# Why would a language be undubbable by universal (machine) translator?

A hypothetical speculative fiction setting uses the universal translator/babelfish/translator microbes/whathaveyou as a convenient plot device. However, certain languages are simply untranslatable by this method.

What property would make such languages untranslatable by the universal translator?

EDIT: Perhaps untranslatable is the wrong description. What property would make a language impractical to translate if the intention is to dub in a translation it in real time? It's still possible for anyone to learn. At first I figured the most likely candidate would be something like Ithkuil, but I don't know if the idea is workable.

• If you have a magical translation device but it doesn't work for some languages that seems more like a plot device than world building to me. Unless you actually define these languages for us we can't answer for you...and if you define them you would probably answer your own question. – James Nov 21 '16 at 21:33
• I'd say "because there's no such thing as a universal translator in the first place", but that's not enough to make an answer. Some of the answers below go into the reasons why without saying so outright. – Monty Harder Nov 21 '16 at 22:48
• Darmok, when the walls fell... – Monica Cellio Nov 22 '16 at 3:59
• You should strike the babelfish from your examples, because the way it works is by circumventing the "language" barrier and infusing thoughts into the "listener"... – I'm with Monica Nov 22 '16 at 11:45
• @MonicaCellio Darmok, on the ocean... Shaka, when the walls fell. Nerd card, please. – J... Nov 22 '16 at 12:39

# Language that requires more than sound

If your translator works via sound, it would struggle with visual communication. Imagine a language where hand motions indicate the past vs. present, verb conjugations, etc.

Update:

• Smells could also add to the language.
• Touch, also (let's say you hold your hand against theirs, and press different fingers to conjugate a verb).
• Taste would be a weird one—maybe these aliens have a long tail, and to communicate you put the other guy's tail into your mouth, and it secretes different tastes onto your tongue as he talks).

# Language that overloads our brains

We speak on average 145-160 words per minute. Let's say their communication is like our computers with 1Mb/s download speeds. 1Mb is about 500 pages of writing. If a race spoke with that kind of speed, they could add so many details that seem unnecessary to us, yet it doesn't hinder them at all. They speak for a minute, and we have to search through 30000 pages of data looking for the main point of their conversation.

# Thoughts that don't make sense

Maybe they can see 40 colors, instead of our 7 color range. Maybe they use these colors to represent feelings or maybe each one relates to an ancient parable of theirs. Time could be measured in how long it takes to grow a certain mushroom. They could enjoy what we hate, and find our pleasures painful. Maybe they speak in poetry. All these put together would make a very confusing speech.

• Humans don't have a seven-colour range. At best, we have four, and that's if the person is a rare freak of nature; the average is lower than three because of colour blindness. The idea works though. – Nij Nov 22 '16 at 9:11
• You don't need to leave earth to find cultures with completely different ideas of what the main colours are and where you draw the lines. – Nathan Cooper Nov 22 '16 at 9:28
• Seven colours? At best four? Are you both mad? Humans can see around 10 million distinct colours! I like the answer for the point that a language may be partly (or entirely) non-sound based though, and that the amount of information might be too great to process. – Grimm The Opiner Nov 22 '16 at 11:52
• The “7 colour range” is confusing two distinct but real phenomena. (A): our colour vision is inherently trichromatic, since we (well, most of us) have 3 types of cone cells, the receptors in the eye that colour vision is based on. This doesn’t mean we can see 3 colours; but it means that our perceptual space of colours is essentially 3-dimensional. (B) “basic colour terms”: the way that classification of colours varies depending on culture and language. [cont’d] – Peter LeFanu Lumsdaine Nov 22 '16 at 16:54
• Roughly, most languages have around 5–10 “basic colour terms” — e.g. in English, we think of blue and green as fundamentally different colours, while we think of azure and indigo as shades of blue. (Not that we can’t distinguish them, but we think of them as variants of the same basic colour.) Different languages have different basic colour terms, some more than others, and there are various patterns in what they can be. This is a well-studied and debated topic in linguistics; Guy Deutscher’s excellent book Through the Language Glass uses this as a central example. – Peter LeFanu Lumsdaine Nov 22 '16 at 16:58

Though the universal translator (in Star Trek: TNG) can translate their words, the Tamarians communicate only through allegory, which baffles the Enterprise crew because they do not know the stories to which the Tamarians are referring. Likewise, the Tamarians cannot understand Picard's straightforward use of language.

