Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Reassessing the Role of Chain-of-Thought in Sentiment Analysis: Insights and Limitations

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Chain-of-thought prompting barely changes sentiment analysis results in large language models.

desk verdict The accuracy claims at the center of this paper rest on proxy labels built from the same models being evaluated, so the 'minimal impact of CoT' conclusion isn't grounded—though the counterfactual demonstration analysis is worth a look. read the letter →

arxiv 2501.08641 v1 pith:2HVEDHY6 submitted 2025-01-15 cs.CL cs.AI

Kaiyuan Zheng , Qinghua Zhao , Lei Li This is my paper · ORCID
classification cs.CLcs.AI
keywords chain-of-thoughtpromptingsentimentanalysisaspect-basedin-contextlearninglanguageandthoughtcounterfactualexperimentslargemodelsdemonstrationdependence
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

This paper is an experimental check on whether teaching a large language model to reason step by step, through chain-of-thought prompting, actually improves its understanding of sentiment. The paper's claim is that it does not: across three CoT prompt styles, four model sizes, two standard aspect-based sentiment datasets, and a new dataset designed with frequent emotion shifts, CoT yields at most a small benefit for the smallest model in one-shot settings and essentially no benefit for larger models. The paper also reports that the text the models produce is more similar to aspect terms than to sentiment words, and that flipping the sentiment labels in the few-shot demonstrations changes predictions more than shuffling the words in the input. If correct, the paper shows that sentiment analysis is not propelled by a distinct reasoning chain, and that the demonstrations carry the task signal.

What carries the argument

The central object is the chain-of-thought prompt, instantiated in three forms to avoid prompt-specific bias: a natural-language version (CoT-v1), a hybrid natural-symbolic version (CoT-v2), and a purely symbolic polarity-sequence version (CoT-v3). The mechanism that carries the argument is a pair of perturbation comparisons measured by Cohen's kappa agreement between original and perturbed outputs: shuffling adjacent words in the input tests reliance on word order, and reversing the sentiment polarity labels inside the demonstrations tests reliance on in-context information. The difference in agreement under these two perturbations is what licenses the paper's conclusion that demonstration information, not the reasoning path, is the operative signal for sentiment classification.

What would settle it

Re-score the Laptop and Restaurant test sets with human annotators for overall sentiment and rerun the standard-versus-CoT comparison; if CoT changes accuracy by more than a few points on those human labels for large models in 18-shot settings, the minimal-impact claim fails. Alternatively, if sentiment-reversed demonstrations leave predictions largely unchanged, the demonstration-dependence claim fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that chain-of-thought has minimal impact on sentiment analysis tasks. The evidence includes a notable gain for the smallest model on one hard implicit split (accuracy rising from 0.24 to 0.62), but the gain shrinks as demonstrations increase and essentially disappears for the larger models on most settings. Input-output similarity analysis shows standard and CoT prompts alike align more strongly with aspect terms than with sentiment words, indicating CoT does not shift the model toward sentiment. Counterfactual tests then reverse the sentiment labels in demonstrations and swap adjacent input words: reversing demonstrations substantially changes predictions, while shuffling the input does not. The paper reads these results as supporting the independence of language and thought in this setting, because a purely semantic task does not need an explicit thought-like reasoning path.

Load-bearing premise

The paper assumes that the proxy overall-sentiment labels for Laptop and Restaurant, built by weighted majority voting over the same four models being evaluated, faithfully represent human overall sentiment; if that proxy is wrong, the reported accuracy differences measure model self-agreement rather than task accuracy.

Editorial extensions

If this is right

  • Adding chain-of-thought prompts to sentiment analysis is unlikely to improve accuracy for capable language models, so effort spent designing reasoning prompts for this task has low expected payoff.
  • The sentiment signal in few-shot setups comes mainly from the demonstrations, so curating or correcting the example labels is a more direct lever than changing the prompt's reasoning format.
  • One-shot CoT can help a small model on hard implicit sentiment, but the benefit fades as more demonstrations are supplied, so the apparent CoT effect is partly a small-model and low-shot phenomenon.
  • For the language-and-thought question, the result is a concrete data point: a semantic classification task does not require an explicit reasoning chain, which supports the independence side of the debate.

