Rethinking Higher Education/Chapter 5/en-zh

Language: EN · ZH · EN-ZH · ← Book

Chapter 5: Learning a Foreign Language with and without AI

Martin Woesler

English (Source)	中文 (Target)
== Learning a Foreign Language with and without AI: An Empirical Comparative Study ==	== 使用和不使用人工智能学习外语：一项实证比较研究 ==
Martin Woesler	Martin Woesler
Hunan Normal University	湖南师范大学
Abstract	摘要
This study compares the self-reported learning outcomes, motivations, and attitudes of 133 Chinese university students learning a foreign language — 85 in an AI-assisted group and 48 in a traditional human-teacher group — over a period of approximately one month. Drawing on a comprehensive survey instrument with 126 variables covering demographics, learning methods, sensory modality preferences, attitudes toward AI in education, and self-assessed improvement across ten language skill areas, the study finds a complex picture that challenges both techno-optimistic and techno-pessimistic narratives. The human-teacher group reported higher overall improvement (63.2% vs. 51.9%), yet the AI group reported greater gains in speaking and listening — precisely the interactive skills that AI chatbots are designed to practise. Both groups expressed strong preference for human teachers, but the AI group simultaneously valued AI’s availability, speed, and pressure-free environment. Attitudes toward AI autonomy were cautious in both groups: over 70% agreed that AI needs ethical control, and fewer than 20% endorsed AI dominance over humans. These findings contribute to the growing literature on AI in language education and are discussed in relation to the qualitative findings of Fang Lu (this volume) and the philosophical framework of Ole Döring (this volume).	本研究比较了133名中国大学生——85名在人工智能辅助组，48名在传统人类教师组——在约一个月时间里自我报告的外语学习成果、动机和态度。基于涵盖人口统计学、学习方法、感官模式偏好、对教育中人工智能的态度以及十个语言技能领域自我评估改善的126个变量的综合调查工具，本研究发现了一幅复杂的图景，挑战了技术乐观主义和技术悲观主义的叙事。人类教师组报告了更高的总体改善（63.2%对51.9%），但人工智能组在口语和听力方面报告了更大的进步——恰恰是人工智能聊天机器人被设计用来练习的互动技能。两组都表现出对人类教师的强烈偏好，但人工智能组同时重视人工智能的可用性、速度和无压力环境。两组对人工智能自主性的态度都很谨慎：超过70%的人同意人工智能需要伦理控制，不到20%的人赞成人工智能对人类的主导地位。这些发现为关于教育中人工智能的日益增长的文献做出了贡献，并结合Fang Lu（本卷）的定性研究和Ole Döring（本卷）的哲学框架进行了讨论。
Keywords: AI-assisted language learning, comparative study, foreign language education, human-AI interaction, digital education, sensory modalities, student attitudes, China, European Union, complementarity thesis	关键词：人工智能辅助语言学习、比较研究、外语教育、人机交互、数字教育、感官模式、学生态度、中国、欧盟、互补性论题
1. Introduction	1. 引言
The integration of artificial intelligence into language education has moved from speculative futurism to daily practice with remarkable speed. Chinese university students in 2025 routinely use AI chatbots — ChatGPT, Kimi, DeepSeek, Doubao — as conversation partners, pronunciation coaches, grammar checkers, and vocabulary tutors. Yet the empirical evidence for whether AI-assisted language learning produces better outcomes than traditional human instruction remains surprisingly thin. Most existing studies are small-scale, focus on a single AI tool, or measure outcomes over very short periods. What is missing is a comparative study that examines not only learning outcomes but also the motivational, attitudinal, and perceptual dimensions of AI-assisted versus human-taught language learning.	将人工智能整合到语言教育中已以惊人的速度从推测性的未来主义转变为日常实践。2025年的中国大学生日常使用人工智能聊天机器人——ChatGPT、Kimi、DeepSeek、豆包——作为对话伙伴、发音教练、语法检查器和词汇辅导工具。然而，关于人工智能辅助语言学习是否比传统人类教学产生更好成果的实证证据仍然出人意料地薄弱。大多数现有研究规模小、只关注单一人工智能工具或只衡量非常短期的成果。缺乏的是一项不仅考察学习成果，还考察人工智能辅助与人类教学语言学习的动机、态度和感知维度的比较研究。
This study addresses that gap. We surveyed 133 Chinese university students — 85 who chose or were assigned to learn a foreign language with AI assistance, and 48 who learned with human teachers — after approximately one month of study. The survey instrument, comprising 126 variables, captures demographics, prior language knowledge, daily study time, reasons for group choice, AI usage methods, feedback quality perceptions, self-assessed improvement across ten specific skill areas, the importance of twelve sensory and social modalities in learning, and attitudes toward fourteen aspects of AI in education and society.	本研究弥补了这一空白。我们调查了133名中国大学生——85名选择或被分配使用人工智能辅助学习外语，48名与人类教师学习——在约一个月的学习之后。调查工具包含126个变量，涵盖人口统计学、先前语言知识、每日学习时间、选组原因、人工智能使用方法、反馈质量感知、十个具体技能领域的自我评估改善、十二种感官和社会模式在学习中的重要性，以及对教育和社会中人工智能十四个方面的态度。
Our findings are situated within a growing body of work on digital education in China and Europe, including the qualitative case studies of Fang Lu (this volume), who examined AI’s effects on critical thinking in Chinese language courses at Boston College, and the philosophical analysis of Ole Döring (this volume), who interrogates the conceptual foundations of „artificial intelligence“ in pedagogical contexts. Where Fang Lu provides depth through individual cases and Döring provides philosophical breadth, we contribute breadth through quantitative comparison across a substantial participant pool.	我们的发现置于越来越多的中欧数字教育研究文献中，包括Fang Lu（本卷）的定性案例研究——她考察了人工智能对波士顿学院中文课程批判性思维的影响——以及Ole Döring（本卷）的哲学分析——他探究了教育学语境中"人工智能"的概念基础。Fang Lu通过个案提供了深度，Döring提供了哲学广度，而我们通过对相当数量参与者群体的定量比较做出了广度上的贡献。
2. Literature Review	2. 文献综述
2.1 AI in Language Education: The State of the Art	2.1 语言教育中的人工智能：技术现状
The application of technology to language learning has a long history, from language laboratories in the 1960s through Computer-Assisted Language Learning (CALL) in the 1990s to the current generation of AI-powered tools. Chapelle (2001) provided an early framework for evaluating technology in second language acquisition, emphasising the importance of language learning potential, learner fit, and practical considerations. Golonka et al. (2014) reviewed 350 studies on technology types in foreign language learning and found that while technology shows promise for vocabulary acquisition and reading comprehension, evidence for speaking and writing gains was limited.	将技术应用于语言学习有着悠久的历史，从1960年代的语言实验室到1990年代的计算机辅助语言学习（CALL），再到当前一代的人工智能工具。Chapelle（2001）为评估第二语言习得中的技术提供了早期框架，强调语言学习潜力、学习者适合度和实际考虑的重要性。Golonka等人（2014）回顾了350项关于外语学习中技术类型的研究，发现虽然技术在词汇习得和阅读理解方面显示出前景，但口语和写作方面的证据有限。
The emergence of large language models (LLMs) — ChatGPT, Claude, and their Chinese counterparts Kimi, DeepSeek, and Doubao — has fundamentally changed the landscape. Unlike earlier chatbots that relied on scripted dialogues and keyword matching, LLM-based chatbots can sustain open-ended, contextually appropriate conversations across virtually any topic. Huang, Hew, and Fryer (2022) conducted a systematic review of chatbot-supported language learning and found positive effects on vocabulary acquisition and speaking confidence, but noted that most studies suffered from small sample sizes, short durations, and lack of control groups.	大型语言模型（LLM）——ChatGPT、Claude及其中国对应产品Kimi、DeepSeek和豆包——的出现从根本上改变了格局。与依赖脚本对话和关键词匹配的早期聊天机器人不同，基于LLM的聊天机器人可以在几乎任何主题上维持开放式的、语境适当的对话。Huang、Hew和Fryer（2022）对聊天机器人支持的语言学习进行了系统综述，发现对词汇习得和口语自信有积极影响，但指出大多数研究存在样本量小、持续时间短和缺乏控制组的问题。