• This was going to be my answer too. It's basically communication through memes, where nothing that is said makes any sense unless you know what the original context is. – AndyD273 Nov 21 '16 at 21:17
• Actually we use that to some extent in our language as well. For example when you say that something is someone's Waterloo, someone without the relevant knowledge would wonder why an event should be some Belgian municipality. Likewise, an universal translator would not help with understanding what a "Catch 22" is. – celtschk Nov 21 '16 at 21:31
• @celtschk I think the best part about this is that many don't even know the original context anymore, we just accept it and eventually learn to understand its meaning. Many people I've talked to don't even know that catch-22 is a book, nor do they know that pulling out all the stops refers to Pipe organs. – Delioth Nov 21 '16 at 22:58
• A sufficiently advanced universal translator would be able to translate allegories and simple phrases like "catch-22". It would be mostly useless otherwise in an even slightly complicated scenario. One can assume that a universal translator can translate anything that a speaker who knows both languages can translate (and probably more than that) since there is no technical problem with translating such phrases. – Sesdun Nov 22 '16 at 10:43
• @Deolater: after reading this question, I hazard to guess that the UT was using a centuries out of date dictionary. To the Tamarians it's all standard idioms (like "the sun rises" is to us) while other species speak in idiosyncratic medievalisms. Most Tamarians may not even be aware their common phrases are idioms from mythology. Such a language is still translatable using an up-to-date dictionary. – Anonymous Nov 23 '16 at 14:22

This response assumes a perfect translator defined as follows:

The perfect translator is a machine doing it's best work possible. It is not limited by technical errors or construction mistakes and it can be assumed to have had opportunity to study all languages in question thoroughly. Further I assume that both ends of the conversation either uses their own perfect translator or both interface appropriate with it - using what sensors are necessary (i.e, a microphone and camera for humans since they communicate with voices and body language/visual signs).

The answer below thus attempts to answer what languages are not translatable even though there is no lack of proper equipment. (i.e creatures communicating through unusual modalities (chemicals, radio-waves etc.) are not a problem as sensors to detect the modality can be assumed to be present).

# Languages untranslatable into certain other languages

A singular real language is actually translated at least twice as part of its normal use.

Speaker A's internal concepts > Language symbols > Speaker B's internal concepts.

As such it requires that both speakers have the same concepts and can express and understand them from the language. Some languages are more apt to expressing certain concepts than other languages because it was developed by speakers wanting to communicate certain concepts.

When looking at a translation between two languages the concepts encoded by the origin language must find symbols or sets of symbols in the target language which gets translated to the same symbols in the minds of listeners.

Even normal human languages fail at this, the meaning of some phrases are very hard to translate and more subtle details are often lost or changed even in very good (human made) translations, thus the phrase "lost in translation".

Considering more alien languages where the original speakers are not even the same species, the underlying concepts of the speakers will differ greatly and the language constructs to express said concepts will be tuned to expressing these special concepts.

To translate between such languages might be very hard to do well - with heavy information loss. If the speakers are too different, communication would be impossible.

## In practice:

Given a perfect translator (a machine doing the best possible) it seems reasonable that languages could be grouped into families where translation would be easy within the family and hard between them. One would expect that families would be grouped depending on the nature of the speakers and their natural environment - more similar creatures would communicate better. In a society where such translators were commonplace - language families would drift closer to each other, adopt loan words and such - reinforcing the familiarity.

When speaking to an out-of-language-family creature one would expect very general notions to translate better, while emotional and sensory vocabulary would be more lacking. Probably they could both do common tasks like trade using simplistic language constructs that translates good enough only for that activity.

To learn a out-of-family-language might be impossible, or might require large changes in mindset or even sensory augmentations.

# Languages intentionally untranslatable

If the availability of artificial translators were common, it is perfectly reasonable for two speakers to converse using a spoken language none of them understands. The translator translates language A into X (understood by none) and the receiver translates it back to A (or to some other language the receiver understands).

Since this is possible, it would be a simple matter to speak in an encrypted language. Each translator installs the same cryptographic-key and produces an encrypted language which can only be decrypted by translators with the same key.

Certain organisations or even cultures could thus have languages impossible to understand by anyone lacking the correct keys.

