Large Language Models Explain Political Rhetoric Unevenly: A Structured Critical Review and Operational ACIL Protocol Across Languages
DOI:
https://doi.org/10.69760/jales.2026004007Keywords:
large language models, political rhetoric, ; political discourse analysis, calibration, explanation faithfulness, multilingual NLP, pragmatics, sociolinguisticsAbstract
Large language models (LLMs) are increasingly asked to identify and explain political metaphor, framing, epistemic stance, presupposition, implicature, and claim–evidence relations. Fluent output, however, is not evidence that an interpretation is correct, appropriately uncertain, grounded in the text, behaviorally supported, or culturally valid. This structured critical review maps peer-reviewed evidence published through 20 August 2026 and introduces an operational ACIL protocol integrating Accuracy, Calibration, Interpretability, and Language parity. Searches of ACL Anthology metadata and OpenAlex produced 972 unique title records after date filtering and deduplication; 15 reports were retained from database screening and five were added through citation chaining. The final evidence map comprises 18 empirical and two conceptual or methodological reports. Direct political-rhetoric evidence is concentrated in Italian, German, English, and Korean, while multilingual calibration and faithfulness evidence is largely indirect. Results vary sharply by task: constrained multiple-choice recognition can be strong, whereas open explanation, implicature reconstruction, false-presupposition rejection, and source attribution remain fragile. We clarify ACIL's contribution relative to general multidimensional, multilingual, calibration, and faithfulness frameworks; remove composite scoring; specify claim-level calibration under multiple admissible interpretations; distinguish textual grounding, behavioral sensitivity, and mechanistic faithfulness; and treat robustness as a control layer rather than a fifth dimension. A worked re-analysis of a published Italian example illustrates how to report unavailable components without manufacturing scores. The review concludes that current evidence supports LLMs as auditable assistants for political discourse research, not autonomous interpreters. ACIL remains a proposed protocol requiring preregistered validation across languages, political genres, and expert coding workflows.
References
Ahuja, S., Aggarwal, D., Gumma, V., Watts, I., Sathe, A., Ochieng, M., et al. (2024). MEGAVERSE: Benchmarking large language models across languages, modalities, models and tasks. In Proceedings of NAACL 2024, 2598–2637. https://doi.org/10.18653/v1/2024.naacl-long.143
Azime, I. A., Tonja, A. L., Belay, T. D., Chanie, Y., Balcha, B. F., Abadi, N. H., et al. (2025). ProverbEval: Exploring LLM evaluation challenges for low-resource language understanding. In Findings of NAACL 2025, 6265–6281. https://doi.org/10.18653/v1/2025.findings-naacl.350
Bender, E. M., Gebru, T., McMillan-Major, A., and Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? In Proceedings of FAccT ’21, 610–623. https://doi.org/10.1145/3442188.3445922
Blodgett, S. L., Barocas, S., Daumé III, H., and Wallach, H. (2020). Language (technology) is power: A critical survey of ‘bias’ in NLP. In Proceedings of ACL 2020, 5454–5476. https://doi.org/10.18653/v1/2020.acl-main.485
Charteris-Black, J. (2011). Politicians and Rhetoric: The Persuasive Power of Metaphor (2nd ed.). Palgrave Macmillan.
Choi, G., and Kim, H. (2025). Automated claim–evidence extraction for political discourse analysis: A large language model approach to Rodong Sinmun editorials. In Proceedings of the Eighth Fact Extraction and VERification Workshop, 1–17. https://doi.org/10.18653/v1/2025.fever-1.1
Conneau, A., Khandelwal, K., Goyal, N., Chaudhary, V., Wenzek, G., Guzmán, F., et al. (2020). Unsupervised cross-lingual representation learning at scale. In Proceedings of ACL 2020, 8440–8451. https://doi.org/10.18653/v1/2020.acl-main.747
Desai, S., and Durrett, G. (2020). Calibration of pre-trained transformers. In Proceedings of EMNLP 2020, 295–302. https://doi.org/10.18653/v1/2020.emnlp-main.21
Entman, R. M. (1993). Framing: Toward clarification of a fractured paradigm. Journal of Communication, 43, 51–58. https://doi.org/10.1111/j.1460-2466.1993.tb01304.x
Gregori, L., Paci, W., and Panunzi, A. (2026). Evaluating the abilities of LLMs and SpeechLMs in discovering implicit contents of Italian political speeches. In Proceedings of PoliticalNLP 2026, 165–170. https://doi.org/10.63317/4y7ki8iw4mwe
Guo, C., Pleiss, G., Sun, Y., and Weinberger, K. Q. (2017). On calibration of modern neural networks. In Proceedings of the 34th International Conference on Machine Learning, 1321–1330.