Jeon (2022) explored AI chatbot affordances with young Korean EFL learners and found that students appreciated the chatbot’s patience, availability, and non-judgmental nature — findings that our data strongly corroborate. Kim (2019) reported that AI chatbot interaction improved English grammar skills among Korean university students, a finding that our data only partially support (grammar improvement was actually lower in our AI group).	Jeon（2022）探索了韩国年轻EFL学习者使用人工智能聊天机器人的可供性，发现学生欣赏聊天机器人的耐心、可用性和非评判性——我们的数据有力地证实了这些发现。Kim（2019）报告说，人工智能聊天机器人互动提高了韩国大学生的英语语法技能——我们的数据仅部分支持这一发现（我们的人工智能组语法改善实际上更低）。
2.2 Foreign Language Anxiety	2.2 外语焦虑
The psychological dimension of language learning has been extensively studied since Horwitz, Horwitz, and Cope (1986) developed the Foreign Language Classroom Anxiety Scale (FLCAS). MacIntyre and Gardner (1994) demonstrated that language anxiety has measurable effects on cognitive processing in the second language: anxious learners process information more slowly, recall less vocabulary, and produce less complex utterances. Krashen’s (1982) „affective filter“ hypothesis posits that negative emotional states — anxiety, self-doubt, boredom — create a mental barrier that impedes language acquisition.	语言学习的心理维度自Horwitz、Horwitz和Cope（1986）开发外语课堂焦虑量表（FLCAS）以来得到了广泛研究。MacIntyre和Gardner（1994）证明了语言焦虑对第二语言认知加工的可测量影响：焦虑的学习者信息处理更慢，词汇回忆更少，产出的语句复杂性更低。Krashen（1982）的"情感过滤器"假说认为，消极的情绪状态——焦虑、自我怀疑、无聊——创造了阻碍语言习得的心理障碍。
The relevance for AI-assisted learning is direct. If AI chatbots can lower the affective filter by providing a judgment-free practice environment, they may enable learners to process and produce language more effectively than they would in the anxiety-producing context of a human classroom. Our data suggest that this mechanism is operative: the AI group’s most highly rated advantage was „no fear of making mistakes“ (76.6%), and the AI group reported greater improvement in precisely those skills — speaking, listening, communicative confidence — that are most inhibited by anxiety.	这与人工智能辅助学习直接相关。如果人工智能聊天机器人能够通过提供无评判的练习环境来降低情感过滤器，它们可能使学习者比在产生焦虑的人类课堂中更有效地处理和产出语言。我们的数据表明这种机制正在起作用：人工智能组最高评价的优势是"不怕犯错"（76.6%），而且人工智能组恰恰在那些受焦虑抑制最强的技能——口语、听力、交际自信——上报告了更大的改善。
2.3 The Chinese Context	2.3 中国背景
China‘s educational AI landscape is distinctive. The Chinese government’s „New Generation Artificial Intelligence Development Plan“ (2017) and „Education Modernization 2035“ plan both identify AI as a strategic priority for educational reform. Chinese students have access to a range of domestically developed AI tools — including Kimi (Moonshot AI), DeepSeek, Doubao (ByteDance), and Ernie (Baidu) — in addition to international tools like ChatGPT (accessible via VPN). The cultural context is also relevant: Chinese classroom culture traditionally emphasises teacher authority, student deference, and face-saving behaviours that can inhibit oral participation — precisely the conditions under which AI’s judgment-free environment may offer the greatest benefit.	中国的教育人工智能格局是独特的。中国政府的"新一代人工智能发展规划"（2017年）和"教育现代化2035"计划都将人工智能确定为教育改革的战略优先事项。中国学生可以使用一系列国产人工智能工具——包括Kimi（月之暗面）、DeepSeek、豆包（字节跳动）和文心一言（百度）——以及ChatGPT等国际工具（通过VPN访问）。文化背景也很相关：中国课堂文化传统上强调教师权威、学生尊重和保全面子的行为，这些恰恰可能抑制口头参与——正是人工智能无评判环境可能提供最大裨益的条件。
3. Study Design and Methodology	3. 研究设计与方法
2.1 Participants	3.1 参与者
A total of 133 Chinese university students participated in the study. The AI group comprised 85 participants (74% female, 26% male; mean age 23.8 years, range 19–38). The human-teacher group comprised 48 participants (89% female, 11% male; mean age 23.1 years, range 20–32). All participants were enrolled at Chinese universities, predominantly studying English (AI: 38%, Human: 29%) or German (AI: 16%, Human: 25%) as their foreign language major. The gender imbalance — more pronounced in the human group — reflects the general demographics of foreign language departments at Chinese universities.	共有133名中国大学生参与了本研究。人工智能组包括85名参与者（74%女性，26%男性；平均年龄23.8岁，范围19-38岁）。人类教师组包括48名参与者（89%女性，11%男性；平均年龄23.1岁，范围20-32岁）。所有参与者均就读于中国高校，主要学习英语（人工智能组：38%，人类组：29%）或德语（人工智能组：16%，人类组：25%）作为外语专业。性别失衡——在人类组中更为明显——反映了中国大学外语系的一般人口构成。
Participants were not randomly assigned. Some chose their group; others were assigned (44.7% of the human group reported passive assignment). This self-selection introduces a potential confound: students who chose the AI group may have been more technologically curious or more dissatisfied with traditional instruction. We address this limitation in Section 5.	参与者不是随机分配的。一些人选择了自己的组；另一些是被分配的（人类组中44.7%报告为被动分配）。这种自我选择引入了潜在的混淆因素：选择人工智能组的学生可能对技术更好奇或对传统教学更不满意。我们在第5节中讨论了这一局限性。
2.2 Survey Instrument	3.2 调查工具
The survey was administered in Chinese via an online questionnaire platform (问卷星) on 28 March 2025. It comprised the following sections:	调查于2025年3月28日通过在线问卷平台（问卷星）以中文进行。包括以下部分：
(a) Demographics: name (anonymised before analysis), date of birth, gender (5 items). (b) Prior language proficiency: self-assessed CEFR levels for Chinese, English, German, French, Japanese, Korean, and up to three additional languages (9 items). (c) Study language and starting level: same structure as (b) but for the language being studied in the experiment (9 items). (d) Study habits: daily study time in minutes, group assignment, daily AI usage time in minutes (3 items). (e) Reasons for group choice: 5–6 reasons rated by relative importance (percentage, totalling approximately 100%) (6–10 items depending on group). (f) AI learning methods (AI group only): chatting with AI, task completion, VR classroom, AI teacher — each rated by usage share (5 items). (g) Reasons for interest in current learning method: 9–10 reasons rated by importance (10 items). (h) AI feedback quality and handling (AI group only): categorical rating and yes/no response (2 items). (i) Self-reported overall improvement: percentage estimate (1 item). (j) Sensory modality importance: 21 items covering visual, auditory, textual, gestural, spatial, tactile, olfactory, gustatory, social (3 sub-items), emotional (2 sub-items), VR immersion (2 sub-items), and AI immersion (2 sub-items), each rated 0–100%. (k) Sensory modality ability: same 21 items, rated for personal capacity (0–100%). (l) Group satisfaction and willingness to switch (4 items). (m) Attitudes toward AI: 14 statements rated 0–100% agreement. (n) Improvement areas: 10 language skill areas rated by relative improvement (percentage, totalling approximately 100%) (11 items).	（a）人口统计学：姓名（分析前匿名化）、出生日期、性别（5项）。（b）先前语言水平：中文、英语、德语、法语、日语、韩语及最多三种附加语言的自评CEFR等级（9项）。（c）学习语言和起始水平：与（b）结构相同但针对实验中学习的语言（9项）。（d）学习习惯：每日学习时间（分钟）、组别分配、每日人工智能使用时间（分钟）（3项）。（e）选组原因：按相对重要性评分的5-6个原因（百分比，总计约100%）（6-10项，取决于组别）。（f）人工智能学习方法（仅人工智能组）：与人工智能聊天、任务完成、VR课堂、人工智能教师——各按使用份额评分（5项）。（g）对当前学习方法感兴趣的原因：按重要性评分的9-10个原因（10项）。（h）人工智能反馈质量和处理（仅人工智能组）：分类评分和是/否回答（2项）。（i）自我报告的总体改善：百分比估计（1项）。（j）感官模式重要性：21项，涵盖视觉、听觉、文本、手势、空间、触觉、嗅觉、味觉、社会（3个子项）、情感（2个子项）、VR沉浸（2个子项）和人工智能沉浸（2个子项），各评分0-100%。（k）感官模式能力：相同的21项，评分为个人能力（0-100%）。（l）组别满意度和转组意愿（4项）。（m）对人工智能的态度：14个陈述，评分0-100%同意度。（n）改善领域：10个语言技能领域按相对改善评分（百分比，总计约100%）（11项）。
2.3 Data Processing	3.3 数据处理
