Localization Failure: Temperature Is Hard | Hacker News Reader

Localization Failure: Temperature Is Hard

randomascii.wordpress.com

177 pointsThe_suffocated152 comments

If you want "twice as hot" to be meaningful, you have to use an absolute scale, either Kelvin or Rankine. (1 K = 1.8 °R) Doubling absolute temperature doubles the pressure (at constant volume) or volume (at constant pressure) of an ideal gas. This gets really important when it comes to compressing and expanding air and other gasses.

Room temperature is approximately 293 K, twice that is 586 K / 595 °F / 313 °C. Hotter than your typical oven cooking temperature.

We often don't realize how warm the world we live in truly is, from a physics standpoint.

Kelvin doesn't use degrees

Why do we use degrees for other temperatures by the way? Is that related to some kind of absolute vs relative metric? (Kelvin, I guess, is based on absolute zero as a reference point, doesn't it?)

Incidentally, why is the degree immune to disappearance in division unlike all other units? That is, if I divide a distance (m) by a speed (m/s) I get a duration (s). From sheer memory, degree doesn't work like that. Any clue why?

Degrees aren't a 'unit' indeed because the zero point isn't a real zero. Hence you don't get correct scaling.

Except for the degrees that measure an angle.

No - An angle can be measured from an arbitrary point, and thus "0°" is not a true zero.

"0° on a circle" (for example) doesn't make sense. It has to be 0° _from somewhere._

A length can be measured from an arbitrary point too, no? And in some cases must be.

The perimeter of a circle for example.

- - -

Further, Radians is another measurement of angle, but it is not "degrees radian"

180° on a circle makes a lot of sense, it’s half the circle. Just like a meter makes sense, it’s a unit of distance.

That is true, but if you're talking about a change in temperature (the temperature is rising by 2 degrees per hour, or the temperature falls by so many degrees per km of altitude) then there's a real zero, of course.

> Incidentally, why is the degree immune to disappearance in division unlike all other units?

I understand why one could say that, but I’d say it’s more misleading than helpful. (And the reason why is directly connected to the problem TFA discusses.)

Instead of temperature, let’s look at time first.

If you’re measuring the temporal extent of an event, then it absolutely makes sense to divide 30s by 15s or 6 days by 3 days and obtain a dimensionless 2: one thing takes twice as long as another, and the factor 2 that doesn’t depend on the choice of units, so it’s dimensionless.

Now consider dates. The year AD 2022 denotes (not very precisely) a point in time by specifiying its distance from a fixed reference in years; as another example, astronomers use the “Julian day”, which is a (possibly non-integer) number of days from a different reference point.

And you could technically divide AD 2022 by AD 1011—the ratio between the distances to the reference point is indeed a dimensionless 2, and if you counted days from it instead of years it’d still be the same. But it’s just not a very useful statement. And if you divide the Julian days instead, you’ll get a different number, because the reference point has moved 4712 years. So you can divide the time coordinates as long as the chosen origin stays in place, but you can’t really divide points in time.

Similarly, when you’re talking about differences in temperature, degrees work like any other unit, but temperatures themselves cannot be divided in the standard scales (without tying yourself to the scale): you’ll get the same result in Celsius and Réaumur (using the same origin) but a different result in Fahrenheit (using a different origin).

Of course, the silly part is that there is a natural origin for temperature, it’s just that the usual scales don’t use it. But then there’s a natural origin for time as well (and you’ll occasionally see it used in cosmology), it’s just that it’s metrologically inconvenient (being known only very approximately) in addition to being very far away.

(Library assignment: what defines a coordinate system on an affine line?)

The root of the word "degree" basically means "units" or "subdivisions" or "measurements" or "amounts".

We didn't really have a better word for units of temperature, as "Celsiuses" or "Farhenheits" didn't become a thing, so "degrees (of) Celsius" or "degrees (of) Fahrenheit" is kind of what we're left with. We could have chosen to say "degrees (of) kelvin", but we decided to go with "kelvins" (like "meters" or "seconds") instead. It's just a choice of language.

Temperature does have dimensionality, but it's weird because it's a bulk property of a particular material. If you consider two stars, you can add their masses, or their volumes, and the resulting values are meaningful. But you can't add their temperatures. You can't even average their temperatures meaningfully without knowing the relative sizes of the stars and a bunch of other information either.

