The Foreign-Language Song Version

Today I wanted to share one of my favorite musical/linguistic phenomena: the foreign-language version!

It used to be common for musicians in non-English speaking countries to cover songs in their native language. Artists may even release songs in languages they don't speak to cross over into more markets. 

ABBA themselves recorded in English, Spanish, Swedish, German, and French. They also made Swedish versions of some of their songs to be recorded by other artists on their record label, Polar. 

Some artists would make their own translations when covering songs, such as this one from Brigitta Wollgård covering Mamma Mia.

When Mamma Mia! the musical was translated into Swedish, this version was not used. Instead a new translation was made for the musical.

Bjorn and Benny of ABBA found some early success with some of their songs in Japan, and in 1972 the earliest Japanese-language cover of an ABBA song I've found so far was released: "She's My Kind of Girl" by LOVE STORY. (More Japanese-language covers collected here.)

And what is likely my favorite foreign translation of an ABBA song: In 1981, sisters Salma and Sabina Agha released an album titled "Sing the Hits of ABBA in Hindi." It is exactly what it sounds like, with the two sisters covering ABBA songs in Hindi and different vocal stylings.

Of course, ABBA were not the only ones to do this. The Beatles recorded two of their hits in German before refusing to do any more. 

Germany used to be a major market for music, prompting artists to translate their songs into German. ABBA recorded several songs into German, though they stopped after Waterloo. Agnetha Falktskog from ABBA also recorded heavily in German.

The foreign translation declined in popularity over time. Perhaps record labels saw they weren't worth the effort since foreign audiences would consume the English versions anyway. If the concern was understanding the lyrics, then it might be cheaper to include a booklet with translations of lyrics rather than write a new lyric that scans, is approved by the writers, and is then recorded by the performer (likely with phonetic lyrics or a coach for pronunciation).

In the 21st century, the only non-English language I've ever seen English-speaking artists record in is Spanish for the Latin American market. Beyonce noticeably released an EP "Irreemplazable" with 5 Spanish-language translations of her songs. It recalls ABBA recording an entire album of Spanish-language translations, and is unusual in its dedication.

Avril Lavigne deserves some credit for recording her song "Girlfriend" 2007 in 8 different languages: English,  Japanese, Portuguese, German, Chinese, Italian, Spanish, and French.

While the English-language music industry, for the most part, doesn't bother with translations anymore, some other music industries do still attempt translations. The k-pop music industry attempts to target China and Japan, and to that end release some Chinese and Japanese versions of their songs.

Hong Kong and the rest of China have historically had separate music industries. Faye Wong, who is a native Mandarin speaker, originally broke through singing in Cantonese. Her early career is in Cantonese, while her latter career is in Mandarin. Some of her songs have both languages available, such as "Love Letter To Myself."

Link Haul - September 2026

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."

If you are not familiar with the song, I recommend listening to it - it's an all-time banger.

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.

Code-Switching for Robots - Language Mixing

I promise I'll write about things besides Japanese and LLMs. But first, a brief word on "language mixing."

I have occasionally encountered situations where LLMs suddenly and without warning begin mixing English and non-English words. Usually, the non-English language also has a non-English script. I've encountered Arabic, Korean, Japanese, and Chinese. The Japanese may be understandable in contexts where I've previously included Japanese text in the conversation for whatever reason, but the Arabic, Korean, and Chinese are quite head-scratching.

Chinese inserted. Image mine. 

Korean inserted. From source.

I've jokingly referred to this as "code-switching," the term for swapping between languages or dialects within a single discourse context. I say "jokingly" because "code-switching" is used by humans with each particular language selected for a reason. The term for it in academia is apparently "language mixing." Unlike human code-switching, language mixing is often "unintentional" in that it is an unwanted artifact. LLMs can produce text that swaps between two languages in a naturalistic way, which we could perhaps refer to as "code-switching," but "language mixing" is broader.

 

Apparently language mixing can actually help the models reason better. Suppressing bilingual chain-of-thought can degrade performance, which makes me think of how people who code-switch are actually people who are strong in both languages. (Of course, that is not to suggest we can generalize about LLM performance from human performance.)

At the same time, because the models are trained so aggressively on English, the chain-of-thought itself may be in English, and trying to force it to reason in a language other than English may also degrade performance, even if the output is meant to be in another language.

Users tend to feel unsettled when they see language mixing. Language mixing breaks the illusion that the LLM you are speaking to is a human-like companion because it makes a speech decision that no human ever would. Because it is so unexpected, it gives the impression that something has gone wrong with the LLM. Most language mixing today is minor slippage of a single word. However, there have been catastrophic instances of language mixing in the past.