This week’s story starts with an instinct. I’ve been learning Swedish, which, as a Norwegian speaker, has advantages and disadvantages. One downside is the need to fight the assumption that the vocabulary of each matches up exactly with an identical etymology, when this is so often patently untrue.
In fact, Norwegian and Swedish have walked separate paths long enough for all sorts of things to happen to their individual vocabularies. For instance, take trist and ledsen, both meaning sad in Norwegian and Swedish respectively. Adding ledsen to my list of Swedish differences (I’m using my Swedish Anki deck just for the differing words), I started wondering about the etymology of both. Norwegian trist, clearly, I thought, is a French borrowing, probably via Danish. On the other hand, ledsen looks like it was inherited from the North Germanic parent language.
Since I’m exploring the use of AI for language learning both personally and professionally at the moment, it seemed like a good test case for a chat. I went straight in with it: is the Norwegian word trist a borrowing from French?
But shockingly, ChatGPT was resolute in its rejection of that hypothesis. The AI assistant insisted that it’s from a Nordic root þrjóstr, the same that gives us þrjóstur (stubborn) in Modern Icelandic, with the variant þristr which seems to have evolved into Modern Norwegian trist.
Now, the thing with ChatGPT is that it can be so convincing. That’s entirely thanks to the very adept use of natural language in a conversational format. The bot simply speaks with an authoritative voice like it knows what it’s talking about.
So it must be true, right?
At this point, it all felt a bit off. I just had to do some manual digging to check. In Bokmål cases like these, my first port of call is the Norsk Akademi Ordbok. If there is an authority on Norwegian words, there’s little that comes close.
So I key in trist, and – lo and behold – it is a French borrowing.
There’s no mention of Danish, just the French and the Latin that comes from. I suspect, with a bit of digging, it might turn out to have been borrowed into Danish first, but NAOB is definitive. Not a hint of Norse etymology.
Now there’s a chance ChatGPT knows something that NAOB doesn’t, although I doubt it. More likely, it’s just the innate talent the emergent AI has for winging it, and making best guesses. That’s what makes it so powerful, but, like human guesses, it’s also what makes it fallible just now. It’s a timely reminder to double-check AI-generated facts for the time being.
And maybe, to just trust your own instinct.