generalizable clinical improvement. 'Disjoint from the training split' is insufficient. Please provide source-stratified results (public-only test, in-house-only test) or re-split so the test set is representative of the combined training distribution.
  2. [Table 2, Eq. (A.8)] Several exact-match accuracies are implausibly low relative to trivial baselines. For binary regional attributes, a majority-class predictor on the reported test prevalences would achieve roughly 0.49–0.54 accuracy (e.g., maxillary GCBL prevalence 54.2%, mandibular 49.2%), yet GPT-5.4 scores 0.418/0.419 and MAIRA-2 scores near 0.10. Many tooth-index attributes also show near-zero accuracy for some models. These numbers suggest the LLM-based normalizer π_j may be systematically mis-mapping model outputs, or that the metric is computed incorrectly. The paper does not report the majority-class baseline, per-attribute positive/negative counts, or a manual analysis of normalization errors. Without this, the conclusion that MLLMs 'fail' at clinical consistency is not trustworthy.
  3. [Sec. 3.3, Sec. 4.2] The 'structure-language consistency' evaluation is partially circular. Ground-truth free-text reports are generated from the same structured annotations using a template bank, and the normalized-report NLG metrics rewrite model outputs into the same template-aligned format before comparing against these template references. This largely measures template mimicry, not clinical understanding. The attribute-level exact-match protocol is anchored to expert-validated labels and is the valid clinical measure, but the paper's language-level 'clinical consistency' claims rely on the more circular NLG comparison. Please either de-emphasize the NLG-based consistency conclusions, or validate them against independent expert-written reports.
  4. [Sec. 4.1, Table 2] The highlighted attribute set used for the headline clinical accuracy does not include several important tooth-level findings, such as missing teeth, partially erupted/impacted teeth, furcation radiolucency, and angular bone loss, even though these are annotated and appear in the dataset. Rare findings have 0% prevalence in the test split, so they cannot be evaluated at all. Consequently, the claim of evaluating 'comprehensive tooth-level structure-language consistency' is overstated. Please either expand the evaluated attribute set or qualify the claim to reflect the attributes actually measured.
minor comments (3)
  1. [Throughout] There are typos and informal draft remnants: 'involvment' (Sec. 3.1), 'teet #16' (Table 3), and Appendix A contains the placeholder sentence 'The final version can directly replace the current draft with the exact implementation details...' This is not appropriate for a submission-ready manuscript.
  2. [Sec. 4.2] The manual audit of only 200 normalized reports is small relative to the 500-case test set, and the normalization prompt explicitly suppresses negative findings not stated in the input. Please report inter-auditor agreement and the rate of normalization-induced errors.
  3. [Table 2] No confidence intervals or significance tests are provided for the fine-tuning gains. With N=500, differences of a few points may be within noise; please report bootstrap intervals or at least per-attribute variances.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the central clinical-accuracy claims are anchored to expert-validated labels and do not reduce to the template-generated text.

full rationale

The paper's main derivation chain is empirical rather than deductive: it constructs a benchmark from expert-validated structured labels, generates free-text reports from those labels via a disclosed template bank, and then evaluates MLLMs along two separate tracks. The clinical-accuracy track (Table 2) uses attribute-targeted prompts and exact-match against the expert-annotated ground-truth values z_{i,j} (Appendix A, Eq. A.8), not against the template verbalizations, so the claims about 'visual localization accuracy and diagnostic correctness' are externally anchored to the expert labels and are not equivalent to the template construction by definition. The NLG track explicitly measures surface-form agreement ('These metrics measure lexical overlap and sequence-level similarity... higher scores indicate stronger surface-form agreement'), and the paper does not present BLEU/ROUGE/METEOR as evidence of clinical factuality. The template-driven report construction is transparently described as a deliberate choice 'rather than capturing reporting diversity,' and any concern that NLG scores reward template mimicry is a limitation of that metric, not a circular derivation of the central result. The self-citations in the introduction (e.g., Liu et al. 2024a,b; Gao et al. 2026) support general background statements and are not load-bearing. The possible test-set composition shift inferred from Table 4 (e.g., missing teeth 88.4% vs 68.7%, maxillary GCBL 54.2% vs 30.8%) raises external-validity concerns about benchmark fairness and domain shift, but that is a data-sampling issue, not a circularity of the kind defined here. Overall, the paper's substantive findings—that zero-shot MLLMs make fine-grained tooth-level errors and that fine-tuning on PanDent improves attribute-level accuracy—are independent of the template-generated reports used for NLG scoring.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

This is a benchmark paper, so there are no fitted physical parameters or invented entities. The load-bearing assumptions are about label correctness, template-based ground truth, LLM normalization fidelity, and test-set representativeness; the test-set assumption is visibly violated by the paper's own statistics.