That said, I can't think of a situation where degrees (of temperature?) would disappear in a mathematical operation. Can you give a specific example?

> That said, I can't think of a situation where degrees (of temperature?) would disappear in a mathematical operation.

I regret that the decades that have elapsed since I studied physical chemistry & thermodynamics don’t permit me to make a proper answer, but if you want this to make sense I believe you should look many orders of magnitude smaller- Boltzman’s derivation of the ideal gas law from Newtonian mechanics using statistical methods.

Temperature is basically thermal density; it’s not really a fundamental unit. Chemical engineers use steam tables all the time to convert temperature to heat or other quantities.

> Incidentally, why is the degree immune to disappearance in division unlike all other units?

It's not. If you divide temperature by temperature (T/T) you get unitless value. But it doesn't make sense (what does it mean Kelvins per Kelvin?) Your mistake is to take a scalar value for temperature: 300 K / 2 = 150 K (T/scalar = T).

Degrees are used for many things besides temperature. Angles obviously, but is also common in some countries for alcohol concentration (same as %ABV) and other specialized units. Essentially, "degree" means "unit", and "degree Celsius" means "the unit of Mr.Celsius". SI dropped it for Kelvin probably for consistency and because it is redundant information, we already know that Kelvin is a unit, no need to specify "degree". It also adds a funny character to the abbreviation, and would be the only unit in two words.

It isn't because it is a relative scale, Rankine uses degrees and it is an absolute temperature scale just like Kelvin. The difference, I'd say, is that the Rankine scale didn't go though the SI standardization process.

There is nothing special with division, "degree Celsius" is the unit, you can't separate the "degree" from the "Celsius". You can measure a thermostat dial in degrees Celsius per degree (of angle). The resulting unit is just that, not "Celsius".

You’re right in modern usage, but it is common to see °K in older documents.

You don't need an absolute scale, you just need a sensible origin. I suggest taking human perception into account.

When talking about the weather or everyday objects, one could argue that the border between "hot" and "cold" is somewhere at room temperature, so around 21°C. At that temperature, it's neither hot or cold. If it's 30°C, it's 9°C of "hotness", so 39°C would be twice as hot. I don't know if it feels twice as hot, as that's quite subjective, but if you want more precision we should switch to another vocabulary entirely.

36.6C?

How given temperature feels depends on humidity, airflow, local culture, individual human and specific condition of that human (sitting/moving/sleeping, caffeinated, just ate a meal, bored, enthusiastic).

- I know someone who wears a puffy jacket indoors if it's 24 degrees.

- You will find sitting in 20 degrees with 88% humidity much more miserable than 20 degrees and 33% humidity.

- 42deg with 20% humidity and slight breeze is fine, 38deg with 90% humidity can make you pass out in minutes if you need to walk anywhere in the sun.

Water-freezing temperature is an absolutely sensible zero. The difference between e.g. 2 °C and -2 °C is really quite obvious for the exposed skin: while it's merely "yeah, it's cold" feeling when it's above zero, it almost feel like the frost is biting, almost literally, when it's below zero.

This is not true, and it's why we developed separate scales for wind chill (which factors in wind) and heat index (which factors humidity).

Looking at the wind chill chart[0] for frostbite times, there is nothing particularly special around the freezing point temperature, even at absurd wind speeds. Ironically, actually, the 0 point for Fahrenheit is far more relevant threshold for frostbite.

[0] https://www.weather.gov/safety/cold-wind-chill-chart

Neither of those are as important as “when water freezes” (ice snow weather effects), boils, and having a nice 100d in between

The freezing point of water is indeed one reasonable origin point but the temperature at which water boil (at sea level) isn't very relevant to most people's day to day experiences much less having a nice interval between the two points. You have to use Kelvin as an absolute scale for a lot of scientific and engineering calculations anyway although it would be pretty silly for day-to-day use.

I'd argue that boiling water (or water-based liquids) is something we encounter pretty often in day to day life (eg. while cooking, brewing tea, ), making it a reasonable reference

But then this is also valid if the zero is at 21°C. There is also nothing special at 0°F, it’s just the colour scale that makes it look like that. And if we based the scale on human perception we’d have to define many other parameters besides wind and humidity, like body shape and metabolism rate. It’s more of an argument against basing a temperature scale on human perception (and then I agree completely, which is why I find the centigrade scale much better, except for serious business that demands kelvins).

