pith. sign in

arxiv: 2604.17255 · v1 · submitted 2026-04-19 · 💻 cs.CL

Are Emotion and Rhetoric Neurons in LLM? Neuron Recognition and Adaptive Masking for Emotion-Rhetoric Prediction Steering

Li Zheng , Xin Zhang , Shuyi He , Fei Li , Chong Teng , Jiangming Yang , Donghong Ji , Zhuang Li This is my paper

Pith reviewed 2026-05-10 06:15 UTC · model grok-4.3

classification 💻 cs.CL
keywords LLM neuronsemotion categoriesrhetorical devicesneuron maskingcausal validationcontrollable generationmodel interpretability
0
0 comments X p. Extension

The pith

Large language models contain identifiable neurons for six emotion categories and four rhetorical devices that can be isolated and regulated through targeted screening and masking.

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

The paper sets out to show that LLMs internally represent emotions and rhetoric at the level of specific neurons rather than only through external prompting or fine-tuning. It develops a screening process that examines multiple dimensions of neuron behavior and pairs it with an adaptive masking procedure that uses dynamic filtering, gradual attenuation, and iterative feedback to test whether those neurons actually drive the relevant outputs. When the identified neurons are regulated, the model produces directed non-target sentences and shows improved performance on emotion tasks by leveraging rhetoric-related neurons. This internal mechanism supplies a route to fine-grained, neuron-level control over expressive qualities in generated text. A reader would care because current steering methods remain coarse, and reliable neuron-level access could make LLMs more predictable and editable in their emotional and stylistic behavior.

Core claim

By applying multi-dimensional screening to locate neurons associated with six emotion categories and four core rhetorical devices, followed by an adaptive masking procedure that incorporates dynamic filtering, attenuation masking, and feedback optimization, the work demonstrates reliable causal links between these neurons and model outputs. Regulation of the identified neurons produces directed induction of non-target sentences and enhancement of emotion tasks through rhetoric neurons, with the overall approach validated across five standard datasets.

What carries the argument

Adaptive masking method that combines dynamic filtering, attenuation masking, and feedback optimization to causally validate the roles of neurons identified through multi-dimensional screening.

If this is right

  • Rhetoric neurons can be leveraged to improve performance on separate emotion-related tasks.
  • Neuron-level regulation supports directed generation of non-target sentences.
  • The combination of screening and adaptive masking supplies a repeatable way to confirm neuron functionality.
  • Fine-grained steering of both emotion and rhetoric expressions becomes feasible inside the model rather than through external methods.

Where Pith is reading between the lines

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

  • The observed interaction between rhetoric and emotion neurons suggests that linguistic style and affective content share internal circuitry that could be edited modularly in future models.
  • The method could be tested on whether masking emotion neurons reduces unwanted affective leakage in otherwise neutral generations.
  • Extending the same screening and masking pipeline to additional linguistic features such as persuasion or narrative structure might reveal broader organizational principles in LLM representations.

Load-bearing premise

That multi-dimensional screening together with adaptive masking can isolate specific neurons tied to emotion and rhetoric without introducing confounding effects or post-hoc selection bias.

What would settle it

If masking the screened neurons produces no measurable change in emotion or rhetoric content of generated sentences, or if the same neurons alter unrelated tasks at comparable rates, the causal isolation claim would be refuted.

Figures

Figures reproduced from arXiv: 2604.17255 by Chong Teng, Donghong Ji, Fei Li, Jiangming Yang, Li Zheng, Shuyi He, Xin Zhang, Zhuang Li.

Figure 1
Figure 1. Figure 1: Emotion recognition accuracy comparison on DailyDialogue (Li et al., 2017): external optimiza￾tion vs. emotion and rhetoric neurons steering. “Exter￾nal” denotes prompt engineering or fine-tuning. “Emo￾tion” indicates the injection of emotion neurons. “Emo￾tion+Rhetoric” denotes the injection of emotion neurons augmented with rhetoric neurons. et al., 2025; Zhong et al., 2024), and customer service (Su et … view at source ↗
Figure 2
Figure 2. Figure 2: Observation. Performance changes of emo￾tion and rhetoric tasks under traditional target neuron masking methods. predictions, nor do they enable targeted neuron￾level steering. In addition, the potential interaction between emotion and rhetoric representations in￾side LLMs is underexplored, limiting opportuni￾ties to leverage rhetorical cues to assist emotion recognition and to build more controllable syst… view at source ↗
Figure 3
Figure 3. Figure 3: The overall architecture of our framework. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Distribution of emotion and rhetoric neurons. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Experimental results of injecting rhetoric neu [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison results of different layers of mask [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Comparison results before and after neurons [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
read the original abstract