Responses were recorded on a 0–100% scale, with 0% indicating „not at all“ and 100% „completely“ or „exclusively.“ For items requiring percentage allocation across multiple options (e.g., reasons for group choice, improvement areas), respondents were instructed that their ratings should sum to approximately 100%. Not all respondents achieved exact summation; we report the raw percentages without normalisation. Missing values were excluded pairwise. All statistical analyses were conducted using Python (descriptive statistics, no inferential tests given the exploratory nature and self-selection design).	回答以0-100%量表记录，0%表示"完全不是"，100%表示"完全"或"排他地"。对于需要在多个选项间分配百分比的项目（如选组原因、改善领域），受访者被指示其评分总和应约为100%。并非所有受访者都达到了精确求和；我们报告原始百分比而不进行标准化。缺失值按成对排除。所有统计分析使用Python进行（描述性统计，鉴于探索性和自我选择设计不做推论测试）。
3. Results	3. 结果
3.1 Daily Study Time and AI Usage	3.1 每日学习时间和人工智能使用
Both groups reported similar daily study times: AI group mean 106 minutes (median 60, SD 103), human group mean 96 minutes (median 60, SD 90). The high standard deviations reflect wide variation: some students studied 10 minutes daily, others 360 minutes. Within the AI group, mean daily AI usage was 32 minutes (median 15), suggesting that AI constituted roughly 30% of total study time, with the remainder spent on textbooks, exercises, or other non-AI methods.	两组报告了相似的每日学习时间：人工智能组平均106分钟（中位数60，标准差103），人类组平均96分钟（中位数60，标准差90）。高标准差反映了广泛的变异：一些学生每天学习10分钟，另一些则360分钟。在人工智能组内，平均每日人工智能使用时间为32分钟（中位数15），表明人工智能约占总学习时间的30%，其余时间用于教科书、练习或其他非人工智能方法。
3.2 Self-Reported Overall Improvement	3.2 自我报告的总体改善
The human-teacher group reported higher overall improvement after one month: mean 63.2% (median 70%, SD 27.5%, n=42) versus the AI group’s mean 51.9% (median 50%, SD 18.1%, n=82). This finding is notable: despite similar study times, students learning with human teachers perceived greater progress. However, the human group’s higher standard deviation (27.5% vs. 18.1%) indicates more heterogeneous experiences — some human-group students reported very high improvement (up to 100%), while others reported as low as 5%.	人类教师组报告了更高的一个月后总体改善：平均63.2%（中位数70%，标准差27.5%，n=42）对比人工智能组的平均51.9%（中位数50%，标准差18.1%，n=82）。这一发现值得注意：尽管学习时间相似，与人类教师学习的学生感知到了更大的进步。然而，人类组更高的标准差（27.5%对18.1%）表明经验更加异质——一些人类组学生报告了非常高的改善（高达100%），而另一些则低至5%。
3.3 AI Feedback Quality	3.3 人工智能反馈质量
Among AI-group participants, perceptions of AI feedback quality were generally positive: 38% rated it as „very pertinent“ (75–100 points), 54% as „okay“ (50–74 points), and only 4% as „average“ (25–49 points). None rated it as poor. Three-quarters (76%) reported handling AI feedback promptly, while 18% did not.	在人工智能组参与者中，对人工智能反馈质量的感知普遍积极：38%评为"非常切题"（75-100分），54%评为"尚可"（50-74分），仅4%评为"一般"（25-49分）。无人评为差。四分之三（76%）报告及时处理人工智能反馈，18%则没有。
3.4 AI Learning Methods	3.4 人工智能学习方法
The most popular AI learning methods were chatting with AI software (mean usage share 68.6%) and asking AI to complete tasks (66.3%). AI teacher functionality received moderate use (51.3%), while VR classroom was the least used (31.9%). This pattern suggests that conversational AI — the free-form chatbot interaction — dominates current AI-assisted language learning, with structured pedagogical AI tools playing a secondary role.	最受欢迎的人工智能学习方法是与人工智能软件聊天（平均使用份额68.6%）和让人工智能完成任务（66.3%）。人工智能教师功能获得了中等使用（51.3%），而VR课堂使用最少（31.9%）。这种模式表明，对话式人工智能——自由形式的聊天机器人互动——主导了当前的人工智能辅助语言学习，结构化的教学人工智能工具发挥次要作用。
3.5 Motivations	3.5 动机
Reasons for choosing the AI group (rated by importance):	选择人工智能组的原因（按重要性评分）：
1. Novelty / trying new things: 75.4%	1. 新颖性/尝试新事物：75.4%
2. Learn anytime, anywhere: 72.5%	2. 随时随地学习：72.5%
3. Immersive learning experience: 66.9%	3. 沉浸式学习体验：66.9%
4. Bored with traditional methods: 60.8%	4. 对传统方法感到厌倦：60.8%
5. Cheaper than human teachers: 59.9%	5. 比人类教师便宜：59.9%
The top two motivations — novelty and flexibility — suggest that early AI adopters are driven more by curiosity and convenience than by dissatisfaction with traditional teaching.	前两个动机——新颖性和灵活性——表明早期人工智能采用者更多是被好奇心和便利性驱动，而非对传统教学的不满。
What makes AI learning attractive (rated by importance):	人工智能学习的吸引力（按重要性评分）：
1. No fear of making mistakes / reduced pressure: 76.6%	1. 不怕犯错/压力减少：76.6%
2. Large knowledge base / diverse topics: 74.7%	2. 大量知识库/多元话题：74.7%
3. Learn anytime, anywhere: 71.9%	3. 随时随地学习：71.9%
4. Fast response speed: 70.4%	4. 反应速度快：70.4%
5. Adaptive difficulty matching: 67.8%	5. 自适应难度匹配：67.8%
6. Adjustable speed, volume, voice: 65.3%	6. 可调节速度、音量、声音：65.3%
7. More encouragement: 64.5%	7. 更多鼓励：64.5%
8. Much cheaper: 59.4%	8. 便宜得多：59.4%
9. More accurate pronunciation correction: 58.5%	9. 更准确的发音纠正：58.5%
The highest-rated advantage — „no fear of making mistakes“ at 76.6% — aligns with a substantial body of research on foreign language anxiety. The AI chatbot creates what language educators call a „low-anxiety practice environment“ in which learners can experiment without social embarrassment.	最高评价的优势——"不怕犯错"（76.6%）——与大量关于外语焦虑的研究一致。人工智能聊天机器人创造了语言教育者所称的"低焦虑练习环境"，学习者可以在其中进行实验而不会感到社交尴尬。
Reasons for choosing the human group:	选择人类组的原因：
1. Prefer learning with real people: 65.7%	1. 更喜欢与真人学习：65.7%
2. Stimulates deeper thinking: 63.8%	2. 激发更深入的思考：63.8%
3. Better at detecting learning problems: 63.6%	3. 更善于发现学习问题：63.6%
4. More precise level assessment: 61.2%	4. 更精确的水平评估：61.2%
5. More diverse feedback methods: 60.5%	5. 更多样化的反馈方式：60.5%
6. Emotional communication in feedback: 58.2%	6. 反馈中的情感交流：58.2%
7. Trust traditional teaching: 52.4%	7. 信任传统教学：52.4%
8. Don’t want to change methods: 52.3%	8. 不想改变方法：52.3%
9. AI not mature yet: 45.3%	9. 人工智能尚未成熟：45.3%
10. Passively assigned: 44.7%	10. 被动分配：44.7%
The human group’s top reasons centre on relational and cognitive depth: human teachers offer personal connection, deeper thinking, and more nuanced assessment. This contrasts with the AI group’s emphasis on convenience and psychological comfort.	人类组选择的主要原因集中在关系和认知深度上：人类教师提供个人联系、更深入的思考和更细致的评估。这与人工智能组强调便利性和心理舒适形成对比。
3.6 Improvement Areas	3.6 改善领域
Students assessed their improvement across ten specific language skill areas. The results reveal a striking complementarity:	学生评估了在十个具体语言技能领域的改善。结果揭示了显著的互补性：
Areas where the AI group reported greater improvement: - Speaking: +12.6 percentage points (AI 58.4%, Human 45.8%) - Listening: +10.2 pp (AI 53.6%, Human 43.5%) - Confidence in communication: +8.3 pp (AI 55.2%, Human 46.9%) - Synonyms/varied expressions: +5.6 pp (AI 56.8%, Human 51.2%)	人工智能组报告更大改善的领域：口语：+12.6个百分点（人工智能58.4%，人类45.8%）；听力：+10.2个百分点（人工智能53.6%，人类43.5%）；交际自信：+8.3个百分点（人工智能55.2%，人类46.9%）；同义词/多样表达：+5.6个百分点（人工智能56.8%，人类51.2%）。
Areas where the human group reported greater improvement: - Reading: +14.0 pp (Human 63.7%, AI 49.8%) - Grammar: +10.1 pp (Human 57.0%, AI 46.9%) - Syntax: +9.3 pp (Human 57.1%, AI 47.8%) - Vocabulary: +5.2 pp (Human 60.7%, AI 55.5%) - Writing: +5.0 pp (Human 51.5%, AI 46.5%)	人类组报告更大改善的领域：阅读：+14.0个百分点（人类63.7%，人工智能49.8%）；语法：+10.1个百分点（人类57.0%，人工智能46.9%）；句法：+9.3个百分点（人类57.1%，人工智能47.8%）；词汇：+5.2个百分点（人类60.7%，人工智能55.5%）；写作：+5.0个百分点（人类51.5%，人工智能46.5%）。
The pattern is clear: AI-assisted learning appears to strengthen interactive, oral skills (speaking, listening, communicative confidence), while human-taught learning produces greater gains in structural, analytical skills (reading, grammar, syntax). This finding has direct pedagogical implications: AI and human instruction may be most effective not as substitutes but as complements, each addressing different aspects of language competence.	模式是清晰的：人工智能辅助学习似乎加强了互动性、口头技能（口语、听力、交际自信），而人类教学在结构性、分析性技能（阅读、语法、句法）方面产生了更大的进步。这一发现具有直接的教学意义：人工智能和人类教学作为互补而非替代品可能最为有效，各自针对语言能力的不同方面。
