Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Hidden in Plain Sight: Evaluation of the Deception Detection Capabilities of LLMs in Multimodal Settings

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Fine-tuned open-source LLMs reach state-of-the-art accuracy on textual deception detection, while multimodal LLMs fail to exploit the visual and acoustic cues that classical models use.

desk verdict A broad empirical survey of LLM/LMM deception detection whose SOTA claim evaporates on the deconfounded control, but the data and analysis are worth referee time. read the letter →

arxiv 2506.09424 v1 pith:V7KWRUOS submitted 2025-06-11 cs.CL

Md Messal Monem Miah , Adrita Anika , Xi Shi , Ruihong Huang This is my paper
classification cs.CL
keywords deceptiondetectionlargelanguagemodelsmultimodalfew-shotpromptselectionin-contextlearningfine-tuningopinionspamreasoninggeneration
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to determine whether modern LLMs, both text-only and multimodal, can detect deception in courtroom interviews, controlled interpersonal videos, and online hotel reviews. Its central claim is that fine-tuned open-source LLMs beat classical deep-learning baselines on text-only deception detection, reaching new state-of-the-art results, whereas vision-language and audio-language models fall behind simpler CNN and BiLSTM baselines when the task requires reading faces, voices, and body language. The paper argues this multimodal gap stems from low temporal resolution and pretraining objectives that describe scenes rather than track deception-relevant micro-behaviors, and it shows that asking LLMs to reason before classifying does not reliably improve accuracy. A sympathetic reader would care because the results stake out where current LLM technology can be trusted in a high-stakes task and where it cannot, with direct implications for deployment in security screening, consumer protection, and legal settings.

What carries the argument

The load-bearing instrument is the experimental protocol itself: three domain-diverse datasets evaluated under a shared stratified 4-fold design, with four classical baselines (RoBERTa-ft, BiLSTM+Attention, CNN, and PECL) built on fixed pretrained encoders (RoBERTa for text, Whisper for audio, and CLIP with 30 fps sampling for video), pitted against LLMs in zero-shot and few-shot setups with three example-selection strategies (random, sim-top, sim-pair), two response strategies (direct label versus post-hoc reasoning), and fine-tuning of LLaMA 3.1, Gemma 2, and Qwen2VL. The supporting machinery is the cue-level analysis of model reasoning, which classifies justifications by specificity, vagueness, filler words, justification, and emotion, linking LLM judgments to classical verbal lie-detection theories and exposing systematic biases such as equating vagueness with deception in emotionally charged speech.

What would settle it

Retrain the text baselines on OpSpam using the paper's stated recipe plus standard early stopping and hyperparameter search, or run the same prompts through released baseline checkpoints; if a properly tuned RoBERTa-ft or CNN reaches or exceeds the 92.25 F1 reported for fine-tuned LLaMA 3.1, the state-of-the-art claim falls.

Watch

Extended reading notes

Core claim

Across three deception datasets spanning courtroom testimony (RLTD), instructed interpersonal deception (MU3D), and hotel review spam (OpSpam), the paper establishes an empirical benchmark with a clear split: on text-only deception detection, fine-tuned open-source LLMs are state of the art. LLaMA 3.1 reaches 92.25 F1 on OpSpam, beating the paper's reproduced RoBERTa-ft baseline (88.09 F1), BiLSTM+Attention (90.45 F1), and CNN (86.36 F1), with Gemma 2 close behind at 90.18 F1, and both fine-tuned models also beat the baselines on the harder Prolific 'Pure Veracity' data where truthful and deceptive reviews come from the same platform. On multimodal data the picture reverses: zero-shot LMMs using video stay at or near chance (LLaVA-NeXT reaches only 52.06 accuracy on RLTD video), and even after fine-tuning they trail the CNN baselines that process video at 10 frames per second, which the paper attributes to the LMMs' coarse temporal sampling (Qwen2VL is pretrained at 2 fps) and to their trained behavior of describing scenes rather than tracking annotated behavioral cues such as repeated nods or gaze direction. The paper also finds that similarity-based in-context example selection consistently beats random selection, that this can leak dataset label biases in OpSpam, and that auxiliary features such as non-verbal gesture names, video summaries, and audio summaries give only marginal and inconsistent gains without fine-tuning.

Load-bearing premise

The claim that fine-tuned LLMs are state of the art for textual deception detection rests on the paper's in-house reproductions of the classical baselines being strong and faithful, since those baselines are the yardstick and were trained with minimal documented detail and no released code.

