This survey provides a comprehensive overview of the research progress and challenges in the hallucination problem in Natural Language Generation (NLG). Hallucination refers to the generation of unfaithful or nonsensical text by NLG models, which can degrade system performance and fail to meet user expectations in real-world scenarios. The survey is organized into three parts: (1) a general overview of metrics, mitigation methods, and future directions; (2) an overview of task-specific research progress on hallucinations in abstractive summarization, dialogue generation, generative question answering, data-to-text generation, machine translation, and visual-language generation; and (3) hallucinations in large language models (LLMs). The survey aims to facilitate collaborative efforts among researchers in tackling the challenge of hallucinated texts in NLG. Hallucination in NLG can be categorized into intrinsic and extrinsic types. Intrinsic hallucination occurs when the generated output contradicts the source content, while extrinsic hallucination occurs when the generated output cannot be verified from the source content. The survey discusses the contributors to hallucination, including source-reference divergence, imperfect representation learning, erroneous decoding, exposure bias, and parametric knowledge bias. It also presents various metrics for measuring hallucination, including statistical metrics, model-based metrics, and human evaluation. The survey also explores mitigation methods for hallucination, including data-related methods, modeling and inference methods, and post-processing methods. These methods aim to reduce the occurrence of hallucination by improving the faithfulness of generated text. The survey highlights the importance of developing effective metrics and mitigation methods for hallucination in NLG, and suggests future research directions in this area. The survey concludes that further research is needed to address the challenges of hallucination in NLG and to improve the performance and reliability of NLG systems.This survey provides a comprehensive overview of the research progress and challenges in the hallucination problem in Natural Language Generation (NLG). Hallucination refers to the generation of unfaithful or nonsensical text by NLG models, which can degrade system performance and fail to meet user expectations in real-world scenarios. The survey is organized into three parts: (1) a general overview of metrics, mitigation methods, and future directions; (2) an overview of task-specific research progress on hallucinations in abstractive summarization, dialogue generation, generative question answering, data-to-text generation, machine translation, and visual-language generation; and (3) hallucinations in large language models (LLMs). The survey aims to facilitate collaborative efforts among researchers in tackling the challenge of hallucinated texts in NLG. Hallucination in NLG can be categorized into intrinsic and extrinsic types. Intrinsic hallucination occurs when the generated output contradicts the source content, while extrinsic hallucination occurs when the generated output cannot be verified from the source content. The survey discusses the contributors to hallucination, including source-reference divergence, imperfect representation learning, erroneous decoding, exposure bias, and parametric knowledge bias. It also presents various metrics for measuring hallucination, including statistical metrics, model-based metrics, and human evaluation. The survey also explores mitigation methods for hallucination, including data-related methods, modeling and inference methods, and post-processing methods. These methods aim to reduce the occurrence of hallucination by improving the faithfulness of generated text. The survey highlights the importance of developing effective metrics and mitigation methods for hallucination in NLG, and suggests future research directions in this area. The survey concludes that further research is needed to address the challenges of hallucination in NLG and to improve the performance and reliability of NLG systems.