"If a lion could speak, we could not understand him" - Wittgenstein

@P-Mouse

5 ай бұрын

+

I work in tourism, and very often tourists smack a phone in my face with a google translating and expecting me to understand what they mean. It's very annoying, just either learn some basic phrases in the local language if you go on holiday, or maybe staff might even know some basics of your language. People are always better in understanding people than computers without context.

@melissahdawn

5 ай бұрын

@@saektalt it is like not learning math because a calculator can do it for them.... just lazy!

@myfaceismyshield5963

5 ай бұрын

​@@melissahdawn even math works better than language... as long as you fully understand how the calculator works. If you fully understand the calculator and its different commands, you can calculate pretty much anything BUT even if you completely understand the ins and outs of Google Translate or even if you personally made the Google translator software, you would still not be able to use it to translate complex sentences from almost any language to almost any other language... You can calculate correctly with a calculator but ironically you pretty much can't correctly translate with Google Translate, even if you're "good at using it"

@melissahdawn

5 ай бұрын

@myfaceismyshield5963 good point, math does not change to fit it's users and because it constant it can be extrapolated in a way that language cannot be predicted, quantified, and programmed.

@IshCaudron

5 ай бұрын

That's interesting. Where I live whoever works in tourism needs to speak multiple languages to...well, help tourists who don't.

@Frahamen

5 ай бұрын

@@IshCaudron We do too. But tourists sometimes just assume we don't. Plus we can't speak every language.

Even two native speakers of the same language can run into difficulties. Recently I misunderstood a younger relative who told me that something or other had recently "dropped". I took it to mean cancelled, but in modern usage I now understand that it means "released", i.e. made available. Aside from changes to a language over time, culture, pragmatics and prosody are likely to be lost on computers for a long time.

@fariesz6786

5 ай бұрын

recently on a yt video i saw that had brits and an american in it.. brit: "you know Dennis the Menace?" american: "ah, yes" ..they _thought_ they were on the same page but i am _pretty sure_ they were not xD

@lakrids-pibe

5 ай бұрын

Something "blows up" means its successful. Not that it was destroyed in an explosion.

Love it. I always find it funny when people try to directly translate from Welsh. "Nes i ddod o hyd gwas y neidr" directly translates to "I did come of length a servant the snake" but actually means "I found a dragonfly". Google translate is pretty good at picking up on the idioms, but sometimes it's choices are very unusual, especially when translating into Welsh. Not even talking about the register/formality/dialect differences which it seems to have zero respect for and use whatever it wants to willy-nilly even in the same sentence.

@JanoTuotanto

5 ай бұрын

Google translate is also thinking in English. It fails the SWE-FIN "jag torde orka att hinna - jaksanen ehtiä" test. This sentence means exactly the same thing in Finnish and Swedish (it is likely that my stamina is sufficient for keeping up with the schedule ), but it has 2 words and a grammatical construction that do not exist in English.

@makkurokokkuri

5 ай бұрын

i'd love to hear more welsh idioms if you feel like sharing!

@ariadne4720

5 ай бұрын

to say nothing of Duolingo's Welsh. Truly abominable.

@danielmeleaku1061

5 ай бұрын

Welsh just isn't prioritized by google right now. But it will get there

Darmok and Jalad at Tanagra. That episode was my first introduction to the epic of Gilgamesh

@Reziac

5 ай бұрын

And I will always think of it as someone's tagline: "Shaka, when the sewers backed up."

@P-Mouse

5 ай бұрын

+

Makes me think about the Yautja (predator) species that uses its mandibles in the language, meaning if you cant see their face you can basically only speek slang.

I always liked the "translator microbes" in Farscape as far as the universal translator thing goes, because at least there was some effort put into it not being "universal" or infallible.

@williambranch4283

5 ай бұрын

Babble fish of HHGTTG.

@RLRSwanson

5 ай бұрын

@@williambranch4283 Yes of course. It's been a long while since I last read the novels. Come to think, I don't think I've read the last one, since Mostly Harmless isn't it.

@fariesz6786

5 ай бұрын

i loved the way how in later episodes of Farscape (spoiler!) Aeryn still tries to _learn_ the English language old-school style. it's actually kind of endearing and touching to see her struggle with it, yet still pulling through, not just because of Crichton but because she actually wants to grow intellectually.