Editorial extensions

If this is right

  • On text-only deception detection, fine-tuned open-source LLMs set new state-of-the-art results on OpSpam and outperform classical baselines on the harder Prolific 'Pure Veracity' set, so text-based screening pipelines can plausibly be built on fine-tuned LLMs.
  • Multimodal LLMs currently underperform classical video and audio baselines on both real and scripted deception, so deployments that rely on visual or acoustic cues should not assume LMMs can replace trained CNNs and recurrent models.
  • Asking models to generate reasoning, whether post-hoc or chain-of-thought, does not consistently improve accuracy and often degrades it, so interpretability is a measurable cost that designers must trade against performance.
  • Similarity-based few-shot example selection improves performance over random selection across all three datasets, but in OpSpam it amplifies dataset-specific label biases, meaning curated in-context examples need auditing before deployment.
  • Fine-tuning with non-verbal gesture features or generated video and audio summaries yields only marginal gains, indicating that auxiliary cues are not yet a reliable route to better multimodal deception detection.

Reading between the lines

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

  • Because the text state-of-the-art claim is measured against the paper's own baseline reproductions, and those baselines were trained with minimal documented detail and no released code, the SOTA result should be re-checked against released checkpoints or a standard tuning protocol before being used in procurement decisions.
  • The temporal-resolution mismatch that the paper identifies (LMMs at 2 fps or 16 frames per video versus CNNs at 10 fps) suggests a concrete testable fix: fine-tune video LLMs on higher frame-rate inputs or add a lightweight temporal adapter that tracks face and hand trajectories, rather than relying on current vision encoders' frame sampling.
  • The cue analysis implies a natural division of labor: LLMs already mirror classical verbal lie-detection heuristics, so a hybrid system combining LLM text judgment with classical signal-processing audio and video models could outperform either approach alone on multimodal datasets.
  • The OpSpam sim-top finding warns that retrieval-based few-shot prompting can leak label-distribution artifacts into predictions, so a system designer should measure the label balance of retrieved examples before trusting few-shot F1 gains as evidence of genuine deception detection ability.
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 / 5 minor

Summary. The paper presents an empirical evaluation of LLMs and LMMs for deception detection on three datasets (RLTD, MU3D, OpSpam) plus a Prolific control dataset intended to de-confound the OpSpam data-origin bias. The authors compare zero-shot and few-shot prompting, random versus similarity-based in-context example selection, direct label generation versus post-hoc reasoning, and auxiliary non-verbal features, video summaries, and audio summaries. They also fine-tune LLaMA 3.1, Gemma 2, and Qwen2VL. The central claims are that fine-tuned LLMs achieve state-of-the-art performance on textual deception detection and that LMMs underperform on multimodal deception cues.

Significance. If substantiated, the paper would provide a useful reference point for LLM-based deception detection across textual, audio, and visual modalities, and its analysis of reasoning cues is informative. The inclusion of the Prolific control dataset (Table 8, Appendix I) is a genuine strength, as is the candid discussion of the OpSpam data-origin confound. The paper also ships a reasonably broad model coverage and several practical findings, e.g., that post-hoc reasoning does not consistently beat direct label prediction. However, the headline SOTA claim is not currently supported by the reported statistics: no confidence intervals, error bars, or significance tests are given, and the comparison is only against in-house reproduced baselines rather than published results. The empirical scope is a benchmark study rather than a theoretical contribution, so circularity is not a concern, but the statistical grounding must be strengthened for the central claim to stand.

major comments (4)
  1. [Section 5, Table 1] The claim that fine-tuned LLaMA 3.1 achieves state-of-the-art performance on OpSpam rests on a 1.79-point F1 margin (92.24 vs. 90.45) over an in-house BiLSTM+Attention baseline, with no error bars, confidence intervals, or significance tests reported. On the Prolific de-confounded control (Table 8), the margin over RoBERTa-ft shrinks to 0.66 F1 (73.54 vs. 72.88). Given the small absolute differences and 3-seed averaging, the observed advantage may be sampling noise. Please report per-fold or per-seed results and perform a paired significance test (e.g., McNemar) on the relevant comparisons, or temper the SOTA claim accordingly.
  2. [Section 5 and Appendix J] The 'state-of-the-art' label is asserted only against four baselines reproduced by the authors, and no comparison is made to published SOTA numbers on OpSpam or the other datasets. If the reproduced baselines are undertrained, the claimed LLM superiority is an artifact of a weak comparator. Appendix J.1 gives only a learning rate and epoch count, with no code released. Please provide more complete training details, release code, or compare against published results from prior work on these datasets.
  3. [Appendix J.2 and Section 4.1.2] The few-shot configuration is selected by exploring n in {2, 4, 6, 8, 10} and reporting the best results. If this selection is performed on the test folds, the reported few-shot numbers are optimistically biased. Please clarify whether the selection was made on a development split, and if not, either provide corrected numbers or report the full curve with error bars so the reader can assess the selection effect.
  4. [Table 3 and Section 5.3] The conclusion that sim-top is superior on OpSpam due to label-imbalanced retrieval is based on a small F1 difference and lacks significance testing. Table 3 shows, e.g., LLaMA 3.1 sim-top F1 62.47 vs. sim-pair 57.78, but the corresponding accuracy difference is 4.89 points on 1,600 examples; please report whether the difference is consistent across seeds and folds, and provide a statistical test or confidence intervals.
minor comments (5)
  1. [Appendix I] The sentence beginning 'This setting is particularly challenging' is repeated verbatim in consecutive sentences; please remove the duplication.
  2. [Appendix J.2] The phrase 'All the results reported are an average of 3 seeds' is useful, but it would be clearer to also state whether the same seed set is used for all models and whether the reported F1 is macro or weighted.
  3. [Table 4] The GPT-4o row for non-verbal features on MU3D contains empty cells; please add '--' or a note explaining why those configurations were not run.
  4. [Section 1] There is a typographical error: 'and and generate step-by-step reasoning' should read 'and generate step-by-step reasoning'.
  5. [Appendix K] The statement about using ChatGPT for writing assistance is appreciated for transparency, but it may be more appropriate in the acknowledgements section than as a standalone appendix.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an external benchmark evaluation whose claims rest on public datasets and pretrained models, not on fitted parameters or self-citation chains.

