Alucinações em Modelos de Linguagem de Grande Escala: Uma Revisão Sistemática da Literatura
Resumo
Modelos de Linguagem de Grande Escala (LLMs) geram texto fluente e coerente, mas frequentemente produzem conteúdo factualmente incorreto ou fabricado, fenômeno denominado alucinação. Este artigo apresenta uma revisão sistemática da literatura abrangendo 32 estudos publicados entre 2020 e 2025, orientada por quatro questões de pesquisa que cobrem tipos de alucinação, estratégias de mitigação, lacunas estruturais e cobertura de línguas de menor recurso. Os resultados são organizados em cinco categorias: geração aumentada por recuperação (RAG), aprendizado por reforço com feedback humano (RLHF), mecanismos de auto-verificação, intervenções arquiteturais e benchmarks de avaliação. Alucinações factuais concentram aproximadamente 78% do esforço de pesquisa, enquanto os tipos semântico e linguístico permanecem subexplorados. Três lacunas críticas são identificadas: ausência de benchmarks padronizados para comparação entre estudos, carência de integrações multi-estratégia e escassez de trabalhos voltados a línguas de menor recurso, como o português, representando oportunidades concretas para a comunidade brasileira de PLN.Referências
Agrawal, G. et al. (2023) Do LLMs know when they're hallucinating? A study of LLM self-awareness. arXiv preprint arXiv:2305.11747.
Bang, Y. et al. (2023) A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity. arXiv preprint arXiv:2302.04023.
Borgeaud, S. et al. (2022) Improving language models by retrieving from trillions of tokens. In: ICML 2022, Proceedings of Machine Learning Research, v. 162, p. 2206-2240.
Chuang, Y.-S. et al. (2023) DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models. arXiv preprint arXiv:2309.03883.
Dhuliawala, S. et al. (2023) Chain-of-Verification Reduces Hallucination in Large Language Models. arXiv preprint arXiv:2309.11495.
Du, Y. et al. (2023) Improving Factuality and Reasoning in Language Models through Multiagent Debate. arXiv preprint arXiv:2305.14325.
Gao, L. et al. (2023) RARR: Researching and Revising What Language Models Say, Using Language Models. arXiv preprint arXiv:2210.08726.
Guo, Z. et al. (2022) Calibrating Sequence Likelihood Improves Conditional Language Generation. arXiv preprint arXiv:2210.00045.
Han, X. and Tsvetkov, Y. (2022) ORCA: interpreting prompted language models via locating supporting data evidence. CoRR, abs/2205.12600.
Huang, L. et al. (2023) A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions. arXiv preprint arXiv:2311.05232.
Ji, Z. et al. (2023) Survey of hallucination in natural language generation. ACM Computing Surveys, v. 55, n. 12, p. 248:1-248:38.
Kadavath, S. et al. (2022) Language Models (Mostly) Know What They Know. arXiv preprint arXiv:2207.05221.
Kitchenham, B. and Charters, S. (2007) Guidelines for performing systematic literature reviews in software engineering. EBSE Technical Report, Keele University.
Lazaridou, A. et al. (2022) Internet-Augmented Language Models through Few-Shot Prompting for Open-Domain Question Answering. arXiv preprint arXiv:2203.05115.
Lewkowycz, A. et al. (2022) Solving Quantitative Reasoning Problems with Language Models. Advances in Neural Information Processing Systems, v. 35.
Li, K. et al. (2023) Inference-Time Intervention: Eliciting Truthful Answers from a Language Model. arXiv preprint arXiv:2306.03341.
Lin, S. et al. (2022) TruthfulQA: Measuring How Models Mimic Human Falsehoods. In: ACL 2022, p. 3214-3252.
Manakul, P.; Liusie, A. and Gales, M. J. F. (2023) SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models. arXiv preprint arXiv:2303.08896.
Maynez, J. et al. (2020) On Faithfulness and Factuality in Abstractive Summarization. Proceedings of ACL 2020.
Medeiros, L. S. F. and Oliveira, H. T. A. (2025) Comparação de Modelos de Embeddings e LLMs para Geração Aumentada por Recuperação em Português. In: Anais do SEMISH 2025, p. 429-440.