Even using in or on for traveling in a vehicle or at/in a house can cause trouble for me learning a new language. For a celtic sad example. Tà brón orm=The sadness is upon me.

I remember taking a class of anthropology which went over west apache language, which almost conveys semantics completely through references of locations; one needs to know what happened at that location, at least the popular version, to understand the basic dialogues in that language. It's really interesting to see this phenomenon in other cultures!

Sort of related: once I arrived early to a meeting in Seattle and I was asked to save a seat near the front, but I was told that saving seats was not allowed, so when my friend arrived he was upset that I was in the back and I replied, "well, I didn't like to have gotten a seat." to which he responded, "you don't even have to sit if you do not want to." I had just moved from Mississippi where such a phrase is understood to mean, "it was very difficult to get this seat". Instead, I communicated that I did not want to sit. I always have problems when trying to translate simple, conversational quips like "What's up?"

@melissahdawn

5 ай бұрын

Regarding translating, many languages are endearing to me because of the idioms, and other poetic ways of saying things (like kennings). In the Sagas in particular things can only be understood because they are not translated verbatim. So, with limited vocabulary one can figure out what a "loftskip" is in Icelandic for example.

Babel fish tho. Little fish in your ear translating to your brain for you. Also worth noting that as an anglophone Canadian I have heard "hold" used in that manner, for staying in the same lane (can't remember where, maybe British? not common tho). And winter is generally October/Novemberish to March-May (depends on location in the Great White North, and from my subjective perspective winters were longer when I was a kid. Certainly had Halloweens with ice and May-birthdays with snow). So even in the same language some of these meanings can change, how would a computer handle that, even if it was an AI?

One of my favorite Star Trek TNG episodes. Even though, as you mention, there's a bit of a plot hole with their language requiring words with some meaning in order to convey the metaphor, it's still fascinating to me. I would have bet money this reference would have appeared in a Ranieri video before Dr. Crawford's, I'm pleasantly surprised! (Unless of course I have missed one of Luke's videos that mentions this)

@fariesz6786

5 ай бұрын

i feel like Luke had at least pop it up as an easter egg somewhere, but i may very well be mistaken. who did have a video almost exclusively dedicated to it was Language Jones.

On celtic languages, a mega-ton of phrases and basic communication is done through things being at/on/in/behind verbs/nouns/adjectives/pronouns. So much so that we have prepositional-pronouns. Tá x agam - X is at me, to mean you HAVE something. (also to know as in Tá a fhios agam) This also extends to saying X is on/at someone for them having/doing something i.e love "Tá grá agam duit" (I love you, lit. Love is at me for you) or hair colour "Tá gruaig bhán ort" (Your hair is white, lit. White hair is on you). Or to lay you're living somehwere "Táim i mo chónaí i mBéal Feirste" (I live in Belfast, lit. I am in my home in Belfast) This often means that direct Irish-to-English translations are difficult/impossible for machines, and vice-versa, leading to a lot of embarrassing screw ups.

@brothertaddeus

5 ай бұрын

Came here to say this. When Jackson mentioned "do" as a "dummy word" in English and some possessives as "at me", my mind went straight to Gàidhlig. And then in his example of "I don't know" in English vs "I know not" in others, it gets real fun with "Chan eil fios agam" (lit. "Is not knowledge at me").

So much conversation doesn't need to be had, but so much understanding has to be introduced.

The spark that first got me interested in linguistics was messing around on Google translate in 9th grade and then quickly realizing I'd have to consult other sources

Linguists have written articles about the Tamarian language. The basic consensus is the show writer had a cool idea but didn't know whay they were talking about when it comes to linguistics. The authors agree the Tamarians must have a language for communicating basic ideas, instructions, and concepts, otherwise they couldn't function as a society. One linguist theorized the Tamarians use the metaphorical language in formal situations. Some of the Star Trek online groups have created memes based on the Tamarian metaphorical model, demonstrating how we communicate in memes. People also respond to scammer texts with random Tamarian phrases, with hilarious results. Check out Star Trek: Lower Decks. In season 2, they introduced a Tamarian security officer named Kayshon. The show has expanded Tamarian metaphors. Thanks for the insightful discussion on universal translators.

