Pith. sign in

REVIEW 2 major objections 7 minor 32 references

Overview of FinMMEval 2026 Task 2: Multilingual Financial Short-Answer Question Answering

T0 review · 2 major / 7 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read This paper reports a 256-item hidden test for multilingual financial short-answer QA and shows the 12 submitted systems cluster at the top, with the best score at 31.18% macro-averaged ROUGE-1 F1.

desk verdict A straightforward shared-task overview with a clean 256-item held-out split; the hidden-test dedup procedure is under-specified and the abstract oversells the leaderboard, but the paper is honest and deserves review. read the letter →

arxiv 2607.19867 v1 pith:ORLVYPGJ submitted 2026-07-22 cs.CL cs.AIcs.CE

Zhuohan Xie , Xueqing Peng , Georgi Georgiev , Dimitar Dimitrov , Yuyang Dai , Rania Elbadry , Vanshikaa Jani , Lingfei Qian
show 13 more authors
Fan Zhang Jimin Huang Jiahui Geng Yankai Chen Ye Yuan Haolun Wu Yuxia Wang Ivan Koychev Veselin Stoyanov Mingzi Song Yu Chen Xue Liu Preslav Nakov
This is my paper
classification cs.CLcs.AIcs.CE
keywords FinMMEvalfinancialquestionansweringmultilingualevaluationshort-answerQAhiddentestsetROUGE-1F1retrieval-augmentedgenerationanswercompression
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

FinMMEval 2026 Task 2 sets out to measure how well systems answer concise financial questions when the evidence spans five languages. To do this, the organizers built a 256-question held-out test by taking 204 candidate items per difficulty tier from the PolyFiQA resource and deleting the 76 items per tier that matched the public release, leaving 128 easy and 128 expert items with gold answers withheld. Twelve systems submitted answers, ranked by macro-averaged item-level ROUGE-1 F1 against the hidden references; the winner scored 31.18% and the next three systems were within 0.79 points. The paper argues this leaderboard supports comparisons among the submitted systems, while being careful that lexical-overlap scoring does not measure factual correctness and that no tier-level or statistical significance analysis was released.

What carries the argument

The mechanism doing the work is the hidden-test construction rule plus the scoring function. For each of the two tiers, the organizers removed the 76 candidate items that matched the corresponding public PolyFiQA release on source task identifier or full query, retaining 128 items per tier drawn from 32 company-report groups. Scoring is macro-averaged item-level ROUGE-1 F1: after NFKC normalization and lowercasing, a script-aware regular expression tokenizes words, numbers, CJK characters, kana, Arabic-script characters, and Greek/Latin ranges, and unigram precision and recall are computed against the organizer-held reference; the item F1 is 2PR/(P+R), with 0 when P+R=0, and the leaderboard

What would settle it

Download the public PolyFiQA release and run near-duplicate detection between its rows and the 256 held-out items using paraphrase embeddings or sentence-level overlap, in addition to exact identifier and query matching; finding even one retained item whose question or reference answer appears in public data under a different identifier or a rephrased form would falsify the hidden-test claim. A second check would be to rescore all submissions against two independently written reference answers; if the leaderboard order changes materially, the metric is not capturing stable answer quality.

Watch

Extended reading notes

Core claim

The paper's central claim is that a hidden evaluation of multilingual financial short-answer QA is feasible and useful: a 256-item test, half easy and half expert, was created by excluding all PolyFiQA items whose source task identifier or full query appeared in the public release, so that reference answers could be withheld. Systems answered English questions over financial statements and news in English, Chinese, Japanese, Spanish, and Greek, and were scored by macro-averaged item-level ROUGE-1 F1 after Unicode normalization and a script-aware tokenizer. The resulting 12-system leaderboard is tightly clustered at the top, with the first-place system at 31.18% and only 0.79 percentage point

Load-bearing premise

The evaluation stands or falls on the assumption that every item excluded from the public PolyFiQA release by task-identifier or full-query match is exactly the set of items whose reference answers were exposed, so no retained held-out question is answerable from public data under a paraphrase or a different identifier.