Accurate comprehension and controllable generation of emotion and rhetoric are pivotal for enhancing the reasoning capabilities of large language models (LLMs). Existing studies mostly rely on external optimizations, lacking in-depth exploration of internal representation mechanisms, thus failing to achieve fine-grained steering at the neuron level. A handful of works on neurons are confined to emotions, neglecting rhetoric neurons and their intrinsic connections. Traditional neuron masking also exhibits counterintuitive phenomena, making reliable verification of neuron functionality infeasible. To address these issues, we systematically investigate the neurons representation mechanisms and inherent associations of 6 emotion categories and 4 core rhetorical devices. We propose a neuron identification framework that integrates multi-dimensional screening, and design an adaptive masking method incorporating dynamic filtering, attenuation masking, and feedback optimization, enabling reliable causal validation of neuron functionality.Through neuron regulation, we achieve directed induction of non-target sentences and enhancement of emotion tasks via rhetoric neurons. Experiments on 5 commonly used datasets validate the effectiveness of our method, providing a novel paradigm for the fine-grained steering of emotion and rhetoric expressions in LLMs.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper claims to systematically investigate neurons associated with 6 emotion categories and 4 rhetorical devices in LLMs. It proposes a multi-dimensional screening framework for neuron identification and an adaptive masking method (dynamic filtering, attenuation masking, feedback optimization) to achieve reliable causal validation. Through neuron regulation, the authors report directed induction of non-target sentences and enhancement of emotion tasks via rhetoric neurons, with effectiveness validated on 5 datasets.

Significance. If the causal claims hold after addressing validation concerns, the work would offer a meaningful advance in mechanistic interpretability by enabling fine-grained, neuron-level steering of both emotion and rhetoric in LLMs, moving beyond external optimizations. The joint treatment of emotion-rhetoric connections and the adaptive masking design represent a potentially useful paradigm for controllable generation.

major comments (2)
  1. [Abstract] Abstract: The central claim of 'reliable causal validation of neuron functionality' and 'experiments on 5 commonly used datasets validate the effectiveness' is unsupported by any quantitative results, baselines, controls, error analysis, or specific metrics. Without these, the effectiveness of the multi-dimensional screening and adaptive masking cannot be evaluated.
  2. [Proposed method] Proposed method (neuron identification and adaptive masking): Multi-dimensional screening combined with feedback optimization in the masking procedure appears to occur on the same 5 datasets used for final evaluation. This creates a risk of post-hoc selection bias and circular validation, where apparent causal effects may partly result from tuning masks to observed performance rather than true intervention effects. Explicit held-out splits or pre-registered selection criteria are needed to substantiate the causality claims.
minor comments (1)
  1. [Abstract] The abstract does not list the specific 6 emotion categories or 4 rhetorical devices, which would clarify the scope of the investigation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our work. We address the two major comments point by point below, clarifying the quantitative support in the full manuscript and the data handling procedures while committing to revisions that strengthen the presentation of our causal validation claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim of 'reliable causal validation of neuron functionality' and 'experiments on 5 commonly used datasets validate the effectiveness' is unsupported by any quantitative results, baselines, controls, error analysis, or specific metrics. Without these, the effectiveness of the multi-dimensional screening and adaptive masking cannot be evaluated.

    Authors: We agree that the abstract would be strengthened by including concrete quantitative highlights rather than high-level claims alone. The full manuscript reports specific metrics (including accuracy and F1 improvements on emotion and rhetoric tasks), comparisons to baseline neuron masking approaches, ablation studies on each component of the adaptive masking method, and error analyses across the five datasets. We will revise the abstract to incorporate key numerical results and explicitly reference the controls and validation procedures detailed in the experiments section. revision: yes

  2. Referee: [Proposed method] Proposed method (neuron identification and adaptive masking): Multi-dimensional screening combined with feedback optimization in the masking procedure appears to occur on the same 5 datasets used for final evaluation. This creates a risk of post-hoc selection bias and circular validation, where apparent causal effects may partly result from tuning masks to observed performance rather than true intervention effects. Explicit held-out splits or pre-registered selection criteria are needed to substantiate the causality claims.