full rationale

This is an empirical evaluation, not a derivation, so the standard circularity patterns do not apply. The central claims—fine-tuned LLMs reach high F1 on OpSpam and LMMs underperform on RLTD/MU3D—are measured against public datasets (RLTD, MU3D, OpSpam) and publicly released pretrained models, with stratified 4-fold cross-validation for fine-tuning. No parameter is fitted to a subset of data and then reported as a prediction of the same quantity, and no equation defines one claimed result in terms of another. The one self-citation (Miah et al., 2023) is used only to support that Whisper-base is robust in audio tasks and is not load-bearing for any headline claim. The acknowledged data-origin confound in OpSpam is mitigated by an external Prolific control, which is additional independent evidence rather than circular support. The concerns raised by the reader—undertrained baselines, best-of-n few-shot selection on the test folds, and absence of significance tests—are methodological validity issues that could weaken the SOTA comparison, but they do not make the claim equivalent to its inputs by construction. Therefore no circular step is identified.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

Two hand-chosen design parameters and three unverified premises carry the results: the best-of-n few-shot selection and 16-feature non-verbal subset are free choices that move scores; the RLTD label conflation, the fidelity of in-house baselines, and the reliability of AI-generated summaries are domain assumptions. If any of these is wrong, the corresponding conclusions weaken.

free parameters (2)
  • In-context example count n_shot = Best value chosen from {2,4,6,8,10} per model/dataset; exact values not reported
    Appendix J.2 reports results as the best across n values; selecting on the test folds inflates reported performance.
  • Non-verbal feature subset = 16 of 40 MUMIN features
    Section 4.1.3 and Appendix B: authors hand-filtered 40 annotated features to 16 'most relevant' ones, which may bias the auxiliary-feature comparison.
assumptions (3)
  • domain assumption RLTD trial-outcome labels (guilty = deceptive, not guilty = truthful) reflect true deception.
    Section 3.2 states labels are assigned based on trial outcomes; this conflates legal verdicts with lying behavior. All RLTD results inherit this assumption.
  • domain assumption The in-house baseline implementations are faithful, strong reproductions of prior work (RoBERTa-ft, BiLSTM+Attention, CNN, PECL).
    Section 3.3 describes the baselines; no code is released and hyperparameters are minimal (Appendix J.1), so the comparisons assume the baselines are not under-trained.
  • domain assumption LLM-generated video and audio summaries preserve deception-relevant cues.
    Section 4.1.3 uses summaries from LLaVA-NeXT and Qwen2-Audio as auxiliary inputs; if these summaries omit micro-expressions or prosody, the multimodal evaluation is weakened. The authors themselves observe LLaVA misses such cues in Section 5.1, partially contradicting this assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hidden in Plain Sight: Evaluation of the Deception Detection Capabilities of LLMs in Multimodal Settings." pith.science (2026). https://pith.science/paper/V7KWRUOS

@misc{pith2026250609424,
  author       = {Pith},
  title        = {Pith review of: Hidden in Plain Sight: Evaluation of the Deception Detection Capabilities of LLMs in Multimodal Settings},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/V7KWRUOS}},
  note         = {Machine review of arXiv:2506.09424}
}
read the original abstract