Editorial extensions

If this is right

  • If the hidden test is uncontaminated, the leaderboard gives a direct comparison of 12 systems on multilingual financial short-answer QA, with the top four effectively within one point.
  • Because the metric is lexical, a correct answer phrased differently from the reference can lose points; the paper itself notes ROUGE-1 is not a proxy for factual correctness.
  • The tight top cluster means small engineering choices such as evidence reranking, answer compression, or schema prompting can shift ranks, though the official results do not isolate which component causes the differences.
  • The same 32 retained company groups appear in both easy and expert tiers, so releasing tier-wise scores in future editions could separate question difficulty from company effects.
  • The overlap-removal recipe of excluding items by identifier or full-query match offers a reusable way to convert a public benchmark into a hidden test, assuming the public data is the exact same version.

Reading between the lines

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

  • The paper leaves implicit that the 76-item-per-tier exclusion is a completeness assumption, not a demonstrated fact: if any retained item appears in public PolyFiQA under a different identifier or with a paraphrased query, the hidden test is contaminated and the leaderboard would need re-auditing.
  • Given the 0.79-point spread among the top four and no paired significance testing, a fair reading is that ranks 1-4 are likely indistinguishable; the robust comparison is top cluster versus lower half, not first versus second.
  • ROUGE-1's sensitivity to exact unigram phrasing suggests answer compression trades against recall: the team with the highest recall (40.44%) ranked third and the team with the highest precision (36.47%) ranked fourth, so F1 alone may hide a systematic precision-recall frontier in short-answer generation.
  • A testable extension would be to rescore submissions against reference-free factual metrics or multiple independent references; if rank order changes substantially, the current leaderboard reflects wording more than financial understanding.
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

2 major / 7 minor

Summary. This paper presents the organization and results of FinMMEval 2026 Task 2, a shared task on multilingual financial short-answer QA. The final test set comprises 256 items (128 easy, 128 expert) built from PolyFiQA by excluding 76 items per tier that matched public PolyFiQA items. Systems submitted one answer per item; scores are macro-averaged item-level ROUGE-1 F1. The paper reports a 12-system leaderboard and describes participant approaches. The authors acknowledge the absence of an organizer baseline, tier-specific scores, and paired uncertainty analysis.

Significance. If the hidden-test construction is sound, the paper provides a useful, compact shared task for multilingual financial QA, with a transparent evaluation formula and a published leaderboard. Strengths: the item-construction arithmetic is clear (Table 1), the ROUGE scoring is specified precisely (NFKC normalization, tokenization), and the paper is candid about limitations (no baseline, no tier-specific scores, no statistical tests). The main value is as a reference point for future editions and for participants' system descriptions. However, the validity of the benchmark hinges on the completeness of the public-item exclusion, which is not demonstrated.

major comments (2)
  1. [§3.1, Table 1] The hidden-test validity depends entirely on the claim that 76 public-overlap items per tier were excluded. The matching procedure is not specified: is it exact string match after NFKC normalization, case folding, or raw string comparison? Does 'source task identifier or full query matched' mean either field suffices, and are task identifiers in the held-out set guaranteed to be absent from the public release? No manual or fuzzy check is reported for paraphrased questions, renumbered identifiers, or the same company-report group appearing under a different label. Since PolyFiQA is public and includes reference answers, any near-duplicate contamination would invalidate the entire leaderboard. The authors should detail the matching algorithm and report a concrete leakage audit (e.g., all 256 retained items reviewed against the public release, plus a near-duplicate search on queries and gro
  2. [§4.2] The task is framed as 'short-answer' QA, but the scorer processes the complete answer string and the 100-word limit is only a formatting guideline. This means submissions are not penalized for length; a system can increase recall (and thus F1) by copying extended evidence passages. Given the leaderboard shows recall values up to 40.44% (IGT), length variation likely influenced rankings. The paper should either enforce the word limit in scoring (e.g., truncate or apply length normalization) or report answer-length statistics and their correlation with F1, and discuss how the results reflect short-answer ability.
minor comments (7)
  1. [Abstract] The phrase 'closely clustered' is used in the abstract and §5.1, but §5.2 correctly notes that no paired uncertainty analysis exists. Please soften the abstract or add a statistical test to avoid overstating the separation.
  2. [§3.2] Typo: 'Nonereserved' should be 'None reserved'.
  3. [§5.2] Missing spaces: 'IGTrecorded' and 'AI_TLfanClubrecorded' should be 'IGT recorded' and 'AI_TLfanClub recorded'.
  4. [§5.3] Missing space: 'pjmathematicianTask 2' should be 'pjmathematician Task 2'.
  5. [References] Reference [1] (Xie et al., storytelling) appears unrelated to the claim about lexical overlap vs. factual correctness. Please verify and replace with an appropriate citation.
  6. [Table 3] The alignment of checkmarks with column headers is ambiguous; please reformat so that each column corresponds clearly to the listed component.
  7. [Title page] The email address in the header appears corrupted ('envel⌢pe-⌢penzhuohan...'). Please fix the LaTeX/formatting.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: leaderboard is an external evaluation on a held-out subset; only minor self-source provenance.