    Authors: We recognize the importance of this methodological concern for substantiating causality. Neuron identification via multi-dimensional screening was performed on training portions of the datasets, with adaptive masking effects and final performance measured on held-out test splits; feedback optimization was applied only during the intervention phase on the evaluation data. To eliminate any ambiguity, we will revise the method section to explicitly document the train/test partitioning, cross-validation folds, and any pre-defined selection criteria. If space permits, we will also report supplementary results on an additional held-out dataset to further isolate intervention effects from selection bias. revision: partial

Circularity Check

0 steps flagged

No circularity detected in empirical neuron identification pipeline

full rationale

The paper describes an empirical framework consisting of multi-dimensional screening for neuron identification followed by adaptive masking (dynamic filtering, attenuation masking, feedback optimization) and experimental validation on 5 datasets. No equations, derivations, first-principles results, or mathematical reductions appear in the provided text. The central claims rest on experimental outcomes rather than any self-definitional structure, fitted parameters renamed as predictions, or load-bearing self-citations. No step reduces by construction to its own inputs; the work is self-contained as an applied identification-plus-intervention pipeline.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no explicit free parameters, axioms, or invented entities; full manuscript required to audit any implicit assumptions about neuron independence or masking causality.

pith-pipeline@v0.9.0 · 5504 in / 1106 out tokens · 35151 ms · 2026-05-10T06:15:11.545829+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

102 extracted references · 4 canonical work pages

  1. [1]

    Issa Annamoradnejad and Gohar Zoghi. 2024. Colbert: Using bert sentence embedding in parallel neural networks for computational humor. Expert Systems with Applications, 249:123685

    2024

  2. [2]

    Vinamra Benara, Chandan Singh, John X Morris, Richard J Antonello, Ion Stoica, Alexander G Huth, and Jianfeng Gao. 2024. Crafting interpretable embeddings for language neuroscience by asking llms questions. Advances in neural information processing systems, 37:124137

    2024

  3. [3]

    Julia Birke and Anoop Sarkar. 2006. A clustering approach for nearly unsupervised recognition of nonliteral language. In 11th Conference of the European chapter of the association for computational linguistics, pages 329--336

    2006

  4. [4]

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. Advances in neural information processing systems, 33:1877--1901

    2020

  5. [5]

    Carlos Busso, Murtaza Bulut, Chi-Chun Lee, Abe Kazemzadeh, Emily Mower, Samuel Kim, Jeannette N Chang, Sungbok Lee, and Shrikanth S Narayanan. 2008. Iemocap: Interactive emotional dyadic motion capture database. Language resources and evaluation, 42(4):335--359

    2008

  6. [6]

    Santiago Castro, Devamanyu Hazarika, Ver \'o nica P \'e rez-Rosas, Roger Zimmermann, Rada Mihalcea, and Soujanya Poria. 2019. Towards multimodal sarcasm detection (an \_obviously\_ perfect paper). In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4619--4629

    2019

  7. [7]

    what do you call a dog that is incontrovertibly true? dogma

    Alessio Cocchieri, Luca Ragazzi, Paolo Italiani, Giuseppe Tagliavini, and Gianluca Moro. 2025. “what do you call a dog that is incontrovertibly true? dogma”: Testing llm generalization through humor. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 22922--22937

    2025

  8. [10]

    Yuzhe Ding, Kang He, Bobo Li, Li Zheng, Haijun He, Fei Li, Chong Teng, and Donghong Ji. 2025. Zero-shot conversational stance detection: Dataset and approaches. In Findings of the Association for Computational Linguistics: ACL 2025, pages 3221--3235

    2025

  9. [12]

    Susan R Fussell and Mallie M Moss. 2014. Figurative language in emotional communication. In Social and cognitive approaches to interpersonal communication, pages 113--141. Psychology Press

    2014

  10. [13]

