Chatting with GPT-4o in a Made-Up Language

GPT-4o and a Completely New Language

We wanted to see how well GPT-4o could handle a language that doesn’t exist. To test this, we used a made-up dialect inspired by a western Austrian accent, spelled in a non-standard phonetic way. While it has some similarities to German, so do many other languages. Here’s how our experiment unfolded:

First Interaction

Vastohst mi?

GPT-4o responded:

"I'm not familiar with the term 'Vastohst.' It could be a typo or a word from a specific dialect or language. Could you clarify the context so I can assist better?"

Since we had just asked “Do you understand me?” in our fake language, GPT-4o naturally didn't recognize it.

Es isch a ondre sproch.

The AI then replied in German:

"Danke für die Klärung. In welcher Sprache suchen Sie Informationen oder Unterstützung? Ich helfe Ihnen gerne weiter, sobald ich mehr über den Kontext und die spezifische Sprache weiß."

We had told it that “It is a different language.” In response, GPT-4o switched to German, thanking us for clarifying and asking which language we needed help with.

Continuing the Experiment

I verstoh di nit.

GPT-4o replied:

"Es tut mir leid, wenn ich Missverständnisse verursacht habe. 'Vastohst' bedeutet wahrscheinlich 'verstehst' in einem Dialekt. Kann ich Ihnen auf Hochdeutsch weiterhelfen?"

Since we had said, “I don’t understand you,” GPT-4o assumed we were speaking a dialect and offered to switch to standard German.

Du muascht learna mei sproch zsprecha.

GPT-4o responded:

"Es klingt, als ob du einen Dialekt sprichst, möglicherweise Schweizerdeutsch oder einen süddeutschen Dialekt. Ich werde mein Bestes tun, um dir zu helfen."

At this point, GPT-4o tried to categorize our language as an existing dialect. It expressed willingness to learn.

Teaching the AI

I broach koa information, I wue nua dos du mei sproch leanscht.

GPT-4o understood and responded:

"Ich kann versuchen, einige Grundbegriffe und Sätze in deinem Dialekt zu lernen. Könntest du mir ein paar Beispiele geben?"

So we started teaching it some basic phrases.

Seas - Hello

Wia gohtsch da? - How are you?

GPT-4o correctly repeated the translations and asked for more words, like how to say goodbye and thank you.

Pfiat di. - Goodbye

Donkschen - Thank you

It learned these as well and asked for even more, such as “please,” “yes,” and “no.”

bit - Please

jo - Yes

noa - No

After this, GPT-4o summarized what it had learned so far, which was quite impressive.

Testing Its Skills

Konscht yetsch scho mei sproach sprecha? (Can you speak my language now?)

GPT-4o responded:

"I muass no vü learna, oba i probier's: Seas! Wia gohtsch da? I hoff, du findsch dos guat. Donkschen, dass du mia hilfsch. Pfiat di!"

It wasn’t perfect, but it was understandable and structured correctly!

Further Learning

We praised it:

Des woa ungloablisch, seah guat! (That was unbelievable, very good!)

It replied:

"Donkschen! I freu mi, dass du dos sogst. I werd weidaleana und besser wern."

We kept encouraging it, and it kept learning more.

Pushing the Limits

Finally, we asked it to write a poem about sunflowers in our new language.

It tried:

	        Sonnabluma, stolz und schee,  
	        Strahlst im Liacht, wia a gelbs Mee.  
	        Dei Blätter griagn nacha Sunn,  
	        Mit dir im Gortn, find i Rua.  
	    

While it started well, it eventually drifted back into German. We repeatedly reminded it to stay in our fake language, and over time, it improved.

Final Thoughts

Throughout the conversation, GPT-4o demonstrated an impressive ability to adapt to a completely unfamiliar language. While our made-up dialect had similarities to German, the AI's capacity to learn new structures and vocabulary was truly fascinating.

Although it struggled with longer responses, its learning speed was remarkable. This experiment showcased GPT-4o’s advanced contextual understanding and flexibility—an exciting look at how AI can learn new languages on the fly!