• +1 I imagine language families would adopt loan idioms as well. – Nathan Cooper Nov 22 '16 at 9:26
• I read once about a people who had no concept of relative directions (left/right), only compass directions. They always gave directions in absolute terms, performed all their dances facing north, etc. It would be impossible to translate left and right into their language. en.wikipedia.org/wiki/Guugu_Yimithirr_language#Grammar – Tavian Barnes Nov 22 '16 at 18:07
• If you found a group of tirbal humans with no idea what a microwave was, you can still communicate "microwave". You just need to show them it and they can make a word. – noɥʇʎԀʎzɐɹƆ Nov 23 '16 at 22:34
• @uoɥʇʎPʎzɐɹC a "microwave oven" is actually a much simpler concept than "left". Most of our advanced concepts are actually closer to explanations in terms of more base ideas than stand alone concepts. It is when the very base of the conceptual tree is lacking or different communication and understanding becomes difficult. Try to explain the different concepts of musical sounds to a deaf person - that is something in a completely different league from "its a box where you put food, press a button and it becomes hot." – Sesdun Nov 23 '16 at 23:40
• @Mattias north + left = west; south + left = east; east + left = north, etc. – noɥʇʎԀʎzɐɹƆ Nov 23 '16 at 23:43

It is hard to translate on-the-fly from german to other languages, because in german language predicates are easy to stack at the end of long sentences. Make your language to stack important words at the end of the sentences. Translator must wait until whole sentence is told to translate it. The easier it is to swap the meanings, the harder it is to translate fluently.

Other method is language full of homophones. If the Last Word in the sentence defines which meaning is actually used there is no way but to wait till the end. Out of context, following czech sentence, written phonetically, has plenty of meanings:

[ʒenu hɒliː strɒj]

1: Dative form of "woman" - "ženu" (žena); present continuous verb "I am propelling/driving" - "ženu" (hnát);
2: Nominative form of "naked / bare" - "holý"; Instrumental form of "stick / staff / rod" - "holí" (hůl); present continuous verb "[he/she/it] is shaving" - "holí" (holit);
3: Nominative form of "engine / machine" - "stroj"; imperative verb "dress" - "stroj" (strojit); imerative verb "make three" - "ztroj" (ztrojit).

It is hard to translate on-the-fly if the languages have different "rhythms". Languages do have short phrases that cannot be translated but must be explained etc. If the translator must spend 15 seconds to translate (explain) 1s word and the original sentence is full of such "dictionary bombs", the speaker is several sentences ahead. On the other hand, if the target language has single word for a long story in the original, the translated speech is full of "dictionary voids". If the bombs and voids are spread well, the translation is full of silences and rushed speech.

Creating bablefish-proof language is impossible since it does not make you hear them speaking your language rather you understand what they mean. That's why it caused all the conflicts; there was no "lost in translation" excuse.

Non-spoken languages, like sign language, or odor-based languages (humanoid ants colony, for instance).

• More generally, “a modality that the translator was not designed for”. – JDługosz Nov 21 '16 at 21:35
• This is excellent information and may be harder to translate than other answers but there isn't a lot of information here. Consider expanding upon these ideas – Zxyrra Nov 21 '16 at 21:48
• This is the best answer. – Tony Ennis Nov 23 '16 at 12:21
• Odor can be converted to sound. – noɥʇʎԀʎzɐɹƆ Nov 23 '16 at 22:38
• @Zxyrra: There probably was a lot of excellent information that didn't survive the universal translator used to make this post. :-) – celtschk Nov 24 '16 at 8:05

One possibility is that there is a preferred structure of language, in the sense that normally language evolution is convergent, that is, it follows path that leads to certain common characteristics even though the evolution was independent. Note that language evolution doesn't happen in a vacuum; language evolves due to necessity, just like everything else. Moreover, language structures follow structures in the brain (or brain equivalent of the alien species), and those structures also evolved under certain evolutionary pressures.

Now the universal translator knows these universal patterns, and uses them to decode the language. Since those patterns are shared by almost every intelligent being throughout the universe, the translator is truly universal.

Well, almost. Because the evolution doesn't necessarily go that way, but is only very likely to. That is, one one in a thousand planets, evolution of language took a different path, and the language therefore does not share the same universal pattern. Since the universal translator depends on that pattern to be present, it cannot translate such an "outlier" language.