Detecting deception in an increasingly digital world is both a critical and challenging task. In this study, we present a comprehensive evaluation of the automated deception detection capabilities of Large Language Models (LLMs) and Large Multimodal Models (LMMs) across diverse domains. We assess the performance of both open-source and commercial LLMs on three distinct datasets: real life trial interviews (RLTD), instructed deception in interpersonal scenarios (MU3D), and deceptive reviews (OpSpam). We systematically analyze the effectiveness of different experimental setups for deception detection, including zero-shot and few-shot approaches with random or similarity-based in-context example selection. Our results show that fine-tuned LLMs achieve state-of-the-art performance on textual deception detection tasks, while LMMs struggle to fully leverage cross-modal cues. Additionally, we analyze the impact of auxiliary features, such as non-verbal gestures and video summaries, and examine the effectiveness of different prompting strategies, including direct label generation and chain-of-thought reasoning. Our findings provide key insights into how LLMs process and interpret deceptive cues across modalities, highlighting their potential and limitations in real-world deception detection applications.

Figures

Figures reproduced from arXiv: 2506.09424 by the authors.

Figure 1
Figure 1. Average number of matching examples in 10-shot for sim-top strategy. [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. F1 score across different beam sizes on RLTD dataset [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗
Figure 3
Figure 3. F1 score across n-shots in few shot learning [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Examples of LLM Reasoning on RLTD Dataset [PITH_FULL_IMAGE:figures/full_fig_p019_4.png]
Figure 5
Figure 5. Figure 5: Examples of LLM Reasoning on MU3D and OpSpam Dataset [PITH_FULL_IMAGE:figures/full_fig_p020_5.png]

Discussion (0). Sign in 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. Semantics of Subterfuge: Benchmarking Legal Deception Detection Against General-domain State-of-the-Art

    cs.CL 2026-07 conditional novelty 5.0 of 10

    On deception detection benchmarks, fine-tuned transformers beat LLMs on data-rich datasets, few-shot GPT-4o wins the small legal corpus, and chain-of-thought prompting frequently reduces F1.

Reference graph

Works this paper leans on

68 extracted references · 34 canonical work pages · cited by 1 Pith paper

  1. [1]

    Jens Allwood, Loredana Cerrato, Kristiina Jokinen, Costanza Navarretta, and Patrizia Paggio. 2007. http://www.jstor.org/stable/30204706 The mumin coding scheme for the annotation of feedback, turn management and sequencing phenomena . Language Resources and Evaluation, 41(3/4):273--287

    work page arXiv 2007

  2. [2]

    Burgoon, Chao Chen, Norah E

    Chongyang Bai, Maksim Bolonkin, Judee K. Burgoon, Chao Chen, Norah E. Dunbar, Bharat Singh, V. S. Subrahmanian, and Zhe Wu. 2019. https://api.semanticscholar.org/CorpusID:160009724 Automatic long-term deception detection in group interaction videos . 2019 IEEE International Conference on Multimedia and Expo (ICME), pages 1600--1605

    work page 2019

  3. [3]

    Bakdash, Murat Kantarcioglu, Daniel C

    Vibha Belavadi, Yan Zhou, Jonathan Z. Bakdash, Murat Kantarcioglu, Daniel C. Krawczyk, Linda Nguyen, Jelena Rakic, and Bhavani Thuriasingham. 2020. https://doi.org/10.1109/TPS-ISA50397.2020.00023 Multimodal deception detection: Accuracy, applicability and generalizability . In 2020 Second IEEE International Conference on Trust, Privacy and Security in Int...

    arXiv 2020

  4. [4]

    Tuck, Rakesh M

    Dainis Boumber, Bryan E. Tuck, Rakesh M. Verma, and Fatima Zahra Qachfar. 2024. https://doi.org/10.1145/3643651.3659898 Llms for explainable few-shot deception detection . In Proceedings of the 10th ACM International Workshop on Security and Privacy Analytics, IWSPA '24, page 37–47, New York, NY, USA. Association for Computing Machinery

    arXiv 2024

  5. [5]

    Buller and Judee K

    David B. Buller and Judee K. Burgoon. 1996. https://doi.org/10.1111/j.1468-2885.1996.tb00127.x Interpersonal deception theory . Communication Theory, 6(3):203--242

    work page arXiv 1996

  6. [6]

    Charles F

    Jr. Charles F. Bond and Bella M. DePaulo. 2006. https://doi.org/10.1207/s15327957pspr1003\_2 Accuracy of deception judgments . Personality and Social Psychology Review, 10(3):214--234. PMID: 16859438

    work page doi:10.1207/s15327957pspr1003 2006

  7. [7]

