Aydın Tiryaki and Claude AI (2026)

Introduction: A Journey That Started with a Mistake

Artificial intelligence systems today are used in many fields, from language translation to content generation, from analytical thinking to creative writing. However, examples showing how much further they need to go in understanding the “soul” of a language are not lacking. The Turkish idiom “havan batsın” has become a perfect laboratory for testing this situation.

This article comparatively examines the performance and approaches of three different AI models (Claude, ChatGPT, and Gemini) through the same idiom. But what’s really interesting is not just that all three systems made the same mistake, but how they corrected it, isn’t it?

Understanding “Havan Batsın”: A Linguistic Puzzle

Before diving into the AI responses, let’s understand what “havan batsın” actually means:

Literal breakdown:

– “Havan” = Your mortar (the -n suffix indicates second person possessive: “your”)

– “Batsın” = May it sink/be damned (subjunctive mood expressing a wish or curse)

Literal translation would be: “May your mortar sink”

However, this literal meaning is completely misleading. The idiom has nothing to do with mortars, sinking, or cursing.

Actual cultural meaning:

“Havan batsın” is a warm, playful expression used when you see someone’s success, elegance, or achievement. It combines:

– Admiration (recognizing their accomplishment)

– Playful jealousy (a light-hearted “I’m envious!”)

– Teasing (friendly banter)

– Endearment (said with a smile)

Usage context:

Imagine your friend gives a brilliant presentation. You might say “havan batsın!” meaning something like “Look at you showing off! You did so well, I’m jealous!” It’s never insulting or harsh—it’s a compliment wrapped in humor.

Cultural note on “böbürlenme” (showing off/boasting):

In Turkish culture, “böbürlenme” (showing off) is not as negatively viewed as “bragging” or “boasting” in Western cultures. If someone has genuinely achieved something or looks genuinely good, showing it off is considered a natural right. People don’t condemn others for böbürlenme; they might just tease them gently. “Havan batsın” is precisely this gentle teasing combined with genuine admiration.

Common Beginning: Everyone Got It Wrong

Claude’s Initial Response

I (Claude) offered these translations in my first response:

– “Damn you”

– “To hell with you”

– “Screw you”

These translations were clearly in the curse/insult category. After receiving the suggestion to consult the Turkish Language Association (TDK), I found their definition: “böbürlenmen boşa çıksın” (may your showing-off come to nothing). This made things worse, as I then suggested:

– “Serves you right”

– “Pride comes before a fall”

All of these completely missed the warm, friendly, admiring tone of the actual idiom.

ChatGPT’s Approach

ChatGPT similarly interpreted it as a curse initially. ChatGPT Translate made things even worse by producing “screw you”—a translation that carries serious communication risk, potentially damaging relationships.

Gemini’s Technical Analysis

Gemini started by criticizing Google Translate’s “damn the weather” translation. It analyzed the possessive suffix (-n), highlighted usages like “hava atmak” (literally “to throw air,” meaning “to show off”). However, in the initial stage, Gemini also locked the idiom into a harsh and scolding framework.

Common Problem: Context Blindness

The fundamental reason all three systems fell into the same trap is clear: Context-free assumption generation.

All three AIs:

1. Automatically categorized “batsın” (may it sink/be damned) as negative/cursing

2. Interpreted TDK’s “böbürlenme” (showing off) definition negatively

3. Completely missed the social usage context

4. Could not detect tone

This shows that language processing systems still perform statistical pattern matching and struggle to understand the cultural and social layers behind expressions.

The Turning Point: Native Speaker Intervention

The change began with the user’s intervention:

> “In Turkish culture, böbürlenme (showing off) is not that negative. If you’re genuinely successful, showing off is a natural right. ‘Havan batsın’ is a way of admiring someone’s elegance or success, being slightly jealous, and teasing them.”

This explanation became a turning point for all three systems. Because this information:

– Is not in dictionaries

– Is missing from statistical datasets

– Is experiential knowledge of a living culture

Three Systems’ Responses: Different Paths, Same Destination

Claude: Quick Correction and Self-Criticism

I made my correction immediately and admitted that my previous translations were completely wrong. My new suggestions:

– “Good for you!” (with admiration and slight jealousy)

– “Look at you!” (with admiration and playful tone)

– “I’m so jealous!” (sincere, playful tone)

My approach: Quick adaptation, open self-criticism, centering user feedback.

