在人类和LLM生成的新闻文本中对比语言模式
Alberto Muñoz-Ortiz1, Carlos Gómez-Rodríguez1, David Vilares1
1Universidade da Coruña, CITIC, Departamento de Ciencias de la Computación y Tecnologías de la Información, Campus de Elviña s/n, A Coruña, 15071 A Coruña Spain.
概括
人类和人工智能生成的新闻文本在语言特征上有很大差异. 大型语言模型 (LLM) 与人类写作相比,在句子结构,词汇,情感和偏见方面显示出不同的模式.
科学领域:
- 计算语言学 计算语言学
- 自然语言处理自然语言处理.
- 人工智能伦理学 人工智能伦理学
背景情况:
- 大型语言模型 (LLM) 越来越多地用于文本生成.
- 了解人类和人工智能生成的文本之间的语言差异对于评估人工智能能力和社会影响至关重要.
研究的目的:
- 量化比较人类撰写的新闻文本与六种不同的大型语言模型的输出.
- 通过形态学,语法学,心理学和社会语言学维度识别可测量的语言区别.
主要方法:
- 对人写的英语新闻文本和可比的大型语言模型 (LLM) 输出的分析.
- 跨多个语言维度的评估:句子长度分布,词汇多样性,依赖性和构成结构,依赖性距离,情感表达,毒性,数字/符号/辅助词的使用,代词的使用和性别歧视.
主要成果:
- 人类文本显示出更多样化的句子长度,更丰富的词汇,不同的语法结构和优化的依赖距离.
- 士学位的输出显示了更客观的语言 (数字,符号,辅助词) 和更多的代词使用.
- 人类文本显示出比LLM输出更强烈的负面情绪和更少的快乐;LLM的毒性随着尺寸的增加而增加.
- 在人类文本中,LLM复制并经常放大存在的性别歧视偏见.
结论:
- 人类和大型语言模型 (LLM) 生成的新闻文本之间存在显著的,可测量的语言差异.
- 在客观性,情绪调和偏见放大方面,LLM表现出独特的特征.
- 在LLM输出和人类写作之间的分歧比各种LLM本身的差异更明显.
更多相关视频
09:09Foreign Accent and Forensic Speaker Identification in Voice Lineups: The Influence of Acoustic Features Based on Prosody
Published on: September 27, 2024
416
08:32Examining Online Syntactic Processing of Spoken Complex Sentences in Chinese Using Dual-Modal Interference Tasks
Published on: September 5, 2019
5.6K
相关概念视频
Lateralization
314
Brain lateralization refers to the division of mental processes and functions between the two hemispheres of the brain, a phenomenon that optimizes neural efficiency and underpins complex abilities in humans. This specialization allows each hemisphere to perform tasks where it has a comparative advantage, facilitating more refined cognitive capabilities across different domains.
314
Genetic Lingo
102.0K
Overview
102.0K
Reporter Genes
11.3K
Reporter genes are a type of protein-coding gene that are often tagged to a gene of interest. Once inside a target cell, reporter genes usually produce visually identifiable characteristics like fluorescence and luminescence when expressed along with the gene of interest. Thus, reporter genes “report” the presence or absence of genes of interest in an organism, determine the gene expression pattern, or track the physical location of a DNA segment or protein in the cell.
11.3K
LTR Retrotransposons
17.4K
LTR retrotransposons are class I transposable elements with long terminal repeats flanking an internal coding region. These elements are less abundant in mammals compared to other class I transposable elements. About 8 percent of human genomic DNA comprises LTR retrotransposons. Some of the common examples of LTR retrotransposons are Ty elements in yeast and Copia elements in Drosophila.
The internal coding region of LTR retrotransposons and their mechanism of transposition closely resembles a...
The internal coding region of LTR retrotransposons and their mechanism of transposition closely resembles a...
17.4K
Sources of Law
1.1K
Laws form the essential rules set by governing authorities to shape and control societal behavior. In nursing, laws guide actions, safeguard patient rights, define nurses' scope of practice, and maintain professional standards. Understanding the legal framework governing nursing involves recognizing four primary sources of law: constitutional, statutory, administrative (regulatory), and common law.
Constitutional law is foundational, deriving from federal and state constitutions, and...
Constitutional law is foundational, deriving from federal and state constitutions, and...
1.1K
Lumber
108
Lumber is derived from logs which are harvested, debarked, and processed into long pieces with a rectangular cross-section. The transformation of logs into lumber involves multiple steps, beginning with an automated saw that slices the log into slabs. These slabs are then transported via a conveyor belt to smaller saws, where they are cut into square-edged pieces of specific widths.
Initially, the surfaces of these lumber pieces are rough, and their dimensions may vary slightly from one end to...
Initially, the surfaces of these lumber pieces are rough, and their dimensions may vary slightly from one end to...
108