    Safa Chebbi and Sofia Ben Jebara. 2021. https://api.semanticscholar.org/CorpusID:237792741 Deception detection using multimodal fusion approaches . Multimedia Tools and Applications, 82:13073--13102

    work page 2021

  8. [8]

    Zebang Cheng, Zhi-Qi Cheng, Jun-Yan He, Kai Wang, Yuxiang Lin, Zheng Lian, Xiaojiang Peng, and Alexander Hauptmann. 2024. https://proceedings.neurips.cc/paper_files/paper/2024/file/c7f43ada17acc234f568dc66da527418-Paper-Conference.pdf Emotion-llama: Multimodal emotion recognition and reasoning with instruction tuning . In Advances in Neural Information Pr...

    work page 2024

Show all 68 references
  1. [9]

    Yunfei Chu, Jin Xu, Qian Yang, Haojie Wei, Xipin Wei, Zhifang Guo, Yichong Leng, Yuanjun Lv, Jinzheng He, Junyang Lin, Chang Zhou, and Jingren Zhou. 2024. https://arxiv.org/abs/2407.10759 Qwen2-audio technical report . Preprint, arXiv:2407.10759

    2024 arXiv

  2. [10]

    Yunfei Chu, Jin Xu, Xiaohuan Zhou, Qian Yang, Shiliang Zhang, Zhijie Yan, Chang Zhou, and Jingren Zhou. 2023. https://api.semanticscholar.org/CorpusID:265157993 Qwen-audio: Advancing universal audio understanding via unified large-scale audio-language models . ArXiv, abs/2311.07919

    2023 arXiv

  3. [11]

    Chi, Jeff Dean, Jacob Devlin, Adam Roberts, Denny Zhou, Quoc V

    Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Alex Castro-Ros, Marie Pellat, Kevin Robinso...

    2022 arXiv

  4. [12]

    Mingyu Ding, An Zhao, Zhiwu Lu, Tao Xiang, and Ji-Rong Wen. 2019. https://doi.org/10.1109/CVPR.2019.00799 Face-focused cross-stream network for deception detection in videos . In 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 7794--7803

    2019

  5. [13]

    Cheng Fan, Heming Zhao, Xueqin Chen, Xiaohe Fan, and Shuxi Chen. 2015. https://doi.org/10.1109/ICICIP.2015.7388181 Distinguishing deception from non-deception in chinese speech

    2015

  6. [14]

    Tiantian Feng and Shrikanth Narayanan. 2023. https://doi.org/10.1109/ACII59096.2023.10388152 PEFT-SER: On the Use of Parameter Efficient Transfer Learning Approaches For Speech Emotion Recognition Using Pre-trained Speech Models . In 2023 11th International Conference on Affec...

    2023

  7. [15]

    Tommaso Fornaciari, Leticia Cagnina, Paolo Rosso, and Massimo Poesio. 2020. https://doi.org/10.1007/s10579-020-09486-5 Fake opinion detection: how similar are crowdsourced datasets to real data? Language Resources and Evaluation, 54

    2020 doi

  8. [16]

    Shuai Gao, Lin Chen, Yuancheng Fang, Shengbing Xiao, Hui Li, Xuezhi Yang, and Rencheng Song. 2024. https://doi.org/10.1109/OJCS.2024.3485688 Video-based deception detection via capsule network with channel-wise attention and supervised contrastive learning . IEEE Open Journal ...

    2024

  9. [17]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Art...

    2024 arXiv

  10. [18]

    Xiaobao Guo, Nithish Muthuchamy Selvaraj, Zitong Yu, Adams Wai-Kin Kong, Bingquan Shen, and Alex Chichung Kot. 2023. https://api.semanticscholar.org/CorpusID:257663960 Audio-visual deception detection: Dolos dataset and parameter-efficient crossmodal learning . 2023 IEEE/CVF I...

    2023

  11. [19]

    Xiaobao Guo, Zitong Yu, Nithish Muthuchamy Selvaraj, Bingquan Shen, Adams Wai-Kin Kong, and Alex C. Kot. 2024. https://arxiv.org/abs/2405.06995 Benchmarking cross-domain audio-visual deception detection . Preprint, arXiv:2405.06995

    2024 arXiv

  12. [20]

    Viresh Gupta, Mohit Agarwal, Manik Arora, Tanmoy Chakraborty, Richa Singh, and Mayank Vatsa. 2019. https://doi.org/10.1109/CVPRW.2019.00016 Bag-of-lies: A multimodal dataset for deception detection . In 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Worksh...

