摘要
当今以深度学习为核心的人工智能催生了科学研究的第五范式,本文呼吁,语言学研究也要拥抱数据/计算密集型的第四/五范式,并通过介绍和评论人工智能研究与技术开发过程中,有关专家对于跟语言学有关的问题的见解,来支持上述倡议。本文主张或赞成的主要观点为:(1)语言的主要功用是交际,思考往往是内心的对话与问答。(2)语言和思维并不完全等同,基于语言运用的智能水平的图灵测试并不完全有效。(3)语法这种智能并不是人类独有的,语法的原理还适用于语言之外的躯体、物品及其图像等组合性系统。(4)现有能力超强的大规模语言模型不能用作人与机器人交谈的技术界面,有效的语言运用必须是一种具身智能,包括具身认知的词语接地和环境可供性等内容;相应地,必须考虑不同类型的具身性图灵测试。(5)ChatGPT的成功说明了大模型、大数据和强算力能够捕获语言的统计规律和运用模式,语言学要采用数据/计算密集型的第四/五范式来探索语言的统计结构。
Abstract
In view that artificial intelligence (AI), with deep learning as its core technology, is bringing about the fifth paradigm in scientific research, this paper argues that linguistic studies also need to embark upon the fourth or the fifth paradigm for pursuing a data-intensive and computation-intensive approaches. To further this suggestion, insights from linguistic experts on linguistics-related issues in AI study, which are often encountered in the scientific research and technical development of AI, are introduced and examined with an aim to support my argument. To engage with this topic, the current paper advocates the following points of view, namely: (1) While language functions primarily to communicate, thoughts usually happen in form of dialoguing and questioning-answering in the mind. (2) As language and reasoning are not totally equivalent, the Turing test cannot be fully valid. (3) Grammaticalization is not an intelligence unique to mankind, so the functional principles of grammaticalization also apply to composite systems other than language, including bodies, objects, and images. (4) Any existing super-capacity large language model (LLM) is not adequate to be used as a technological interface for the communication between human beings and robots, because any effective use of language should be an embodied intelligence, including lexical grounding and environmental affordance in embodied cognition; accordingly, different types of embodied Turing tests should be taken into consideration. (5) The success of ChatGPT shows that large models, big data and powerful computability work to capture statistical rules and operating modes of language. In this sense, linguistic studies should adopt data- and computation-intensive methodologies as a fourth or a fifth paradigm to probe into the algorithmic structure of language.
关键词
深度学习 /
人工智能 /
第四/五范式 /
语言与思维 /
(具身性)图灵测试
Key words
deep learning /
artificial intelligence (AI) /
fourth/fifth paradigm /
language and mind /
(embodied) Turing test
袁毓林.
人工智能大飞跃背景下的语言学理论思考[J]. 语言战略研究. 2023, 8(4): 7-18 https://doi.org/10.19689/j.cnki.cn10-1361/h.20230401
Yuan Yulin.
Theoretical Reflections on Linguistic Studies Against the Background of AI Great Leap Forward[J]. Chinese Journal of Language Policy and Planning. 2023, 8(4): 7-18 https://doi.org/10.19689/j.cnki.cn10-1361/h.20230401
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
基金
澳门大学讲座教授研究与发展基金(CPG2023-00004-FAH)和启动研究基金(SRG2022-00011-FAH)及国家社会科学基金专项项目“新时代中国特色语言学基本理论问题研究”(19VXK06)。承蒙编辑部和匿名审稿专家的指正,谨此谢忱。