Pith. sign in

REVIEW 4 major objections 6 minor 31 references

SonicVerse: Multi-Task Learning for Music Feature-Informed Captioning

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read Multi-task music features improve AI captions and key accuracy

desk verdict SonicVerse is a reasonable open-source multi-task music captioning model, but its central ablation is statistically indistinguishable from zero and the feature evaluation lacks human validation. read the letter →

arxiv 2506.15154 v1 pith:SUZ5PF4Q submitted 2025-06-18 cs.SD cs.AIcs.CLcs.MMeess.AS

classification cs.SDcs.AIcs.CLcs.MMeess.AS
keywords musiccaptioningmulti-tasklearninginformationretrievalaudio-languagemodelsfeatureextractionLLMchainingMERTMIRFLEX
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

SonicVerse is a music captioning model that tries to make generated captions more detailed and accurate by training the model to detect concrete musical features at the same time as it writes captions. The audio is encoded by MERT, and a multi-task projector maps both the raw music content and the detected features (key, instrumentation, genre, mood, vocals, and vocal gender) into language tokens that feed a frozen Mistral-7B language model. The paper claims that this feature-guided token input improves caption quality and detail over a content-only projector, and it reports gains on BLEU, ROUGE, and BERT scores, plus better accuracy for naming the correct key. The same pipeline is chained with an LLM to produce time-stamped captions for full-length songs. A careful reader would care because richer captions could upgrade music databases, training data for text-to-music models, and access to music-theory information.

What carries the argument

The load-bearing component is the multi-task projector: a shared backbone computes a learned layer-weighted, time-averaged representation of MERT embeddings, feeds it to $K$ task heads that predict music features, and projects each feature vector through an MLP ($\mathrm{MLP}_{\mathrm{feat}}$) into language tokens. Those tokens are concatenated with content tokens and the textual query before entering the frozen LLM. The learned layer weights $\alpha_\ell$ and $\beta_\ell$ let the model choose which MERT layers carry which musical information, and the task heads make feature extraction part of the captioning pipeline rather than an external pre-processing step.

What would settle it

Take the feature-augmented model and retrain it with feature labels randomly shuffled across training clips; if caption scores stay at the same level, the improvement is not caused by feature content. Alternatively, have human annotators compare feature-augmented and content-only captions against ground-truth audio for key, instrumentation, and vocal presence; if the augmented captions are not factually closer to the audio, the central claim fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that injecting predicted music features into the language-token stream improves music captioning. In the proposed architecture, $\mathrm{Caption} = \mathrm{LLM}([z_{\mathrm{content}} \parallel z_{\mathrm{feature}} \parallel q])$, where $z_{\mathrm{content}}$ comes from a learned weighted average of MERT layers and $z_{\mathrm{feature}}$ comes from parallel task heads that predict key, instrument, mood/genre, vocals, and vocal gender, each head's output projected by an MLP into language tokens. The feature heads are trained with binary cross-entropy losses and the captioning head with cross-entropy. In the authors' ablation, the feature-augmented model (Baseline B) exceeds the content-only control (Baseline A) on BLEU, BLEU-4, ROUGE, and BERT-Score, ties on METEOR, and outperforms open-data baselines on the music-feature accuracy metrics, especially key detection. The authors also claim that, by captioning 10-second chunks and chaining the captions with an LLM, the model produces a coherent long caption that tracks the musical evolution of a full piece.

Load-bearing premise

The load-bearing premise is that the auxiliary feature labels, generated by the authors' MIRFLEX tool rather than by human annotators, are accurate enough that the feature heads and projected feature tokens carry true musical information.

Editorial extensions

If this is right

  • If the central claim holds, captioning models can be trained on smaller open datasets by adding auxiliary feature-detection tasks, reducing dependence on massive private audio-text corpora.
  • Feature-augmented captions should name the musical key more often and more accurately; the paper reports the best key-match score among the compared models on the MusicBench test set.
  • Because audio and features share one projection stage, no external feature extractor is needed at inference time, which simplifies deployment and keeps the pipeline fully open-source.
  • Chunk captioning plus LLM chaining yields time-stamped long-form captions, so the same model can describe both a 10-second fragment and the full temporal arc of a song.