I remember a scifi short story where the alien language changed so quickly that the universal translator could not keep up.

This explains why my babel fish 🐠 isn't working

An accessible book to learning to read Old Norse would be a godsend. I hope, for you, that the subject gains more popularity.

I'm not sure I fully agree--stuff like Spanish not needing the pronoun, or English having do as a dummy verb is already pretty well handled in google translate. Where it breaks down is when the two languages are encoding different amounts of information (like spanish has different formality and singular/plural forms of you, so if I say "you should sit down", the translator has to guess whuh you to pick), and then smaller less common languages that just don't get the same level of attention. As for AI, I've chatted with ChatGPT in English and French, and it's grammar is always impeccable, but I doubt Icelandic has got anything like that level of quality. The bots need huge amounts of training data, so smaller languages that don't self generate amounts of text content to feed the AI will struggle--and new unknown languages that no machine has been trained on, that's just nonsense.

I'm a native Russian speaker. A long time ago, Google Translate was terrible for English to Russian (or vice versa). Russian is highly inflected and shares very little vocabulary and grammar with English, same as Icelandic. Now, it's good enough for me to send google-translated articles to my English-speaking friends. It's great for news and factual articles, but rapidly disintegrates when it gets something resembling literature.

Your video opened Google Translate on my phone. It heard itself.

This was fascinating. Thank you so much!

One of the most entertaining features of the Universal Translator is that it so well knows when *not* to translate Klingon for dramatic effect. 😂

4:13 the "hold it" example is also interesting because while can mean stop, it can ALSO mean to continue continue doing something depending on context. So yea.

I've enjoyed the work you've done with the old norse language, so i purchased the 3pk of books from hackett publishing and can't wait till they arrive so i can dig deeper into apart of my heritage into understanding the norse culture better.

I think about this stiff all the time. I actually mentioned something similar commenting on that supposed "leak" about aliens. We make so many assumptions based on our narrow perceptions.

Great video.

Dude, I LOVED that episode of TNG

Vyvyan: Why can't we have some decent food? Neil: Because we haven't got any bread. Vyvyan : Well, why don't we get some bread? Neil: Because, Vyvyan, we haven't got any bre-e-ead!!! (The young ones)

Great video!!!

Your stuffs are awesome. I'm learning Icelandic. I know the basics and use BÍN a lot. I have a good book with a little Oldnorse. Can you tell me some hints how to memorize those huge amount of seemingly random and similar flectated forms? Mother tongue is Hungarian.

All your base are belong to us.

07:10 an interesting point about the year being split into two seasons vs. four. In the American Southwest (specifically Arizona), there are five seasons, not four. This four season thing is a cultural appropriation from Europe that does not translate directly into the reality of seasons in Iceland, Arizona, or the Amazon.

I've invented the term "Babelrybski" for a translation that could only be done by a computer. The original example was "at the bottom of birth" for "на дне рождения". A human using a dictionary would look up "день", which means "day", not "дно", which means "bottom". The word refers to AltaVista Babelfish, which was in common use at the time, rather than the HHGG Babel fish it was named for.

So in the last few years, I've seen people start to play old Japanese adventure games with a plugin that runs whatever text the game displays through a translator automatically. And what keeps making it really awkward is that due to Japanese being a pro-drop language that lacks subject-verb agreement, it has to keep taking wild guesses as to which pronouns to use, which means it keeps shifting between first, second, and third person in a way that's hard to read and probably incoherent if you aren't aware of the underlying grammatical problem that's tripping up the translator.

Always been fascinated by the natural mental faculty for spoken language. It's a way to translate abstract symbols in the brain into sounds that another person can decode. I think you're spot on saying a machine couldn't replicate this, computers have no equal capacity even within artificial "intelligence," they just copy and generate based on structure, without any internal understanding.

this commentary is in direct line with my current project. Positive you'd be interested in what i have been doing lately. Think about it.

In Star Trek The Original Series, the Universal Translator scanned brainwaves to do its translations, but it didn't really make too much sense.

Crawford on a mountain, that44rdv4rk, his ears open.