3.7 Sensory and Social Modality Preferences	3.7 感官和社会模式偏好
Participants rated the importance of twelve sensory and social modalities for their language learning. Several large differences emerged between groups:	参与者评估了十二种感官和社会模式对其语言学习的重要性。两组之间出现了几个显著差异：
Modalities rated higher by the AI group: - Auditory perception: +40.7 pp (AI 79.6%, Human 38.9%) - Written text: +37.4 pp (AI 74.5%, Human 37.1%) - Intrinsic motivation: +35.1 pp (AI 77.5%, Human 42.4%) - Extrinsic motivation: +30.0 pp (AI 69.1%, Human 39.1%) - Visual perception: +29.3 pp (AI 74.6%, Human 45.2%) - Emotions/motivation: +29.0 pp (AI 72.6%, Human 43.6%) - Environmental immersion: +20.6 pp (AI 69.9%, Human 49.3%) - Group dynamics: +17.7 pp (AI 64.6%, Human 46.9%)	人工智能组更高评价的模式：听觉感知：+40.7个百分点（人工智能79.6%，人类38.9%）；书面文本：+37.4个百分点（人工智能74.5%，人类37.1%）；内在动机：+35.1个百分点（人工智能77.5%，人类42.4%）；外在动机：+30.0个百分点（人工智能69.1%，人类39.1%）；视觉感知：+29.3个百分点（人工智能74.6%，人类45.2%）；情感/动机：+29.0个百分点（人工智能72.6%，人类43.6%）；环境沉浸：+20.6个百分点（人工智能69.9%，人类49.3%）；群体互动：+17.7个百分点（人工智能64.6%，人类46.9%）。
Modalities rated higher by the human group: - Taste: +32.1 pp (Human 76.3%, AI 44.2%) - AI teacher immersion: +31.7 pp (Human 83.9%, AI 52.2%) - VR immersion: +29.3 pp (Human 83.0%, AI 53.7%) - VR ethics: +29.3 pp (Human 81.3%, AI 52.0%) - AI chatbot immersion: +27.2 pp (Human 79.4%, AI 52.2%) - Social impressions: +21.5 pp (Human 81.5%, AI 59.9%) - Smell: +16.0 pp (Human 59.8%, AI 43.8%)	人类组更高评价的模式：味觉：+32.1个百分点（人类76.3%，人工智能44.2%）；人工智能教师沉浸：+31.7个百分点（人类83.9%，人工智能52.2%）；VR沉浸：+29.3个百分点（人类83.0%，人工智能53.7%）；VR伦理：+29.3个百分点（人类81.3%，人工智能52.0%）；人工智能聊天机器人沉浸：+27.2个百分点（人类79.4%，人工智能52.2%）；社会印象：+21.5个百分点（人类81.5%，人工智能59.9%）；嗅觉：+16.0个百分点（人类59.8%，人工智能43.8%）。
These results require careful interpretation. The AI group placed significantly greater importance on the primary language-learning modalities — visual, auditory, and textual — as well as on motivational factors. The human group, paradoxically, rated AI and VR immersion as more important than the AI group did. One interpretation is that human-group students, not having experienced AI immersion directly, may idealise it, while AI-group students, having used AI tools daily, are more measured in their assessment.	这些结果需要谨慎解读。人工智能组对主要语言学习模式——视觉、听觉和文本——以及动机因素赋予了显著更大的重要性。矛盾的是，人类组对人工智能和VR沉浸的重要性评价高于人工智能组。一种解释是，没有直接体验过人工智能沉浸的人类组学生可能将其理想化，而每天使用人工智能工具的人工智能组学生则更加审慎。
The human group’s higher rating of social impressions (81.5% vs. 59.9%) is consistent with their stated preference for learning with real people and reflects the importance of social presence in language education — a factor that current AI tools, despite rapid advances, cannot fully replicate.	人类组对社会印象的更高评价（81.5%对59.9%）与他们所表达的偏好与真人学习的立场一致，反映了社会临场感在语言教育中的重要性——这一因素虽然人工智能工具进步迅速，但当前仍无法完全复制。
3.8 Attitudes toward AI in Education and Society	3.8 对教育和社会中人工智能的态度
Fourteen attitude statements were rated on a 0–100% agreement scale. The results reveal a nuanced picture:	十四个态度陈述以0-100%同意量表评分。结果揭示了细致入微的图景：
Both groups strongly like human teachers: AI group 77.7%, Human group 83.6%. Even after a month of AI-assisted learning, AI-group students retain strong appreciation for human instruction.	两组都强烈喜欢人类教师：人工智能组77.7%，人类组83.6%。即使经过一个月的人工智能辅助学习，人工智能组学生仍保持对人类教学的强烈欣赏。
The AI group is more positive toward AI teaching: current AI teacher approval was 57.3% (vs. 38.2% in human group), and future advanced AI teacher approval was 66.4% (vs. 53.3%). However, even in the AI group, current AI teacher approval (57.3%) is substantially lower than human teacher approval (77.7%).	人工智能组对人工智能教学更积极：当前人工智能教师认可度为57.3%（人类组为38.2%），未来高级人工智能教师认可度为66.4%（人类组为53.3%）。然而，即使在人工智能组中，当前人工智能教师认可度（57.3%）也大大低于人类教师认可度（77.7%）。
Both groups express fear of AI dependency: - „Fear AI replaces thinking ability“: AI 60.1%, Human 61.0% - „Fear knowledge/skills decline“: AI 60.6%, Human 66.5% - „Fear losing independence / AI addiction“: AI 59.6%, Human 71.6%	两组都表达了对人工智能依赖的恐惧："担心人工智能取代思考能力"：人工智能组60.1%，人类组61.0%。"担心知识/技能退化"：人工智能组60.6%，人类组66.5%。"担心失去独立性/人工智能成瘾"：人工智能组59.6%，人类组71.6%。
The human group consistently reports higher fear of AI dependency, with the largest gap on addiction (71.6% vs. 59.6%). The AI group, perhaps through direct experience, has developed a more moderate but still cautious view.	人类组一贯报告更高的人工智能依赖恐惧，最大差距在成瘾方面（71.6%对59.6%）。人工智能组，或许通过直接经验，发展出了更温和但仍然谨慎的看法。
Both groups strongly endorse AI ethics: „Need to control AI with ethics“ received 72.8% (AI) and 68.7% (Human) agreement.	两组都强烈支持人工智能伦理："需要用伦理控制人工智能"获得72.8%（人工智能组）和68.7%（人类组）的同意。
Both groups reject AI dominance: „Let AI control humans“ received only 14.4% (AI) and 21.5% (Human) agreement. „Only AI robots, no humans, is enough“ received 15.2% and 19.3%. These findings suggest that Chinese university students in 2025 maintain a firmly humanist orientation: they welcome AI as a tool but reject it as a master.	两组都拒绝人工智能主导："让人工智能控制人类"仅获得14.4%（人工智能组）和21.5%（人类组）的同意。"只有人工智能机器人、不需要人类就够了"仅获得15.2%和19.3%的同意。这些发现表明，2025年的中国大学生保持着坚定的人文主义取向：他们欢迎人工智能作为工具，但拒绝它作为主宰。
Romantic attachment to AI or teachers is minimal: „Fell in love with an AI“ averaged approximately 20% in both groups, and „fell in love with a human teacher“ averaged 20–33%. These low figures suggest that immersive AI interaction has not, for this cohort, produced the emotional dependency that some commentators have predicted. The Chinese cultural context may be relevant here: the pragmatic orientation toward AI as a tool rather than a companion, combined with clear social norms around human relationships, may provide a cultural buffer against the parasocial attachment that has been reported in some Western studies of human-AI interaction.	对人工智能或教师的浪漫依恋极少："爱上了人工智能"在两组中平均约为20%，"爱上了人类教师"平均约为20-33%。这些低数字表明，沉浸式人工智能互动并没有——至少对于这一群体——产生一些评论者所预测的情感依赖。中国的文化背景可能在此有所关联：将人工智能视为工具而非伴侣的务实取向，加上围绕人类关系的明确社会规范，可能提供了文化缓冲，防止一些西方人机互动研究中报告的类社交依恋。
The willingness to use AI as a labour-saving device was moderate (approximately 39% in both groups), suggesting that most students do not view AI primarily as a shortcut. Combined with the strong endorsement of ethical AI control, this pattern indicates a cohort that views AI as useful but limited — a sophisticated position that contradicts stereotypes of Chinese students as uncritical technology adopters.	使用人工智能作为省力工具的意愿中等（两组约39%），表明大多数学生并不将人工智能主要视为捷径。结合对人工智能伦理控制的强烈支持，这一模式表明这是一个认为人工智能有用但有局限的群体——一种精细的立场，驳斥了将中国学生视为不加批判的技术采用者的刻板印象。
3.9 Detailed Attitude Analysis	3.9 详细态度分析
To understand the nuanced attitudes more clearly, we can group the fourteen attitude items into thematic clusters:	为更清晰地理解这些细致的态度，我们可以将十四个态度项目分为主题群组：
Cluster A — Teaching preference: - „I like human teacher teaching me“: AI 77.7%, Human 83.6% - „I like current AI teacher teaching me“: AI 57.3%, Human 38.2% - „I’d like future advanced AI teacher“: AI 66.4%, Human 53.3%	群组A——教学偏好："我喜欢人类教师教我"：人工智能组77.7%，人类组83.6%。"我喜欢当前的人工智能教师教我"：人工智能组57.3%，人类组38.2%。"我希望未来有更先进的人工智能教师"：人工智能组66.4%，人类组53.3%。
Both groups prefer human teachers, but the AI group shows significantly greater openness to both current and future AI instruction. The 20-point gap between human teacher approval (77.7%) and current AI teacher approval (57.3%) in the AI group — after direct experience with AI tools — suggests that familiarity breeds qualified appreciation rather than enthusiasm.	两组都偏好人类教师，但人工智能组对当前和未来人工智能教学都表现出明显更大的开放性。人工智能组中人类教师认可度（77.7%）与当前人工智能教师认可度（57.3%）之间的20个百分点差距——在直接体验人工智能工具之后——表明熟悉带来的是有限度的欣赏而非热情。