    Palaash Goel, Dushyant Singh Chauhan, and Md Shad Akhtar. 2025. Target-augmented shared fusion-based multimodal sarcasm explanation generation. In Findings of the Association for Computational Linguistics: NAACL 2025, pages 8480--8493

    2025

  11. [14]

    Nabil Hossain, John Krumm, Tanvir Sajed, and Henry Kautz. 2020. Stimulating creativity with funlines: A case study of humor generation in headlines. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: System Demonstrations, pages 256--262

    2020

  12. [15]

    Zhiting Hu, Zichao Yang, Xiaodan Liang, Ruslan Salakhutdinov, and Eric P Xing. 2017. Toward controlled generation of text. In International conference on machine learning, pages 1587--1596. PMLR

    2017

  13. [16]

    Yuanxiang Huangfu, Peifeng Li, Yaxin Fan, and Qiaoming Zhu. 2025. Non-emotion-centric empathetic dialogue generation. In Proceedings of the 31st International Conference on Computational Linguistics, pages 989--999

    2025

  14. [17]

    Hiruni Maleesa Jayasinghe, Kok Wai Wong, and Anupiya Nugaliyadde. 2025. A systematic review of interpretability and explainability for speech emotion features in automatic speech emotion recognition. Pattern recognition, page 112122

    2025

  15. [18]

    Minseo Kim, Taemin Kim, Thu Hoang Anh Vo, Yugyeong Jung, and Uichin Lee. 2025. Exploring modular prompt design for emotion and mental health recognition. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, pages 1--18

    2025

  16. [19]

    David Konstan. 2007. Rhetoric and emotion. A companion to Greek rhetoric, 411:25

    2007

  17. [20]

    emotion neurons

    Jaewook Lee, Woojin Lee, Oh-Woog Kwon, and Harksoo Kim. 2025. Do large language models have “emotion neurons”? investigating the existence and role. In Findings of the Association for Computational Linguistics: ACL 2025, pages 15617--15639

    2025

  18. [21]

    Yanran Li, Hui Su, Xiaoyu Shen, Wenjie Li, Ziqiang Cao, and Shuzi Niu. 2017. Dailydialog: A manually labelled multi-turn dialogue dataset. In Proceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 986--995

    2017

  19. [22]

    Zhipeng Li, Binglin Wu, Yingyi Zhang, Xianneng Li, Kai Li, and Weizhi Chen. 2025. Cusmer: Multimodal intent recognition in customer service via data augment and llm merge. In Companion Proceedings of the ACM on Web Conference 2025, pages 3058--3062

    2025

  20. [23]

    Jian Liao, Yu Feng, Yujin Zheng, Jun Zhao, Suge Wang, and Jianxing Zheng. 2025. My words imply your opinion: Reader agent-based propagation enhancement for personalized implicit emotion analysis. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 16156--16172

    2025

  21. [24]

    Zhiwei Liu, Kailai Yang, Qianqian Xie, Tianlin Zhang, and Sophia Ananiadou. 2024. Emollms: A series of emotional large language models and annotation tools for comprehensive affective analysis. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 5487--5496

    2024

  22. [25]

    Michael Mohler, Mary Brunson, Bryan Rink, and Marc Tomlinson. 2016. Introducing the lcc metaphor datasets. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16), pages 4221--4227

    2016

  23. [26]

    Shereen Oraby, Vrindavan Harrison, Lena Reed, Ernesto Hernandez, Ellen Riloff, and Marilyn Walker. 2016. Creating and characterizing a diverse corpus of sarcasm in dialogue. In Proceedings of the 17th annual meeting of the special interest group on discourse and dialogue, pages 31--41

    2016

  24. [27]

    Jiao Ou, Junda Lu, Che Liu, Yihong Tang, Fuzheng Zhang, Di Zhang, and Kun Gai. 2024. Dialogbench: Evaluating llms as human-like dialogue systems. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 6137--6170

    2024

  25. [28]

    Alec Radford, Rafal Jozefowicz, and Ilya Sutskever. 2018. Learning to generate reviews and discovering sentiment

    2018

  26. [29]

    Rohit Rajakumar and Mihai Boicu. 2025. Evaluating lightweight transformer models for rhetorical element classification in student essays. Journal of Student-Scientists' Research, 7

    2025

  27. [30]