I've often thought that you could base an entire course in the philosophy of language on analyzing the famous Star Trek: The Next Generation episode "Darmok". W. V. O. Quine argued in "Word and Object" (1960) that "radical translation", the decipherment of a completely foreign language from only the verbal and non-verbal behaviour of its speakers and the observable context of that behaviour, is impossible.

Written text is one thing, speech is another. Text cannot convey the full meaning of spoken language, for many reasons - prosody, pragmatics, accent etc. The advances in voice recognition are pretty amazing. But it has a long way to go. I could not get my android phone to recognize the phrase "router bits" (the woodworking tool). I finally realized that my Canadian "ou" pronunciation was the problem. When I say "router" with a nasal US twang, voice recognition works. But it should have guessed from the phrase: "small pieces of a helicopter rotor" is surely a lower probability guess.

I asked Bing chat and got this answer: The translation of “I can’t make a buck” into Icelandic is "Ég get ekki grætt".

@h734802

5 ай бұрын

Ask an Icelandic.

I only speak English and if you told me "hold to the left" I understand so 'hold left' would make sense to me.

Some languages are spoken different to written, so it would be even more difficult to train ai. In additipn many words dont exist in different languages and need substitute words. Local doalicts change this further and a slow penetration of english and french words makes for some very complicated algorithms. In addition cultural sarcasm and emotion can also change sentence meaning. Google is pretty bad.

My experience is that such online translators as Google Translate work better if you have an advanced knowledge of both languages. At least you may be able to figure out the meaning if your knowledge is at a higher level. if you have little knowledge of both languages it is just about useless.

Maybe a universal translation engine would have to be capable of making requests for clarification.

Thinking about the problem from a cognitive science point of view I'd don't think idiom is a high barrier to machine learning. After all what is idiom but vocabulary extended. A greater barrier is context. Whe we interpret text or speech we progress rapidly through subconscious phases of listing all possible meanings, shortlisting the ones that work, and then selecting the most likely from contextual clues. It is that context which will be hardest for any AI system because we lack a basis to train one with what is largely an intuitive skill for us. Still I don't think this is insurmountable. Where any Universal Translator is likely to come unstuck is when we ask it to go where no human has gone before. How do we train on what we do not understand? All attempts to create AI systems are derivative of our own creations, none have been able to truly innovate and we have no model of how to make them so.

My partner has been studying German on Duolingo for nearly three years. I am a native German speaker. What Duolingo serves up is 80% correct, 10% close, and 10% utter bullsh`t.

I wonder if the idea of universal translators first appeared in books or television. Either way, I'm pretty sure they were introduced as a story-telling convenience rather than for any scientific reason.

@maddockemerson4603

5 ай бұрын

Almost certainly books. While the oldest form of an actual universal translator device I know of is in Star Trek, that show was itself a throwback to the pulp scifi genre that had started decades before the invention of television. Most pulp scifi doesn’t bother with an explanation for why races or cultures completely alien to each other can speak unimpeded, but surely some of them thought to write in a reason.

Has anyone who speaks another language that English entered "I can't make a buck" into Google translate to see if it can handle this idiom in other languages? I only speak English, or I'd do it myself.

@JanoTuotanto

5 ай бұрын

I tested a few . The translation is correct in Finnish, Swedish , German, Estonian , Italian, Afrikaans, Japanese. Dutch, Greek... Also Finnish gets the colloquial "tienata" instead of proper formal "ansaita" for "make money".

@jrpipik

5 ай бұрын

@@JanoTuotanto Interesting! Thanks!

seeing how much machine translation has improved in recent years i do not share your skepticism in regards to idiomatic expressions and such. i'm not even too skeptical about an AIs ability to produce a consistent localisation of a body of text with potential continuation (for instance with sequels of a book) although we are probably a few decades away. and we are definitely at least several decades away from AI being halfway capable of coming up with wordplays in a somewhat reliable fashion. what i really _am_ skeptical off are a) the idea of decoding a language it has never perceived before when there's not even some protocol for like broadcasting some corpus of your literature and/or recordings of typical conversations.. like, if you have that protocol then you probably also have a mutual intergalactic friend and so probably know about each other in some shape or form anyway, so the setup is completely different to begin with. and b) the idea of real-time translation to the degree that is often portrayed. real-time translation as done by humans is _notoriously_ difficult and usually requires protocols and techniques specifically designed for the combination of the source and target language. often times the language register used will also have a certain formality to it and/or the parties will interact with the translators on a meta level by asking about how a certain phrase should be understood, or by having various other fall-backs like in world politics when in reality they already know what is going on before a conference but will react only during the conference as a form of dramatic display. the way how the universal translator has zero issues when someone for instance makes a dramatic pause is completely unrealistic. you might also add c) how the duck the translator knows that a certain (usually Klingon) expression is supposed to not be translated when it's not a loanword and the source character is already speaking Klingon so it should be translated like everything else.