Min, S. et al. (2023) FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation. arXiv preprint arXiv:2305.14251.
Mundler, N. et al. (2023) Self-contradictory Hallucinations of Large Language Models: Evaluation, Detection and Mitigation. arXiv preprint arXiv:2305.15852.
Nakano, R. et al. (2021) WebGPT: Browser-assisted question-answering with human feedback. arXiv preprint arXiv:2112.09332.
Ouyang, L. et al. (2022) Training language models to follow instructions with human feedback. Advances in Neural Information Processing Systems, v. 35.
Peng, B. et al. (2023) Check Your Facts and Try Again: Improving Large Language Models with External Knowledge and Automated Feedback. arXiv preprint arXiv:2302.12813.
Shuster, K. et al. (2021) Retrieval Augmentation Reduces Hallucination in Conversation. arXiv preprint arXiv:2104.07567.
Stiennon, N. et al. (2020) Learning to summarize with human feedback. Advances in Neural Information Processing Systems, v. 33.
Sun, W. et al. (2023) Contrastive learning reduces hallucination in conversations. In: AAAI 2023, p. 13618-13626.
Tapajós, R. et al. (2025) Avaliação de comportamento de modelos de linguagem pré-treinados em português: um estudo comparativo. In: Anais do STIL 2025.
Tonmoy, S. M. et al. (2024) A Comprehensive Survey of Hallucination Mitigation Techniques in Large Language Models. arXiv preprint arXiv:2401.01313.
Vaswani, A. et al. (2017) Attention Is All You Need. Advances in Neural Information Processing Systems, v. 30.
Wei, J. et al. (2022) Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. Advances in Neural Information Processing Systems, v. 35.
Weng, M. et al. (2023) Large Language Models Are Better Reasoners with Self-Verification. arXiv preprint arXiv:2212.09561.
Wolf, T. et al. (2020) Transformers: State-of-the-art natural language processing. In: EMNLP 2020 System Demonstrations, p. 38-45.
Yao, S. et al. (2023) Tree of Thoughts: Deliberate Problem Solving with Large Language Models. arXiv preprint arXiv:2305.10601.
Yu, W. et al. (2023) Improving language models via plug-and-play retrieval feedback. CoRR, abs/2305.14002.
Zhao, Z. et al. (2021) Calibrate Before Use: Improving Few-Shot Performance of Language Models. Proceedings of ICML 2021.
Bang, Y. et al. (2023) A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity. arXiv preprint arXiv:2302.04023.
Borgeaud, S. et al. (2022) Improving language models by retrieving from trillions of tokens. In: ICML 2022, Proceedings of Machine Learning Research, v. 162, p. 2206-2240.
Chuang, Y.-S. et al. (2023) DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models. arXiv preprint arXiv:2309.03883.
Dhuliawala, S. et al. (2023) Chain-of-Verification Reduces Hallucination in Large Language Models. arXiv preprint arXiv:2309.11495.
Du, Y. et al. (2023) Improving Factuality and Reasoning in Language Models through Multiagent Debate. arXiv preprint arXiv:2305.14325.
Gao, L. et al. (2023) RARR: Researching and Revising What Language Models Say, Using Language Models. arXiv preprint arXiv:2210.08726.
Guo, Z. et al. (2022) Calibrating Sequence Likelihood Improves Conditional Language Generation. arXiv preprint arXiv:2210.00045.
Han, X. and Tsvetkov, Y. (2022) ORCA: interpreting prompted language models via locating supporting data evidence. CoRR, abs/2205.12600.
Huang, L. et al. (2023) A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions. arXiv preprint arXiv:2311.05232.
Ji, Z. et al. (2023) Survey of hallucination in natural language generation. ACM Computing Surveys, v. 55, n. 12, p. 248:1-248:38.
Kadavath, S. et al. (2022) Language Models (Mostly) Know What They Know. arXiv preprint arXiv:2207.05221.
Kitchenham, B. and Charters, S. (2007) Guidelines for performing systematic literature reviews in software engineering. EBSE Technical Report, Keele University.