    Elvis Saravia, Hsien-Chi Toby Liu, Yen-Hao Huang, Junlin Wu, and Yi-Shin Chen. 2018. Carer: Contextualized affect representations for emotion recognition. In Proceedings of the 2018 conference on empirical methods in natural language processing, pages 3687--3697

    2018

  28. [31]

    Stephen D Smith and Jennifer Kornelsen. 2011. Emotion-dependent responses in spinal cord neurons: a spinal fmri study. NeuroImage, 58(1):269--274

    2011

  29. [32]

    Kevin Stowe, Tuhin Chakrabarty, Nanyun Peng, Smaranda Muresan, and Iryna Gurevych. 2021. Metaphor generation with conceptual mappings. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 6724--6736

    2021

  30. [33]

    Hanchen Su, Wei Luo, Yashar Mehdad, Wei Han, Elaine Liu, Wayne Zhang, Mia Zhao, and Joy Zhang. 2025. Llm-friendly knowledge representation for customer support. In Proceedings of the 31st International Conference on Computational Linguistics: Industry Track, pages 496--504

    2025

  31. [35]

    Enrica Troiano, Carlo Strapparava, G \"o zde \"O zbal, and Serra Sinem Tekiro g lu. 2018. A computational exploration of exaggeration. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 3296--3304

    2018

  32. [36]

    Yuchi Wang, Junliang Guo, Jianhong Bai, Runyi Yu, Tianyu He, Xu Tan, Xu Sun, and Jiang Bian. 2025. Instructavatar: Text-guided emotion and motion control for avatar generation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 8132--8140

    2025

  33. [37]

    Zhikun Wu, Thomas Weber, and Florian M \"u ller. 2025. One does not simply meme alone: Evaluating co-creativity between llms and humans in the generation of humor. In Proceedings of the 30th International Conference on Intelligent User Interfaces, pages 1082--1092

    2025

  34. [38]

    Zhenyu Xu, Hailin Xu, Zhouyang Lu, Yingying Zhao, Rui Zhu, Yujiang Wang, Mingzhi Dong, Yuhu Chang, Qin Lv, Robert P Dick, et al. 2024. Can large language models be good companions? an llm-based eyewear system with conversational common ground. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 8(2):1--41

    2024

  35. [39]

    Senqi Yang, Dongyu Zhang, Jing Ren, Ziqi Xu, Xiuzhen Jenny Zhang, Yiliao Song, Hongfei Lin, and Feng Xia. 2025 a . Cultural bias matters: A cross-cultural benchmark dataset and sentiment-enriched model for understanding multimodal metaphors. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers),...

    2025

  36. [40]

    Yang Yang, Xunde Dong, and Yupeng Qiang. 2025 b . Mse-adapter: A lightweight plugin endowing llms with the capability to perform multimodal sentiment analysis and emotion recognition. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 25642--25650

    2025

  37. [41]

    Jiecheng Zhang, CL Philip Chen, Shuzhen Li, and Tong Zhang. 2025 a . Incongruity-aware tension field network for multi-modal sarcasm detection. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 14499--14508

    2025

  38. [42]

    Xinjie Zhang, Tenggan Zhang, Lei Sun, Jinming Zhao, and Qin Jin. 2025 b . Exploring interpretability in deep learning for affective computing: a comprehensive review. ACM Transactions on Multimedia Computing, Communications and Applications

    2025

  39. [43]

    Yunxiang Zhang and Xiaojun Wan. 2022. Mover: Mask, over-generate and rank for hyperbole generation. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 6018--6030

    2022

  40. [44]

    Li Zheng, Boyu Chen, Hao Fei, Fei Li, Shengqiong Wu, Lizi Liao, Donghong Ji, and Chong Teng. 2024 a . Self-adaptive fine-grained multi-modal data augmentation for semi-supervised muti-modal coreference resolution. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 8576--8585

    2024

  41. [45]

    Li Zheng, Hao Fei, Lei Chen, Bobo Li, Fei Li, Chong Teng, Liang Zhao, and Donghong Ji. 2025 a . Stpar: A structure-aware triaffine parser for screenplay character coreference resolution. Transactions of the Association for Computational Linguistics, 13:923--937

    2025

  42. [46]

