You can get lost in some translations. An example: The local language "Limburgish" (south east of the Netherlands and north east Belgium and even some neighbouring parts of Germany) may be recognized as a language, but the dialect variations within the small geographic region are huge. The differences are not mere pronunciation. The words used differ greatly every few kilometres from town to town. Even greater differences are heard depending on which side of the river Meuse you are. It is a true and fascinating story how small local communities evolve(d) their local communication.

Dialects are always a nightmare for linguists, so forget about computers.

Exactly the same applies to the Basque language Euskara -- although there is a modern "standardised" version, developed in the late 60s. My partner, who is Basque (from Navarra), though not a native Euskara speaker, is fluent in the standardised version. She has difficulty understanding some native speakers who grew up less than 50km away, while Euskara from some of the French provinces is effectively incomprehensible to her.

As a child growing up in the West RIding in the sixties, it was common to be able to place people to within a couple of streets based on their accent and dialect (word choice). Placing them to a town was _easy_.

(My parents moved to Skye in the eighties and spoke Gaelic to A-level standard in two years. I wish I had inherited their facility for languages but it seems I don't have those genes!)

Whilst I think it was always a bit fishy that the THHGTGG suggested it caused a lot of wars, when altavista came up with the babelfish branding for (I think) one of the first online translation engines, it was a stroke of genius.

The fact the babelfish was aurally inserted was, at least in australasia at that time, a stroke of Adam's humourous genius.

The rhetorical question "what's that hanging out your ear" was clearly understood by the subject that the questioner was asserting it was the instrument that had recently fornicated with the subject's brain and was still evident.*

I think that, possibly too subtle, verbal abuse has now passed into oblivion.

*FITH

Elon Musk is geen verlies vir Suid-Afrika nie .

But the only afrikaans I know is the little found in John Buchan's stories so I cannot say if the translation is valid although I am sure the sentiment is.

I have noticed the latin can be a bit peculiar.

Navaho isn't on the list so the US WW2 code talkers' exchanges are still safe from the chocolate factory's grasp.

> Elon Musk is geen verlies vir Suid-Afrika nie.

As an ex-fluent Afrikaans speaker*, yes, that sounds about right. And the sentiment too**.

*Not a native speaker, but learned at an early age - and then pretty much forgot it through disuse after emigrating over 40 years ago.

**Afrikaans, especially the street dialect of my native Cape Town, is a superb language for invective... some particularly fruity phrases pop into mind vis à vis Musk, but for reasons of basic propriety I couldn't possibly reproduce them here.

Navaho isn't on the list

Do the native American languages have written forms? I know the oral traditions are impressively detailed, but I do wonder how such a computer translation would work if no easy way to write it down.

Cherokee has its own script, and it's in Unicode.

Unlike South America, written forms did not begin until the arrival of Europeans, but a number developed subsequently. The most famous is probably for [1]Cherokee . There's also a system for [2]Cree . But given there were at least 300 languages in North America and settlers were not, in general, fond of promoting native literacy, it's probably only a minority that have a written form that has seen a great deal of use.

[1] https://en.wikipedia.org/wiki/Cherokee_syllabary

[2] https://en.wikipedia.org/wiki/Cree_syllabics

There is a system for first nations languages here, it uses numbers and backwards letters as punctuation.

Nobody uses it day-day but it becomes a bit of a political football, some government agency doesn't use it because their dot matrix bus stop display doesn't have the characters means they are racist.

A couple of schools have been renamed in first nations languages but they objected to having the pronunciation in IPA because that's imperialist and no two bands can agree on who is the correct pronunciation

I remember my grandfather doing something similar: he was RAF chap stationed in North Africa in WW2. He was a bit of a luvvie (stage actor & BBC employee after the RAF) but still managed to put on an incredibly strong Geordie accent when using voice communications to confound the Germans who mostly knew RP. It's been decades now, but IIRC he mentioned one of the eavesdroppers actually butted in to ask him to talk more clearly!

No Arapaho either, so I won't be able to easily take the advice offered in Spitting Image's The Chicken Song.

Just from experience using translate with some languages like Chinese, Vietnamese and so on. It's sort of fine if a sentence is used, often fails if its more than that.

I download TV programs with their subtitles and run them through an online service from French into English - it uses Google to do the work. The translation engine really struggles with short natural language dialogue. It consistently makes the same error in language that even the most basic learner wouldn't make. It's better than nothing, but I then need to go through the subtitles and edit them myself correcting a lot of the strange translations. I've found that DeepL has consistently better translations than Google - at least for English to French and French to English - however I can't use DeepL as the back end with the tool that speaks VTT files.

DeepL is generally very good but, like even human translators, it depends on context to clear up ambiguities. Some of that will be supplied by the previous subtitles but, in TV programmes, we're also inferring context from the picture which the machine translator can't see. Also, given TV subtitles are often outsourced to the lowest bidder, the source material may not be of tremendous quality.

Incidentally, the [1]DeepL document translation API does claim to support .srt files and so if you can live with plain text you should be able to convert these to and from VTT.

[1] https://developers.deepl.com/docs/api-reference/document

The English subtitles for "Summer of 84" on Amazon Prime are what appears to be the Spanish subtitles dragged through Google Translate. Wrong tenses all over the place and English sayings replaced with literally translated Spanish sayings which wouldn't make any sense unless the viewer already understood Spanish.

Paying for enshittified accessibility features, classy move Amazon.

My neural network guesses that "Gura mie ayd" parallels the Irish "Go raibh maith agat" or "Thank you".

[Author here]

Yep, it is what I found as the Manx equivalent of GRMA. :-)

Manx orthography is wildly different from other forms of Gaelic, I think because the ancient Manx chopped down all the trees and therefore couldn't make paper and so couldn't write. I believe the modern form was invented by missionaries from another European country.

But I am not a Manx native and don't speak it. I just know a few phrases.

You can get yourself in some right bother with just Latin based languages so christ knows what it's translating for stuff with no Latin roots.

It'll be a few years before the word "fluent", in this context, earns the right to drop the quotation marks.