free parameters (2)
  • Highlighted attribute set for headline clinical accuracy
    Radiologists selected 7 primary + 4 secondary attributes, but Table 2 reports only 9; this hand-chosen subset drives the average accuracy and is not ablated, so the headline number depends on an arbitrary evaluation design choice.
  • LLM normalization and attribute-specific prompts
    Attribute-level accuracy relies on hand-written prompts (Appendix B) and a frozen LLM normalizer to map outputs to canonical labels; no ablation or error analysis is provided, so absolute accuracies depend on these choices.
assumptions (5)
  • domain assumption The FDI permanent dentition universal set Ω of 32 teeth and the derived missing-teeth set M = Ω \ (E ∪ P) is a valid clinical model for all cases.
    Sec. 3.2; assumes no supernumerary or ectopic teeth are handled and that missing teeth can be inferred from eruption labels.
  • domain assumption Annotations produced by four dentists are clinically correct for all 9,524 cases even though the senior radiologist validated only a randomly sampled subset.
    Sec. 3: 'validated by an assistant professor in oral and maxillofacial radiology on a randomly sampled subset'; no inter-observer agreement statistics are reported.
  • ad hoc to paper Template-generated free-text reports are a valid ground truth for 'clinically consistent radiology reports.'
    Sec. 3.3: reports are synthesized from labels via a radiologist-authored template bank, not written independently; evaluating clinical consistency against them measures alignment with template logic rather than with free-form expert prose.
  • ad hoc to paper The LLM-based report normalizer faithfully preserves clinical facts and maps outputs to canonical labels without introducing or omitting findings.
    Sec. 4.2 and Appendix B; only 200 normalized reports were manually audited, and the below-chance binary accuracies in Table 2 suggest the normalizer may corrupt outputs.
  • domain assumption The held-out 500-case test set is representative of the same distribution as the training split.
    Sec. 3/Fig. 3/Tab. 4: the release contains 505 in-house and 9,019 public cases while the test set has 500 cases, and Table 4 shows large prevalence shifts, indicating the test set is likely the in-house batch.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PanDent: Toward Comprehensive Tooth-Level Structure-Language Consistency in Dental Radiology." pith.science (2026). https://pith.science/paper/YAUFULS4

@misc{pith2026260727378,
  author       = {Pith},
  title        = {Pith review of: PanDent: Toward Comprehensive Tooth-Level Structure-Language Consistency in Dental Radiology},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YAUFULS4}},
  note         = {Machine review of arXiv:2607.27378}
}
read the original abstract

Accurate evaluation of multimodal large language models (MLLMs) in dental panoramic radiography (orthopantomogram, OPG) is limited by the lack of fine-grained, clinically reliable benchmarks that reflect expert interpretation. This work introduces PanDent, a large-scale, clinically grounded OPG benchmark built upon fine-grained, expert-validated tooth-level annotations. The dataset comprises 9,524 high-quality OPGs, each associated with comprehensive structured annotations produced by experienced dentists and further validated by an oral and maxillofacial radiologist, providing clinically reliable supervision for tooth-level diagnosis and reasoning. Clinically consistent radiology reports are constructed from expert-validated findings using clinician-defined reporting logic, establishing explicit correspondence between structured clinical evidence and free-text descriptions. This design enables evaluation of whether MLLMs generate reports that are not only linguistically coherent but also clinically consistent with expert-validated tooth-level findings. Experiments are conducted on diverse MLLMs, including state-of-the-art (SOTA) proprietary models, general-domain open-source models, and medical-specific models. Results show that current MLLMs can generate fluent reports, yet fail to produce clinically consistent descriptions, exhibiting substantial errors in fine-grained localization and tooth-level diagnosis. Fine-tuning on PanDent significantly improves structure-language consistency, substantially enhancing visual localization accuracy and diagnostic correctness, and bringing model outputs closer to expert dental interpretation. These results establish PanDent as a rigorous benchmark for evaluating tooth-level clinical reasoning in MLLMs and a valuable resource for clinically grounded dental AI.

Figures

Figures reproduced from arXiv: 2607.27378 by the authors.

Figure 1
Figure 1. Overview of the PanDent curation pipeline. The dataset is constructed from multi-source OPG records, followed by expert-validated tooth-level structured annotation, clinician-guided report construction, and final release for benchmark evaluation and supervised training. diagnostic systems and studying expert interpretation behavior. OdontoAI (Silva et al., 2023) releases a panoramic radiograph benchmark with baselin… view at source ↗
Figure 2
Figure 2. Multi-level annotation of an OPG image in Pa [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Overview of the dataset composition and attribute distribution. Left: total number of cases for each attribute across the entire dataset. Middle: overall [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Visualization of two data examples from Pa [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: Split-aware text-level statistics of the released reports. [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 1 linked inside Pith

  1. [1]

    Expert Syst

    Nripendra Kumar Singh and Khalid Raza , title =. Expert Syst. Appl. , volume =. 2022 , doi =