full rationale

The paper's central result is a leaderboard of 12 external systems, computed by macro-averaged item-level ROUGE-1 F1 between submitted answers and organizer-held reference answers (Section 4.2). This is an empirical evaluation, not a derivation from fitted parameters or from the paper's own equations; no parameter is fitted to the leaderboard scores and no 'prediction' is generated from the inputs. The test set is a held-out subset of PolyFiQA, constructed by excluding 76 items per tier whose source task identifier or full query matched the public PolyFiQA release (Section 3.1); the 256 remaining items and their references were withheld until submission. That is a standard split, not a circular construction. The paper's self-references (MultiFinBen/PolyFiQA, companion FinMMEval overviews) are data provenance and lab context, not load-bearing proof of the empirical outcome, and the data source is a public benchmark with external participant systems. The one genuinely load-bearing assumption—that the exact-match deduplication fully separates held-out from public items—is a data-contamination risk, not a circularity, and the paper explicitly acknowledges its analytical limits in Section 5.2 (no baseline, no tier-specific scores, no paired uncertainty analysis). Therefore the circularity score is minimal.

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

The paper's central claim rests on the integrity of the hidden test and the validity of ROUGE-1 F1 as the measure. No new entities or fitted parameters are introduced.

assumptions (4)
  • domain assumption The PolyFiQA gold answers are correct and appropriate for this task.
    The evaluation scores all systems against these organizer-held references; errors in the gold answers would directly corrupt the leaderboard (Section 3.2).
  • domain assumption Excluding items whose source task identifier or full query matched the public PolyFiQA release removes all test-set leakage.
    The hidden-test design depends on this being complete; no additional leakage check is described (Section 3.1).
  • domain assumption Macro-averaged item-level ROUGE-1 F1 is a suitable ranking metric for the task.
    The paper notes it captures lexical overlap, not factual correctness, and without a baseline or significance tests the ranking may not reflect meaningful system differences (Sections 1, 5.2).
  • domain assumption The tokenizer's regular expression defining unigram units across English, CJK, kana, and Greek/Latin ranges provides an appropriate cross-lingual matching scheme.
    This tokenization defines the metric and is arbitrary; the paper does not compare alternatives (Section 4.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Overview of FinMMEval 2026 Task 2: Multilingual Financial Short-Answer Question Answering." pith.science (2026). https://pith.science/paper/ORLVYPGJ

@misc{pith2026260719867,
  author       = {Pith},
  title        = {Pith review of: Overview of FinMMEval 2026 Task 2: Multilingual Financial Short-Answer Question Answering},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ORLVYPGJ}},
  note         = {Machine review of arXiv:2607.19867}
}
read the original abstract

FinMMEval 2026 Task 2 evaluates short-answer financial question answering over multilingual evidence. Each final-test item pairs an English question with financial statements and news in English, Chinese, Japanese, Spanish, and Greek. Participating systems submit one concise answer per item in JSONL format. The final-test set contains 256 items, split evenly between easy and expert tiers; each tier contains four question templates instantiated over 32 company-report groups. Gold answers were withheld during submission, and systems were ranked by macro-averaged item-level ROUGE-1 F1 against organizer-held reference answers. The final leaderboard includes 12 ranked submissions. The strongest systems are closely clustered, with the top four separated by less than one percentage point in ROUGE-1 F1. The submitted system papers document retrieval-augmented generation, cross-lingual evidence handling, structured prompting, answer compression, and validation strategies.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

32 extracted references · 3 canonical work pages

  1. [1]

    Z. Xie, T. Cohn, J. H. Lau, The next chapter: A study of large language models in storytelling, in: C. M. Keet, H.-Y. Lee, S. Zarrieß (Eds.), Proceedings of the 16th International Natural Language Generation Conference, Association for Computational Linguistics, Prague, Czechia, 2023, pp. 323–

    2023

  2. [2]

    Z. Xie, Y. Dai, R. Elbadry, V. Jani, X. Peng, L. Qian, G. Georgiev, D. Dimitrov, F. Zhang, J. Huang, J. Geng, Y. Chen, Y. Yuan, H. Wu, Y. Wang, I. Koychev, V. Stoyanov, M. Song, Y. Chen, X. Liu, P. Nakov, Overview of FinMMEval 2026: Multilingual and multimodal financial evaluation, in: Experimental IR Meets Multilinguality, Multimodality, and Interaction,...