Reading between the lines

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

  • Editorial inference: if demonstration dependence is as strong as the counterfactual results suggest, sentiment predictions can be steered by flipping labels in the prompt, which turns demonstration selection into a practical control knob for deployment.
  • Editorial inference: the similarity finding points to aspect detection rather than polarity judgment as the binding constraint in aspect-based sentiment analysis; a testable next step is whether interventions that make aspects salient improve accuracy more than reasoning prompts do.
  • Editorial inference: because the proxy gold labels come from weighted voting over the same models, the cleanest extension is a human-annotated overall-sentiment benchmark; without it, the paper's accuracy comparisons are partly measures of model self-agreement.
  • Editorial inference: the input-shuffling and demonstration-reversal design could be carried to other semantic tasks such as natural language inference or factual consistency to ask whether demonstration-driven behavior is specific to sentiment or a general property of in-context learning.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper examines whether chain-of-thought (CoT) prompting improves sentiment analysis in large language models, using this as an indirect test of whether semantic understanding depends on reasoning, which the authors equate with thought. Experiments compare a standard prompt with three CoT prompt variants across four models (Gemma-2 2B/9B/27B and LLaMA-3 8B), multiple shot counts, and three datasets: filtered subsets of SemEval Laptop/Restaurant data and a manually constructed multi-emotion-shift (MES) dataset. The paper reports accuracy, input-output token similarity, and perturbation experiments (word-order shuffling and counterfactual demonstrations). It concludes that CoT has minimal impact on sentiment analysis, that both standard and CoT prompts focus on aspect terms, and that model behavior depends primarily on demonstration information, thereby supporting the independence of language and thought.

Significance. If the central claims were established, the paper would contribute a useful empirical datapoint to the discussion of CoT in sentiment analysis and to the broader language-and-thought debate. Its strengths include testing three distinct CoT formats, using multiple model families and sizes, and attempting a new emotion-shift dataset. However, the primary quantitative evidence is currently compromised by a self-referential evaluation setup: the proxy ground-truth labels for overall sentiment are derived from weighted majority voting over the same models being evaluated. The MES dataset lacks any reported human validation. These problems undermine the paper's main conclusions as they stand, though they are potentially addressable through additional human annotation and re-analysis.

major comments (4)
  1. [Section 4.1, Figures 2 and 3] The accuracy results for RQ-1 and RQ-2 (e.g., Gemma2-2b implicit Laptop 0.24 vs 0.62) are computed against proxy ground-truth labels obtained by weighted majority voting over the models under test. Because the proxy labels are a function of the models' own outputs, these numbers measure agreement with a model-derived consensus rather than accuracy relative to human overall sentiment. The paper explicitly states that Laptop and Restaurant lack ground-truth overall sentiment, and the reported 'accuracy' is therefore not task accuracy. This invalidates the central 'minimal impact' claim as currently formulated. The authors should obtain human-annotated overall sentiment labels for their test samples (or an independently validated external gold standard) and report accuracy, Cohen's kappa, or a similar metric against that standard.
  2. [Section 4.2, MES dataset] The manually constructed multi-emotion-shift dataset consists of only 100 texts, and the paper reports no annotation procedure, no number of annotators, no inter-annotator agreement, and no validation of the emotion labels or shift counts. Consequently, the accuracy trends in Figure 4 (e.g., Gemma2-9b decreasing from 0.92 to 0.78 with more shifts) are not verifiable. Without such validation, the MES results cannot support the RQ-2 conclusion that the findings are 'consistent' with the Laptop and Restaurant results.
  3. [Section 4.4 and abstract] The counterfactual demonstration test measures the sensitivity of predictions to reversed sentiment labels in the demonstrations, but it does not measure the relative contribution of pre-training knowledge versus demonstration information. The abstract's claim that 'the model's handling of sentiment tasks primarily depends on information from demonstrations' requires a quantitative comparison of the magnitude of demonstration effects against a measure of pre-training knowledge; the current experiment provides only the former. RQ-4's conclusion is therefore overstated and should be reframed as a demonstration-sensitivity result, not a decomposition of knowledge sources.
  4. [Section 5] The discussion contains an explicit contradiction: it states that the findings 'challenge Wittgenstein's view that language limits the boundaries of thought' and 'support the independence of language and thought,' then immediately says 'the authors still align with Wittgenstein's perspective.' This unresolved inconsistency makes the philosophical conclusion unclear. The authors need to state their actual position and explain how the experimental evidence bears on it, rather than asserting both sides in consecutive sentences.
minor comments (6)
  1. [Section 3.2] The text refers to 'SemEval-2014' Laptop and Restaurant datasets, but the cited reference [13] is SemEval-2016 Task 5; please correct the citation or the dataset version.
  2. [Figure 2] Figure 2 contains twelve panels without individual panel labels; please add labels or subfigure numbering so that the reader can map the described results to the correct panel.
  3. [Figure 7] The caption mentions 'upper part' and 'lower part,' but the figure contains multiple panels per row; please clarify which panels correspond to word-order shuffling and which to sentiment reversal.
  4. [Section 4.3] The similarity heatmaps in Figures 5 and 6 are visually dense, and the claims about 'explicit split demonstrates a stronger similarity' are not supported by a quantitative summary or statistical test; please add numerical aggregates.
  5. [Section 4.2] The MES dataset has only 100 samples, which is small even for exploratory analysis; please include this as an explicit limitation and report confidence intervals or error bars if possible.
  6. [Table 2] The prompt examples in Table 2 refer to colored highlights, but the manuscript is not in color; please replace colors with formatting that is visible in grayscale print.