Yeah, it's not like Fahrenheit or any temperature scale is literally calibrated to human perception (they're all now defined as transformations from kelvin anyways), that's just a (minor) ancillary impact that can be discussed.

Wind chill is a mostly stupid number that weather reports like to trot out so they have a amped up "wow it's really cold" factor. Yes, a number of factors p[lay into our subjective perceptions of temperature but you're probably not walking around in a bathing suit when it's below freezing with a howling wind.

Sure, same with heat index. They are tools for communicating the impact of additional axes (wind speed for cold and humidity for heat) on subjective perception. I won't wander out without a scarf, or neck covering, if it's freezing and 40mph winds, but I don't mind to if it's freezing and no wind.

I don't think they are aiming for the wow factor. Most people arguably watch the weather report to find out how cold or warm it feels outside, which heat index and wind chill are far better at reporting.

To take ratios you need an absolute zero because otherwise ratios aren‘t even defined.

See also my other comment here: https://news.ycombinator.com/item?id=37928268

Dealing with any temperature difference requires an absolute scale as well.

The difference between 20°C and 25°C is 5K.

Twice as hot is also useful in Celsius. For every increase of 10C chemical reactions happen twice as fast, it will literally, thermodynamically be twice as hot.

Which will probably even match your subjective feeling.

Celsius scales thermodynamically logarithmically. You can’t just double the numbers. (The same with sound etc.)

Reaction rates are approximately exponential in temperature, but there is nothing special about 10C. The doubling temperature difference depends on what reaction you look at.

This is as opposed to decibel for sound, where +10 decibel means X10 power exactly.

To be clear, that's true of power decibels for everything, not just sound.

>For every increase of 10C chemical reactions happen twice as fast

Except for the ones that go at different rates.

The Celsius scale is an "interval scale" while the Kelvin scale is a "ratio scale". One cannot take ratios of Celsius temperature values as ratios are not even defined for that. It’s possible for differences of temperatures on the Celsius scale, though.

https://en.wikipedia.org/wiki/Level_of_measurement

> We often don't realize how warm the world we live in truly is, from a physics standpoint.

Or cold? There is a big ball in our solar system sitting at tens of millions of degrees K, and you think 273 degrees is warm?

The Python package "pint" (to manage units) implements that with different units for absolute temperature vs delta temperatures.

https://pint.readthedocs.io/en/stable/user/nonmult.html

A useful habit (which I think I picked up while studying physics at sixth-form) is to describe temperature intervals in “Celsius degrees” (and specific temperatures in “degrees Celsius”).

So a temperature of 43 degrees Celsius is 18 Celsius degrees hotter than 25 degrees Celsius.

32 Fahrenheit degrees are equivalent to 18 Celsius degrees; but 32 degrees Fahrenheit is equivalent to 0 degrees Celsius.

(There's no real need to do this when using absolute units, but I'd still pluralise the interval and not the absolute value: 316 kelvin is 25 kelvins hotter than 291 kelvin.)

Temperature difference used to have a separate unit with the symbol ‘deg C’ in an old edition of the SI system.

Modern SI no longer bothers, and just calls the unit ‘kelvin’. Why should you?

https://en.wikipedia.org/wiki/Level_of_measurement

I learned this relatively late in my life, but it was so an insight. From one hand it is just obvious facts, but from the other hand they are nicely categorized and so feels like you have learned something completely new.

It's not hard at all, the localization team just made a mistake.

The solution is to not localize for Americans.

Yes just ignore the vast majority of native English speakers. I wonder why an English-language newspaper (which heavily markets itself outside the UK) didn’t think of that.

Or we can just use a calculator. I'm sure most people's phones can even translate it by a voice command "What's 20 degrees Celsius in Fahrenheit?"

That will give the same wrong answer this post is discussing.

Yep https://www.google.com/search?q=%22What%27s+20+degrees+Celsi...

There's a decent chance that the author of the incorrect headline did exactly that.

"vast majority" is a bit of a hyperbole; there are more English speakers outside of the USA than in it, though it's close to 50/50, and it's indeed the country with the most English speakers.

I did say native English speakers, but regardless, the point stands that the US is an extremely important market for any media targeting itself to the broader Anglosphere.

> Anglosphere

If you're moving the target to the Anglosphere, we have to count the whole population consumming content in English as their second of third language, so the US becomes a way tinier portion of it.