Cluster B — Fear of AI: - „Fear: AI replaces thinking ability“: AI 60.1%, Human 61.0% - „Fear: knowledge/skills decline“: AI 60.6%, Human 66.5% - „Fear: lose independence, AI addiction“: AI 59.6%, Human 71.6% - „Not afraid: focus on other areas“: AI 55.7%, Human 53.4%	群组B——对人工智能的恐惧："担心人工智能取代思考能力"：人工智能组60.1%，人类组61.0%。"担心知识/技能退化"：人工智能组60.6%，人类组66.5%。"担心失去独立性/人工智能成瘾"：人工智能组59.6%，人类组71.6%。"不担心：专注于其他领域"：人工智能组55.7%，人类组53.4%。
Both groups harbour substantial anxiety about cognitive atrophy — a concern that Fang Lu’s qualitative data make vivid. The human group’s higher fear of addiction (71.6% vs. 59.6%) may reflect a less differentiated understanding of what AI interaction actually involves: the unknown is often more frightening than the known.	两组都对认知萎缩有相当大的焦虑——Fang Lu的定性数据使这一担忧更加生动具体。人类组对成瘾的更高恐惧（71.6%对59.6%）可能反映了对人工智能互动实际涉及内容的不够分化的理解：未知往往比已知更令人恐惧。
Cluster C — AI governance: - „Need to control AI with ethics“: AI 72.8%, Human 68.7% - „Give AI freedom to develop next gen“: AI 47.5%, Human 50.0% - „Let AI control humans“: AI 14.4%, Human 21.5% - „Only AI robots, no humans, is enough“: AI 15.2%, Human 19.3%	群组C——人工智能治理："需要用伦理控制人工智能"：人工智能组72.8%，人类组68.7%。"给人工智能自由发展下一代"：人工智能组47.5%，人类组50.0%。"让人工智能控制人类"：人工智能组14.4%，人类组21.5%。"只有人工智能机器人、不需要人类就够了"：人工智能组15.2%，人类组19.3%。
The governance attitudes reveal a clear hierarchy: strong endorsement of ethical control, ambivalence about AI autonomy, and firm rejection of AI supremacy. The consistency across both groups suggests that these attitudes reflect a broader generational consensus rather than group-specific effects.	治理态度揭示了一个清晰的层次：强烈支持伦理控制，对人工智能自主性持矛盾态度，坚决拒绝人工智能霸权。两组之间的一致性表明，这些态度反映的是更广泛的代际共识而非特定于组别的效应。
3.10 Group Satisfaction and Switching Willingness	3.10 组别满意度和转组意愿
Both groups reported high satisfaction with their assignment: AI group 80.9% (median 80%), human group 76.7% (median 85%). However, willingness to switch groups tells a different story: 47% of the AI group and a remarkable 68% of the human group expressed willingness to switch. The human group’s high switching rate suggests that many human-group students are curious about AI-assisted learning even while satisfied with their current experience — consistent with the broader cultural moment in which AI is perceived as novel and attractive.	两组都报告了较高的分配满意度：人工智能组80.9%（中位数80%），人类组76.7%（中位数85%）。然而，转组意愿讲述了不同的故事：47%的人工智能组和高达68%的人类组表示愿意转组。人类组的高转组率表明，许多人类组学生对人工智能辅助学习感到好奇，即使对当前体验感到满意——这与人工智能被视为新颖和有吸引力的更广泛文化时刻一致。
Among AI-group respondents who described their switching preference, the most common response was „AI group: convenient“ (便利), suggesting that those who would remain valued practical accessibility above all. Among human-group respondents, several articulated thoughtful positions: „AI is not yet mature“ (AI不完善), „human teaching methods are more suited to me“ (human组的教学方法比较适合我), and notably: „I prefer exploring on my own. Humans will never be replaced by AI“ (我更喜欢自己探索。人类永远不会被AI取代) — a statement that encapsulates the humanist position shared by the majority of respondents.	在描述转组偏好的人工智能组受访者中，最常见的回答是"人工智能组：方便"（便利），表明那些愿意留下的人将实际的可及性置于首位。在人类组受访者中，一些人表达了深思熟虑的立场："人工智能尚不完善"（AI不完善），"人类组的教学方法比较适合我"，以及值得注意的是："我更喜欢自己探索。人类永远不会被AI取代"——这一表述概括了大多数受访者所持有的人文主义立场。
4. Discussion	4. 讨论
The results paint a nuanced picture that resists simple conclusions. We organise our discussion around five themes: the complementarity of AI and human instruction, dialogue with the companion essays in this volume, the anxiety-reduction mechanism, modality differences, and implications for European-Chinese comparative education.	结果描绘了一幅抗拒简单结论的细致入微的图景。我们围绕五个主题组织讨论：人工智能和人类教学的互补性、与本卷配套论文的对话、焦虑减少机制、模式差异以及对欧中比较教育的启示。
4.1 The Complementarity Thesis	4.1 互补性论题
Our central finding — that AI-assisted learning strengthens interactive oral skills while human teaching strengthens structural analytical skills — supports what we call the Complementarity Thesis: AI and human instruction are not substitutes but complements, each better suited to different dimensions of language competence. This finding challenges both the techno-optimist position (that AI will replace human teachers) and the techno-pessimist position (that AI cannot teach effectively).	我们的核心发现——人工智能辅助学习加强互动式口头技能，而人类教学加强结构性分析技能——支持我们所称的互补性论题：人工智能和人类教学不是替代品而是互补品，各自更适合语言能力的不同维度。这一发现挑战了技术乐观主义（人工智能将取代人类教师）和技术悲观主义（人工智能无法有效教学）两种立场。
The mechanism is plausible and grounded in established SLA theory. AI chatbots provide unlimited, patient, judgment-free conversation practice — precisely the conditions that promote speaking fluency and listening comprehension. This aligns with Long’s (1996) Interaction Hypothesis, which posits that conversational interaction — including negotiation of meaning, recasts, and comprehension checks — drives language acquisition. AI chatbots provide abundant interaction, albeit without the human interactional moves that Long emphasised. Human teachers provide structured instruction, error analysis, and metalinguistic explanation — precisely the conditions that promote grammatical accuracy, reading comprehension, and syntactic awareness. This aligns with Swain’s (2000) Output Hypothesis, which argues that learners need not only comprehensible input but opportunities to produce language and receive corrective feedback that pushes them beyond their current competence.	其机制是合理的，且建立在已确立的第二语言习得（SLA）理论基础上。人工智能聊天机器人提供无限制的、耐心的、无评判的对话练习——恰恰是促进口语流利性和听力理解的条件。这与Long（1996）的互动假说一致，该假说认为会话互动——包括意义协商、重述和理解检查——驱动语言习得。人工智能聊天机器人提供了大量互动，尽管缺少Long所强调的人类互动特征。人类教师提供结构化教学、错误分析和元语言解释——恰恰是促进语法准确性、阅读理解和句法意识的条件。这与Swain（2000）的输出假说一致，该假说认为学习者不仅需要可理解的输入，还需要产出语言并接受推动他们超越当前能力的纠正性反馈的机会。
The Complementarity Thesis has practical implications: rather than debating whether AI should replace human teachers (a question our data clearly answer: no), educators should ask how AI and human instruction can be orchestrated to serve different learning objectives within a unified curriculum.	互补性论题具有实际意义：教育者不应争论人工智能是否应该取代人类教师（我们的数据明确回答了这个问题：不应该），而应询问如何在统一课程中协调人工智能和人类教学以服务于不同的学习目标。
4.2 Dialogue with Fang Lu	4.2 与Fang Lu的对话
Fang Lu’s qualitative study (this volume) identifies a critical risk of AI-assisted language learning: the potential erosion of critical thinking, creativity, and independent judgment. Her case studies — an elementary student whose AI-assisted writing was structurally perfect but intellectually shallow, and an advanced student whose AI-assisted translation was fluent but lacked cultural nuance — illustrate the „pulling seedlings to help them grow“ (拔苗助长) phenomenon: AI accelerates surface-level performance while undermining deeper cognitive development.	Fang Lu的定性研究（本卷）确定了人工智能辅助语言学习的一个关键风险：批判性思维、创造力和独立判断力可能遭到侵蚀。她的案例研究——一名初级学生的人工智能辅助写作在结构上完美但智力上肤浅，以及一名高级学生的人工智能辅助翻译虽流利但缺乏文化细微差别——说明了"拔苗助长"现象：人工智能加速了表面层面的表现同时削弱了更深层的认知发展。