    2026

  3. [3]

    Z. Xie, Y. Dai, R. Elbadry, V. Jani, G. Georgiev, D. Dimitrov, F. Zhang, X. Peng, L. Qian, J. Huang, J. Geng, Y. Chen, Y. Yuan, H. Wu, Y. Wang, I. Koychev, V. Stoyanov, M. Song, Y. Chen, X. Liu, P. Nakov, Overview of FinMMEval 2026 Task 1: Multilingual Financial Multiple-Choice Question Answering, in: CLEF 2026 Working Notes, CEUR Workshop Proceedings, CE...

    2026

  4. [4]

    Z. Xie, L. Qian, G. Georgiev, D. Dimitrov, Y. Dai, R. Elbadry, V. Jani, X. Peng, F. Zhang, J. Huang, J. Geng, Y. Chen, Y. Yuan, H. Wu, Y. Wang, I. Koychev, V. Stoyanov, M. Song, Y. Chen, X. Liu, P. Nakov, Overview of FinMMEval 2026 Task 3: Live Financial Decision-Making Agents, in: CLEF 2026 Working Notes, CEUR Workshop Proceedings, CEUR-WS.org, Jena, Ger...

    2026

  5. [5]

    X. Peng, L. Qian, Y. Wang, R. Xiang, Y. He, Y. Ren, M. Jiang, V. J. Zhang, Y. Guo, J. Zhao, H. He, Y. Han, Y. Feng, Y. Jiang, Y. Cao, H. Li, Y. Yu, X. Wang, P. Gao, S. Lin, K. Wang, S. Yang, Y. Zhao, Z. Liu, P. Lu, J. Huang, S. Wang, T. Papadopoulos, P. Giannouris, E. Soufleri, N. Chen, Z. Deng, H. Fu, Y. Zhao, M. Lin, M. Qiu, K. E. Smith, A. Cohan, X.-Y....

    2026

  6. [6]

    M. Maia, S. Handschuh, A. Freitas, B. Davis, R. McDermott, M. Zarrouk, A. Balahur, WWW’18 open challenge: Financial opinion mining and question answering, in: P. Champin, F. Gandon, M. Lalmas, P. G. Ipeirotis (Eds.), Companion Proceedings of The Web Conference 2018, Association for Computing Machinery, Lyon, France, 2018, pp. 1941–1942. URL: https://doi.o...

    arXiv 2018

  7. [7]

    Z. Chen, W. Chen, C. Smiley, S. Shah, I. Borova, D. Langdon, R. Moussa, M. Beane, T.-H. Huang, B. Routledge, W. Y. Wang, FinQA: A dataset of numerical reasoning over financial data, in: M.-F. Moens, X. Huang, L. Specia, S. W.-t. Yih (Eds.), Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, Association for Computationa...

    2021

  8. [8]

    F. Zhu, W. Lei, Y. Huang, C. Wang, S. Zhang, J. Lv, F. Feng, T.-S. Chua, TAT-QA: A question answering benchmark on a hybrid of tabular and textual content in finance, in: C. Zong, F. Xia, W. Li, R. Navigli (Eds.), Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural ...

    2021

Show all 32 references
  1. [9]

    Z. Chen, S. Li, C. Smiley, Z. Ma, S. Shah, W. Y. Wang, ConvFinQA: Exploring the chain of numerical reasoning in conversational finance question answering, in: Y. Goldberg, Z. Kozareva, Y. Zhang (Eds.), Proceedings of the 2022 Conference on Empirical Methods in Natural Language...