Reading between the lines

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

  • The measured n-gram gains are small (BLEU 0.3484 vs 0.3456), so the headline contribution may be feature-specific accuracy and descriptive detail rather than general language fluency; a human evaluation of informativeness would be the decisive test.
  • Because all feature labels come from MIRFLEX, the same author-group tool, part of the benefit could be the model learning to echo label patterns rather than true acoustic facts; training on independently human-annotated features would separate these.
  • The chaining prompt already instructs the LLM to distrust chunk descriptions that conflict with their neighbors; this could be turned into an explicit per-chunk confidence score, letting the pipeline flag unreliable segments rather than silently blending them.
  • The learned layer weights $\alpha_\ell$ and $\beta_\ell$ are an interpretable by-product: inspecting them could reveal which MERT layers encode key, genre, or vocals, and possibly guide other music-understanding systems.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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. SonicVerse is a multitask music captioning system that feeds MERT audio embeddings through a "music content projector" and, simultaneously, through "music feature extractor heads" that predict concrete attributes (key, instrument, mood/theme, genre, vocals, vocal gender). The predicted feature vectors are projected into language tokens and concatenated with content tokens and query tokens for a frozen Mistral-7B LLM, which produces the caption. The authors extend the Magna-Tag-A-Tune, Jamendo, and MusicBench datasets with MIRFLEX-generated feature labels, pretrain on the extended Jamendo/Magna-Tag-A-Tune data, finetune on the extended MusicBench set, and compare a content-only baseline (Baseline A) with the feature-augmented SonicVerse (Baseline B). They also propose an LLM-chaining method that turns per-10-second clip captions into long, temporally aware descriptions of full tracks. The central claim is that incorporating auxiliary music features improves the quality and detail of generated captions, supported by an ablation (Table 1), a comparison to prior models (Table 2), and GPT-4-based music-feature agreement metrics (Table 3), plus a single qualitative case study.

Significance. If the central claim is borne out, the paper contributes a practical way to inject explicit MIR knowledge into an LLM-based captioning pipeline while training on relatively small, open, creative-commons datasets. The open release of model code and weights is a notable strength, as several prior systems (LLaRK, SALMONN, FUTGA parts) do not release their weights. The chaining mechanism for long-form captions is also a useful direction. However, the core evidence for feature-informed improvement is currently too fragile: the reported metric gains in Table 1 are within the range of metric noise, the feature labels are generated by a tool from the same research group, and the music-specific metrics depend on an unvalidated GPT-4 judge. These issues need to be resolved before the abstract's headline claim can be accepted. The paper is otherwise a reasonable system description with a clear architecture, and the identified problems are addressable through additional statistical validation and label-quality analysis rather than requiring a conceptual rework.