Another possibility is that the language is not actually an evolved one, but an artificial one that at one point got adopted (think Esperanto). If the invented language, following abstract rules instead of being based on similarity to existing languages, differs too much from the universal structure, that language might also be non-translatable by the universal translator.

Indeed, there's even the possibility that a language was actually designed to not be translatable by the universal translator. Think for example some military use, where you are not interested in the enemy listening (see also: Navajo code).

Even if not completely untranslatable, having a species with a very strange culture or mindset, or a very different sense of reality, would make their language very difficult to understand and only partially translatable, as most words would cover concepts we might completely lack.

A good example for it is the Orz language:

My *fingers* reach through into *heavy space* and you *see* *Orz bubbles*
but it is really *fingers*.
Maybe you do not even *smell*? That is sad.
*Smelling* *pretty colors* is the best *game*.

*Space* is many. *Colors* are many. You are so *sticky*.
You cannot *slide* like Orz from *outside* to *inside* and *in between*.
It is sad, but Orz can *pull* the *campers* after being *connected*.


"What property would make a language impractical to translate if the intention is to dub in a translation it in real time?"

I'd say the most common thing that would make it impractical is speed of communication. Already we see this to some extent in movies where what is said in one language only takes 3 or 4 words and only a couple of seconds to say, but in the translation it takes many more words and much long and so you get dubs where the audio stops but the person's lips keep moving for awhile. Or the opposite effect that the audio is something really long past when the person stopped moving their lips.

And if you're dealing with a very complex language that has words for many very complex ideas that one language really doesn't have any comparable words for, it might take entire paragraphs to translate. For example, consider trying to translate the word "Kafkaesque". Sure, you could give a definition of "similar in tone or theme to the work of Kafka", but without further explanation of what Kafka's work is like, it's not an adequate translation.

Or you could have a species whose speech is just very fast. It's not that their language can communicate more in few words, but that they say their words much faster than normal human speech. 300 words per minute is about as fast as some one can talk and still be mostly understandable to a native speaker, but suppose an alien language was normally spoken at 1200 words per minute.

And if you combine alll of these, you could end up in a situation where an alien talks for one and a half seconds, and then you have to sit there for 10 minutes listening to the translation. And then when you talk back, you might have to talk for 10 minutes just to produce another 1.5 second blurb for the alien to listen to. Or if it doesn't wait to buffer all your speech, it might sound painfully annoying to listen to someone talk that slow. Imagine trying to pay attention to what some one was saying if they're only saying a single word every 20 seconds. That's hardly practical for real time dubbing to have a conversation.

Some reasons (still) not mentioned:

• Unknown concepts. Sami (Northern European natives) have words for snow, African tribes have words for sand desert (I do not debate the amount of words, but the thing is that the native language cannot have a direct translation of the other concept because it simply does not exist). This is normally easy to remedy: You need to explain/circumscribe the concept in a sentence with the known words or the people experience it itself. Being clumsy as tool sooner or later either the people define own words for the phenomenon (white sand, grainy landscape), add adjectives or they incorporate loanwords. Amok (Indonesian), Angst (German), Deja vu (French) are words which have a distinctive meaning which cannot be expressed as single word in the original English language before their adoption.

• Ambigous concepts: There are many words in languages like the English "smart" which have extremely different meanings with the same word, sometimes even contrary ones if ironic undertones are replacing the original meaning. It gets even worse if the person deliberately chooses the word to use the ambiguity for exactly this purpose. This means jokes/puns/hidden meanings cannot be translated because normally different languages have not both a word with the same ambiguity definitions.

• Strange concepts: There is always a tension between the exact meaning of a word and its social context. "No" seems easy to understand, but is in different cultures completely taboo or polite/expected. So if I mean "No" when saying "no", the universal translator must decide whether he risks an affront by directly translating it or try to guess my intention and disarm with an literally incorrect, but most fitting sentence.

Any language designed to be untranslatable given a deep understanding of the inner workings of the universal translator.

In other words, similar to the spear and shield arms race, a race could take place to have an untranslatable language for multiple reasons: . Secret societies: understanding the secret language is a proof you are part of the group . Intellectual challenge: similar to can biology beat the AI-translator . Anti-colonialism: making aliens unwelcome in a world

While most likely the forces behind the universal translator would thrive to identify always faster those new languages and provide translation. Many parameters would be considered: a new language to be untranslatable would need to have never been identified, be significantly remote from any known one ( which over time becomes increasingly difficult).