    2022 doi

  2. [10]

    Islam, A

    P. Islam, A. Kannappan, D. Kiela, R. Qian, N. Scherrer, B. Vidgen, FinanceBench: A new benchmark for financial question answering, 2023. URL: https://arxiv.org/abs/2311.11944. doi:10.48550/arX iv.2311.11944.arXiv:2311.11944

    2023 doi arXiv

  3. [11]

    Z. Xie, D. Orel, R. Thareja, D. Sahnan, H. Madmoun, F. Zhang, D. Banerjee, G. N. Georgiev, X. Peng, L. Qian, J. Huang, J. Su, A. Singh, R. Xing, R. Elbadry, C. Xu, H. Li, F. Koto, I. Koychev, T. Chakraborty, Y. Wang, S. Lahlou, V. Stoyanov, S. Ananiadou, P. Nakov, FinChain: A ...

    2026

  4. [12]

    Y. Dai, Y. Lin, Z. Xie, Y. Wang, RealFin: How well do LLMs reason about finance when users leave things unsaid?, in: M. Liakata, V. P. Moreira, J. Zhang, D. Jurgens (Eds.), Findings of the Association for Computational Linguistics: ACL 2026, Association for Computational Lingu...

    2026 doi

  5. [13]

    Elbadry, S

    R. Elbadry, S. Ahmad, A. Heakl, D. Bouch, M. Ahsan, M. AlMahri, M. E. Khalil, Y. Wang, S. Lahlou, S. Ananiadou, V. Stoyanov, J. Huang, X. Peng, P. Nakov, Z. Xie, SAHM: A benchmark for Arabic financial and shari’ah-compliant reasoning, in: M. Liakata, V. P. Moreira, J. Zhang, D...

    2026

  6. [14]

    Z. Liu, D. Huang, K. Huang, Z. Li, J. Zhao, FinBERT: A pre-trained financial language representation model for financial text mining, in: C. Bessiere (Ed.), Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI-20, International Joint...

    2020 doi

  7. [15]

    S. Wu, O. Irsoy, S. Lu, V. Dabravolski, M. Dredze, S. Gehrmann, P. Kambadur, D. Rosenberg, G. Mann, BloombergGPT: A large language model for finance, 2023. URL: https://arxiv.org/abs/2303.17564. doi:10.48550/arXiv.2303.17564.arXiv:2303.17564

    2023 doi arXiv

  8. [16]

    Q. Xie, W. Han, X. Zhang, Y. Lai, M. Peng, A. Lopez-Lira, J. Huang, PIXIU: A comprehensive benchmark, instruction dataset and large language model for finance, in: A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, S. Levine (Eds.), Advances in Neural Information Processing...

    2023

  9. [17]

    Yang, X.-Y

    H. Yang, X.-Y. Liu, C. D. Wang, FinGPT: Open-source financial large language models, 2023. URL: https://arxiv.org/abs/2306.06031. doi:10.48550/arXiv.2306.06031.arXiv:2306.06031

    2023 doi

  10. [18]

    Y. Zhou, F. Zhang, Y. Chen, H. Zhang, P. Nakov, Z. Xie, FinCARDS: Card-based analyst reranking for financial document question answering, in: M. Liakata, V. P. Moreira, J. Zhang, D. Jurgens (Eds.), Findings of the Association for Computational Linguistics: ACL 2026, Associatio...

    2026 doi

  11. [19]

    Zhang, M

    F. Zhang, M. Song, R. Elbadry, Y. Chen, S. Wang, Y. Zhou, X. Zheng, Y. He, Y. Dai, G. N. Georgiev, A. Gull, M. U. Safder, F. Wu, L. Meng, F. Ji, J. Zhao, X. Peng, J. Huang, Y. Chen, X. Liu, P. Nakov, Z. Xie, FinReporting: An agentic workflow for localized reporting of cross-ju...

    2026

  12. [20]

    X. Peng, Z. Xie, Y. Cao, H. Li, L. Qian, Y. Wang, V. J. Zhang, H. He, X. Ai, L. Ma, R. Xiang, Y. He, Y. Han, S. Wang, Y. Guo, M. Jiang, Y. Zhao, Y. Dong, X. Wang, Y. Chen, Y. Yuan, Q. Zhang, F. Lyu, H. Wu, Y. Yang, Z. Zhao, Y. Dai, F. Zhang, R. Elbadry, A. Gull, M. U. Safder, ...