major comments (4)
  1. [§5.1, Table 1] The abstract's claim that "incorporating features in this way improves the quality and detail of the generated captions" rests entirely on Table 1, but the reported differences are BLEU 0.3456 versus 0.3484, BLEU-4 0.1799 versus 0.1824, METEOR 0.2507 versus 0.2506, ROUGE 0.2621 versus 0.2622, and BERT 0.8716 versus 0.8723. No standard deviations, confidence intervals, number of seeds, or significance tests are reported. Given that both baselines share the same frozen encoder and LLM and differ only in the projector heads, these deltas are well within typical run-to-run and bootstrap noise for n-gram and embedding-similarity metrics. In particular, the METEOR result numerically regresses. The paper must report error bars over at least three seeds and a paired bootstrap test (or a similar significance check) before the improvement claim is testable. Without this, the central contribution is not established by the evidence as presented.
  2. [§4.3, MIRFLEX labels] The auxiliary training labels for key, genre, instrumentation, vocals, and other features are generated by MIRFLEX, a tool developed by the same research group, and the paper itself acknowledges in Section 4.3 that "using the MIRFLEX feature extractors rather than ground truth music features might introduce some noise and bias into the dataset." This is not a minor caveat: if MIRFLEX is systematically wrong on some feature (for example, key detection), the multi-task heads and the projected tokens z_feature will encode that bias, and any improvement or degradation in the ablation could be an artifact of label noise rather than a genuine benefit of feature-informed projection. The paper does not report MIRFLEX's accuracy on independently annotated data, nor does it analyze how label noise affects the feature heads. I request a validation of MIRFLEX against a standard human-annotated benchmark (e.g., key/genre/instrument tags) and, ideally, a sensitivity analysis or a small-scale experiment with clean human labels to confirm that the feature pathway is beneficial when the labels are reliable.
  3. [§4.4, Table 3] The music-feature match metrics in Table 3 are produced by a GPT-4 prompt that compares the generated caption and the reference caption on key, instrumentation, genre, mood, vocal presence, and vocal gender. The paper provides no human validation of this LLM judge, no inter-annotator agreement, and no confidence intervals on the reported proportions. Since the claim that SonicVerse "outperforms all other models" for key (0.3048 versus 0.2966) rests on differences of roughly one percent in an unvalidated automatic judgment, the result should be treated with caution. The authors should sample a subset of judgments, compute agreement with human annotators, and report the judge's reliability (e.g., Cohen's kappa) and bootstrap confidence intervals for Table 3.
  4. [§5.2, Table 2] The state-of-the-art comparison in Table 2 mixes numbers "taken from the respective papers" with numbers obtained from re-running BLAP and QWEN2-Audio. Because the other models may have been evaluated on different MusicCaps splits, with different decoding settings or metric implementations, the claim that SonicVerse "can outperform most models" is not directly supported unless all rows come from an identical evaluation protocol. The paper calls this a "rough benchmark," which weakens the competitive claim. I suggest either re-evaluating the open-weight baselines under the same harness as SonicVerse, or clearly reporting the exact split, decoding parameters, and metric version used to produce each row.
minor comments (6)
  1. [§3, Eq. (1)] Equation (1) and the surrounding text define N as "the number of tokens representing the textual query," but N is elsewhere used for the total number of music-feature tokens and later N_k for per-task token counts in Section 4.3. Please fix this notation so that the query tokens and feature tokens are named consistently.
  2. [§4.3] The last sentence of Section 4.3 refers to evaluations of music features "in Section 3," but the feature evaluation is described in Section 4.4 and reported in Section 5.2. Please correct the cross-reference.
  3. [Throughout] There are several typographical and formatting issues: "SonicV erse" in the title, "Figure!2," "with with," "ground trust caption" (should be "ground truth"), "temporal evaluation" (should be "temporal evolution"), and inconsistent spellings of "MosiacML/MosaicML" and "LLARK/LLaRK." A careful proofreading pass is needed.
  4. [§4.3] Section 4.3 states that MIRFLEX also generates chords and downbeat timings, but the model's feature heads cover only key, instrument, mood/theme, genre, vocals, and vocal gender. Please clarify whether chords and downbeats are used in any part of the training or evaluation, or remove them from the dataset description.
  5. [§5.3] The long-caption case study is a single anecdote. Consider adding a small human evaluation of chained long captions, or at least qualitative comparison with FUTGA, which also targets temporally structured music descriptions, so that the chaining contribution can be assessed beyond one example.
  6. [Figure 2] The caption of Figure 2 labels the per-task token counts as "N1 Tokens" and "NK Tokens" without a defined relationship to the total token budget M + N. Please state the total token budget and the per-task allocation directly in the figure caption.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the captioning improvement claim rests on an empirical ablation measured against reference captions, not on a definitional or self-citation chain.