@ymirfrostgiant

5 ай бұрын

In episodes with Klingons in the TNG era, my head canon is that they're actually all speaking some Lingua Franca version of Klingon and that all Star Fleet officers have some familiarity with it. Klingon has got to be one of the major trade languages of that side of the galaxy considering how big the Klingon Empire is, and Klingons are the kind of supremacists that wouldn't bother to learn other peoples' languages or to mess around with translators. It kind of tracks with how culturally influential Klingon culture seems to be in DS9.

Even with perfect technology, the differences in grammar would make real instantaneous translations impossible o super akward. Those German phrases that have the verb in the end, or Latin o Quechua basically free order, you'd need to end the sentence before translating.

Even the little things can be a problem. The Japanese, for instance, say "hai" (yes) in the same way we say "uhuh" while someone else is speaking. They also use it with a slight rising tone to indicate "confusion but continue speaking". And, of course, it acts as "yes" too. It's hard enough to always know how it's being used when you've been immersed in that culture for a decade, let alone as a machine. Sure, this specific one is a surmountable problem, but you have to surmount tonnes of problems like that, in many languages. At multiple levels of formality. Outputting an equivalent tone of voice and register for the listeners too. All in real time. And even if you manage that, there's enough regional variation that things can still get tricky. For instance, in the US "to table an idea" means "to set it aside for now", while in the UK it means the precise opposite, "we put the idea on the agenda for present discussion". Sure, over time all English speakers would get used to the differences and settle on a single unified form of English to make the translator work properly; until your Aladaran diplomat goes to the Moon of Turnipus and greets the English speaking colonists who split off 100 years ago and offends them by telling them he'll "make sure your proposal isn't tabled".

All the issues you raise can be programmed into a translator (or taught to an AI), but that doesn’t solve the issue that the translator or AI has to KNOW about them before it can translate. The fantasy comes in with the idea of a translator that can understand and translate a language without knowing anything about it.

Etymology app???? It has fractions of words to put together... because why wouldn't I be the only freak to do this... it also helps if you know of you know ancient symbolism....

Looking at GPT (and related LLMs), I don't think this is as far away as you make it out to be. LLMs have some of the exact capabilities you (I think correctly) state that such a translator would need: for example being able to generate novel content -- the G in GPT stands for Generative. Similarly, I think Chat-GPT clearly understands intents. I don't think we are there yet, but a LLM-based universal translator that truly understands all of languages it translates between does not seem that far away. I do agree that a universal translator that translates between a known and unknown language is total fantasy. LLMs need to be trained on large amounts of data.

@ymirfrostgiant

5 ай бұрын

The generative aspect of GPT is largely an illusion though. It can mash together words from its training data partly at random, but it's basically playing a giant game of Madlibs. If there are holes in it's data, as with Icelandic and Google Translate, the illusion falls apart.

Always found this idea a lil goofy, but it has never fully taken me out of a story.

@beepboop204

5 ай бұрын

Quine and radical translation, and some other philosophy stuff, try to explore this kind of stuff, and it would probably take a long time to figure out a series of appropriate expressions for appropriate contexts. just knowing "word A means word B" isnt enough, dictionaries arent enough, you need to figure all that out and then figure out the visual and or auditory cues that prompt words. hope that makes sense

Samhain to Beltaine- winter. Just saying

Look at ai today, add 400 years, add thousands of alien languages to its database. With all that combined, it wouldn't be impossible at all.