    2022

  2. [2]

    Agaian , title =

    Karen Panetta and Rahul Rajendran and Aruna Ramesh and Shishir Paramathma Rao and Sos S. Agaian , title =. IEEE J. Biomed. Health Inform. , volume =. 2022 , doi =

    2022

  3. [3]

    Boosting Research on Dental Panoramic Radiographs: A Challenging Data Set, Baselines, and a Task-Central Online Platform for Benchmark , journal =

    Bernardo Silva and La. Boosting Research on Dental Panoramic Radiographs: A Challenging Data Set, Baselines, and a Task-Central Online Platform for Benchmark , journal =

  4. [4]

    Menze , title =

    Ibrahim Ethem Hamamci and Sezgin Er and Enis Simsar and Anjany Sekuboyina and Mustafa Gundogar and Bernd Stadlinger and Albert Mehl and Bjoern H. Menze , title =. CoRR , volume =. 2023 , doi =

    2023

  5. [5]

    Ai and Lun M

    Jing Hao and Yuxuan Fan and Yanpeng Sun and Kaixin Guo and Lizhuo Lin and Jinrong Yang and Qi Yong H. Ai and Lun M. Wong and Hao Tang and Kuo Feng Hung , title =. CoRR , volume =

  6. [6]

    Alistair E. W. Johnson and Tom J. Pollard and Seth J. Berkowitz and Nathaniel R. Greenbaum and Matthew P. Lungren and Chih-Ying Deng and Roger G. Mark and Steven Horng , title =. CoRR , volume =. 2019 , doi =

    2019

  7. [7]

    Kohli and Marc B

    Dina Demner-Fushman and Marc D. Kohli and Marc B. Rosenman and Sonya E. Shooshan and Laritza Rodriguez and Sameer Antani and George R. Thoma and Clement J. McDonald , title =. J. Am. Med. Inform. Assoc. , volume =. 2016 , doi =

    2016

  8. [8]

    Medical Image Anal

    Aur. Medical Image Anal. , volume =. 2020 , doi =

    2020

Show all 47 references
  1. [9]

    Hasan and Vivek V

    Asma Ben Abacha and Sadid A. Hasan and Vivek V. Datla and Joey Liu and Dina Demner-Fushman and Henning M. 2019 , url =

    2019

  2. [10]

    IEEE Int

    Bo Liu and Li-Ming Zhan and Li Xu and Lin Ma and Xiao. IEEE Int. Symp. Biomed. Imaging (. 2021 , doi =

    2021

  3. [11]

    CoRR , volume =

    Xiang He and Yuhong Zhang and Liansheng Mou and Eric Xing and Pengtao Xie , title =. CoRR , volume =. 2020 , doi =

    2020

  4. [12]

    Summers , title =

    Xiaosong Wang and Yifan Peng and Le Lu and Zhiyong Lu and Mohammadhadi Bagheri and Ronald M. Summers , title =. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (. 2018 , doi =

    2018

  5. [13]

    Li and Xiaodan Liang and Zhiting Hu and Eric P

    Christy Y. Li and Xiaodan Liang and Zhiting Hu and Eric P. Xing , title =. AAAI Conf. Artif. Intell. (. 2019 , doi =

    2019

  6. [14]

    Summers , title =

    Yixiao Zhang and Xiaosong Wang and Zhiyong Lu and Le Lu and Mohammadhadi Bagheri and Ronald M. Summers , title =. AAAI Conf. Artif. Intell. (. 2020 , doi =

    2020

  7. [15]

    Zhihong Chen and Yaling Shen and Yan Song and Xiang Wan , title =. Conf. Empir. Methods Nat. Lang. Process. (. 2020 , doi =

    2020

  8. [16]

    Chunyuan Li and Cliff Wong and Sheng Zhang and Naoto Usuyama and Haotian Liu and Jianwei Yang and Tristan Naumann and Hoifung Poon and Jianfeng Gao , title =. Adv. Neural Inf. Process. Syst. (. 2023 , doi =

    2023

  9. [17]

    Xing and Pranav Rajpurkar , title =

    Michael Moor and Qian Huang and Sushrut Wu and Mike Yasunaga and Yannis Dalmia and Jure Leskovec and Cyril Zakka and Eric P. Xing and Pranav Rajpurkar , title =. Mach. Learn. Health (. 2023 , url =

    2023

  10. [18]

    CoRR , volume =

    Zhihong Chen and Maya Varma and Jean. CoRR , volume =. 2024 , doi =

    2024

  11. [19]

    Saahil Jain and Ashwin Agrawal and Adriel Saporta and Steven Q. H. Truong and Du Nguyen Duong and Tan Bui and Pierre Chambon and Yuhao Zhang and Matthew P. Lungren and Andrew Y. Ng and Curtis P. Langlotz and Pranav Rajpurkar , title =. 2021 , url =