Our quantitative data both support and complicate Fang Lu’s findings. The human group’s greater improvement in grammar and syntax — skills requiring analytical reasoning rather than pattern reproduction — is consistent with her concern that AI may bypass rather than develop cognitive skills. However, the AI group’s greater improvement in communicative confidence suggests that AI serves a genuine and important function that human instruction often fails to provide: creating a psychologically safe space for oral practice.	我们的定量数据既支持又使Fang Lu的发现复杂化。人类组在语法和句法方面的更大改善——需要分析推理而非模式复制的技能——与她对人工智能可能绕过而非发展认知技能的担忧一致。然而，人工智能组在交际自信方面的更大改善表明，人工智能服务于一种真正重要的功能，而人类教学往往未能提供：创造一个心理安全的口语练习空间。
The implication is not that AI should be avoided but that its role should be carefully defined. AI appears most beneficial for fluency development and anxiety reduction; human instruction appears most beneficial for accuracy development and analytical thinking. A well-designed curriculum would deploy both.	其含义不是应该避免人工智能，而是应该谨慎界定其角色。人工智能似乎最有利于流利性发展和焦虑减少；人类教学似乎最有利于准确性发展和分析性思维。一个精心设计的课程将同时部署两者。
4.3 Dialogue with Ole Döring	4.3 与Ole Döring的对话
Döring’s philosophical paper (this volume) challenges the very concept of „artificial intelligence“ as applied to teaching, arguing that the German philosophical tradition’s distinction between Vernunft (reason, judgment) and Verstand (understanding, calculation) reveals a fundamental category error in claims that machines can „teach.“ What machines do, Döring argues, is process — not understand, not judge, not care.	Döring的哲学论文（本卷）挑战了"人工智能"这一概念应用于教学的合理性，认为德国哲学传统中理性（Vernunft，判断）与知性（Verstand，计算）的区分揭示了声称机器可以"教学"的根本范畴错误。Döring认为，机器所做的是处理——而非理解、判断或关怀。
Our attitudinal data resonate with Döring’s analysis. When students say they „like“ human teachers at 78–84% but only „like“ AI teachers at 38–57%, they may be responding to precisely the distinction Döring identifies: the human teacher offers Vernunft — judgment, care, understanding of the individual learner — while the AI offers Verstand — calculation, pattern-matching, information retrieval. Both are useful, but they are not equivalent.	我们的态度数据与Döring的分析相呼应。当学生以78-84%"喜欢"人类教师但仅以38-57%"喜欢"人工智能教师时，他们可能正是在回应Döring所确定的区别：人类教师提供理性——判断、关怀、对个体学习者的理解——而人工智能提供知性——计算、模式匹配、信息检索。两者都有用，但不等价。
The students’ strong endorsement of ethical AI control (70%+) and strong rejection of AI dominance (<20%) further support Döring’s humanist position. These 133 Chinese university students, while enthusiastically using AI tools, maintain a clear conceptual boundary between human and machine agency.	学生对人工智能伦理控制的强烈支持（70%以上）和对人工智能主导的强烈拒绝（不到20%）进一步支持了Döring的人文主义立场。这133名中国大学生虽然热情使用人工智能工具，但在人类和机器能动性之间维持着清晰的概念边界。
4.4 The Pressure-Free Environment	4.4 无压力环境
The highest-rated advantage of AI learning — „no fear of making mistakes“ at 76.6% — deserves particular attention. Foreign language anxiety is one of the most extensively documented barriers to language acquisition. Traditional classroom settings, with their inherent social dynamics of performance, judgment, and face, create anxiety that inhibits practice, particularly oral practice. The AI chatbot circumvents this entirely: there is no audience, no judgment, no loss of face.	人工智能学习的最高评价优势——"不怕犯错"（76.6%）——值得特别关注。外语焦虑是语言习得中记录最为广泛的障碍之一。传统课堂环境中固有的表演、评判和面子的社交动态产生的焦虑会抑制练习，特别是口语练习。人工智能聊天机器人完全规避了这一点：没有观众、没有评判、没有丢面子。
This finding suggests that AI’s primary educational contribution may be not as a teacher but as a practice partner — a tireless, patient interlocutor who never judges, never loses patience, and never generates social anxiety. If this is correct, the optimal educational model is not „AI instead of human teachers“ but „AI as supplement to human teachers,“ specifically for the practice component of language learning where anxiety most inhibits performance.	这一发现表明，人工智能对教育的主要贡献可能不是作为教师而是作为练习伙伴——一个永不疲倦、耐心、永不评判、永不失去耐心、永不产生社交焦虑的对话者。如果这是正确的，最优的教育模式不是"用人工智能代替人类教师"，而是"用人工智能补充人类教师"，特别是在焦虑最能抑制表现的语言学习练习环节。
4.5 Modality Differences and Their Implications	4.5 模式差异及其启示
The large differences in sensory modality preferences between groups — AI students valuing visual, auditory, and textual input more highly, human students valuing social impressions, VR immersion, and physical senses more highly — suggest that the two groups may have fundamentally different learning orientations. AI-group students appear to be cognitively oriented learners who prioritise information input channels. Human-group students appear to be socially and physically oriented learners who prioritise relational and embodied experience.	两组之间感官模式偏好的显著差异——人工智能组学生更看重视觉、听觉和文本输入，人类组学生更看重社会印象、VR沉浸和身体感官——表明两组可能有根本不同的学习取向。人工智能组学生似乎是认知取向的学习者，优先考虑信息输入渠道。人类组学生似乎是社会和身体取向的学习者，优先考虑关系和具身体验。
Whether these differences are causes or consequences of group choice is unclear. Students who prefer cognitive input channels may have selected the AI group because AI tools deliver precisely those channels. Alternatively, a month of AI-assisted learning may have habituated students to valuing cognitive input over social experience. Longitudinal research would be needed to disentangle these possibilities.	这些差异究竟是组别选择的原因还是结果尚不清楚。偏好认知输入渠道的学生可能因为人工智能工具恰恰提供这些渠道而选择了人工智能组。另一种可能是，一个月的人工智能辅助学习可能使学生习惯于重视认知输入而非社会体验。纵向研究将有助于理清这些可能性。
5.6 Implications for European-Chinese Comparative Education	4.6 对欧中比较教育的启示
Our findings have specific relevance for the European-Chinese educational dialogue that this volume addresses. European language education, shaped by the Common European Framework of Reference for Languages (CEFR) and the communicative approach, has traditionally emphasised oral competence, interaction, and task-based learning. Chinese language education, shaped by examination-driven culture and grammatical-translation pedagogy, has traditionally emphasised reading, writing, grammar, and vocabulary. The emergence of AI as a practice partner may help bridge this gap: Chinese students who lack opportunities for authentic oral practice with human speakers can use AI to develop the communicative skills that European pedagogical approaches prioritise.	我们的发现对本卷所涉及的欧中教育对话具有特殊相关性。欧洲语言教育受《欧洲语言共同参考框架》（CEFR）和交际教学法的影响，传统上强调口语能力、互动和任务式学习。中国语言教育受应试文化和语法-翻译教学法的影响，传统上强调阅读、写作、语法和词汇。人工智能作为练习伙伴的出现可能有助于弥合这一差距：缺乏与人类说话者进行真实口语练习机会的中国学生可以使用人工智能来发展欧洲教学方法优先考虑的交际技能。
At the same time, the European emphasis on critical thinking, learner autonomy, and reflective practice — values articulated in the EU Digital Education Action Plan (2021-2027) — provides a necessary counterweight to the risk that AI practice may develop fluency without depth. Fang Lu’s case studies illustrate this risk concretely: the student whose AI-assisted writing was fluent but intellectually empty had developed surface competence without the deeper cognitive engagement that human interaction promotes.	与此同时，欧洲对批判性思维、学习者自主性和反思性实践的强调——这些价值在《欧盟数字教育行动计划》（2021-2027）中有所体现——为人工智能练习可能发展流利性而非深度的风险提供了必要的对冲。Fang Lu的案例研究具体说明了这一风险：那位人工智能辅助写作流利但智力空洞的学生发展了表面能力，而没有人类互动所促进的更深层认知参与。
A European-Chinese model of AI-integrated language education might therefore combine Chinese students’ enthusiastic adoption of AI tools with European pedagogical frameworks that insist on critical thinking and reflective practice. The technology provides the medium; the pedagogy provides the purpose.	因此，一种欧中融合的人工智能整合语言教育模式可以将中国学生对人工智能工具的热情采用与强调批判性思维和反思性实践的欧洲教学框架相结合。技术提供媒介；教学法提供目的。
5.7 Recommendations for Practice	4.7 实践建议