Circularity Check

1 steps flagged · score 6.0 of 10

Laptop/Restaurant accuracy rests on proxy labels built from the same models being evaluated, making the central 'minimal impact' claim partially self-referential.

  1. self definitional [Section 4.1 (RQ-1) and Figure 3 caption]
    "It is important to note that our experimental design required an additional step due to the absence of ground-truth overall sentiment in the Laptop and Restaurant datasets. To address this, we implemented a post-hoc analysis using a weighted majority voting method to establish proxy ground-truths. ... based on the agreement results from the above experiments, a weight is assigned to each model's voting."

    The proxy overall-sentiment labels are constructed by weighted majority voting over the same models whose standard-vs-CoT accuracies are then reported; the weights are each model's Cohen's Kappa agreement with aspect-level labels. Accuracy is therefore agreement with a model-derived ensemble consensus rather than with an external human overall-sentiment label. The headline differences (e.g., Gemma2-2b 0.24 vs 0.62) measure how far a prompting condition moves a model from that consensus. If the consensus diverges from human overall sentiment, RQ-1 cannot establish that CoT has 'minimal impact' on sentiment analysis; the target labels are fitted from the evaluated models' outputs, so the evaluation loop is self-definitional.

full rationale

The clearest circular step is the construction of proxy ground-truth overall-sentiment labels in Section 4.1 from the very models under evaluation. Because the Laptop and Restaurant datasets lack overall-sentiment labels, the paper assigns each model a voting weight based on its agreement with aspect-level ground truth and uses the weighted majority of the models' outputs as the target for accuracy. All RQ-1 accuracy numbers and the 'minimal impact' conclusion for these datasets therefore measure self-consistency with an ensemble consensus, not task accuracy against an external standard. The manual MES dataset in RQ-2 has labels defined by construction and is not part of this loop, and the RQ-4 counterfactual demonstration analysis follows the independent method of Madaan et al. [10], so those parts are not circular. The Section 5 statement that the findings support independence while 'the authors still align with Wittgenstein's perspective' is internally inconsistent but is a coherence problem, not circularity. Overall, one central evaluation loop affects the paper's headline claim, giving partial circularity rather than full circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claims rest on three kinds of assumptions: the conceptual identification of thought with CoT and language with SA, the ad hoc proxy ground-truth construction, and the perturbational logic for demonstration reliance. All are stated in the paper, but none is independently validated.