I'll give you that for The Guardian, a US newspaper with US subscribers, the US market is extremely important. The broader case is harder to make.

> If you're moving the target to the Anglosphere, we have to count the whole population consumming content in English as their second of third language, so the US becomes a way tinier portion of it.

By “Anglosphere” I don’t mean everyone who speaks English. I mean the countries whose primary language is English and whose majority culture originally descends from England: at least the US, UK, Australia, Canada, and New Zealand. (A few other countries like Ireland or parts of South Africa don’t fit the strict definition but have a long and close association with England so can be thought of similarly. My point here is not to debate to what extent Ireland is culturally similar to England which is of course a controversial subject).

People in these countries consume much of the same media and are much more culturally similar to each other than they are to people in e.g. India or the Philippines, even the ones who speak native-level English.

> The Guardian, a US newspaper with US subscribers

The Guardian is British.

> The Guardian is British.

Touché.

> By “Anglosphere” I don’t mean everyone who speaks English

Thanks, point taken.

The stats I gave was for English as a first language.

If you included non-native speakers, the stats would be much more lopsided than nearly 50:50.

You're absolutely right and I concede that I misspoke initially. However, I think the main thrust of my point is correct: that the US represents a huge chunk of the Guardian's target audience and therefore suggesting that the Guardian not localize its articles for Americans is absurd.

Why would it be absurd? What exactly prevents an American from reading British English?

Maybe it helps to push for Murica to finally adapt SI units.

All American countries outside the USA already use °C

https://www.cia.gov/the-world-factbook/references/weights-an...

> Maybe it helps to push for Murica to finally adapt SI units.

Nitpick: Celsius is not an SI unit.

Answering the substantive point: it’d be nice if we switched to SI units (plus Celsius); personally my phone is already set to display those and I wish we’d join the rest of the world and do standard stuff. However, I think it’s extremely unlikely to ever happen. The deadlocked US political system makes it impossible to enact any reform on any topic that anyone can create controversy about, and that includes switching to international standard units.

It's on the way :-)

https://en.wikipedia.org/wiki/Metrication_in_the_United_Stat...

Some stuff in the US is metric, and some isn’t, but as far as I can tell the situation of metric units in daily life is static and not changing much.

And it was ok the way and if regan hadn’t canceled it America would have switched around the same time Canada did

Per wikipedia

> The degree Celsius is ... one of two temperature scales used in the International System of Units (SI), the other being the Kelvin scale.

Edit: To expand, within SI Celsius is defined as a derived unit, and Kelvin is a base unit.

It seems you’re absolutely right and my nitpick was not correct.

Pretty much ALL countries use C - 97% of people on the planet use C

C is the global standard

An affine space is a tuple (𝗔, 𝗩) where 𝗩 is a vector space and 𝗔 has the same structure as 𝗩 but has "forgotten its origin"; ie. unlike a vector space, has no privileged zero point. As such, multiplication by a scalar is not meaningful, and neither is the addition of two elements of 𝗔. The elements of 𝗔 are called points, and the elements of 𝗩 are translation vectors, or simply translations. The fundamental operations are:

(−): 𝗔 × 𝗔 → 𝗩 — gives the translation from a point to another point, and

(+): 𝗔 × 𝗩 → 𝗔 — applies a translation to a point.

(For convenience we can also write 𝒂 − 𝒗 = 𝒂 + −𝒗 for all 𝒂 ∊ 𝗔, 𝒗 ∊ 𝗩).

The Celsius and Fahrenheit scales are affine, and technically should have separate units for temperatures and temperature differences so that mishaps like in TFA would be less likely to happen.

> it is not immune to errors

They’re pretty much part of the guardian’s dna:

> Frequent typographical errors during the age of manual typesetting led Private Eye magazine to dub the paper the "Grauniad" in the 1970s, a nickname still occasionally used by the editors for self-mockery

My father still refers to it as The Grauniad

c.f. Private Eye's "The Getelarph".

Hard? So is using the degree symbol apparently: not a single one in the whole article.

(68F is not a temperature — it’s a seat number.)

https://degreeswhat.com/?68

"To be fair", there is ASCII in the name

"To be fair..." (chorus of Letterkenny voices)

I appreciate the ASCII-only excuse, but I've corrected the omission anyway. Non-ASCII characters in randomascii - what will they think of next?

Thanks for that!

Pssst... You over-corrected

> 278 °K to 283 °K,

There's no ° symbol before the kelvin unit, just a regular non-breaking space if you want to be SI-compliant.

You may have overcorrected a bit; 300 Kelvin is written 300K, not 300 °K. (The system is a mess.)

It is written "300 K", with a space in the middle. Possibly even a non-breaking space.

Definitely a non-breaking space.

To be double fair, the page says in the bottom "And worry about whether this blog should have been called randomutf-8"

    68:degrees:F

No, 68 farads is a capacitance. Pretty massive one, too.

It's another question how they obtain that from coulombs...

That would be 68 F [1]. I agree with parent, 68F is a seat number.

[1] https://physics.nist.gov/cuu/Units/checklist.html (number 15).

After using siunitx[1], I have taken great care to ensure the spacing of measures and their corresponding units.

[1]: https://ctan.org/pkg/siunitx?lang=en

That would be 68 F, not 86F.

This is taking being a grammar nazi to a whole new level.

That's grammar Nazi

I don't know how to tell you, but language prescription is not actually morally equivalent to genocide.

Way back in high school precalculus, I had a particularly hard (and hard-ass) teacher. I generally scored in the 85-95 range, but one exam I was shocked to get a grade in the 60s. When I asked the teacher about it they pointed out that since I hadn't written my angles with the degree symbol (e.g. 45˚) that meant they were in radians, and thus very (very) wrong.

Never did that again.

Delta's like temperature are standard issue to resolve in globalization. Another fun one are complex units that are a combination of more base units like luminance (nits = candela / m*2 or mph.

I worked on a package to globalize a consumer-oriented fintech reporting product. With ESG reporting this has become increasingly important to get the unit conversions and Delta's right. You end up mapping the data source units to locale targets.

In Python, the Pint framework plus Babel locales are excellent.

How do frameworks deal with the miles/gallon -> liters/100km conversion? I've always found that one particularly interesting as one that requires inverting the units also.

It’s even worse, the US has different gallons from the rest of the world.

To be fair, imperial gallons are based on 10 lbs of water (with very specific atmospheric conditions), which while a nice round number, isn’t as easy to recreate as the U.S. gallon, which is 231 cubic inches.

Of course, liters make way more sense than either of these.

Out of curiosity, how dit it end up being that weird 231 number? Why not 230? Or a nice 200?

Wikipedia says it was defined as cylinder of d=6” and h=7”, and they later redefined it to use the approximation of pi as 22/7. 22/7 * 7” * (7”/2)^2 = 231 in^3. Not approximately equal, precisely equal. It’s an interesting math oddity that I hope someone else will explain.

Also, I’ve no idea if that history is true, but it sounds plausible.

Which gallon? There are two different ones (and for even more fun Canada and us don’t use the same)

I blogged about that also:

https://randomascii.wordpress.com/2016/03/01/cars-in-canada-...

In Pint you can define custom units in a unit registry. So you make your complex unit out of predefined units and the underlying unit conversion is built-in. Pint also works with pandas/numpy so you you get vectorized operations too for performance.

Well, I get how mathematically you do it, I'm talking about how to make it contextually aware like the OP. A sentence like "The latest generation doubled MPG from 15 MPG to 30 MPG" should be localized to "The latest generation halved L/100km from 16 L/100km to 7.8 L/100km".

The unit registry maps units to locales (eg EN_US, DE_DE, NL_BE) you pass the configured locale to your getter call for the unit. If you need unit strings, you can pass the locale to the string formatter.

I feel like we're talking cross purposes. A sentence like "MPG increased by 20%" should be translated to "L/100km decreased by 16.6%", a sentence like "MPG increase by 5" is untranslatable without more context.

I understand if you are getting a human to translate each sentence you can just point them to the right phrases but my understanding is that you're translating text in bulk via machine translation. For scales in which the conversion is linear and the starting point is zero, this task is relatively trivial. For scales like temp where the zero points are different, you need to treat deltas different from absolute values. For scales where one is the inverse of the other, there's a lot of very tricky edge cases that make machine translation hard.

As far as I can tell from googling, nobody ever says it’s “twice as hot” except as the figurative name for something or to say that it’s wrong to say “twice as hot”.

GPT-4 correctly converts the difference to 36F. GPT-3.5 on the other hand gets crazy and decides that the difference is 154.4F. Same prompt. At this point I'm starting to believe GPT3.5 is dangerously stupid.

I worded the question as: "What's the temperature difference between 43°C and 23°C in Fahrenheit?" and GPT3.5 answered correctly.

I put in the original sentence and got 122°F first. If I tell it to rewrite and correct C to F I randomly get 20 and 68 every time I ask it.

If I say "recalculate the celcius to farenheit rewriting as little as possible in this sentence: Malawi swelters in record heat with temperatures nearly 20C above average"

Attempt 1. Malawi swelters in record heat with temperatures nearly 36°F above average.

Attempt 2. Malawi swelters in record heat with temperatures nearly 68°F above average.

The book Inventing Temperature by Hasok Chang is a wonderful account of how hard it has been to grapple with temperature scientifically.

My favorite anecdote is the scientist who carried arround and shook a bottle for two weeks to distill the water for experiments in finding when water boiled. When the water exploded after heating he found that not only did he not know when water boiled, but now he wasn't even sure what boiling was.

From a software engineering perspective, this shows that physical units are not a fool-proof typesystem, and you need alternative units (types) for differences.

Isn't ths following incorrect.... somehow? Something doesn't line up.

----

Reading the article I found this:

parts of Malawi saw a maximum temperature of 43C (109F), compared with an average of nearly 25C (77F)

As I expected the actual temperature increase was 32 °F, not 68 °F.

---

Except the error should have been 32 (which you go on to explain, because we add 32 to start a conversion) and the correct temp difference should have been 36 (20 * 1.8) which is shown later in the article.

I can see getting that wrong, but the actual temps quoted seems to correctly portray 32 opposed to 36 like we would expect.

Just rounding errors or some other oddness?

The original article started by rounding an 18 degree difference up to 20.

Yeah, I was a bit imprecise there. I didn't expect a difference of exactly 32, but I did expect something of that magnitude.

Looks like the Guardian have already amended the temperature differences in the article in question according to the footnote: https://www.theguardian.com/world/2023/oct/17/malawi-swelter...

It says 20C for me and I’m in the US/have everything set to US English. Is my browser telling it I prefer Celsius somehow (I do have the weather app on my phone set to Celsius) or did they change it to Celsius for everyone?

> “The temperature is 20 °C” translates to “The temperature is 78 °F”

Correction, 68°F.

> So 20 °C is either 78 °F

It's still 68°F.

A bit related, but I hate how many errors I get because I want my system language to be English instead of my native language (Norwegian) on my laptop, phone etc. Because I don't want a badly translated experience. However, I still want units in meters, Celsius, dates written sensibly, time in 24h format etc. But not all apps understand this, which is quite annoying. It's like they only look at the language and make a decision, even though on OS level there's more fine grained locale settings.

Your browser doesn't give away this information. It's inconvenient in cases like yours but on the other hand it would be yet another signal in browser fingerprinting

Browsers always send the accept language you configured.

It drives me nuts, most sites are geolocating me instead and serving me Spanish content. My browser is always sending accept language, and it's most often ignored than not, even frommthe big old web players (Google and Wikipedia to name a few)

This used to work back in the old days, but more and more sites decides to ignore this a few years ago. The reasoning is that the percentage of users with misconfigured browsers (who don't know how to fix this and also don't know how to click a "change language" button on the website) is bigger than the percentage of users who intentionally use a language that's different from their geolocated one. So the ad-revenue maximizing choice is to ignore the user's configuration.

That’s why I like travelling: I’ll get all my web content in English (the language I do know) and the ads in a language I don’t know (perfect).

It used to work on Facebook to just change your location to a weird country and you would get their advertisements. Or none if there weren't any like for North Korea. Then I got banned on Airbnb (where I used Facebook login) because they thought I'm from North Korea and they would break sanctions by renting me a bnb...

I can select English UK and that seems to work pretty well.

That's my solution in some places too. I generally want US spellings of words (especially in anything related to programming!) so "en_US language, Norwegian units/time/etc, but US number format" is my preferred set-up, but "UK everything" is an alright approximation of that for software which doesn't have that much specificity.

Try Canadian English: en_CA. That might be an even closer approximation of what you want. (They still spell some words the British way, but a lot the American way).

It really is incredible how many translated apps/websites talk like a drunken finn, when i set language to finnish. Only the things made by native finns seem to have good Finnish on them.

It's the same in Dutch. In the past, Microsoft had very good translations, but even they seem detoriating.

Open source is also quite reasonable, except from the latest features that didn't get translated yet.

Apart from that, only local software seems to understand the language.

There is hope, however. ChatGPT does quite well translating. You can still see the original English words in the Dutch text, but at least it doesn't descend into total gibberish.

GPT-4 is actually quite good with languages. You can even get it to do some local Dutch dialects. We use it (via the code gpt plugin) for helping us with localization files for apps. We just give it whole chunks of localization files (mozilla fluent) and task it with translating those while preserving the id structure. Aside from minor formatting issues, I've not caught it making any mistakes. I doubt we'll be working with external contractors for localizations again.

It's also really good at doing things like synonyms. We used this to improve our icon search by letting it come up with comma separated lists of alternatives for each icon name. Worked really well too.

And it can do fictional languages. We did a pirate speak localization a few weeks ago for international pirate day. Just for giggles but it works.

I think the problem stems from translators being sent a bunch of strings with little context. At least even when we hired professionals it could sometimes be a bit wonky, especially with shorter texts.

For instance the word "speaker" can mean both someone that presents, or a loud-speaker playing music ("høgtaler"). If you tell a translator to translate the string "Speakers:" for a heading, they can't know the context. I once saw a conference program bragging about their loud-speakers in my language..

IMO the worst is that cal starts weeks on Sundays for some weird reason.

As long as the boundary doesn’t happen in the middle of the work week, I find it completely arbitrary whether the official first day is Saturday, Sunday, or Monday. Why’s it any more weird to start on one or the other?

It just "feels wrong". Like, why is the weekend split in half? Doesn't map to my mental model of the world (which of course is influenced by how I'm used to seeing calendars).

Related, but a few years ago the Norwegian broadcaster NRK asked people how the visualize a year. https://nrkbeta.no/2018/01/01/this-is-what-the-year-actually... (English version)

Quite interesting how different people are. Mostly it's a clock of sorts. To me however it's more of a sine wave, where each year is one phase (summer on bottom), but it's kinda fractally so if I zoom in it got bumps for the months, smaller bumps on there again for the weeks etc.

That is interesting. I don’t really have a mental visualization for the concept of “year” at all. If I had to try I guess I’d just imagine a calendar (twelve month-pages).

I wonder if visualizing a year as a circle is a particularly Norwegian thing, or if I’m the odd one out anywhere…

Back to the subject of which day the week begins: I’d say the fact that the week begins on Sunday doesn’t have much of a meaning in US culture, if any. If I say “next week” at work, it clearly means “the next M-F period”. If I wanted to refer to Saturday and Sunday I’d say “this weekend”. So if what I normally mean by a week begins on Monday and ends on Friday, you have to slot the weekend on one of those ends, and it feels slightly wrong whichever way you do it.

So, the only practical, concrete way in which “the week starts on Sunday” means anything to Americans is that we are used to seeing calendars printed with Sunday on the left, and Saturday on the right. I continue to maintain that this is arbitrary! But it does make me really confused when I have to use a calendar widget in an interface that wasn’t localized, and starts with a day other than Sunday, so I get your frustration.

It's not, but it's annoying when there's a difference, and you won't know which day "Sunday week 45" refers to without doing a thorough background check on the guy who said it.

It's a bit like decimal point vs. decimal comma. Neither is inherently better, but man would it have been nice if we'd all picked the same one.

> man would it have been nice if we'd all picked the same one.

No disagreement there!

I personally like this (at least when displaying the calendar) -- on Sunday, I can have a look what is happening "this week" :)

Blame the bible. God created the world in six days, and rested on the seventh, which was Saturday.

The real question is why the weekend was defined as straddling the last and first days of the week. Weird.

I'm Italian and I have the same problem. You just have to set the system to English (UK) and all your problems will be fixed :)

Perhaps not all of your problems! I have my phone in en-gb and live in Switzerland. When I travel to the US it starts showing me temperatures in farenheit

When I travel to the US/UK I _do_ actually want to start seeing distances in miles (I'm "bilingual" with m/km, it's useful to think in distances that locals understand and that match signage). So "just use my home locale's units all the time" wouldn't work (especially for speeds!). So yeah I guess localization is indeed hard!

Honestly I've now just started trying to learn Farenheit! 90 is hot, 70 is nice, 50 is cold, 30 is ice.

I was also pleasantly surprised to see that the "C" locale seems to default to metric units and a 24-hours date format.

Except now the web browser will offer to translate every Italian website you visit.

You can easily prevent that by choosing "Never translate Italian"

I usually set my system to English / Ireland. I get English but with normal units and € as a default currency.

I think en_IE is a rather undersold/underpromoted "international/european english" locale that more people should know about!

I was recently recommended English (Canadian) for that reason but I've not yet managed to get Google Maps on my mobile to show me km instead of miles, no matter what setting. (It's not the account, because a desktop browser on a Windows with en_US and accept-language en/US will show km with English text). It is indeed infuriating.

Many software also have a "English/Belgium", "English/Germany", or "English/Europe" locale.

A fun trick is that if you install Windows 11 with the "English/Europe" locale, it doesn't install a lot of the bloatware because it's not defined for that location.

You don't know the pain until you move to Switzerland. Geolocalisation to German websites or language, numerical format is a mess (sometimes thousand separators are either ' or space, huge mess in excel), cloud Saas providers believe only fr-FR or de-DE exists. Just awful...

MS Visio refused to run in anything other than English on a PC I used because Windows was set to a specific flavor of German which Visio didn't have in its language list. As far as I could tell there was no way to change the program's language other than changing the OS language and reinstalling Visio.

Now Firefox added stupid popups offering translating websites written in my native language because the OS is set to English. And they didn't see a need to create a checkbox for that outside of about:config (Mozilla, if you want to get rid of me you can just say so instead of spamming me with popups)

Pretty sure you can just click the dotted menu button on that popup and select "do not translate from $LANG".

But it still shows an icon on every non-english page despite me disabling the entire feature. Like back then with the unremovable Pocket button. What's the point of the program letting me unclutter the UI with the "customize toolbar" feature when it's constantly blocked for Mozilla's annoying feature of the month?

Hello, fellow mountain ape. I get Norwegian text in /some/ apps on the Mac even though Norwegian is 2nd in "preferred languages." I guess some use "Region" to decide instead of the logical setting. And setting the region to something else will cock up in other interesting ways by showing wrong units instead of the wrong language.

Even the keyboard on my MBP is physically US ANSI, because if you've ever coded more than a hundred lines using brackets on an ISO keyboard you know why.

I managed to get everything right in KDE on my PC though, by some weird way of using "Danish (UK)" or whatever it's called in some specific settings -- I don't even remember how or why but everything worked just right. Sorting, decimal point, app languages, everything. Could never replicate it though.

Yeah, one additional thing is that for Norwegian apps, I want the Norwegian version and not the English one. It's hard to get right, I guess. But what I generally want could be summarized as "give me the original version of the app if in Norwegian, if not give me the English version".

Same here,

> However, I still want units in meters, Celsius, dates written sensibly, time in 24h format etc.

but also sensible sorting. I'm currently mixing en_DK, en_IE, en_US, C and cs_CZ (my native language) locales for various LC_*, but it's still not perfect and some apps (such as Thunderbird) ignore this anyway and others (Slack) have only one language setting, so I can select either US with 12-hour time or UK with UK spelling. Does anyone have a solution? Should a write a new locale?

One thing that really annoys me is youtube translating video titles - why would I ever want the title of an english language video to be translated?

If I didn't speak English, I'm sure I would appreciate it. E.g. to swap the roles, while I'm sure some wouldn't mind but I'd guess that the majority of people who watch anime appreciate the anime titles have English translations, even if they watch in Japanese (with subtitles).

But YouTube uses some weird translations, this behaviour is not fully consistent and also YouTube knows that I don't watch videos of those channels with subtitles.

Um, not sure it has any point except if there are available subtitles in your language though

It does annoy me when I see it at my sister's computer, but I speak English. I can see how somebody who doesn't speak English would enjoy seeing some video titles in their own language.

Some videos really only need the title translated and can be now enjoyed by more people: Top 5 Mike Tyson KOs, amazing woodworking techniques, best trick shots of 2023 compilation, dash cam accidents.

If you visit YouTube with your account logged out, these videos are popular and always show up as a recommendation.

Also, some language combinations have pretty good "auto translate captions" experience, so you can turn on the subtitles, too, then you can even enjoy a bit more sophisticated content.

At least you get English videos. A couple of times I've gotten videos in... Hindu, I think, by searching in English and clicking on a video with an English title. And it was a technical video, not much language-independent content.

Reply on news.ycombinator.com