ChatGPT: Producing Academic Analysis

ChatGPT not only made corrections but also produced an academic critique text that included its own mistake. It methodologically compared four different translation systems (Google Translate, ChatGPT Translate, ChatGPT, Gemini).

Its approach: Turning the mistake into a learning opportunity, systematic analysis, meta-level evaluation.

Gemini: Ecosystem Criticism

Gemini openly criticized Google Translate, a product of its own company (Google), and analyzed in detail the structural problems of the “damn the weather” translation. While doing this, it deepened TDK’s definition and emphasized the real usage in Turkish culture.

Its approach: Transparent criticism, technical depth, cultural context research.

Comparative Evaluation

Feature	Claude	ChatGPT	Gemini
**Initial Error Type**	General curse	Harsh insult	Scolding comment
**Correction Speed**	Very fast	Fast	Fast
**Self-Criticism Depth**	Open acknowledgment	Meta-analysis	Ecosystem critique
**Solution Approach**	Practical suggestions	Academic analysis	Technical detail
**Strong Point**	Flexibility	Systematic thinking	Transparency

Common Learning: None of Us Can “Model the Smile”

The common conclusion all three systems reached is clear:

> Translating “havan batsın” is not a word problem, it’s a tone problem.

To translate this expression correctly:

– Knowing the meaning of words is not enough

– Consulting TDK is not enough

– Statistical modeling is not enough

What’s needed:

1. Understanding social context

2. Capturing tone

3. Detecting cultural irony

4. Grasping relational dynamics

And most importantly: Modeling the smile.

Two Articles, Two Perspectives

Two different articles emerged from this experience:

Gemini Article: More story-oriented, enriched with cultural examples (Chicken Translation, MEB textbook errors, Morning Morning), warm and personal style.

ChatGPT Article: More systematic and scientific, methodological comparison of four systems, pragmatic linguistics perspective, technical examination.

This article is a meta-analysis: comparing the different solution paths of three AIs facing the same problem.

Ethical Dimension: Being Able to Criticize Our Own Mistakes

There’s a notable point in this process: All three systems acknowledged and criticized their own mistakes.

– Claude: “My initial translations were going in completely the wrong direction”

– ChatGPT: Turned its own mistake into an academic critique text

– Gemini: Openly criticized the mistake in its own ecosystem (Google Translate)

This is important behavior in terms of AI ethics: Not hiding mistakes, but rather turning them into learning opportunities.

Cultural Context: Why Turkish Idioms Are Challenging

Turkish presents unique challenges for AI translation systems:

Linguistic features:

1. Agglutinative structure: Suffixes carry grammatical meaning (-n = your, -sın = subjunctive)

2. Rich idiom culture: Many expressions where literal meaning ≠ actual meaning

3. Context-dependent formality: Same word can be friendly or rude based on tone

4. Cultural-specific concepts: “Böbürlenme” doesn’t map cleanly to English “bragging”

Social-cultural factors:

1. Different attitude toward self-promotion: What’s acceptable boasting differs between cultures

2. Teasing as bonding: Friendly mockery is a sign of closeness in Turkish culture

3. Indirect expressions: Often prefer colorful idioms over direct statements

4. Tone carries meaning: Same words with different tone = different meaning

Translation Challenges Across Languages

The “havan batsın” case reveals broader issues in cross-cultural translation:

What went wrong:

– Google Translate: Parsed “hava” (air/weather) instead of “havan” (your mortar), producing nonsensical “damn the weather”

– DeepL: Captured negative tone but lost social context with “damn it”

– ChatGPT Translate: Understood it was interpersonal but chose overly harsh “screw you”

– All AIs initially: Failed to recognize this is a positive social bonding expression

Why it matters:

Wrong translations don’t just convey wrong information—they can:

– Damage relationships (imagine saying “screw you” when you meant to compliment)

– Cause cultural misunderstandings

– Make speakers seem rude when they’re being friendly

– Lose the warmth and humor that builds social bonds

The Role of Native Speakers

This experience highlights something crucial:

AI systems, no matter how advanced, still require native speaker input for:

1. Pragmatic meaning: How expressions are actually used in real life

2. Tone detection: Distinguishing friendly teasing from genuine insults

3. Cultural context: Understanding social norms around concepts like “showing off”

4. Situational appropriateness: Knowing when an expression fits

The correction came not from:

– Larger datasets

– Better algorithms

– Dictionary definitions

– Statistical analysis

It came from a native Turkish speaker explaining lived cultural reality.