    2019

  13. [21]

    Chen, and Ai Ti Aw

    Yingxu He, Zhuohan Liu, Shuo Sun, Bin Wang, Wenyu Zhang, Xunlong Zou, Nancy F. Chen, and Ai Ti Aw. 2025. https://arxiv.org/abs/2412.09818 Meralion-audiollm: Bridging audio and language with large language models . Preprint, arXiv:2412.09818

    2025 arXiv

  14. [22]

    Brenier, Fiona Enos, Sarah Friedman, Susan Gilman, Cynthia Girand, Marc Graciarena, Andreas Kathol, Laura Michaelis, Blanton L

    Julia Hirschberg, Stefan Benus, Jason M. Brenier, Fiona Enos, Sarah Friedman, Susan Gilman, Cynthia Girand, Marc Graciarena, Andreas Kathol, Laura Michaelis, Blanton L. Pellom, Elizabeth Shriberg, and Andreas Stolcke. 2005. https://doi.org/10.21437/Interspeech.2005-580 Disting...

    2005 doi

  15. [23]

    Shun-Wen Hsiao and Cheng-Yuan Sun. 2022. https://doi.org/10.1109/BigData55660.2022.10020331 Attention-aware multi-modal rnn for deception detection . In 2022 IEEE International Conference on Big Data (Big Data), pages 3593--3596

    2022

  16. [24]

    Loukas Ilias, Felix Soldner, and Bennett Kleinberg. 2022. https://arxiv.org/abs/2210.03080 Explainable verbal deception detection using transformers . Preprint, arXiv:2210.03080

    2022 arXiv

  17. [25]

    Manvi Kamboj, Christian Hessler, Priyanka Asnani, Kais Riani, and Mohamed Abouelenien. 2021. https://doi.org/10.1109/MMUL.2020.3048044 Multimodal political deception detection . IEEE MultiMedia, 28(1):94--102

    2021

  18. [26]

    Hamid Karimi, Jiliang Tang, and Yanen Li. 2018. https://doi.org/10.1109/BigData.2018.8621909 Toward end-to-end deception detection in videos . In 2018 IEEE International Conference on Big Data (Big Data), pages 1278--1283

    2018

  19. [27]

    Mohan Karnati, Ayan Seal, Anis Yazidi, and Ondrej Krejcar. 2022. https://doi.org/10.1109/TCDS.2021.3086011 Lienet: A deep convolution neural network framework for detecting deception . IEEE Transactions on Cognitive and Developmental Systems, 14(3):971--984

    2022

  20. [28]

    Gangeshwar Krishnamurthy, Navonil Majumder, Soujanya Poria, and Erik Cambria. 2023. A deep learning approach for multimodal deception detection. In Computational Linguistics and Intelligent Text Processing, pages 87--96, Cham. Springer Nature Switzerland

    2023

  21. [29]

    Shanglin Lei, Guanting Dong, Xiaoping Wang, Keheng Wang, Runqi Qiao, and Sirui Wang. 2024. https://arxiv.org/abs/2309.11911 Instructerc: Reforming emotion recognition in conversation with multi-task retrieval-augmented large language models . Preprint, arXiv:2309.11911

    2024 arXiv

  22. [30]

    Timothy R. Levine. 2014. https://doi.org/10.1177/0261927X14535916 Truth-default theory (tdt): A theory of human deception and deception detection . Journal of Language and Social Psychology, 33(4):378--392

    2014 doi

  23. [31]

    Sarah Ita Levitan, Angel Maredia, and Julia Hirschberg. 2018. https://doi.org/10.18653/v1/N18-1176 Linguistic cues to deception and perceived deception in interview dialogues . In Proceedings of the 2018 Conference of the North A merican Chapter of the Association for Computat...

    2018 doi

  24. [32]

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. 2023. Visual instruction tuning

    2023

  25. [33]

    Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. https://arxiv.org/abs/1907.11692 Roberta: A robustly optimized bert pretraining approach . Preprint, arXiv:1907.11692

    2019 arXiv

  26. [34]

    Paige Lloyd, Jason Deska, Kurt Hugenberg, Allen McConnell, Brandon Humphrey, and Jonathan Kunstman

    E. Paige Lloyd, Jason Deska, Kurt Hugenberg, Allen McConnell, Brandon Humphrey, and Jonathan Kunstman. 2018. https://doi.org/10.3758/s13428-018-1061-4 Miami university deception detection database . Behavior Research Methods, 51

    2018 doi

  27. [35]

    Riccardo Loconte, Roberto Russo, Pasquale Capuozzo, Pietro Pietrini, and Giuseppe Sartori. 2023. https://doi.org/10.1038/s41598-023-50214-0 Verbal lie detection using large language models . Scientific Reports, 13(1):22849. Published on 2023/12/21

    2023 doi

  28. [36]

    Matari\' c

    Leena Mathur and Maja J. Matari\' c . 2020. https://doi.org/10.1145/3382507.3418864 Introducing representations of facial affect in automated multimodal deception detection . In Proceedings of the 2020 International Conference on Multimodal Interaction, ICMI '20, page 305–314,...