free parameters (3)
  • Per-model voting weights for proxy ground-truth = Cohen's Kappa values (examples: 0.73 for Gemma2-27b, 0.01 for Gemma2-2b on Laptop explicit)
    In Section 4.1, each model is assigned a weight based on its agreement with the 'ground truth', and these weights are then used in a weighted majority vote to define the proxy ground-truth labels for accuracy. The weights are post hoc and directly affect all reported accuracy numbers.
  • Test-sample filtering thresholds = Unreported counts; thresholds include at least two aspects and two sentiment changes
    Section 3.2 filters SemEval-2014 samples by text length, sentiment dynamics, complexity, and explicit/implicit split following [7]. The resulting subset sizes are not given, so the degree of selection is unknown.
  • MES dataset construction choices = 100 texts; six emotion categories; at least two emotion shifts
    Section 4.2 introduces a manually constructed dataset with author-chosen emotions and transitions. It is not released and has no inter-annotator agreement, so it functions as an ad hoc evaluation set for RQ-2.
assumptions (4)
  • domain assumption CoT prompting is a valid operationalization of thought, and sentiment analysis is a valid operationalization of language.
    Section 1 maps 'thought as reasoning, CoT as reasoning technique, SA as language comprehension'; the philosophical conclusion relies on this mapping being meaningful.
  • domain assumption Cognitive Load Theory implies that activating reasoning abilities should deepen sentiment understanding.
    Section 1 motivates the experimental hypothesis with psychological Cognitive Load Theory; if this bridge fails, the null result says nothing about reasoning and semantics.
  • ad hoc to paper Weighted majority voting over model predictions yields a valid proxy for ground-truth overall sentiment.
    Section 4.1: 'we implemented a post-hoc analysis using a weighted majority voting method to establish proxy ground-truths.' The proxy is constructed from the models being evaluated, so it is not an independent benchmark.
  • domain assumption Output changes under word-order disruption and sentiment-reversal perturbations isolate the model's reliance on demonstrations rather than prompt sensitivity.
    Section 4.4 adopts the word-order disruption method from [24] and the counterfactual method from [10]; the interpretation assumes these perturbations change semantics, not just surface form.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Reassessing the Role of Chain-of-Thought in Sentiment Analysis: Insights and Limitations." pith.science (2026). https://pith.science/paper/2HVEDHY6

@misc{pith2026250108641,
  author       = {Pith},
  title        = {Pith review of: Reassessing the Role of Chain-of-Thought in Sentiment Analysis: Insights and Limitations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2HVEDHY6}},
  note         = {Machine review of arXiv:2501.08641}
}
read the original abstract

The relationship between language and thought remains an unresolved philosophical issue. Existing viewpoints can be broadly categorized into two schools: one asserting their independence, and another arguing that language constrains thought. In the context of large language models, this debate raises a crucial question: Does a language model's grasp of semantic meaning depend on thought processes? To explore this issue, we investigate whether reasoning techniques can facilitate semantic understanding. Specifically, we conceptualize thought as reasoning, employ chain-of-thought prompting as a reasoning technique, and examine its impact on sentiment analysis tasks. The experiments show that chain-of-thought has a minimal impact on sentiment analysis tasks. Both the standard and chain-of-thought prompts focus on aspect terms rather than sentiment in the generated content. Furthermore, counterfactual experiments reveal that the model's handling of sentiment tasks primarily depends on information from demonstrations. The experimental results support the first viewpoint.

Figures

Figures reproduced from arXiv: 2501.08641 by the authors.

Figure 1
Figure 1. The framework of our work. To answer the aforementioned question, we propose an experimental ap￾proach that examines tasks not explicitly reliant on reasoning abilities and eval￾uates whether providing additional reasoning information enhances model per￾formance. The overall framework of our work is shown in [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overall accuracy across datasets and shots. The top row presents results [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Agreement (Cohen’s Kappa value, treated as weights for majority vot [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Model accuracy as a function of emotion shifts and categories. Results [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Similarity between input and output tokens. Top row (Laptop) and bot [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Similarity between input and output tokens. Top row (Laptop) and bot [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Model output agreement before and after input shuffling (upper part), [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. LLaMAs Have Feelings Too: Unveiling Sentiment and Emotion Representations in LLaMA Models Through Probing

    cs.CL 2025-05 conditional novelty 6.0 of 10

    LLaMA models encode binary sentiment most strongly in middle layers and emotions in early layers, and truncating the model at the best layer with a probe head yields efficient sentiment classifiers.