Gupta, A., Blodgett, S. L., Gross, J. H., and O’Connor, B. (2022). Examining political rhetoric with epistemic stance detection. In Proceedings of NLP+CSS 2022, 89–104. https://doi.org/10.18653/v1/2022.nlpcss-1.11
He, Q., Lu, J., Zhang, Y., Yuan, S., Meng, X., Wei, J., et al. (2026). Metaphor reasoning is meta-reasoning. In Proceedings of ACL 2026, 6914–6935. https://doi.org/10.18653/v1/2026.acl-long.314
Hu, J., Ruder, S., Siddhant, A., Neubig, G., Firat, O., and Johnson, M. (2020). XTREME: A massively multilingual multi-task benchmark for evaluating cross-lingual generalization. In Proceedings of ICML 2020, 4411–4421.
Jacovi, A., and Goldberg, Y. (2020). Towards faithfully interpretable NLP systems: How should we define and evaluate faithfulness? In Proceedings of ACL 2020, 4198–4205. https://doi.org/10.18653/v1/2020.acl-main.386
Jiang, Z., Araki, J., Ding, H., and Neubig, G. (2021). How can we know when language models know? On the calibration of language models for question answering. Transactions of the Association for Computational Linguistics, 9, 962–977. https://doi.org/10.1162/tacl_a_00407
Joshi, P., Santy, S., Budhiraja, A., Bali, K., and Choudhury, M. (2020). The state and fate of linguistic diversity and inclusion in the NLP world. In Proceedings of ACL 2020, 6282–6293. https://doi.org/10.18653/v1/2020.acl-main.560
Kuhn, L., Gal, Y., and Farquhar, S. (2023). Semantic uncertainty: Linguistic invariances for uncertainty estimation in natural language generation. In International Conference on Learning Representations.
Lachenmaier, C., Sieker, J., and Zarrieß, S. (2025). Can LLMs ground when they (don’t) know: A study on direct and loaded political questions. In Proceedings of ACL 2025, 14956–14975. https://doi.org/10.18653/v1/2025.acl-long.728
Lakoff, G., and Johnson, M. (1980). Metaphors We Live By. University of Chicago Press.
Liang, P., Bommasani, R., Lee, T., Tsipras, D., Soylu, D., Yasunaga, M., et al. (2023). Holistic evaluation of language models. Transactions on Machine Learning Research.