One last consideration insiders from the organisation driving the universal translator might be best placed for creating back doors rendering certain languages untranslatable.

A language where much of the meaning is dependent on other factors, the time of day, the relative status of the speakers, the (relative) sexes of the speakers.

In current languages there are many examples of situation dependent words Consider the Polish "kurwa" which in many ways is even more versatile than the potential English usage of "fuck"*, it's equally possible to use it, and derivatives thereof, as every word in a sentence and it cannot be translated out of context.

You can have examples of words that just don't translate because the second language just doesn't have that concept, German examples being "verschlimmbesserung" and "schadenfreude", the latter of which English just had to take directly. While these words can be explained, you have to stop the flow of conversation to explain a concept rather than give a straight translation.

Consider also the double entendre (said the actress to the bishop), which meaning is actually intended, or whether both are. It's up to the listener to identify the intent of the sentence, or potentially miss all that's being said because the meanings are overlaid onto each other. Once translated the duplicitous meanings are often lost.

*I'll give you a language warning but it should be taken as read just from the text of the link.

• This isn't just for an intentional double entendre. In a fansub of The Slayers, they referred to one the characters as Mr Gourry when he was undercover as a female. In the original Japanese he was referred to as Gourri San, with San being the traditional translation of both Mr and Miss. Thus the translators had to choose between whether to translate San as Mr or Miss, either way adding meaning that wasn't in the original. – gmatht May 28 '17 at 4:04

What property would make a language impractical to translate if the intention is to dub in a translation it in real time?

A few years ago, there was a study which showed that while human languages vary in information-density per syllable, we typically speak at a rate such that the same amount of overall information is delivered in a minute or so. That is, Mandarin is very information-dense, but spoken slowly, while Spanish is more wordy but typically spoken quickly.

This property makes real-time dubbing possible, although you often get the jarring effect we're familiar with from badly dubbed movies where the speaker's mouth clearly moves as if saying a one word and a long string of sound comes out (or the other way around).

Have your alien language be both dense and spoken quickly, and the real-time translation simply can't keep up. That would probably require more language-processing capabilities in the alien's brains — but hey, aliens, right? It wouldn't need to be a factor of 10 or anything like that — even a 20% increase would make every significant conversation require a lot of "hold on a minute please!".

You could also go the other way around, making the aliens slower (like Ents, I suppose, except without the ability to shift to human time scale). Now the universal translator can't easily render human languages to the aliens without making us wait, and we all know how impatient we are.

• Hurray for the Ents! – Hurda Nov 25 '16 at 18:10

A highly – nay, whollysynthetic language could be impossible to translate without external references. Imagine the two sentences "I like dogs" and "dogs like me" if the words for "dog" and "dogs" are completely different in singular and plural forms, and as the object or subject of a sentence; "like" rendered differently if the actor is a person, singular, or animal, plural; "me" and "I" already exemplify this, but imagine if it's always excluded when it's the subject.

Now take it a step further: "dog" is written differently again depending on the type of sentence, the type of verb, the tense, the first vowel of the first morpheme and so on. Now take "dog" and insert morphemes and entire words into it (like "d li me og ke s").

Even a lengthy sample on a specific topic would be inscrutable without a basic understanding of underlying grammar and syntax governing transformations as above.

• I think your point is that data learned from some previous examples will not help it to understand other sentences. – JDługosz Nov 22 '16 at 0:05
• @JDługosz Without external data (previous samples or foreknowledge of what is being discussed) it's possible no pattern to the language could be discerned. With enough external data the UT could eventually catch on – but it would be learning the language from scratch, as a child would, not translating instantaneously. ("Unlearnable" isn't a condition, thankfully.) – rek Nov 22 '16 at 4:00
• Would a language like that ever survive as a means of communication? We do overly unefficient animals in nature, and neither do we see crazily excessive languages that survive the on the evolutionary landscape. – Contango Nov 22 '16 at 12:20
• @Contango Among humans probably not, but human languages are full of insane things that seem completely implausible and impractical to anyone who doesn't speak it. An alien mind might be better suited to that sort of thing though. – rek Nov 22 '16 at 19:47

I am spacing on the title and author, but one of the old time scifi writers did a couple shorts on this idea (same events from different perspectives). Humans want to colonize a particular world as a way station for a shipping route. The rules are that they must have permission from the local inhabitants if there are any intelligent species involved. The initial survey did not detect any intelligent life, but one member of the follow up certification party is observing the local life forms and, while they are primitive, he believes he is seeing them communicate and act in a manner that indicates intelligence. Unfortunately no one can detect any language or other means of communication.