    2020

  29. [37]

    Leena Mathur and Maja J Matarić. 2021. https://arxiv.org/abs/2108.07897 Affect-aware deep belief network representations for multimodal unsupervised deception detection . Preprint, arXiv:2108.07897

    2021 arXiv

  30. [38]

    Md Messal Monem Miah, Adarsh Pyarelal, and Ruihong Huang. 2023. https://doi.org/10.18653/v1/2023.findings-emnlp.505 Hierarchical fusion for online multimodal dialog act classification . In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 7532--7545,...

    2023 doi

  31. [39]

    OpenAI, :, Aaron Hurst, Adam Lerer, Adam P. Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, Aleksander Mądry, Alex Baker-Whitcomb, Alex Beutel, Alex Borzunov, Alex Carney, Alex Chow, Alex Kirillov, Alex Nichol, Alex Pai...

    2024 arXiv

  32. [40]

    Myle Ott, Yejin Choi, Claire Cardie, and Jeffrey T. Hancock. 2011. https://aclanthology.org/P11-1032/ Finding deceptive opinion spam by any stretch of the imagination . In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language T...

    2011

  33. [41]

    Ver\' o nica P\' e rez-Rosas, Mohamed Abouelenien, Rada Mihalcea, and Mihai Burzo. 2015. https://doi.org/10.1145/2818346.2820758 Deception detection using real-life trial data . In Proceedings of the 2015 ACM on International Conference on Multimodal Interaction, ICMI '15, pag...

    2015

  34. [42]

    Kowsar Hossain Sakib, David Asirvatham, Neethiahnanthan Ari Ragavan, Cesar Sanin, and Edward Szczerbicki

    Shanjita Akter Prome, Md Rafiqul Islam, Md. Kowsar Hossain Sakib, David Asirvatham, Neethiahnanthan Ari Ragavan, Cesar Sanin, and Edward Szczerbicki. 2024. https://doi.org/10.32604/cmc.2024.055311 Ldnet: A robust hybrid approach for lie detection using deep learning techniques...

    2024

  35. [43]

    Khubaib Ahmed Qureshi, Rauf Ahmed Shams Malick, Muhammad Sabih, and Hocine Cherifi. 2022. https://doi.org/10.1016/j.knosys.2022.109649 Deception detection on social media: A source-based perspective . Knowledge-Based Systems, 256:109649

    2022

  36. [44]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. 2021. https://arxiv.org/abs/2103.00020 Learning transferable visual models from natural lan...

    2021 arXiv

  37. [45]

    Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2022. https://arxiv.org/abs/2212.04356 Robust speech recognition via large-scale weak supervision . Preprint, arXiv:2212.04356

    2022 arXiv

  38. [46]

    Sheth, and Amitava Das

    Anku Rani, Dwip Dalal, Shreya Gautam, Pankaj Gupta, Vinija Jain, Aman Chadha, Amit P. Sheth, and Amitava Das. 2023. https://api.semanticscholar.org/CorpusID:265551498 Sepsis: I can catch your lies - a new paradigm for deception detection . ArXiv, abs/2312.00292

    2023 arXiv

  39. [47]

    Rodrigo Rill-Garcia, Hugo Jair Escalante, Luis Villasenor-Pineda, and Veronica Reyes-Meza. 2019. High-level features for multimodal deception detection in videos. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops

    2019

  40. [48]

    Puneet Kumar Sehrawat, Rajat Kumar, Nitish Kumar, and Dinesh Kumar Vishwakarma. 2023. https://doi.org/10.1109/ICIDCA56705.2023.10099779 Deception detection using a multimodal stacked bi-lstm model . In 2023 International Conference on Innovative Data Communication Technologies...

    2023

  41. [49]

    Felix Soldner, Bennett Kleinberg, and Shane D. Johnson. 2022. https://doi.org/10.1371/journal.pone.0277869 Confounds and overestimations in fake review detection: Experimentally controlling for product-ownership and data-origin . PLOS ONE, 17(12)

    2022 doi

  42. [50]

    Javier Sánchez-Monedero and Lina Dencik. 2022. https://doi.org/10.1080/1369118X.2020.1792530 The politics of deceptive borders: ‘biomarkers of deceit’ and the case of iborderctrl . Information, Communication & Society, 25(3):413--430

    2022

  43. [51]

    Taylor, Scott R

    John B. Taylor, Scott R. Beach, and Nicholas Kontos. 2017. https://doi.org/10.1016/j.genhosppsych.2017.03.010 The therapeutic discharge: An approach to dealing with deceptive patients . General Hospital Psychiatry, 46:74--78

    2017 doi

  44. [52]

    Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, Johan Ferret, Peter Liu, Pouya Tafti, Abe Friesen, Michelle Casbon, Sabela Ramos, Ravin Kumar, Charline Le Lan...