Reference graph

Works this paper leans on

27 extracted references · 16 canonical work pages · cited by 1 Pith paper

  1. [19]

    arXiv preprint arXiv:1804.07461 (2018)

    Wang, A.: Glue: A multi-task benchmark and analysis platform for natural lan- guage understanding. arXiv preprint arXiv:1804.07461 (2018)

    arXiv 2018

  2. [26]

    Word Order and World Knowledge

    Zhao, Q., Ravishankar, V., Garneau, N., Søgaard, A.: Word order and world knowl- edge (2024), https://arxiv.org/abs/2403.00876 A Supplement Table 2: CoT Prompts Examples Input Battery life could be better but overall for the price and Toshiba’s reputation for laptops it’s great! Type Content CoT-v1 Because the Battery life is negative, the price is positi...

    work page Pith review arXiv 2024

  3. [1]

    Advances in neural information processing systems33, 1877–1901 (2020)

    Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J.D., Dhariwal, P., Nee- lakantan, A., Shyam, P., Sastry, G., Askell, A., et al.: Language models are few-shot learners. Advances in neural information processing systems33, 1877–1901 (2020)

    2020

  4. [2]

    Cambridge University Press (2002)

    Chomsky, N.: On Nature and Language. Cambridge University Press (2002)

    work page 2002

  5. [3]

    arXiv preprint arXiv:2407.21783 (2024)

    Dubey, A., Jauhri, A., Pandey, A., Kadian, A., Al-Dahle, A., Letman, A., Mathur, A., Schelten, A., Yang, A., Fan, A., et al.: The llama 3 herd of models. arXiv preprint arXiv:2407.21783 (2024)

    arXiv 2024

  6. [4]

    Nature630(8017), 575–586 (2024)

    Fedorenko, E., Piantadosi, S.T., Gibson, E.A.: Language is primarily a tool for communication rather than thought. Nature630(8017), 575–586 (2024)

    work page 2024

  7. [5]

    In: The 61st Annual Meeting Of The Association For Computational Linguistics (2023)

    Fei, H., Li, B., Liu, Q., Bing, L., Li, F., Chua, T.S.: Reasoning implicit sentiment with chain-of-thought prompting. In: The 61st Annual Meeting Of The Association For Computational Linguistics (2023)

    work page 2023

  8. [6]

    RVISA: Reasoning and Verification for Implicit Sentiment Analysis

    Lai, W., Xie, H., Xu, G., Li, Q.: Rvisa: Reasoning and verification for implicit sentiment analysis. arXiv preprint arXiv:2407.02340 (2024)

    work page Pith review arXiv 2024

Show all 27 references
  1. [7]

    arXiv preprint arXiv:2111.02194 (2021)

    Li, Z., Zou, Y., Zhang, C., Zhang, Q., Wei, Z.: Learning implicit sentiment in aspect-based sentiment analysis with supervised contrastive pre-training. arXiv preprint arXiv:2111.02194 (2021)

    2021 arXiv

  2. [8]

    arXiv preprint arXiv:2307.05052 (2023)

    Liu, F., Xu, P., Li, Z., Feng, Y., Song, H.: Towards understanding in-context learning with contrastive demonstrations and saliency maps. arXiv preprint arXiv:2307.05052 (2023)

    2023 arXiv

  3. [9]

    arXiv preprint arXiv:2104.08786 (2021)

    Lu, Y., Bartolo, M., Moore, A., Riedel, S., Stenetorp, P.: Fantastically ordered prompts and where to find them: Overcoming few-shot prompt order sensitivity. arXiv preprint arXiv:2104.08786 (2021)

    2021 arXiv

  4. [10]

    In: Findings of the Association for Computational Linguistics: EMNLP 2023

    Madaan, A., Hermann, K., Yazdanbakhsh, A.: What makes chain-of-thought prompting effective? a counterfactual study. In: Findings of the Association for Computational Linguistics: EMNLP 2023. pp. 1448–1535 (2023)

    2023

  5. [11]

    Ad- vances in neural information processing systems34, 11054–11070 (2021)

    Perez, E., Kiela, D., Cho, K.: True few-shot learning with language models. Ad- vances in neural information processing systems34, 11054–11070 (2021)

    2021

  6. [12]

    (eds.) Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021

    Pham, T., Bui, T., Mai, L., Nguyen, A.: Out of order: How important is the sequential order of words in a sentence in natural language understanding tasks? In: Zong, C., Xia, F., Li, W., Navigli, R. (eds.) Findings of the Association for Computational Linguistics: ACL-IJCNLP 2...