    Li Zheng, Hao Fei, Ting Dai, Zuquan Peng, Fei Li, Huisheng Ma, Chong Teng, and Donghong Ji. 2025 b . Multi-granular multimodal clue fusion for meme understanding. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 26057--26065

    2025

  43. [47]

    Li Zheng, Hao Fei, Fei Li, Bobo Li, Lizi Liao, Donghong Ji, and Chong Teng. 2024 b . Reverse multi-choice dialogue commonsense inference with graph-of-thought. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 19688--19696

    2024

  44. [48]

    Li Zheng, Donghong Ji, Fei Li, Hao Fei, Shengqiong Wu, Jingye Li, Bobo Li, and Chong Teng. 2025 c . Ecqed: emotion-cause quadruple extraction in dialogs. IEEE Transactions on Audio, Speech and Language Processing

    2025

  45. [49]

    Li Zheng, Tengyue Song, Yuzhe Ding, Xiaorui Wu, Fei Li, Dongdong Xie, Jinbo Li, Chong Teng, and Donghong Ji. 2025 d . Improving emotion and intent understanding in multimodal conversations with progressive interaction. IEEE Transactions on Affective Computing

    2025

  46. [50]

    Li Zheng, Sihang Wang, Hao Fei, Zuquan Peng, Fei Li, Jianming Fu, Chong Teng, and Donghong Ji. 2025 e . Enhancing hyperbole and metaphor detection with their bidirectional dynamic interaction and emotion knowledge. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL'25)

    2025

  47. [51]

    Shanshan Zhong, Zhongzhan Huang, Shanghua Gao, Wushao Wen, Liang Lin, Marinka Zitnik, and Pan Zhou. 2024. Let's think outside the box: Exploring leap-of-thought in large language models with creative humor generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13246--13257

    2024

  48. [52]

    Learning to generate reviews and discovering sentiment , author=

  49. [53]

    emotion neurons

    Do large language models have “emotion neurons”? investigating the existence and role , author=. Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    2025

  50. [54]

    arXiv preprint arXiv:2505.16491 , year=

    LLaMAs Have Feelings Too: Unveiling Sentiment and Emotion Representations in LLaMA Models Through Probing , author=. arXiv preprint arXiv:2505.16491 , year=

    work page arXiv

  51. [55]

    IMO : Greedy Layer-Wise Sparse Representation Learning for Out-of-Distribution Text Classification with Pre-trained Models

    Feng, Tao and Qu, Lizhen and Li, Zhuang and Zhan, Haolan and Hua, Yuncheng and Haf, Reza. IMO : Greedy Layer-Wise Sparse Representation Learning for Out-of-Distribution Text Classification with Pre-trained Models. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024. doi:10.18653/v1/2024.acl...

    work page doi:10.18653/v1/2024.acl-long.144 2024

  52. [56]

    Proceedings of the 2018 conference on empirical methods in natural language processing , pages=

    CARER: Contextualized affect representations for emotion recognition , author=. Proceedings of the 2018 conference on empirical methods in natural language processing , pages=

    2018

  53. [57]

    Rephrase and respond: Let large language models ask better questions for themselves

    Rephrase and respond: Let large language models ask better questions for themselves , author=. arXiv preprint arXiv:2311.04205 , year=

    work page arXiv

  54. [58]

    Journal of Student-Scientists' Research , volume=

    Evaluating Lightweight Transformer Models for Rhetorical Element Classification in Student Essays , author=. Journal of Student-Scientists' Research , volume=

  55. [59]

    Advances in neural information processing systems , volume=

    Crafting interpretable embeddings for language neuroscience by asking LLMs questions , author=. Advances in neural information processing systems , volume=

  56. [60]

    Expert Systems with Applications , volume=

    ColBERT: Using BERT sentence embedding in parallel neural networks for computational humor , author=. Expert Systems with Applications , volume=. 2024 , publisher=

    2024

  57. [61]

    Proceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 1: Long Papers) , pages=

    DailyDialog: A Manually Labelled Multi-turn Dialogue Dataset , author=. Proceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 1: Long Papers) , pages=

  58. [62]

    Proceedings of the 17th annual meeting of the special interest group on discourse and dialogue , pages=

