So this is an question I haven't tried to answer/solve too much before posting, mainly because it's more of a game and exercise in creativity and wanted to have many opinions.

So clearly this task of universalizing all languages if it were to be accomplished would be done not by any human or group of but by some insane AI of the future. However if we attempted to solve at least some basic questions like "Can it actually be done?" or "What would this blended grammar be chosen to best represent most languages" it would make me happy :D

To clarify some stuff I mean a mix of all existing languages and not a word more. So I feel like excluding dead languages from this is essential simply due to the fact that including all the evolutions of a single word between year -500 and 1500 would already be a mess. Maybe even excluding unwritten languages if it makes it too complicated to transcribe them. Also for simplicity a word would be pronounced like in the original language(s) and if its meaning is different in those languages, the context or pronunciation would provide clarification.

A sentence might be like "волк Il φοβισμένος dass人类اقتله ແມ່ນ".

Literally "wolf the scared thathumanskillit is".

Of course no one would use so many languages in the same sentence but it would be a 100% correct one and so would be the same sentence with "wolf" replacing "волк". However the random grammatical choices here, like a left to right sentence and spaces between some words, already make me doubt the feasibility of a harmonious or at least permissive composition of the most average worldwide grammar rules, but not too much :D

So to me the biggest obstacle is the grammar, can we make a grammar so that every language can use its words within it and still make sense? Must we use a grammar so permissive in its rules that all the particle words of all the languages with particles are/can be usefully used? Because that's a prerequisite of this all encompassing language.

So what would be the obstacles and workarounds?

Thank you for not dismissing this as silly, cheers.

  • 3
    That is the basic premises of Universal Grammar, but approaching from a different angle. The biggest hurdle is: There is no need. A significant hurdle is the collection of all these words in a (mental) lexicon. You'd end up with so many true homophones, it would just be noise. Ultimately, you cannot import lexemes wholesale, but ignore the grammar. You cannot just add up different grammars, because the result would be inconsistent (do you follow SVO or SOV perhaps? Do you have adjectives, and do you place them before or after nouns? Can you grade the word correct?).
    – vectory
    Commented Jan 12, 2019 at 2:59
  • en.wikipedia.org/wiki/Namespace Commented Jan 12, 2019 at 6:27
  • To expand on the above comment, common lexical items are tied with grammar, so you cannot export one without the other. Take "only" for example: it means very different thing depending on whether it's before or after an article ("only a child" vs "the only child"). If you attach it to a word from an article-less language, what does it mean?
    – jick
    Commented Jan 12, 2019 at 6:28
  • 1
    @sinekonata (1) English and French are related languages; English and, say, Tibetan are not (or, if they are, the relationship goes back so far that any non-universal connections are swamped by random noise). (2) As with most borrowings, the English word "pork" is the Norman French word "porc"—e.g., it cannot be used to mean "pig", only "meat of a pig". (3) Mixed languages usually have nearly identical grammar to one parent, not an arbitrary hybrid. (Or they're pidgins, which have only a simple linear grammar.)
    – abarnert
    Commented Jan 12, 2019 at 10:53
  • 1
    (4) Even ignoring grammar, blending two languages never gives you anything close to the entire lexicon of both languages. Many words from Norman French never made it into English (and many English words quickly died out in favor of replacements); cases like pig/pork are notable because they're unusual. (5) There's no reason to believe that blending languages scales linearly, so it's quite possible that blending 1000 languages at once could be 1000^2, or even 2^1000, more work.
    – abarnert
    Commented Jan 12, 2019 at 10:57

1 Answer 1


Languages are more than just collections of words, and you're going to run into many problems at many levels.

Let's pick one really obvious problem: What counts as a word? The single Yupik word "tuntussuqatarniksaitengqiggtuq" crams an entire clause into it. The root "ssur" just means "to hunt", but after attaching 6 affixes, the full word means "He had not yet said again that he was going to hunt reindeer." Surely you don't want every possible Yupik clause to be a word in your language. But if you want each Yupik lexeme to be a word, then how do they incorporate, or attach to, non-Yupik lexemes that don't follow the same rules?

And even the niggling details will be problems. Do adjectives go before nouns (like English) or after (like Hebrew)? As long as you pick one, that's fine; you can handle all adjectives and nouns from English, Hebrew, Japanese, Persian… but what about Romance adjectives? Some come first, some go second, and some can go on either side and have a different meaning each way. How do you get both the pre-noun and post-noun meanings of "grande" in your language? But you can't just use the Romance system, because that would make phrase boundaries too ambiguous to parse, were it not for the extremely restricted set of word endings in each Romance language (which will, of course, not be true for your language).

But I think there's a more tractable version of your problem that might still be interesting (depending on what interested you in the question).

Real languages borrow words all the time, but they don't do it exactly. Look at French "tenez" (you-plural hold!) to English "tennis" (the familiar racket game) to Japanese "tenisu" (the same game). You can see that the phonological, semantic, and morphological changes are sometimes tiny, sometimes huge ("tenisu" and "tennis" are almost perfect synonyms, but "tennis" and "tenez" aren't even the same part of speech, or about the same kinds of things).

And all of this works just fine. The borrowed word takes on a life of it own in its new language. So, if you could simulate realistic borrowing, and then just feed in every word from every language…

But can you simulate realistic borrowing? Depends on how realistic. The rules for how borrowing works are partly regular (e.g., the same English syllables usually map to the same Japanese syllable sequences), but only partly. More seriously, most of those rules are well below conscious accessibility. And most of them involve both the local context (why someone felt the need to borrow that word), and language-internal context (what existing words it has to fit into its new lexicon with). But you don't need to exactly model what would happen if a real community of X-speakers felt the need to borrow word Y from language Z, just to come up with an algorithm (probably randomizing a lot of the details) that feels realistic, roughly, and (depending on your intended goal) that may be more than interesting enough.

A few notes:

  • You still have to solve the "what is a word" problem, but now you can solve it with just "morphemes are words".
  • Given that the existing lexicon you're borrowing into is part of what affects borrowing, you probably want to iterate over all of the words of all of the languages in random order.
  • Function words and morphemes like "the" and past-tense "-d" are almost certainly going to end up distorted to the point of unrecognizability by any borrowing algorithm—maybe that's acceptable, or even fun, or maybe you just skip them.
  • Almost every part of the core design of your original language will affect how much things get distorted. (Think of the way Japanese has to insert "u" into closed syllables, import most adjectives as nouns, etc.) But this is something you can—and probably want to—play with, not a problem to surmount.

What I think really makes it infeasible in practice is just sourcing the lexicons of all of the world's languages. You can't just "use a dictionary", unless you work out how to parse dictionary-style definitions (in every language in the world, and mostly written in an unusual style that would confuse general-purpose parsers like the ones Google has trained up…) into the kind of representation you need (look at, e.g., an HPSG lexical entry, for something closer to usable by a plausible word-borrower algorithm).

But fortunately, you don't need to do 100% of the world's words to have something interesting. Just manually creating entries for a few hundred words from a dozen languages is probably already enough to start playing with your algorithm. If that's then so promising/exciting/whatever that it's worth embarking on a decade-long dictionary-parsing project, great; if not, go as far as you want to go with it.

Not the answer you're looking for? Browse other questions tagged or ask your own question.