    2021 doi

  7. [13]

    In: ProWorkshop on Semantic Evaluation (SemEval-2016)

    Pontiki, M., Galanis, D., Papageorgiou, H., Androutsopoulos, I., Manandhar, S., Al-Smadi,M.,Al-Ayyoub,M.,Zhao,Y.,Qin,B.,DeClercq,O.,etal.:Semeval-2016 task 5: Aspect based sentiment analysis. In: ProWorkshop on Semantic Evaluation (SemEval-2016). pp. 19–30. Association for Com...

    2016

  8. [14]

    arXiv preprint arXiv:2404.03361 (2024)

    Rusnachenko, N., Liang, H.: nicolay-r at semeval-2024 task 3: Using flan-t5 for reasoning emotion cause in conversations with chain-of-thought on emotion states. arXiv preprint arXiv:2404.03361 (2024)

    2024 arXiv

  9. [15]

    Harcourt, Brace and World (1921)

    Sapir, E.: Language: An Introduction to the Study of Speech. Harcourt, Brace and World (1921)

    1921

  10. [16]

    Oxford University Press (1985)

    Scarry, E.: The Body in Pain: The Making and Unmaking of the World. Oxford University Press (1985)

    1985

  11. [17]

    In: Moens, M.F., Huang, X., Specia, L., Yih, S.W.t

    Sinha, K., Jia, R., Hupkes, D., Pineau, J., Williams, A., Kiela, D.: Masked language modeling and the distributional hypothesis: Order word matters pre-training for little. In: Moens, M.F., Huang, X., Specia, L., Yih, S.W.t. (eds.) Proceedings of the 12 K. Zheng et al. 2021 Co...

    2021

  12. [18]

    arXiv preprint arXiv:2403.08295 (2024)

    Team, G., Mesnard, T., Hardin, C., Dadashi, R., Bhupatiraju, S., Pathak, S., Sifre, L., Rivière, M., Kale, M.S., Love, J., et al.: Gemma: Open models based on gemini research and technology. arXiv preprint arXiv:2403.08295 (2024)

    2024 arXiv

  13. [20]

    Advances in neural information processing systems35, 24824–24837 (2022)

    Wei, J., Wang, X., Schuurmans, D., Bosma, M., Xia, F., Chi, E., Le, Q.V., Zhou, D., et al.: Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems35, 24824–24837 (2022)

    2022

  14. [21]

    Wei, J., Wei, J., Tay, Y., Tran, D., Webson, A., Lu, Y., Chen, X., Liu, H., Huang, D., Zhou, D., Ma, T.: Larger language models do in-context learning differently (2023), https://arxiv.org/abs/2303.03846

    2023 arXiv

  15. [22]

    Wittgenstein, L.: Tractatus logico-philosophicus (2023)

    2023

  16. [23]

    Physical Review 96(1), 191–195 (1954)

    Yang,C.N.,Mills,R.L.:Conservationofisotopicspinandisotopicgaugeinvariance. Physical Review 96(1), 191–195 (1954)

    1954

  17. [24]

    Expert Systems with Applications 241, 122700 (2024)

    Zhao, Q., Li, J., Liu, J., Kang, Z., Zhou, Z.: Is word order considered by foundation models? a comparative task-oriented analysis. Expert Systems with Applications 241, 122700 (2024). https://doi.org/https://doi.org/10.1016/ j.eswa.2023.122700, https://www.sciencedirect.com/s...

    2024

  18. [25]

    Knowledge-Based Systems 277, 110792 (2023)

    Zhao, Q., Liu, J., Kang, Z., Zhou, Z.: Tracenet: Tracing and locating the key elements in sentiment analysis. Knowledge-Based Systems 277, 110792 (2023). https://doi.org/https://doi.org/10.1016/j.knosys.2023.110792, https:// www.sciencedirect.com/science/article/pii/S0950705123005427

    2023

  19. [2913]

    Association for Computational Linguistics, Online and Punta Cana, Domini- can Republic (Nov 2021).https://doi.org/10.18653/v1/2021.emnlp-main.230, https://aclanthology.org/2021.emnlp-main.230

    2021 doi

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.