诠释学视域下GPT语言模型的本质及特征
The Essence and Characteristics of GPT Language Model from Hermeneutical Perspective
刘伟伟1
作者信息
- 1. 山西大学哲学社会学学院,山西太原 030006
- 折叠
摘要
GPT语言模型的设计思路具有自然语言理解的诠释学思维特征,但本质上该模型并不具备诠释学语言理解的属人性基础;GPT语言模型将智能视为一种本体论层面以语言作为媒介的整体性系统"涌现"结果,缺乏诠释的"本体论—主体性"地位和"本体论—整体性"结构;GPT语言模型采用的生成式和预训练的设计思路凸显了诠释学理解和解释的"历史性"特征,而数据训练的强化和"思维链"的对话机制使该模型具有"效果历史"和"视域融合"的语言理解特征;GPT语言模型在模拟人类"偏见性"认知方面取得进步,但与人类的"偏见—个性化"和"偏见—创造性"能力相比存在根本差异;GPT语言模型形成了"诠释学循环"的语言对话机制,但其并不具有自身独立的"诠释学循环"实践基础,因此,难以达成诠释学意义上的语言理解共识.
Abstract
The design idea of the GPT language model embodies the hermeneutic thinking characteristics of natural language understanding,but in essence,the model does not have the humanistic basis of hermeneutic language"understanding".The GPT language model regards"intelligence"as the result of the emergence of a holistic system using language as a medium at the ontological level,but the model still lacks"ontology-subjectivity"status and"ontology-integrity"structure in the hermeneutic sense.The"generative"and"pre-training"design ideas adopted in the GPT language model highlight the"historical"characteristics of hermeneutic understanding and interpretation,while the strengthening of data training and the dialogue mechanism of"thought chain"make the model possess the"understanding"characteristics of"effective history"and"fusion of horizons".The GPT language model has made some progress in simulating human"biased"cognition,but there are still fundamental differences compared to human"biased personalization"and"biased creativity"abilities.The GPT language model forms a linguistic"dialogue"mechanism of"hermeneutic cycle",but it does not have its own independent practical basis of"hermeneutic cycle",making it difficult to reach a"consensus"of language understanding in the hermeneutic sense.
关键词
诠释学/人工智能/理解/解释/自然语言Key words
Hermeneutics/Artificial Intelligence/Understanding/Dialogue/Natural Language引用本文复制引用
基金项目
教育部青年项目(22YJC720008)
国家社会科学基金一般项目(21BZX105)
出版年
2024
NETL
NSTL
万方数据