Lipton, Z. C. (2018). The mythos of model interpretability. Communications of the ACM, 61, 36–43. https://doi.org/10.1145/3233231
Lyu, Q., Havaldar, S., Stein, A., Zhang, L., Rao, D., Wong, E., et al. (2023). Faithful chain-of-thought reasoning. In Proceedings of IJCNLP-AACL 2023, 305–329. https://doi.org/10.18653/v1/2023.ijcnlp-main.20
Mehrazar, M., Yousefi, M. A., Beygi, P. A., and Bahrak, B. (2026). Can LLMs faithfully explain themselves in low-resource languages? A case study on emotion detection in Persian. In Proceedings of LREC 2026, 560–579. https://doi.org/10.63317/4tds2kffmg7j
Mendelsohn, J., and Budak, C. (2025). When people are floods: Analyzing dehumanizing metaphors in immigration discourse with large language models. In Proceedings of ACL 2025, 8079–8103. https://doi.org/10.18653/v1/2025.acl-long.398
Otmakhova, Y., Guida, M., and Frermann, L. (2026). Not all ANIMALs are equal: Metaphorical framing through source domains and semantic frames. In Findings of ACL 2026, 38334–38355. https://doi.org/10.18653/v1/2026.findings-acl.1911
Paci, W., Gregori, L., and Panunzi, A. (2026). Teaching LLMs to unveil tendentious implicit contents of Italian political communication. In Proceedings of CMCL 2026, 221–229. https://doi.org/10.63317/4da4mm3sngm7
Paci, W., Panunzi, A., and Pezzelle, S. (2025). They want to pretend not to understand: The limits of current LLMs in interpreting implicit content of political discourse. In Findings of ACL 2025, 15569–15593. https://doi.org/10.18653/v1/2025.findings-acl.804
Pedersen, B. S., Sørensen, N., Nimb, S., Hansen, D. H., Olsen, S., and Al-Laith, A. (2025). Evaluating LLM-generated explanations of metaphors: A culture-sensitive study of Danish. In Proceedings of NoDaLiDa/Baltic-HLT 2025, 470–479.
Pragglejaz Group. (2007). MIP: A method for identifying metaphorically used words in discourse. Metaphor and Symbol, 22, 1–39. https://doi.org/10.1080/10926480709336752
Reimann, S., and Scheffler, T. (2025). Using large language models to perform MIPVU-inspired automatic metaphor detection. In Proceedings of Analogy-Angle II, 10–21. https://doi.org/10.18653/v1/2025.analogyangle-1.2
Sanchez-Bayona, E., and Agerri, R. (2025). Metaphor and large language models: When surface features matter more than deep understanding. In Findings of ACL 2025, 17462–17477. https://doi.org/10.18653/v1/2025.findings-acl.898
Singh, A. K., Kumar, V., Murthy, R., Sen, J., Mittal, A., and Ramakrishnan, G. (2025). Indic QA Benchmark: A multilingual benchmark to evaluate question-answering capability of LLMs for Indic languages. In Findings of NAACL 2025, 2607–2626. https://doi.org/10.18653/v1/2025.findings-naacl.141
Skadina, I., Bakanovs, B., and Darģis, R. (2025). First steps in benchmarking Latvian in large language models. In Proceedings of RESOURCEFUL-2025, 86–95.
Steen, G. J., Dorst, A. G., Herrmann, J. B., Kaal, A. A., Krennmayr, T., and Pasma, T. (2010). A Method for Linguistic Metaphor Identification: From MIP to MIPVU. John Benjamins. https://doi.org/10.1075/celcr.14
Turpin, M., Michael, J., Perez, E., and Bowman, S. R. (2023). Language models don’t always say what they think: Unfaithful explanations in chain-of-thought prompting. Advances in Neural Information Processing Systems, 36, 74952–74965.
van Dijk, T. A. (1997). What is political discourse analysis? Belgian Journal of Linguistics, 11, 11–52. https://doi.org/10.1075/bjl.11.03dij
Wiegreffe, S., and Pinter, Y. (2019). Attention is not not explanation. In Proceedings of EMNLP-IJCNLP 2019, 11–20. https://doi.org/10.18653/v1/D19-1002
Zaghouani, W. (2026a). Cultural adaptation in large language models for political discourse. In Proceedings of PoliticalNLP 2026, 272–283. https://doi.org/10.63317/2prsxx26bfis
Zaghouani, W. (2026b). Lost in dialect: The annotation gap in multilingual LLM safety. In Proceedings of MeLLM 2026, 1–13. https://doi.org/10.18653/v1/2026.mellm-1.1
Zheng, L., Wang, S., Fei, H., Peng, Z., Li, F., Fu, J., et al. (2025). Enhancing hyperbole and metaphor detection with their bidirectional dynamic interaction and emotion knowledge. In Proceedings of ACL 2025, 489–499. https://doi.org/10.18653/v1/2025.acl-long.23
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Journal of Azerbaijan Language and Education Studies

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.