full rationale

After walking the derivation chain, no load-bearing step reduces to its own inputs. The central claim is an empirical ablation: Baseline B's feature-projector plus multi-task heads is compared against Baseline A's content-only projector on the MusicCaps split (Table 1) using standard captioning metrics against reference captions. This comparison is not derived from the feature labels; it is measured. The paper's auxiliary labels are generated by MIRFLEX, a same-group tool, and the paper itself flags this in Section 4.3: 'using the MIRFLEX feature extractors rather than ground truth music features might introduce some noise and bias into the dataset.' That is a real limitation on label quality, but it does not make the captioning result circular: the Table 1 metrics are computed against ground-truth reference captions, and Table 3's music-feature match is computed by GPT-4 comparing prediction and reference captions, not against MIRFLEX outputs. The self-citation of MIRFLEX is not used to prove the captioning improvement; the improvement stands or falls on the measured deltas. The small effect sizes and lack of significance tests are real experimental concerns, but they concern statistical evidence and experimental rigor, not definitional circularity. No equation in the paper defines the predicted caption in terms of the evaluation labels, and no fitted parameter is renamed as a prediction. The MIRFLEX noise caveat is explicitly acknowledged and does not create a self-fulfilling evaluation loop.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim does not rest on a mathematical derivation or fitted constants, but it does depend on hyperparameter choices and on machine-generated feature labels. The auxiliary losses and token allocations are chosen by hand, and the 'ground truth' features come from the authors' own MIRFLEX extractor.

free parameters (5)
  • Task loss weight λ_k (feature pretraining) = 0.2
    Uniform weight for each feature detection task during the pretraining phase; chosen without reported tuning.
  • Caption loss weight λ_cap = 1.0
    Weight for the captioning loss in the combined objective.
  • Feature loss weight λ_k (caption pretraining) = 0.1
    Weight for individual feature tasks during the captioning pretraining step.
  • Content token count M = 35
    Number of language tokens assigned to the music content projection.
  • Feature token count N_k per head = 5
    Number of tokens per auxiliary task head in the feature projector.
assumptions (4)
  • domain assumption MERT embeddings contain sufficient acoustic and musical information for feature detection and captioning.
    The architecture uses MERT features as the sole audio representation and relies on them to support both the feature heads and the captioning input, without independent verification of feature-specific informativeness.
  • domain assumption MIRFLEX-generated feature labels are accurate enough to serve as training targets.
    The authors extend datasets with MIRFLEX and acknowledge it 'might introduce some noise and bias into the dataset' (Section 4.3); the entire multi-task supervision depends on these labels.
  • domain assumption GPT-4-based evaluation of music feature matches is a valid proxy for ground truth.
    Music metrics are computed by asking GPT-4 to compare prediction and reference captions, with no human validation of this judge-based metric.
  • domain assumption MusicCaps/MusicBench reference captions are reliable ground truth.
    The reference captions include pseudo-captions generated by LLMs; the paper treats them as ground truth for evaluation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SonicVerse: Multi-Task Learning for Music Feature-Informed Captioning." pith.science (2026). https://pith.science/paper/SUZ5PF4Q

@misc{pith2026250615154,
  author       = {Pith},
  title        = {Pith review of: SonicVerse: Multi-Task Learning for Music Feature-Informed Captioning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SUZ5PF4Q}},
  note         = {Machine review of arXiv:2506.15154}
}
read the original abstract

Detailed captions that accurately reflect the characteristics of a music piece can enrich music databases and drive forward research in music AI. This paper introduces a multi-task music captioning model, SonicVerse, that integrates caption generation with auxiliary music feature detection tasks such as key detection, vocals detection, and more, so as to directly capture both low-level acoustic details as well as high-level musical attributes. The key contribution is a projection-based architecture that transforms audio input into language tokens, while simultaneously detecting music features through dedicated auxiliary heads. The outputs of these heads are also projected into language tokens, to enhance the captioning input. This framework not only produces rich, descriptive captions for short music fragments but also directly enables the generation of detailed time-informed descriptions for longer music pieces, by chaining the outputs using a large-language model. To train the model, we extended the MusicBench dataset by annotating it with music features using MIRFLEX, a modular music feature extractor, resulting in paired audio, captions and music feature data. Experimental results show that incorporating features in this way improves the quality and detail of the generated captions.