    2026 doi arXiv

  13. [21]

    Lin, ROUGE: A package for automatic evaluation of summaries, in: Text Summarization Branches Out, Association for Computational Linguistics, Barcelona, Spain, 2004, pp

    C.-Y. Lin, ROUGE: A package for automatic evaluation of summaries, in: Text Summarization Branches Out, Association for Computational Linguistics, Barcelona, Spain, 2004, pp. 74–81. URL: https://aclanthology.org/W04-1013/

    2004

  14. [22]

    Y. Chiu, IGT @ FinMMEval 2026 Task 2: Question-type prompting with targeted extraction for multilingual financial QA, in: CLEF 2026 Working Notes, CEUR Workshop Proceedings, CEUR-WS.org, Jena, Germany, 2026

    2026

  15. [23]

    G. Jain, H. Hitesh, S. Sundriyal, S. Bhardwaj, H. Behl, P. Gautam, S. Zimmerman, Calibrated Signals @ FinMMEval 2026 Task 2: Inference-only strategies for cross-lingual financial question answering, in: CLEF 2026 Working Notes, CEUR Workshop Proceedings, CEUR-WS.org, Jena, Ger...

    2026

  16. [24]

    Thenmozhi, A

    D. Thenmozhi, A. Sivakumar, A. Balasubramanian, A. A, A. Gopinath, TextSentinels @ FinMMEval Task 2: Chronological stabilization of a BM25-grounded multilingual financial question answering pipeline, in: CLEF 2026 Working Notes, CEUR Workshop Proceedings, CEUR-WS.org, Jena, Ge...

    2026

  17. [25]

    D. T. Duc, N. N. Minh, L. T. Huong, AI_TLfanClub at FinMMEval 2026 Task 2: FiSCO-RAG for cross-lingual evidence recall in financial question answering, in: CLEF 2026 Working Notes, CEUR Workshop Proceedings, CEUR-WS.org, Jena, Germany, 2026

    2026

  18. [26]

    D. Liu, S. Li, S. Tian, DS@GT at FinMMEval 2026 Task 2: FinNexus-agentic retrieval-augmented reranking orchestration for multilingual financial question answering, in: CLEF 2026 Working Notes, CEUR Workshop Proceedings, CEUR-WS.org, Jena, Germany, 2026

    2026

  19. [27]

    A. Gill, E. N. Sheikh, F. T. Zahra, A. Samad, F. Alvi, S. Kumar, The Lab Rats @ FinMMEval 2026 Task 2: LLM translation and BAS-tuned chain-of-thought for multilingual financial QA, in: CLEF 2026 Working Notes, CEUR Workshop Proceedings, CEUR-WS.org, Jena, Germany, 2026

    2026

  20. [28]

    P. Rastogi, Pranshu Rastogi @ FinMMEval 2026: Systems for Tasks 1 and 2 – prompt-engineered Gemini for multilingual and cross-lingual financial reports QA, in: CLEF 2026 Working Notes, CEUR Workshop Proceedings, CEUR-WS.org, Jena, Germany, 2026

    2026

  21. [29]

    Affan, M

    M. Affan, M. B. Qureshi, M. H. Shahzad, F. Alvi, A. Samad, HU_LLM_Fin @ FinMMEval 2026 Task 2: A structure-aware hybrid RAG pipeline utilizing mixture-of-experts, in: CLEF 2026 Working Notes, CEUR Workshop Proceedings, CEUR-WS.org, Jena, Germany, 2026

    2026

  22. [30]

    E. L. Pontes, M. Benjannet, TCLabs @ FinMMEval 2026: Systems for Tasks 1, 2 and 3, in: CLEF 2026 Working Notes, CEUR Workshop Proceedings, CEUR-WS.org, Jena, Germany, 2026

    2026

  23. [31]

    Vachharajani, pjmathematician @ FinMMEval 2026: Systems for Tasks 1, 2 and 3, in: CLEF 2026 Working Notes, CEUR Workshop Proceedings, CEUR-WS.org, Jena, Germany, 2026

    P. Vachharajani, pjmathematician @ FinMMEval 2026: Systems for Tasks 1, 2 and 3, in: CLEF 2026 Working Notes, CEUR Workshop Proceedings, CEUR-WS.org, Jena, Germany, 2026

    2026

  24. [351]

    doi:10.18653/v1/2023.inlg-main.23

    URL: https://aclanthology.org/2023.inlg-main.23/. doi:10.18653/v1/2023.inlg-main.23

    2023 doi

Pith tools

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