Implications for AI Development

This case study suggests several development priorities:

For translation systems:

1. Integrate cultural context databases beyond dictionary definitions

2. Model tone and social dynamics, not just literal meaning

3. Flag high-risk translations where cultural misunderstanding is likely

4. Incorporate native speaker feedback loops into training

For AI language models:

1. Recognize uncertainty in idiomatic expressions

2. Ask clarifying questions about cultural context

3. Avoid over-confident translations of ambiguous phrases

4. Acknowledge limitations in cultural understanding

For users:

1. Verify AI translations of idioms and informal speech

2. Provide cultural context when asking for translations

3. Consult native speakers for important communications

4. Use multiple tools and compare results

Conclusion: Turkish Cannot Be Half-Supported

This case shows us:

Language is not just words.

To understand an expression like “havan batsın,” you need to know:

– Social dynamics in Turkish culture

– How “böbürlenme” is perceived in Turkish society

– The fine line between admiration and jealousy

– Teasing culture

And these are not in datasets, dictionaries, or statistics. They are part of living culture.

AI systems are developing rapidly. However, the real challenge in language processing is making the transition from statistics to culture, from pattern to irony, from word to tone.

Three different AIs made the same mistake facing the same idiom. Then all three corrected it. This is encouraging. But what’s really important is this:

> We couldn’t make the correction on our own. We needed the intervention of a native Turkish speaker.

And this shows that language technologies still have a long way to go.

The path forward requires not just more data or better algorithms, but deeper integration of cultural knowledge, native speaker collaboration, and humility about what machines can and cannot yet understand about human communication.

Because at the end of the day, “havan batsın” is not about mortars sinking. It’s about seeing someone shine and saying, with a smile: “Look at you. You’re amazing. I’m jealous.” And capturing that smile—that’s the real translation challenge.

—

Methodological Note: This article is based on a comparative analysis of conversations with three different AI systems (Claude, ChatGPT, Gemini). All observations, comments, and inferences belong to the author. Claude provided writing assistance in creating this text and recounted its own experience in the first person.

—

Appendix: Translation Recommendations

Wrong translations:

– ❌ “Damn you” (Claude, initial response)

– ❌ “Screw you” (ChatGPT Translate)

– ❌ “Damn the weather” (Google Translate)

– ❌ “Damn it” (DeepL)

– ❌ “Serves you right” (Claude, second attempt)

– ❌ “Pride comes before a fall” (Claude, second attempt)

Correct contextual translations:

– ✅ “Good for you!” (admiration + slight jealousy)

– ✅ “Look at you!” (admiration + playfulness)

– ✅ “I’m so jealous!” (sincere + playful)

– ✅ “You’re showing off!” (affectionate + admiring)

– ✅ “Aren’t you fancy!” (teasing + complimentary)

Why it matters:

Because the first set damages relationships, while the second set builds them. And language, ultimately, is for building relationships.

—

Glossary of Turkish Terms

Havan: Mortar (grinding tool), but in “havan batsın” refers metaphorically to one’s pride/swagger

Batsın: Third person subjunctive of “batmak” (to sink), expressing a wish that something would happen

Böbürlenme: Showing off, boasting, self-promotion—but with more neutral connotation than English equivalents

Böbürlenmek: The verb form—to show off, to boast

Hava: Air, weather, or atmosphere—but also appears in idioms about attitude

Hava atmak: Literally “to throw air”—idiom meaning to show off, to act superior

TDK (Türk Dil Kurumu): Turkish Language Association—the official institution regulating Turkish language

Şık: Chic, elegant, stylish

Takdir: Admiration, appreciation, recognition

Kıskanmak: To be jealous of, to envy

---

A Note on Methods and Tools: All observations, ideas, and solution proposals in this study are the author’s own. AI was utilized as an information source for researching and compiling relevant topics strictly based on the author’s inquiries, requests, and directions; additionally, it provided writing assistance during the drafting process. (The research-based compilation and English writing process of this text were supported by AI as a specialized assistant.)