Figures

Figures reproduced from arXiv: 2506.15154 by the authors.

Figure 1
Figure 1. Overview of the proposed caption generation framework. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of the proposed SonicVerse architecture. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 18 canonical work pages

  1. [1]

    L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., et al

    Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F. L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., et al. (2023). Gpt-4 technical report. arXiv:2303.08774

  2. [2]

    I., Borsos, Z., Engel, J., Verzetti, M., Caillon, A., Huang, Q., Jansen, A., Roberts, A., Tagliasacchi, M., et al

    Agostinelli, A., Denk, T. I., Borsos, Z., Engel, J., Verzetti, M., Caillon, A., Huang, Q., Jansen, A., Roberts, A., Tagliasacchi, M., et al. (2023). Musiclm: Generating music from text. arXiv:2301.11325

  3. [3]

    and Lavie, A

    Banerjee, S. and Lavie, A. (2005). Meteor: An automatic metric for mt evaluation with improved correlation with human judgments. In Proceedings of the acl workshop on intrinsic and extrinsic evaluation measures for machine translation and/or summarization , pages 65--72

  4. [4]

    Bogdanov, D., Won, M., Tovstogan, P., Porter, A., and Serra, X. (2019). The mtg-jamendo dataset for automatic music tagging. ICML

  5. [5]

    Chen, S., Wu, Y., Wang, C., Liu, S., Tompkins, D., Chen, Z., and Wei, F. (2022). Beats: Audio pre-training with acoustic tokenizers. arXiv:2212.09058

  6. [6]

    Chopra, A., Roy, A., and Herremans, D. (2024). Mirflex: Music information retrieval feature library for extraction. arXiv:2411.00469

  7. [7]

    Chu, Y., Xu, J., Yang, Q., Wei, H., Wei, X., Guo, Z., Leng, Y., Lv, Y., He, J., Lin, J., et al. (2024). Qwen2-audio technical report. arXiv:2407.10759

  8. [8]

    Dataset, M. S. (2011). Million song dataset

Show all 31 references
  1. [9]

    Deng, Z., Ma, Y., Liu, Y., Guo, R., Zhang, G., Chen, W., Huang, W., and Benetos, E. (2023). Musilingo: Bridging music and text with pre-trained language models for music captioning and query response. arXiv:2309.08730

  2. [10]

    Doh, S., Choi, K., Lee, J., and Nam, J. (2023). Lp-musiccaps: Llm-based pseudo music captioning. In Proceedings of the 24th International Society for Music Information Retrieval Conference (ISMIR) , pages 409--416, Milan, Italy

  3. [11]

    Gardner, J., Durand, S., Stoller, D., and Bittner, R. M. (2023). Llark: A multimodal foundation model for music. arXiv:2310.07160

  4. [12]

    Q., Sablayrolles, A., Mensch, A., Bamford, C., Chaplot, D

    Jiang, A. Q., Sablayrolles, A., Mensch, A., Bamford, C., Chaplot, D. S., Casas, D. d. l., Bressand, F., Lengyel, G., Lample, G., Saulnier, L., et al. (2023). Mistral 7b. arXiv:2310.06825

  5. [13]

    R., and Macha, S

    Kamuni, N., Jindal, M., Soni, A., Mallreddy, S. R., and Macha, S. C. (2024). Exploring jukebox: A novel audio representation for music genre identification in mir. In 2024 3rd International Conference on Artificial Intelligence For Internet of Things (AIIoT) , pages 1--6. IEEE

  6. [14]

    Kuang, Z., Zong, S., Zhang, J., Chen, J., and Liu, H. (2022). Music-to-text synaesthesia: Generating descriptive text from music recordings

  7. [15]

    A., Pinkl, C., Perraudin, N., and Wattenhofer, R

    Lanzend \"o rfer, L. A., Pinkl, C., Perraudin, N., and Wattenhofer, R. (2025). Bootstrapping language-audio pre-training for music captioning. In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages 1--5. IEEE

  8. [16]

    I., Bay, M., and Downie, J

    Law, E., West, K., Mandel, M. I., Bay, M., and Downie, J. S. (2009). Evaluation of algorithms using games: The case of music tagging. In ISMIR , pages 387--392. Citeseer

  9. [17]

    Li, Y., Yuan, R., Zhang, G., Ma, Y., Chen, X., Yin, H., Lin, C., Ragni, A., Benetos, E., Gyenge, N., et al. (2023). Mert: Acoustic music understanding model with large-scale self-supervised training. arXiv:2306.00107

  10. [18]

    Lin, C.-Y. (2004). Rouge: A package for automatic evaluation of summaries. In Text summarization branches out , pages 74--81

  11. [19]

    S., Sun, C., and Shan, Y

    Liu, S., Hussain, A. S., Sun, C., and Shan, Y. (2024). Music understanding llama: Advancing text-to-music generation with question answering and captioning. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages 286--290. IEEE

  12. [20]

    Manco, I., Benetos, E., Quinton, E., and Fazekas, G. (2021). Muscaps: Generating captions for music audio. In 2021 International Joint Conference on Neural Networks (IJCNN) , pages 1--8. IEEE

  13. [21]

    Melechovsky, J., Guo, Z., Ghosal, D., Majumder, N., Herremans, D., and Poria, S. (2024). Mustango: Toward controllable text-to-music generation. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language...

  14. [22]

    MosaicML NLP Team et al. (2023). Introducing mpt-7b: A new standard for open-source, commercially usable llms. DataBricks (May, 2023) www. mosaicml. com/blog/mpt-7b

  15. [23]

    C., Davies, M

    Nieto, O., McCallum, M. C., Davies, M. E., Robertson, A., Stark, A. M., and Egozy, E. (2019). The harmonix set: Beats, downbeats, and functional segment annotations of western popular music. In ISMIR , pages 565--572

  16. [24]

    Papineni, K., Roukos, S., Ward, T., and Zhu, W.-J. (2002). Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics , pages 311--318

  17. [25]

    W., Xu, T., Brockman, G., McLeavey, C., and Sutskever, I

    Radford, A., Kim, J. W., Xu, T., Brockman, G., McLeavey, C., and Sutskever, I. (2023). Robust speech recognition via large-scale weak supervision. In International conference on machine learning , pages 28492--28518. PMLR

  18. [26]

    Roy, A., Liu, R., Lu, T., and Herremans, D. (2025). Jamendomaxcaps: A large scale music-caption dataset with imputed metadata. In Proceedings of the International Joint Conference on Neural Networks (IJCNN) , Rome, Italy

  19. [27]

    Tang, C., Yu, W., Sun, G., Chen, X., Tan, T., Li, W., Lu, L., Ma, Z., and Zhang, C. (2023). Salmonn: Towards generic hearing abilities for large language models. arXiv:2310.13289

  20. [28]

    Wu, J., Novack, Z., Namburi, A., Dai, J., Dong, H.-W., Xie, Z., Chen, C., and McAuley, J. (2024). Futga: Towards fine-grained music understanding through temporally-enhanced generative augmentation. arXiv:2407.20445

  21. [29]

    Wu*, Y., Chen*, K., Zhang*, T., Hui*, Y., Berg-Kirkpatrick, T., and Dubnov, S. (2023). Large-scale contrastive language-audio pretraining with feature fusion and keyword-to-caption augmentation. In IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP

  22. [30]

    Yang, Q., Xu, J., Liu, W., Chu, Y., Jiang, Z., Zhou, X., Leng, Y., Lv, Y., Zhao, Z., Zhou, C., et al. (2024). Air-bench: Benchmarking large audio-language models via generative comprehension. arXiv:2402.07729

  23. [31]

    Q., and Artzi, Y

    Zhang, T., Kishore, V., Wu, F., Weinberger, K. Q., and Artzi, Y. (2019). Bertscore: Evaluating text generation with bert. arXiv:1904.09675

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.