They bring in such a "universal translator" and it reports the sounds the locals are making as being meaningless and repetitive, having no syntaxual value. This is causing some contention between the development company and the certification team because the team refuses to certify the planet, because they think this species is intelligent, but can not communicate with them to prove it. The situation finally resolves with a naval officer being called in to mediate and figuring out that the aliens communicate telepathically, and the sounds they make are an identifier/carrier to allow the communication. Because the vocalizations carry no meaning in and of themselves, the translator was unable to recognize them as a language.

There are a number of possibilities, which could provide various plots:

1. Target language has no related languages (a Rosetta Stone reference is missing)
2. The language does not operate in known basis of measurement (sound, motion, etc) such as Formics (Buggers) communicating via pheromones in the Ender books
3. The language evolved using a natural encryption, and without the key the phrase is gibberish
4. The language is forbidden/dangerous for political/religious reasons, and therefore the translater software prevents translation

My intent behind the idea that a language could EVOLVE with a natural encryption, not necessarily a digital or binary type encryption. While I was considering the idea, I was thinking of a reason why encryption would be required... and I imagined a society that only communicated via sonar or pheromones in a crowded space. In order to separate conversations from the noise of other conversations, a natural encryption system was developed. The parties involved in the conversation had the key to decode, and everything else that was indecipherable was just considered background noise.

Generally, Universal Translators are only able to pick up on sound, and convert it into sound patters we interpret as words. This means that anything else will be untranslatable. If they communicate using anything other than sound waves, like light or touch, it can not be translated.

Even if they do use sound and it can translate their words, If they incorporate these other things that are untranslatable, it is enough to not know what they mean. Take for example Sarcasm. The words can be identical, but the tone might be all it takes to completely reverse the meaning of those words. If the language relies on tone to convey the meaning, then knowing the words is just not enough.

• A universal translator is posited to be universal. If it has trouble with the minor tone shifts found in English-style sarcasm, it would be totally lost with the eastern Asian tonal languages. – Mark Nov 21 '16 at 22:20
• @Mark If it is capable of translating any spoken word into your language, it is still universal. For Tonal languages, the different tones are too universal to everyone to speaks that language, so it would probably still work, but things like sarcasm (or stories, proverbs, history) are optional to know in order to speak the language. The Chinese proverb a frog in a well is 100% translated word for word, but the meaning of the words is not in the words themselves, but the meaning of the situation they describe. so when a frog is actually in a well, its a whole other meaning. – Ryan Nov 22 '16 at 16:18

It very much depends on how your universal translator works, and what you mean by "untranslatable", exactly.

If you just want a delay, the training process for the translator could simply be too long for the story needs. The language would not be actually untranslatable, but for practical purposes, for the duration of the story, it cannot be translated. It could be too complex, require too much source material, whatever.

But if you want a universal translator that cannot translate a few languages, you've made a tall order. It wouldn't be much universal in that case, right? Your way out is the "out of the ordinary" excuse. Maybe all of the species in your universe are carbon-based, more or less similar to life on earth and the translators function is based around common principles. Some species could be exotic, breaking these concepts.

Text in a language can be difficult or impossible to translate with when the required information is absent.

One way is that the grammar of one language requires information which the other does not. An example is tense in English: When translating from a language where the time when something happened need not be specified, the translator must often guess.

Another similar way that this occurs is with how implicit or explicit a language is in practice. In English, one is required to speak full sentences (or nearly so). In Japanese, one cuts a sentence down to the bare minimum. For example where in English one would say "I'm going home", the Japanese equivalent is simply the verb "to return" with all information about who is returning, where to and when left out (even though the grammar does have these things). A listener is supposed to understand the intent from context but it is not always possible to explain with certainty what was meant unless the translation is heavily annotated, especially if the text is obtained out of context. This is why machine translation from Japanese to English should always be viewed with suspicion.

