I come bearing more articles on language-related posts and news!
We'll start by traveling back in time to medieval Britain. I enjoyed this curious video from Gavin the Medievalist about a difficult passage in Beowulf, featuring his attempted explanation.
Gavin also has a video on the earliest known mention of King Arthur, and discusses the unusual way in which he is brought up. I recommend following up with this video on King Arthur by Cambrian Chronicles, which discusses the complicated history of mentions of Arthur. Fun Old Welsh (and I'm wondering if the mention of Arthur is to rhyme with Gwawrddur.
Then looking at Indian music: I looked up the lyrics to one of my favorite Hindi songs recently ("Marjaani"). I found this blog post that provided not just a translation, but some cultural commentary as well. I thought this was pretty interesting:
1. So far as I've understood, the word 'Khasmanukhaye' originally meant a woman who'd eat (bring misfortune to) her husband (khasam). But now it simply mean someone wretched, or someone who brings misfortune.2. I came across this really cute piece on the internet where a girl explains why Punjabis use so many swear words. In her words, "In punjabi culture (at least what I saw of it) complimenting or gushing over someone was supposed to jinx (nazar lag jaati hai). So they go out of their way to be abusive. I know my friends mom always calls her grandaughter Chudail (witch) and jhalli and she loves her to bits. My grandma would say "Kinni soni lagdi hai marjaani" to me all the time if I dressed up and was looking pretty. It is harmless namecalling and very cute too."
And finally, some more on AI stylistics. We discussed language mixing in LLMs last month. This post from LessWrong talks about concerns that LLMs will become unintelligible to humans, whether by developing a new language mix or by becoming so telegraphic and shortened that humans will not be able to understand. The author makes the argument that this is unlikely to happen, and that examples of language mixing or shortening may come from patterns in training data rather than being actually efficient ways to reason. There are some interesting counter examples in the comments if you'd like to delve into that.
And speaking of 'delve' (a common stylistic tic from ChatGPT,), this old tweet from 2024 came to my attention. While it is definitely a very straightforward way of trying to determine authorship, it's not rare. I've talked to a lot of people who are looking for ways to spot that something was AI-written.
However, the opposite phenomenon may well be arising. Using AI-influenced vocabulary may actually serve as a signal that you're clued in to the AI world and not merely posing as a vibe coder. This highly entertaining article from User Mag discusses "Claudlish," the style particular to Anthropic's Claude. "Claudlish" is seen as being closer to how agents actually write, so familiarity with it is useful to understand what Claude is saying, to avoid having to spend unnecessary tokens asking for clarifications, and as a proxy that you're genuinely plugged in to using agentic AI tools:
Being able to speak Claudish has become a meme on X, but fluency has also become a proxy for how heavily engaged with AI software development a person actually is. One startup founder I spoke to said that giving a potential job applicant a paragraph written in Claudish and asking them to translate would be a useful gauge for how heavily they actually use AI coding tools. He noted that a lot of people like to “LARP” as vibe coders, but if you can’t translate Claudish, you haven’t spent enough time in Claude Code.
Being able to understand and speak Claudish directly is also potentially important in a business context, Deng said. “If you ask [Claude Code] to explain or to translate the outputs back into human language, you’re wasting tokens,” he said, which can be costly.
Of course, it's not all fun and games, and there's plenty of ribbing about Claudlish in the article as well.