    2021

  12. [20]

    Findings of the Association for Computational Linguistics:

    Sophie Ostmeier and Justin Xu and Zhihong Chen and Christian Vachette and Sophie Bodduluri and Louis Blankemeier and Maya Varma and Jason Ma and Christian Bluethgen , title =. Findings of the Association for Computational Linguistics:. 2024 , doi =

    2024

  13. [21]

    CoRR , volume =

    OpenAI , title =. CoRR , volume =. 2023 , doi =

    2023

  14. [22]

    CoRR , volume =

    Gemini:. CoRR , volume =. 2023 , doi =

    2023

  15. [23]

    CoRR , volume =

    Yuntao Bai and others , title =. CoRR , volume =. 2022 , doi =

    2022

  16. [24]

    CoRR , volume =

    Xingjun Ma and others , title =. CoRR , volume =. 2026 , doi =

    2026

  17. [25]

    CoRR , volume =

    Shuai Bai and others , title =. CoRR , volume =. 2025 , doi =

    2025

  18. [26]

    2025 , doi =

    CoRR , volume =. 2025 , doi =

    2025

  19. [27]

    CoRR , volume =

    Jinguo Zhu and others , title =. CoRR , volume =. 2025 , doi =

    2025

  20. [28]

    CoRR , volume =

    Zhiyu Wu and others , title =. CoRR , volume =. 2024 , doi =

    2024

  21. [29]

    2024 , doi =

    Pixtral 12B , journal =. 2024 , doi =

    2024

  22. [30]

    2025 , doi =

    Gemma 3 Technical Report , journal =. 2025 , doi =

    2025

  23. [31]

    CoRR , volume =

    Abhimanyu Dubey and others , title =. CoRR , volume =. 2024 , doi =

    2024

  24. [32]

    2025 , doi =

    MedGemma Technical Report , journal =. 2025 , doi =

    2025

  25. [33]

    2025 , doi =

    Lingshu: A Generalist Foundation Model for Unified Multimodal Medical Understanding and Reasoning , journal =. 2025 , doi =

    2025

  26. [34]

    AAAI Conf

    Jiao Xu and others , title =. AAAI Conf. Artif. Intell. (. 2026 , url =

    2026

  27. [35]

    CoRR , volume =

    Shruthi Bannur and others , title =. CoRR , volume =. 2024 , doi =

    2024

  28. [36]

    Xi Zhang and Zaiqiao Meng and Jake Lever and Edmond S. L. Ho , title =. Findings of the Association for Computational Linguistics:. 2025 , url =

    2025

  29. [37]

    CoRR , volume =

    Yuxuan Fan and Jing Hao and Hong Chen and Jiahao Bao and Yihua Shao and Yuci Liang and Kuo Feng Hung and Hao Tang , title =. CoRR , volume =. 2026 , doi =

    2026

  30. [38]

    2026 , month = jan, url =

    2026

  31. [39]

    Kishore Papineni and Salim Roukos and Todd Ward and Wei-Jing Zhu , title =. Annu. Meeting Assoc. Comput. Linguist. (

  32. [40]

    Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization , pages =

    Satanjeev Banerjee and Alon Lavie , title =. Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization , pages =

  33. [41]

    Workshop on Text Summarization Branches Out , pages =

    Chin-Yew Lin , title =. Workshop on Text Summarization Branches Out , pages =

  34. [42]

    Eugen-Olsen Lutescu and Octavia Chețan and Elena Drăghici and Anca Apostu and Marius Ilie and Mădălina Iliescu , title =. Front. Radiol. , volume =

  35. [43]

    Jones and Kalpana Ramchandani and Ahmed H

    Hend Alqaderi and Ceib Phillips and Judith A. Jones and Kalpana Ramchandani and Ahmed H. Mitwalli and Naglaa A. Youssef and S. Elham Emami and Thomaz Teixeira and Paulo Salazar Ferreira and Haithem Yousfi and Deepthi Cheela and Asma Bahaj and Hisham Elgendy and Maha G. Elamin ...

  36. [44]

    Thao-Bao To Pham and Nadar Mohammed Al-Hebshi , title =. Front. Artif. Intell. , volume =

  37. [45]

    2024 , bibsource =

    Liu, Chang and Tian, Yuanhe and Chen, Weidong and Song, Yan and Zhang, Yongdong , booktitle =. 2024 , bibsource =

    2024

  38. [46]

    AAAI , pages =

    Gao, Jiechao and Liu, Chang and Li, Yuangang , title =. AAAI , pages =. 2026 , bibsource =

    2026

  39. [47]

    2311.14199 , archivePrefix=

    Chang Liu and Yuanhe Tian and Yan Song , year=. 2311.14199 , archivePrefix=

    arXiv

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.