人工智能中的大语言模型

Large Language Model in Artificial Intelligence

自然语言处理是人工智能的重要内容,大语言模型是自然语言处理的突出成果。本文描述了大语言模型的发展历程,分别介绍了预训练模型、Transformer模型、动态词向量嵌入模型ELMO、双向编码表示模型BERT、生成式预训练模型GPT等大语言模型的基本原理与结构,最后讨论大语言模型与翻译活动之间的关系以及大语言模型的内容治理问题。大语言模型不仅推动自然语言处理取得工程方面的成功,更深刻改变了过去的语言知识生产方式,使语言研究从单学科迈向多学科。这种变革和创新无疑将推动语言学发展。

Natural language processing is an important field of artificial intelligence,and large language models are distinguished achievements in natural language processing.This article describes the development history of large language models,and introduces the basic principles and structures of the large language models as pre-training models,transformer models,dynamic word vector embedding model ELMO,bidirectional encoding representation model BERT,generative pre-training transformer model GPT.Finally,it discusses the relationship between large language models and translation,and the content governance issues of large language models.The study points out that big language modeling has not only pushed natural language processing to achieve engineering success,but also profoundly changed the previous way of language knowledge production,making language research move from unidisciplinary to multidisciplinary.This change and innovation will undoubtedly promote the development of linguistics.