    2024 arXiv

  45. [53]

    Thannoon, Wissam H

    Harith H. Thannoon, Wissam H. Ali, and Ivan A. Hashim. 2018. https://doi.org/10.1109/SCEE.2018.8684170 Detection of deception using facial expressions based on different classification algorithms . In 2018 Third Scientific Conference of Electrical Engineering (SCEE), pages 51--56

    2018

  46. [54]

    Sushma Krupa Venkatesh, Raghavendra Ramachandra, and Patrick A. H. Bours. 2019. https://api.semanticscholar.org/CorpusID:214730661 Video based deception detection using deep recurrent convolutional neural network . In International Conference on Computer Vision and Image Processing

    2019

  47. [55]

    Aldert Vrij. 2008. Detecting lies and deceit: Pitfalls and opportunities. John Wiley & Sons

    2008

  48. [56]

    Fisher, Samantha Mann, and Sharon Leal

    Aldert Vrij, R. Fisher, Samantha Mann, and Sharon Leal. 2008. https://doi.org/10.1002/jip.82 A cognitive load approach to lie detection . Journal of Investigative Psychology and Offender Profiling, 5(1-2):39--43

    2008 doi

  49. [57]

    Fisher, and Hartmut Blank

    Aldert Vrij, Ronald P. Fisher, and Hartmut Blank. 2017. https://doi.org/10.1111/lcrp.12088 A cognitive approach to lie detection: A meta-analysis . Legal and Criminological Psychology, 22(1):1--21

    2017 doi

  50. [58]

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Yang Fan, Kai Dang, Mengfei Du, Xuancheng Ren, Rui Men, Dayiheng Liu, Chang Zhou, Jingren Zhou, and Junyang Lin. 2024. https://arxiv.org/abs/2409.12191 Qwen2-v...

    2024 arXiv

  51. [59]

    Chi, Quoc V

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc V. Le, and Denny Zhou. 2022. Chain-of-thought prompting elicits reasoning in large language models. In Proceedings of the 36th International Conference on Neural Information Processi...

    2022

  52. [60]

    Hang Zhang, Xin Li, and Lidong Bing. 2023. https://doi.org/10.18653/v1/2023.emnlp-demo.49 Video- LL a MA : An instruction-tuned audio-visual language model for video understanding . In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing: Syst...

    2023 doi

  53. [61]

    Jiaxuan Zhang, Sarah Ita Levitan, and Julia Hirschberg. 2020. https://api.semanticscholar.org/CorpusID:221878474 Multimodal deception detection using automatically extracted acoustic, visual, and lexical features . In Interspeech

    2020

  54. [62]

    Qixuan Zhang, Zhifeng Wang, Dylan Zhang, Wenjia Niu, Sabrina Caldwell, Tom Gedeon, Yang Liu, and Zhenyue Qin. 2024 a . https://doi.org/10.18653/v1/2024.emnlp-main.257 Visual prompting in LLM s for enhancing emotion recognition . In Proceedings of the 2024 Conference on Empiric...

    2024 doi

  55. [63]

    Wenxuan Zhang, Yue Deng, Bing Liu, Sinno Pan, and Lidong Bing. 2024 b . https://doi.org/10.18653/v1/2024.findings-naacl.246 Sentiment analysis in the era of large language models: A reality check . In Findings of the Association for Computational Linguistics: NAACL 2024, pages...

    2024 doi

  56. [64]

    Xuan Zhang and Wei Gao. 2023. https://doi.org/10.18653/v1/2023.ijcnlp-main.64 Towards LLM -based fact verification on news claims with a hierarchical step-by-step prompting method . In Proceedings of the 13th International Joint Conference on Natural Language Processing and th...

    2023 doi

  57. [65]

    Yuanhan Zhang, Bo Li, haotian Liu, Yong jae Lee, Liangke Gui, Di Fu, Jiashi Feng, Ziwei Liu, and Chunyuan Li. 2024 c . https://llava-vl.github.io/blog/2024-04-30-llava-next-video/ Llava-next: A strong zero-shot video understanding model

    2024

  58. [66]

    Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, Zheyan Luo, Zhangchi Feng, and Yongqiang Ma. 2024. http://arxiv.org/abs/2403.13372 Llamafactory: Unified efficient fine-tuning of 100+ language models . In Proceedings of the 62nd Annual Meeting of the Association for Compu...

    2024 arXiv

  59. [67]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  60. [68]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.