Lazaridou, A. et al. (2022) Internet-Augmented Language Models through Few-Shot Prompting for Open-Domain Question Answering. arXiv preprint arXiv:2203.05115.
Lewkowycz, A. et al. (2022) Solving Quantitative Reasoning Problems with Language Models. Advances in Neural Information Processing Systems, v. 35.
Li, K. et al. (2023) Inference-Time Intervention: Eliciting Truthful Answers from a Language Model. arXiv preprint arXiv:2306.03341.
Lin, S. et al. (2022) TruthfulQA: Measuring How Models Mimic Human Falsehoods. In: ACL 2022, p. 3214-3252.
Manakul, P.; Liusie, A. and Gales, M. J. F. (2023) SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models. arXiv preprint arXiv:2303.08896.
Maynez, J. et al. (2020) On Faithfulness and Factuality in Abstractive Summarization. Proceedings of ACL 2020.
Medeiros, L. S. F. and Oliveira, H. T. A. (2025) Comparação de Modelos de Embeddings e LLMs para Geração Aumentada por Recuperação em Português. In: Anais do SEMISH 2025, p. 429-440.
Min, S. et al. (2023) FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation. arXiv preprint arXiv:2305.14251.
Mundler, N. et al. (2023) Self-contradictory Hallucinations of Large Language Models: Evaluation, Detection and Mitigation. arXiv preprint arXiv:2305.15852.
Nakano, R. et al. (2021) WebGPT: Browser-assisted question-answering with human feedback. arXiv preprint arXiv:2112.09332.
Ouyang, L. et al. (2022) Training language models to follow instructions with human feedback. Advances in Neural Information Processing Systems, v. 35.
Peng, B. et al. (2023) Check Your Facts and Try Again: Improving Large Language Models with External Knowledge and Automated Feedback. arXiv preprint arXiv:2302.12813.
Shuster, K. et al. (2021) Retrieval Augmentation Reduces Hallucination in Conversation. arXiv preprint arXiv:2104.07567.
Stiennon, N. et al. (2020) Learning to summarize with human feedback. Advances in Neural Information Processing Systems, v. 33.
Sun, W. et al. (2023) Contrastive learning reduces hallucination in conversations. In: AAAI 2023, p. 13618-13626.
Tapajós, R. et al. (2025) Avaliação de comportamento de modelos de linguagem pré-treinados em português: um estudo comparativo. In: Anais do STIL 2025.
Tonmoy, S. M. et al. (2024) A Comprehensive Survey of Hallucination Mitigation Techniques in Large Language Models. arXiv preprint arXiv:2401.01313.
Vaswani, A. et al. (2017) Attention Is All You Need. Advances in Neural Information Processing Systems, v. 30.
Wei, J. et al. (2022) Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. Advances in Neural Information Processing Systems, v. 35.
Weng, M. et al. (2023) Large Language Models Are Better Reasoners with Self-Verification. arXiv preprint arXiv:2212.09561.
Wolf, T. et al. (2020) Transformers: State-of-the-art natural language processing. In: EMNLP 2020 System Demonstrations, p. 38-45.
Yao, S. et al. (2023) Tree of Thoughts: Deliberate Problem Solving with Large Language Models. arXiv preprint arXiv:2305.10601.
Yu, W. et al. (2023) Improving language models via plug-and-play retrieval feedback. CoRR, abs/2305.14002.
Zhao, Z. et al. (2021) Calibrate Before Use: Improving Few-Shot Performance of Language Models. Proceedings of ICML 2021.
Publicado
19/07/2026
Como Citar
PEREIRA, Andressa Silva; AQUINO, Simone Azevedo Bandeira de Melo; XAVIER, Emmanuel Silva.
Alucinações em Modelos de Linguagem de Grande Escala: Uma Revisão Sistemática da Literatura. In: SEMINÁRIO INTEGRADO DE SOFTWARE E HARDWARE (SEMISH), 53. , 2026, Gramado/RS.
Anais [...].
Porto Alegre: Sociedade Brasileira de Computação,
2026
.
p. 61-73.
ISSN 2595-6205.
DOI: https://doi.org/10.5753/semish.2026.23904.