    Creating and characterizing a diverse corpus of sarcasm in dialogue , author=. Proceedings of the 17th annual meeting of the special interest group on discourse and dialogue , pages=

  59. [63]

    Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing , pages=

    A computational exploration of exaggeration , author=. Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing , pages=

    2018

  60. [64]

    11th Conference of the European chapter of the association for computational linguistics , pages=

    A clustering approach for nearly unsupervised recognition of nonliteral language , author=. 11th Conference of the European chapter of the association for computational linguistics , pages=

  61. [65]

    Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages=

    Dialogbench: Evaluating llms as human-like dialogue systems , author=. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages=

    2024

  62. [66]

    Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    Zero-Shot Conversational Stance Detection: Dataset and Approaches , author=. Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    2025

  63. [67]

    Transactions of the Association for Computational Linguistics , volume=

    STPar: A Structure-Aware Triaffine Parser for Screenplay Character Coreference Resolution , author=. Transactions of the Association for Computational Linguistics , volume=. 2025 , publisher=

    2025

  64. [68]

    Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=

    Can large language models be good companions? An LLM-based eyewear system with conversational common ground , author=. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies , volume=. 2024 , publisher=

    2024

  65. [69]

    Proceedings of the 30th International Conference on Intelligent User Interfaces , pages=

    One does not simply meme alone: Evaluating co-creativity between LLMs and humans in the generation of humor , author=. Proceedings of the 30th International Conference on Intelligent User Interfaces , pages=

  66. [70]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Let's think outside the box: Exploring leap-of-thought in large language models with creative humor generation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  67. [71]

    Proceedings of the 32nd ACM International Conference on Multimedia , pages=

    Self-Adaptive Fine-grained Multi-modal Data Augmentation for Semi-supervised Muti-modal Coreference Resolution , author=. Proceedings of the 32nd ACM International Conference on Multimedia , pages=

  68. [72]

    Proceedings of the 31st International Conference on Computational Linguistics: Industry Track , pages=

    Llm-friendly knowledge representation for customer support , author=. Proceedings of the 31st International Conference on Computational Linguistics: Industry Track , pages=

  69. [73]

    Companion Proceedings of the ACM on Web Conference 2025 , pages=

    CuSMer: Multimodal Intent Recognition in Customer Service via Data Augment and LLM Merge , author=. Companion Proceedings of the ACM on Web Conference 2025 , pages=

    2025

  70. [74]

    A companion to Greek rhetoric , volume=

    Rhetoric and emotion , author=. A companion to Greek rhetoric , volume=. 2007 , publisher=

    2007

  71. [75]

    Social and cognitive approaches to interpersonal communication , pages=

    Figurative language in emotional communication , author=. Social and cognitive approaches to interpersonal communication , pages=. 2014 , publisher=

    2014

  72. [76]

    International conference on machine learning , pages=

    Toward controlled generation of text , author=. International conference on machine learning , pages=. 2017 , organization=

    2017

  73. [77]

    IEEE Transactions on Knowledge and Data Engineering , year=

    DECIDER: A Dual-System Rule-Controllable Decoding Framework for Language Generation , author=. IEEE Transactions on Knowledge and Data Engineering , year=

  74. [78]

    Advances in neural information processing systems , volume=

    Language models are few-shot learners , author=. Advances in neural information processing systems , volume=

  75. [79]

    Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

    Emollms: A series of emotional large language models and annotation tools for comprehensive affective analysis , author=. Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

  76. [80]

    NeuroImage , volume=

    Emotion-dependent responses in spinal cord neurons: a spinal fMRI study , author=. NeuroImage , volume=. 2011 , publisher=

    2011

  77. [81]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    My words imply your opinion: Reader agent-based propagation enhancement for personalized implicit emotion analysis , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  78. [82]

    Pattern recognition , pages=

    A systematic review of interpretability and explainability for speech emotion features in automatic speech emotion recognition , author=. Pattern recognition , pages=. 2025 , publisher=

    2025

  79. [83]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Instructavatar: Text-guided emotion and motion control for avatar generation , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  80. [84]

    Proceedings of the 31st International Conference on Computational Linguistics , pages=

    Non-Emotion-Centric Empathetic Dialogue Generation , author=. Proceedings of the 31st International Conference on Computational Linguistics , pages=

Showing first 80 references.