Based on our findings, we offer four recommendations for educators considering the integration of AI into foreign language teaching:	基于我们的发现，我们为考虑将人工智能整合到外语教学中的教育者提供四项建议：
First, use AI for oral practice, not as a replacement for instruction. The data suggest that AI’s greatest contribution is in developing speaking fluency and communicative confidence through low-anxiety conversational practice. This function complements rather than replaces human instruction.	第一，将人工智能用于口语练习，而非作为教学的替代品。数据表明，人工智能最大的贡献在于通过低焦虑的对话练习发展口语流利性和交际自信。这一功能补充而非取代人类教学。
Second, maintain human teaching for analytical skills. Grammar, syntax, reading comprehension, and writing — the skills that showed greater improvement in the human group — appear to benefit from the structured, explanatory, and corrective instruction that human teachers provide.	第二，保持人类教学用于分析性技能。语法、句法、阅读理解和写作——在人类组中显示出更大改善的技能——似乎受益于人类教师提供的结构化、解释性和纠正性教学。
Third, address students’ AI anxiety proactively. Over 60% of students in both groups expressed fear that AI would replace their thinking ability or erode their skills. These concerns are legitimate and should be addressed through explicit discussion of AI’s limitations, ethical frameworks for AI use, and assignments that require independent critical thinking.	第三，主动应对学生的人工智能焦虑。两组中超过60%的学生表达了对人工智能取代其思考能力或侵蚀其技能的恐惧。这些担忧是合理的，应通过明确讨论人工智能的局限性、人工智能使用的伦理框架以及要求独立批判性思维的作业来加以应对。
Fourth, design assessment that AI cannot shortcut. As Fang Lu’s cases illustrate, AI can produce polished output that masks shallow understanding. Assessments should include oral examinations, spontaneous responses, and tasks that require genuine analytical reasoning — areas where AI assistance is either unavailable or visibly artificial.	第四，设计人工智能无法走捷径的评估。正如Fang Lu的案例所示，人工智能可以产出掩盖肤浅理解的精美输出。评估应包括口试、即兴回答和需要真正分析推理的任务——人工智能协助要么不可用要么明显人工化的领域。
6. Limitations	5. 局限性
Several limitations constrain the interpretation of these results:	几个局限性制约了对这些结果的解读：
First, the study relies entirely on self-reported data. Students’ perceptions of their improvement may not correspond to their actual improvement as measured by standardised tests. A pre-post test design would provide more robust evidence.	第一，本研究完全依赖自我报告数据。学生对其改善的感知可能与标准化测试所测量的实际改善不一致。前后测设计将提供更稳健的证据。
Second, the non-random group assignment introduces self-selection bias. Students who chose the AI group may differ systematically from those who chose or were assigned to the human group — in technological literacy, learning motivation, personality, or other unmeasured variables. The AI group’s higher male percentage (26% vs. 11%) and broader age range suggest some demographic differences, though the practical significance of these differences for language learning outcomes is unclear.	第二，非随机组别分配引入了自选偏差。选择人工智能组的学生可能在技术素养、学习动机、个性或其他未测量变量方面与选择或被分配到人类组的学生存在系统性差异。人工智能组更高的男性比例（26%对11%）和更广的年龄范围表明存在一些人口统计学差异，尽管这些差异对语言学习成果的实际意义尚不明确。
Third, the one-month observation period is short. Language learning is a long-term process, and the relative advantages of AI versus human instruction may shift over longer periods. The AI group’s advantage in speaking may be an early-stage fluency gain that plateaus, while the human group’s advantage in grammar may compound over time.	第三，一个月的观察期较短。语言学习是一个长期过程，人工智能与人类教学的相对优势可能在更长的时期内发生变化。人工智能组在口语方面的优势可能是一种早期的流利性增长，之后趋于平台期，而人类组在语法方面的优势可能随时间累积。
Fourth, the sample is entirely Chinese university students, predominantly female, studying English or German. Generalisability to other cultural contexts, age groups, genders, or target languages is uncertain. The cultural specificity of our findings should be emphasised: Chinese classroom culture’s emphasis on face-saving and teacher authority may amplify the anxiety-reduction benefits of AI in ways that would be less pronounced in cultures with more informal teacher-student relationships.	第四，样本完全是中国大学生，以女性为主，学习英语或德语。对其他文化背景、年龄组、性别或目标语言的推广性不确定。应强调我们发现的文化特殊性：中国课堂文化对保全面子和教师权威的强调，可能以在教师-学生关系更为随意的文化中不那么显著的方式，放大了人工智能的焦虑减少效益。
Fifth, all measurements are self-reported. The „improvement areas“ data (Section 4.6) represent students’ perceptions of where they improved, not objectively measured gains. Students may overestimate improvement in areas they practised most (confusing practice with progress) or underestimate improvement in areas where gains are less consciously perceived.	第五，所有测量均为自我报告。"改善领域"数据（第3.6节）代表的是学生对其改善位置的感知，而非客观测量的进步。学生可能高估了他们练习最多的领域的改善（将练习与进步混淆）或低估了意识感知较弱领域的改善。
Sixth, the survey was conducted at a single time point. Longitudinal data — tracking motivation, attitudes, and outcomes over a full semester or year — would provide a richer picture. A follow-up study with the same participants after six months or one year of continued study would be particularly valuable for testing whether the Complementarity Thesis holds over longer learning periods.	第六，调查在单一时间点进行。纵向数据——跟踪一个完整学期或学年的动机、态度和成果——将提供更丰富的图景。对同一参与者在六个月或一年后继续学习的后续研究将特别有价值，可以检验互补性论题是否在更长的学习期间内成立。
Despite these limitations, the study offers one of the larger-sample comparative investigations of AI-assisted versus human-taught language learning available to date, and the breadth of the survey instrument — covering motivation, modality preferences, attitudes, and skill-specific improvement — provides a multidimensional picture that most existing studies lack.	尽管存在这些局限性，本研究提供了迄今为止关于人工智能辅助与人类教授语言学习的样本量最大的比较调查之一，调查工具的广度——涵盖动机、模式偏好、态度和特定技能改善——提供了大多数现有研究所缺乏的多维图景。
6. Conclusion	6. 结论
This study of 133 Chinese university students learning foreign languages with AI assistance (n=85) and with human teachers (n=48) yields four principal findings:	本研究对133名中国大学生使用人工智能辅助（n=85）和人类教师（n=48）学习外语的调查得出四个主要发现：
First, human-taught students reported higher overall improvement (63.2% vs. 51.9%), but the pattern is skill-specific: AI-assisted students improved more in speaking (+12.6 pp), listening (+10.2 pp), and communicative confidence (+8.3 pp), while human-taught students improved more in reading (+14.0 pp), grammar (+10.1 pp), and syntax (+9.3 pp). This supports a Complementarity Thesis: AI and human instruction serve different, complementary functions in language education.	第一，人类教授的学生报告了更高的总体改善（63.2%对51.9%），但模式是特定于技能的：人工智能辅助学生在口语（+12.6个百分点）、听力（+10.2个百分点）和交际自信（+8.3个百分点）方面改善更大，而人类教授的学生在阅读（+14.0个百分点）、语法（+10.1个百分点）和句法（+9.3个百分点）方面改善更大。这支持了互补性论题：人工智能和人类教学在语言教育中服务于不同的、互补的功能。
Second, the primary perceived advantage of AI learning is not informational but psychological: „no fear of making mistakes“ was rated highest at 76.6%. AI’s greatest contribution to language education may be creating a pressure-free environment for oral practice — addressing one of the most persistent barriers to language acquisition.	第二，人工智能学习的主要感知优势不是信息性的而是心理性的："不怕犯错"以76.6%获得最高评价。人工智能对语言教育最大的贡献可能是为口语练习创造无压力环境——解决语言习得中最持久的障碍之一。
Third, both groups maintain strongly humanist attitudes. Even after a month of AI-assisted learning, AI-group students rate human teachers higher than AI teachers (77.7% vs. 57.3%). Both groups endorse ethical AI control (>68%) and reject AI dominance over humans (<22%).	第三，两组都保持了强烈的人文主义态度。即使经过一个月的人工智能辅助学习，人工智能组学生对人类教师的评价仍高于人工智能教师（77.7%对57.3%）。两组都支持伦理人工智能控制（>68%）并拒绝人工智能对人类的主导（<22%）。
