AI in Language Education
Comparing generative AI with native speakers in terms of request expressions in Japanese
This study by Yijun Chen, Peng Yue, and Henry Davidge, published in Discover Education (2025) 4:478, critically investigates the pragmatic competence of generative AI models in Japanese language interactions. The findings reveal significant divergences between AI and human communication norms, underscoring the need for careful integration of AI tools in language learning.
Executive Impact: Key Pragmatic Gaps Identified
Our analysis highlights critical areas where current GenAI models fall short in replicating native Japanese pragmatic norms, posing significant challenges for learners aiming for authentic communication.
~65%
Point Difference in 'No Address Term' Usage (Gemini vs. Native)
+11%
Point Increase in Non-Conventional Indirectness (GenAI vs. Native)
~100%
GenAI Responses Lacking Native Spoken Features
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Formal Complexity
Directness Strategies
Diversity & Naturalness
Pedagogical Implications
Formal Complexity: Address Terms & Modification Acts
Analysis of formal complexity revealed significant differences in how GenAI models construct Japanese requests, particularly regarding address terms and politeness modifications.
65%
Gemini 2.0's Overuse of Address Terms vs. Native Speakers
Native speakers used address terms sparingly (Table 2), reflecting typical Japanese interactional patterns where context often obviates explicit address. DeepSeekR1 mirrored this low usage, but Gemini 2.0 dramatically overused address terms, particularly familiar and deferential types, suggesting a poor grasp of Japanese sociolinguistic norms.
| Characteristic | Native Speakers | GenAI Models (General) |
|---|---|---|
| Internal Downgraders | Lower frequency | Higher rates, potentially leading to overly mitigated requests |
| External Downgraders | Stronger reliance | Varied, less consistent reliance |
| Upgraders | Specific, context-driven usage | Higher rates (Gemini 2.0, ChatGPT-40 with certain prompts), potentially overly forceful or unnatural |
Directness Strategies: Divergence from Native Norms
The study found that GenAI models deviate significantly from native speakers in selecting appropriate levels of directness for Japanese requests, risking ambiguity or face-threatening communication.
+11%
Point Increase in Non-Conventional Indirectness (Gemini vs. Native)
Native speakers displayed a balanced use of direct and conventionally indirect strategies, with minimal use of nonconventional indirectness (Table 5). In contrast, AI models deviated from this pattern. For instance, Gemini 2.0 and DeepSeekR1 showed a higher tendency towards nonconventionally indirect requests compared to native speakers.
Native Speaker Directness Strategy
Context Evaluation
→
Social Power/Distance
→
Rank of Imposition
→
Balanced Direct/Indirect Strategy
→
Pragmatically Appropriate Request
Diversity & Naturalness: The Gap in Authentic Language Use
Despite generating diverse request forms, GenAI models consistently failed to replicate the qualitative naturalness and subtle variations found in native Japanese spoken language.
~100%
GenAI Responses Lacking Native Spoken Features
While AI models generated a large number of distinct head acts, this diversity differed qualitatively from native speakers. Native speakers uniquely employed dialectal variations and common spoken-language features like particle and verb omission, none of which were consistently replicated by the AIs. The AI models adhered strictly to standard Japanese (hyōjungo) and complete grammatical structures, lacking the flexibility and naturalistic variation found in native speaker data.
Case Study: Omissions & Honorific Nuances
Native speakers often omitted sentence-final question particles (ka) or even main verbs (e.g., repōto wa? for 'Where is the report?'). GenAI models, especially ChatGPT-40, did not replicate these natural spoken omissions. Furthermore, subtle differences in honorific nuances (e.g., desu ka vs. deshō ka) highlight potential inconsistencies in AI's grasp of politeness registers, leading to unnatural or contextually inappropriate phrasing.
Pedagogical Implications: Guiding Learners in GenAI Use
The study highlights a critical gap between GenAI's formal linguistic capabilities and functional pragmatic competence, emphasizing responsible integration of these tools in language education.
Recommended AI Integration for Pragmatics Learning
Controlled Exposure to GenAI Examples
→
Hypothesis Testing (Learner Evaluation)
→
Critical Digital Literacy (Evaluation of AI Output)
→
Authentic Human Interaction & Feedback
→
Explicit Pragmatic Instruction
The findings suggest caution for learners and educators. While GenAI can generate grammatically correct and diverse requests, significant deviations from native speaker norms in directness, complexity, and diversity indicate that learners cannot solely rely on these tools for acquiring pragmatically appropriate Japanese. AI should function as a sophisticated tool rather than a surrogate native speaker.
Quantify Your Enterprise AI ROI
Estimate the potential efficiency gains and cost savings by strategically implementing AI solutions tailored to your organization's needs.
Estimated Annual Savings
$0
Annual Hours Reclaimed
0
Your AI Implementation Roadmap
A structured approach to integrating AI, from initial assessment to ongoing optimization, ensuring sustainable success and alignment with your strategic objectives.
Phase 1: Discovery & Strategy
In-depth analysis of current processes, identification of high-impact AI opportunities, and development of a tailored AI strategy roadmap. Define clear objectives and success metrics.
Phase 2: Pilot & Validation
Implement a targeted AI pilot program. Test solutions in a controlled environment, gather performance data, and refine models based on real-world feedback to ensure efficacy.
Phase 3: Scaled Deployment
Roll out validated AI solutions across relevant departments. Integrate with existing systems, provide comprehensive training, and establish robust monitoring frameworks.
Phase 4: Optimization & Expansion
Continuous monitoring, performance tuning, and identification of new areas for AI application. Stay ahead with evolving AI capabilities and expand impact across the enterprise.
Ready to Transform Your Enterprise with AI?
Unlock the full potential of AI for your business. Schedule a personalized consultation with our experts to design a strategic AI roadmap that drives real results.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat