#lingvistika

[hlas-lidu]

Jak se říká těm věcem v posteli, těm dlouhejm tyčím?????

Špricle????

Šprcle?????

Špecle???

Špicle????

jak se říká těm věcem v posteli, těm dlouhejm tyčím?????

špricle

šprcle

špecle

špicle

myslíš fucking rošt????

nikdy jsem neviděl*a vnitřnosti postele / rumový extrakt

Voting ended onDec 29, 2025

[sire-rizku]

těžký život studentů lingvistiky. i dort vám zkazí.

[twopplinatrenchcoat]

Mám úžasnýho kámoše kterej obecně mluví hodně spisovně správně a na veřejnosti ale bohužel se mu to nedaří přenést do psané konverzace takže jsem od něj dostala "gratulky" k maturitě👍 co to má být jako za slovo lidi

[suspiciouscatastrophe]

An Experiment With Machine Translation/AI

Hello there, my friends! Usually, I'm posting about trans stuff. Today, I'm going to switch it up for translation.

So you see, I'm a student and translation studies are one of the degrees I'm desperately vying for. This week, one of my classes had an interesting assignment: We got an excerpt from a book (The Map of Us by Jules Preston) and our goal was not only to translate it but also create a second version, using a machine translation or ChatGPT. A lot of translation work nowadays is being more or less AI's factchecker and editor, so like it or not (and you can believe me that I don't like it), it's a skill to have.

And I thought I would share the comparisons! They're not surprising. They all show that unedited AI or MT is not by any means ready to replace human translators, the least of all in art. However, the ways they fail are quite interesting!

I was also fully prepared for ChatGPT, being, you know, a language model, to blow poor Google Translate out of the water. And while it was definitely the more creative of the two, it also performed... worse overall...? You can be the judge of that!

And of course, this post is best enjoyed if you're a Czech speaker, but if you happen not to speak Czech or Slovak, don't hesitate to read on anyway, I'll provide additional translation back into English for the fails.

And... if you happen to be a human with a penchant for translation and you're cringing at my solutions, trembling with the need to correct me, you will make me cry, but go ahead.

Example 1:

ENG: She did not ask his opinion, for he had already shown himself to be difficult and ill-tempered when it came to making a choice.

HT (human translator): Na jeho názor se ani neptala, protože už věděla, jak protivný dokázal být, když se musel rozhodovat.

GT (Google Translate): Neptala se ho na názor, protože se už ukázal jako obtížný a špatně naladěný, když došlo na volby.

[She did not ask his opinion, because he has already shown himself to be hard to do and ill-tempered when it came to elections.]

AI (ChatGPT): Neptala se na jeho názor, protože už ukázal, že je těžko rozhodnutelný a náladový, pokud šlo o výběr.

[She did not ask his opinion, because he has already shown himself to be hard to decide and moody in the matter of selection.]

In this one, I think the bots reached a tie in being wrong. Machine translation is notoriously wonky on composite sentences, however ChatGPT genuinely surprised me with somehow managing to "think" a little forward and already connecting "difficult" (těžko) with "making a choice" (rozhodnout se) and laboriously birthing "těžko rozhodnutelný" which would mean "difficult to decide/judge" but in the context of e.g. a competition being tight and it being difficult for the judges to choose the winner. It's not used with sentient beings at all. In my version, I've decided to merge the descriptors "difficult and ill-tempered" into one word "protivný" which is a little more general in meaning, so it encompasses both and drives home the idea of being a neutral way to call out somebody's mean and problem-generating behavior.

Example 2:

ENG: Every time she would have to press a small button and the top of the Royal Quiet Deluxe typewriter would pop open automatically, making the interior of the machine accessible.

HT: Pokaždé, když se to stalo, musela zmáčknout čudlík, aby se horní část jejího psacího stroje Royal Quiet Deluxe automaticky odklopila a ona mohla sáhnout dovnitř.

GT: Pokaždé by musela stisknout malé tlačítko a horní část psacího stroje Royal Quiet Deluxe se automaticky otevřela, čímž se zpřístupnil vnitřek stroje.

(no fail here as far as I can tell)

AI: Každýkrát musela stisknout malé tlačítko a víko psacího stroje Royal Quiet Deluxe se automaticky otevřelo, což umožnilo přístup k vnitřku stroje.

[Everyonetimes she had to press a small button and the top of the Royal Quiet Deluxe typewriter would pop open automatically, making it possible to access the interior of the machine.]

Now, the word "každýkrát" means "every time," however, in SLOVAK, not in Czech. Two completely different languages. In Czech? "Každý" means "everyone" and "krát" means "time(s)" (as in multiplying something), so "everyonetimes." I'm quite sure that passing the word as Czech counts as an anti-Slovak microaggression XD Google Translate slayed here, no mistakes, I would just word it differently because I've got that Human Touch TM and an insufferable superiority complex.

Example 3:

ENG: There was an 'e' in 'nose' and 'ear.'

HT: Nos i ucho měly 'o.'

This one is rigged against both GT and AI because what they do is gluing the most probable translation equivalents together, and they have no way of knowing that the letter the words "nose" (nos) and "ear" (ucho) share in Czech is "o" and not "e". GT performed as expected.

GT: V 'nosu' a 'uchu' bylo 'e'.

(besides making it sound a little like the letter was physically in someone's nose and ear, it's fine)

But ChatGPT? I wouldn't actually expect it to solve the problem and correctly switch the letters, but given that it is supposed to be much more advanced than GT, I had my hopes high and thought that it would show at least a trace of acknowledgment of the part being somehow difficult. AI: Bylo tam 'e' ve slovech 'nos' a 'ucho'.

[There was an 'e' in the words 'nos' and 'ucho'.] (liar)

Despite those words noticeably lacking in the "e" department, it's worth noting that while GT went out of its way to try to make it sound natural (too natural, in fact), its supposedly smarter cousin fully copied the English sentence structure, which is technically gramatically correct but sounds extremely reverse-Czenglish. (I guess if Czenglish is a Czech-speaking learner's English influenced by Czech, that makes an English-speaking learner's Czech... Ech. It sounds extremely Ech.)

Example 4:

ENG: Did it matter what an imaginary explorer's lips looked like or felt like?

HT: Záleželo vůbec na tom, jak vypadaly rty imaginárního dobrodruha, nebo jaké byly na dotek?

GT: Záleželo na tom, jak vypadaly nebo jak vypadaly rty imaginárního průzkumníka?

[Did it matter what an imaginary explorer's lips looked like or looked like?]

AI: Záleželo na tom, jak vypadají nebo jak se cítí rty imaginárního dobrodruha?

[Did it matter what an imaginary explorer's lips looked like or how they were feeling?]

Do facts care about the your (lips') feelings? As someone who likes to play around with GT quite a lot, I'm very familiar with the "when in doubt, just repeat whatever you've last said" strategy. If you fuck with it long enough, you can make the poor tool repeat one phrase so many times it fills the entire line. However, the mistake ChatGPT made perplexes me. It feels very humanizing. It's a mistake I can easily imagine an actual learner making, failing to distinguish between "to feel like" (comparison) and "to feel" (to experience a feeling). It's a mistake I would totally make several years ago, and fuck it, maybe even now if I'm trying to read fast on a very bad day. Good job at doing a bad job, GPT.

Anyway, these 4 were probably the funniest and most interesting of the whole excerpt. I must admit that analyzing them turned out to be a humbling experience because revisiting my work, I've noticed several mistakes I've made as well that I can't take back, because I've already turned the assignment in. Oh well. However, I did have fun!

I hope you had fun, too! Stay těžko rozhodnutelní!

[budutinahrobnositmandarinky]

syntax je královna gramatiky

[tisicduvodu]

9. 4. 2025 Další nedoplatek, tentokrát za vodu. 7 000,–. Další dvojka za přezkoušení elektroměru, zda náhodou není vadný. Účetní majitele bytu špatně rozpočítala zálohy v rámci společných družstevních při předávání bytu v červenci, takže moje měsíční náklady se teď zvednou o další litr a půl. Majitel odmítl přiznat jakoukoliv odpovědnost, takže nájem nám poměrně snížit nehodlá, naopak aktivně vyhrožoval, že pokud se nám to nehodí, může ze dne na den sehnat nový nájemníky. Spolubydlícímu schválili insolvenci, takže brzo ztratí většinu příjmů. It's a white trash life. (Bonusové body všem, kteří to uslyší se správnou melodií.)

Dneska jsem měla pohovor na soukromé gymnázium – k údivu všech (včetně mě) mám totiž odborné kvalifikace učit několikero jazyků. Moje zoufalství z nastalé finanční situace jaksi dlouhodobá pracovní neschopnost a jeden beletristický rukopis pro Prostor nevyřeší (byť jsem za tak ochotnou a rychlou šanci vrátit se zpátky do beletrie velice vděčná). Na stránkách se gymnázium kasalo inkluzivitou, individuálním přístupem k žákům a moderními výukovými postupy. Z pohovoru, na který jsem musela čekat půl hodiny, protože paní z HR (střední školy teď mají HR?) si to „špatně zorganizovala, moc se omlouvám, to se mi normálně nestává“, jsem byla po deseti minutách neurotypicky vypoklonkována, když jsem zmínila, že mám PAS. Ach, kvazi avantgardní pedagogiko, smrdíš povědomou hnilobou.

Život si člověk tak nějak musí zasloužit, ale já jsem evidentně zatím nevyluštila, jakým způsobem. Tak se prozatím alespoň učím hangul, abych se přiblížila nemožné potřebě mého autismu ovládnout všechny dostupné abecedy.

[linka-r9-vysocina]

drazí lingvisté

pokud jsou ty vaše hláskové zákony tak strašně důležité

PROČ se o nich nedá nic pořádného najít?

ptám se samozřejmě pro kamarádku

[ghost-orion]

dělá někdo na českých novorodech? protože někteří z nás by je opravdu ocenili.