Fourth, the human group’s paradoxically higher valuation of AI and VR immersion suggests curiosity about technologies they have not experienced, while the AI group’s more measured assessment reflects the moderating effect of actual use.	第四，人类组对人工智能和VR沉浸的矛盾性更高评价表明了对未体验过的技术的好奇心，而人工智能组更审慎的评估反映了实际使用的调节效应。
These findings carry clear implications for educational design. The evidence does not support replacing human teachers with AI, nor does it support excluding AI from language education. Instead, it points toward an integrated model in which AI serves as a complementary practice partner — providing the unlimited, judgment-free conversational practice that develops oral fluency and communicative confidence — while human teachers provide the structured instruction, analytical guidance, and social presence that develop grammatical competence, reading comprehension, and critical thinking. Such a model would honour both the technological possibilities documented in our data and the philosophical concerns articulated by Döring and the pedagogical warnings articulated by Fang Lu.	这些发现为教育设计带来了明确的启示。证据不支持用人工智能取代人类教师，也不支持将人工智能排除在语言教育之外。相反，它指向一种整合模式，在这种模式中，人工智能作为互补性练习伙伴——提供无限制的、无评判的对话练习以发展口语流利性和交际自信——而人类教师提供发展语法能力、阅读理解和批判性思维的结构化教学、分析指导和社会陪伴。这一模式将尊重我们数据所记录的技术可能性，也尊重Döring所阐述的哲学关切和Fang Lu所提出的教学警示。随着人工智能能力的不断进步，问题将不是是否在语言教育中使用人工智能，而是如何明智地使用它——这是一个需要持续的实证研究、哲学反思和教学创新的问题。
These findings carry clear implications for educational design. The evidence does not support replacing human teachers with AI, nor does it support excluding AI from language education. Instead, it points toward an integrated model that leverages the complementary strengths of both: AI for fluency development and anxiety reduction, human teachers for accuracy development and critical thinking. As AI capabilities continue to advance, the question will be not whether to use AI in language education but how to use it wisely — a question that requires continued empirical research, philosophical reflection, and pedagogical innovation.	致谢
Acknowledgments	由欧盟共同资助。所表达的观点和意见仅代表作者本人，不一定反映欧盟的立场[101126782]。
Co-funded by the European Union. Views and opinions expressed are however those of the author only and do not necessarily reflect those of the European Union [101126782].	我们感谢学生参与者的坦诚回答，以及协助施测的同事们。
We thank the student participants for their candid responses, and the colleagues who administered the survey.	参考文献
References	Chapelle, C. A. (2001). Computer Applications in Second Language Acquisition. Cambridge University Press.
Chapelle, C. A. (2001). Computer Applications in Second Language Acquisition. Cambridge University Press.	Döring, O. (this volume). AI and pedagogy: Between artificial intelligence and human understanding.
Döring, O. (this volume). AI and pedagogy: Between artificial intelligence and human understanding.	Garrett, N. (2009). Computer-assisted language learning trends and issues revisited: Integrating innovation. The Modern Language Journal, 93(s1), 719–740.
Garrett, N. (2009). Computer-assisted language learning trends and issues revisited: Integrating innovation. The Modern Language Journal, 93(s1), 719–740.	Godwin-Jones, R. (2015). Contributing, creating, curating: Digital literacies for language learners. Language Learning & Technology, 19(3), 8–20.
Godwin-Jones, R. (2015). Contributing, creating, curating: Digital literacies for language learners. Language Learning & Technology, 19(3), 8–20.	Golonka, E. M., Bowles, A. R., Frank, V. M., Richardson, D. L., & Freynik, S. (2014). Technologies for foreign language learning: A review of technology types and their effectiveness. Computer Assisted Language Learning, 27(1), 70–105.
Golonka, E. M., Bowles, A. R., Frank, V. M., Richardson, D. L., & Freynik, S. (2014). Technologies for foreign language learning: A review of technology types and their effectiveness. Computer Assisted Language Learning, 27(1), 70–105.	Horwitz, E. K., Horwitz, M. B., & Cope, J. (1986). Foreign language classroom anxiety. The Modern Language Journal, 70(2), 125–132.
Horwitz, E. K., Horwitz, M. B., & Cope, J. (1986). Foreign language classroom anxiety. The Modern Language Journal, 70(2), 125–132.	Huang, W., Hew, K. F., & Fryer, L. K. (2022). Chatbots for language learning — Are they really useful? A systematic review of chatbot-supported language learning. Journal of Computer Assisted Learning, 38(1), 237–257.
Huang, W., Hew, K. F., & Fryer, L. K. (2022). Chatbots for language learning — Are they really useful? A systematic review of chatbot-supported language learning. Journal of Computer Assisted Learning, 38(1), 237–257.	Jeon, J. (2022). Exploring AI chatbot affordances in the EFL classroom: Young learners' experiences and perspectives. Computer Assisted Language Learning, 37(1–2), 1–26.
Jeon, J. (2022). Exploring AI chatbot affordances in the EFL classroom: Young learners’ experiences and perspectives. Computer Assisted Language Learning, 37(1–2), 1–26.	Kim, N. Y. (2019). A study on the use of artificial intelligence chatbots for improving English grammar skills. Journal of Digital Convergence, 17(8), 37–46.
Kim, N. Y. (2019). A study on the use of artificial intelligence chatbots for improving English grammar skills. Journal of Digital Convergence, 17(8), 37–46.	Krashen, S. D. (1982). Principles and Practice in Second Language Acquisition. Pergamon Press.
Krashen, S. D. (1982). Principles and Practice in Second Language Acquisition. Pergamon Press.	Lai, C., & Zheng, D. (2018). Self-directed use of mobile devices for language learning beyond the classroom. ReCALL, 30(3), 299–318.
Lai, C., & Zheng, D. (2018). Self-directed use of mobile devices for language learning beyond the classroom. ReCALL, 30(3), 299–318.	Long, M. H. (1996). The role of the linguistic environment in second language acquisition. In W. C. Ritchie & T. K. Bhatia (Eds.), Handbook of Second Language Acquisition (pp. 413–468). Academic Press.
Long, M. H. (1996). The role of the linguistic environment in second language acquisition. In W. C. Ritchie & T. K. Bhatia (Eds.), Handbook of Second Language Acquisition (pp. 413–468). Academic Press.	Lu, F. (this volume). AI in Chinese teaching: Opportunities and challenges from the perspective of critical thinking.
Lu, F. (this volume). AI in Chinese teaching: Opportunities and challenges from the perspective of critical thinking.	MacIntyre, P. D., & Gardner, R. C. (1994). The subtle effects of language anxiety on cognitive processing in the second language. Language Learning, 44(2), 283–305.
MacIntyre, P. D., & Gardner, R. C. (1994). The subtle effects of language anxiety on cognitive processing in the second language. Language Learning, 44(2), 283–305.	Swain, M. (2000). The output hypothesis and beyond: Mediating acquisition through collaborative dialogue. In J. P. Lantolf (Ed.), Sociocultural Theory and Second Language Learning (pp. 97–114). Oxford University Press.
Swain, M. (2000). The output hypothesis and beyond: Mediating acquisition through collaborative dialogue. In J. P. Lantolf (Ed.), Sociocultural Theory and Second Language Learning (pp. 97–114). Oxford University Press.	World Economic Forum. (2025). The Future of Jobs Report 2025. Geneva: WEF.
World Economic Forum. (2025). The Future of Jobs Report 2025. Geneva: WEF.