A very simple example from Earth is that translating German to English is not possible in real time. It's possible and not uncommon to have long German sentences where the last word completely changes the meaning of the whole sentence, so a translator, human or mechanical, has to wait until the last word of the sentence is spoken until they can start giving the English translation.

It seems this is upsetting some people. @Lostinfrance: They can't translate in real time. As in German speaker starts the sentence, translator starts speaking in English. There are sentences where the translator needs to listen to the complete sentence before being able to translate.

@AlexanderKosubek: I'm not quite clear what you are going on about, but the other way round it's not a problem, because the first few words of the English sentence can end up influencing the very end of the German translation, which is no problem for a real time translation.

• And this never, ever happens the other way round, not! – I'm with Monica Nov 22 '16 at 15:55
• There are probably a dozen people simultaneously interpreting from German to English and vice versa at this moment in the EU Parliament. Probably they are rather bored by what they have to translate, but it doesn't stop them doing their job. It's true that "simultaneous" interpretation is never literally simultaneous, and often involves slight anticipation, guessing, hesitation, and backtracking due to difference in word order, but English and German are closely related languages. It is easy to translate between them compared to pairs of languages from entirely different language families. – Lostinfrance Nov 22 '16 at 16:10
• That's just dumb sentence construction. I find those to be annoying in any language unless it's for some really nice humor. – Joshua Nov 22 '16 at 23:58
• @Lostinfrance My german is a bit "rusty", I haven't spoken it for decades, but I can say sentences on my native languages which coudn't be translated before they are finished for the same reason. – mg30rg Nov 24 '16 at 15:50

In cryptography, a message encrypted via one-time pad1 is impossible to decrypt without access to the shared secret (the "pad") used to encrypt it.

If you think of the speakers of your language as ciphers, the thoughts they wish to express as the input to the cryptographic function ("plaintext") and the words they emit as the output of the cryptographic function ("ciphertext"), then a language with this property would be understandable to anybody who knows the encryption key while being completely unintelligible to even an infinitely powerful translating computer.

Since a one-time pad requires that you never reuse keys, distributing shared secrets is a very hard logistical problem in the real world. If the message you wish to communicate is 10,000 characters long, your encryption key will also need to be 10,000 characters long, it'll need to be random, and you'll never be able to use that encryption key again. Maybe the hypothetical speakers of this language share some sort of telepathy which they subconsciously and instantaneously (and securely because telepathy is secure and *cough*handwave*cough*) use to interpret the language in real time?

A dynamic language or metalanguage

This is close to a couple of the other answers ("make it depend on the time of day", "natural encryption", "talk with allegories", etc), but they don't hit it directly on - what if the language was not fixed, but was created on the fly each time people met up, from a more abstract set of rules?

I don't know for sure that you could create a whole language like this, but if each conversation 'saying the same thing' used different words based on the current situation around you, the universal translator would have to basically be a self-aware participant in the conversation to know what to do.

e.g.

Instead of saying 'They were afraid for their lives' you say "Shaka, when the walls fell", referring to a situation where someone was afraid for their life - but it's still the same saying every time. Instead, your language rule could be not "what to say to communicate fear" but "how to choose what to say to communicate fear".

• "refer to the most recent international news incident where someone was scared for their life", so it would be continuously changing every conversation. One day "The French at Bataclan", another day "New Zealand Earthquake endurers", another day "the flowers" (because a swarm of locusts just appeared yesterday).

• Maybe you couldn't communicate ideas and concepts as precisely, they would have more room for interpretation - less chat, and broader strokes. It could be Less international news based, more immediate surroundings based. Not "are you thirsty?" or "Like Shaka in the heat wave", it might be "like the dogs?" (if you just saw dogs drinking water) or "like the drain?" (if you both just watched rainwater draining away).

Two or more of their species spend time together and build up a language between them on the fly based on global events and local moment by moment shared experiences, which then fades as they split apart. Their species has a 'recency bias', Twitter style, and they have very similar brains to each other, so they tend to get the same emotional or practical feeling as each other in response to the same news story and events. News stories and events are tuned to this to be short and sweet and pack a communicative punch.