Lexical and syntactic translation is more mechanical. Semantic translation fails, because semantics is much less mechanical. To the extent that syntax and pragmatics depend on semantics, they are handicapped too. Google Translate works well only on new sentences closest to the sentences it was trained on. Hence Chat GPT ... simply throws a vast amount of training data at the problem (quantity is its own quality BS). This is why Ranieri is so good, he is generative with the languages he studies.

The way I see it, you could probably have a "Universal Translator" that could translate any KNOWN language to any other. The artificial intelligence is only as good as the data set it is given. So, if it came across an unknown language, it could make "guesses" about the meanings by comparing it to known languages, but it definitely wouldn't be able to translate idioms in the language. Not just comparing vocabulary, but looking for grammatical patterns that are similar to other languages. There are some universals about languages, but if it was a truly alien language like in Star Trek, we don't know if these so-called "universal grammatical patterns" would hold true. The idea of a "Universal Translator" relies heavily on the idea of a "Universal Grammar" of languages, from which all languages simply choose different ways of expressing. But, from what little I have learned by learning foreign languages, this isn't true. For instance, English obligatorily marks the plural form of nouns, like "dog/dogs", whereas Japanese doesn't. How a translator, human or machine, could possibly translate that from English to Japanese, preserving that grammatical distinction on every noun, I have no idea. Possibly the AI would have to tell you beforehand "Hey, English has this grammatical distinction, so I am choosing to explicitly mark it in my translation, even if this isn't how it would usually be said in Japanese".

Not sure about a sci-fi universal translator. Hard to believe it could translate a language it has never encountered before. Also, I wouldn't want to read a novel or research paper that was just translated by google five minutes ago. OTOH, I think online translators are great. As an American I'm used to hearing all levels of English and can still generally figure out what someone means despite the words coming out of their mouth. It's not much different puzzling out what an AI translator meant. If you're really not sure ask the person what they meant or please to it rephrase. AI is improving by leaps and bounds and it probably already understands and speaks English better than most people who can at least speak some English, and better than a lot of native English speakers.

It's not the show's fault, but a lot of people, a LOT of people, especially it seems nerds, watch Star Trek and think it's science. These folks seem to think they are PRO science, but then when actual science says "That's not how any of this works," folks reject it in favor of the fantasy.

Google can translate "buck" into "money" but what it can't do is give an equivalent slang term. That's the difference between a mediocre translation and a good one.

Once we discover the Babel Fish species, we will no longer need this faulty AI nonsense.

I think you underestimate the capabilities of AI to learn at the moment. I would say we are a year away from the first version of a Universal Translator for at least 20 languages. If you had asked me this question a year ago, I would have said we weren’t even close. Exciting times

@az4037

5 ай бұрын

I will come back to this comment in exactly a year to tell you there is still no such thing as a universal translator

@beepboop204

5 ай бұрын

because Quine and radical translation. a dictionary isnt enough, you need to understand the contexts which in you deploy utterances. though i do think Farscapes idea of "translator microbes" was cool but i dunno how realistic it would be compared to any other sort of universal translator.

@jjdinanno4147

5 ай бұрын

@@az4037 see you then! 🤩🤞🏽😎

@hive_indicator318

5 ай бұрын

I wish I wasn't on mobile, so I could grab a link to this thread and set a reminder for five years from now to ask how long you think it will be

@sparshjohri1109

5 ай бұрын

@@az4037 Perhaps not a universal translator, but transformer models (and models like ChatGPT in particular) are able to take context into account in ways that are relatively unprecedented (although extrapolating an entire language from a miniscule dataset will still be a fantasy, of course). I don't think it's infeasible that the majority of issues discussed in this video will end up being solved in a few years' time for most sufficiently-documented languages

No universal translator, not likely to get warp drive either. And we have all these billionaires running around who utterly failed to get the point of Star Trek. Oh well.

Ah yes.... all manner of mistakes to be made... in my... call it a conlang, words are built from particles. ba- (scent) sa (other) -ik (separate or distinct) So what do you think when you see the word "bashik" ?? A straight-up translation might guess at "smell something different". But in actual idiom, it means corpse-eater. Ooops.

meh linear resonators.

Try Google translate for Finnish...hilarious. and a huge waste of time

some thinkers have chosen to see it as a final and clinching proof of the nonexistence of God