This survey provides a comprehensive overview of the research progress and challenges in hallucination in Natural Language Generation (NLG). Hallucination, defined as the generation of nonsensical or unfaithful text, is a significant issue in NLG due to its potential to degrade system performance and raise safety concerns in real-world applications. The survey is organized into two main parts: (1) a general overview of metrics, mitigation methods, and future directions; and (2) task-specific research progress on hallucinations in various downstream tasks, including abstractive summarization, dialogue generation, generative question answering, data-to-text generation, machine translation, and visual-language generation. Additionally, the survey discusses hallucinations in large language models (LLMs). The authors aim to facilitate collaborative efforts among researchers to address the challenge of hallucinated texts in NLG. Key topics covered include the definition and categorization of hallucinations, contributors to hallucination, metrics for measuring hallucination, and mitigation methods. The survey also highlights open challenges and potential future directions, such as the development of fine-grained metrics, fact-checking, and the incorporation of human cognitive perspectives.This survey provides a comprehensive overview of the research progress and challenges in hallucination in Natural Language Generation (NLG). Hallucination, defined as the generation of nonsensical or unfaithful text, is a significant issue in NLG due to its potential to degrade system performance and raise safety concerns in real-world applications. The survey is organized into two main parts: (1) a general overview of metrics, mitigation methods, and future directions; and (2) task-specific research progress on hallucinations in various downstream tasks, including abstractive summarization, dialogue generation, generative question answering, data-to-text generation, machine translation, and visual-language generation. Additionally, the survey discusses hallucinations in large language models (LLMs). The authors aim to facilitate collaborative efforts among researchers to address the challenge of hallucinated texts in NLG. Key topics covered include the definition and categorization of hallucinations, contributors to hallucination, metrics for measuring hallucination, and mitigation methods. The survey also highlights open challenges and potential future directions, such as the development of fine-grained metrics, fact-checking, and the incorporation of human cognitive perspectives.