A rolling, evolving language, cueing from the previous usages, on a scale of seconds and minutes instead of hundreds of years. Like we do when we see a coworker ruin something and the same day we see someone doing the same thing and say "he's just 'done a Richmond'" instantly referring to the event from earlier the same day as a new saying. That, but moreso.

A 'natural encryption' keyed between people as they spend time with each other. A thing that is learnable (maybe) by becoming involved in their lives and their world, but is not learnable by listening to a conversation, or by a mechanical device which has no sense of the 'state of affairs' or the current moment's experience.

(A real universal translator might have to be sentient; in that case, anything you could learn, it could learn).

# An alien's language uses asemetric encryption

Either the aliens have evolved to be really good and fast at math (Our Algebra/Calculus is their basic arithmetic) or the part of their brain that handles language just evolved to use encryption in their language (May be lying and overhearing was a extremely big problem in their species millions of years ago). You could have something where their name is their public key, so no one else who knew the language could know what they are saying, much less a translator. If an alien wanted to talk in the group you just tell everyone in the group (using their name as a public key) the public and private keys of that group. To talk to everyone you use a well known public private key pair. You could also implement encryption signing where you so aliens can't lie about what someone else said. "This [alien name (also the public key)] said [message signed with public key]."

If you aren't familiar with asemetric encryption, in this scenario:

• Knowing the public key (the alien's name) allows you to talk to those with the corresponding private key

• Knowing the private key allows you to hear and understand those who are talking using the corresponding public key

You could say that they might have super computers that can crack their encryption.

You could say they have a really big key (the aliens have really good memory), so they would need a super-duper computer. The aliens might raise the key's length as computers get better. (The longer the key, the harder to crack)

Or, you say that they use and encryption that no one understands and never will. (Of course then people may steal and sell the alien's brains for encrypting). If this is the case no one, except members of their own species, can ever talk to them using their language, but the aliens could speak another language. (This might be very difficult for the aliens if their brains evolved to use encryption.) If they use a known method of encryption or scientists figure their encryption out, then translators can be made, but only work if you and the aliens use your alien-public-key name or a commonly known key pair.

TL;DR The aliens use a language where overhearing is impossible, where you can only hear aliens who are talking to you, due to the nature of the language (e.x. encryption).

Wordplay and irony are two of many features of the English language that I have to deal with every day in Japan.

For instance, they were talking about the U.S. elections on the news the other day, and they showed a protester with a "love trumps hate" sign. That three word phrase is so loaded with context that it would take minutes to explain to somebody who doesn't speak English, and even then they would probably not completely understand it. It was clumsily translated as "I hate trump".

The same thing happens with irony. "Sure, I'd loooooove to meet your parents" is very difficult to translate because irony simply doesn't exist in Japanese, and it would take a lot of time to explain the concept of irony in a real time translation.

Also the other way around, there are lots of words in Japanese that are loaded with cultural significance that accurately translating them to English would require minutes to hours to explain the context.

That's why machine translation between English and Japanese is so poor.

Any language which has sentences with multiple meanings (i.e. every natural language ever created since the dawn of language) would lead to untranslatable bits. One of the following would have to happen:

• The translator would have to pick an interpretation and translate that (hoping it got it right). This is what modern Google translations do. You can see how effective they are.
• The translator could try to convey all possible interpretations. This could be incredibly slow. It could even be exponentially slow if interpretations layered on interpretations.
• The translator would have to limit itself to phrases which have a clear set of matching interpretations in both languages.

# Words based on deterministic input

For any given string of characters, the previous word(not string of characters), mood, body language, situation, and other inputs determine what word that string of characters represent.

This would create a simplified block chain where each word depends and what situation, body language, and word was said before that which depends on the situation, body language, and word... Ad continuum. Each combination of aliens would have a different block chain.

In order for translation to be possible the translator would need to have heard every conversation using that block chain, and know and understand every situation using that block chain, and for every other input that determines a word.

Different dialects might use different inputs, or interpret those inputs differently. You might be able to brute force their word pairs, and see if it makes sense, but that would take lots of AI and computing power. And you'd have to guess which dilect they are using. To combat this the aliens might use the last 10, 100, or all the words to determine the next word. In that case you'd need to guess all possible sentences, paragraphs, or what the group of aliens have said ever in order to brute force. Of course the aliens must have large and exact memory to pull this off, but this is a little more plausible than the aliens evolving with computer encryption.