In other words, Is there an internal measure/index of "cohesion" of language?

I was thinking of the contrast between artificial languages that machines or humans produce, and natural languages. The principal question is: When evaluating a language, can we say that this language is probably natural or artificial? how so?

My guess is to say that we can indeed tell difference; that there is some kind of internal 'harmony' or 'simplicity' or 'strong order' of the artificial language, which I will call here "cohesion". This term is not well defined here obviously, for I can't define what I still not sure what it is, but one can have in mind the principal question outlined above.

The idea here that natural living languages are subject, through the years, to many external (or even internal) "perturbations", thus we end up with language like modern-English that has its vocabulary dramatically influenced by foreign forces. It also has many irregularities of verbs inclinations. I ask: can we discern or quantify this trait just from evaluating the language by itself. The motivation for the measure being internal and independent is at least two-fold:

  1. Our knowledge is limited and might be even wrong. Here the motivation is rather to "create"/"discover" some kind of new information. Actually our knowledge of influences might be used to verify and evaluate the model post-factum, as they are probably correlated.
  2. Maybe a language can incorporate/absorb new terms, vocabulary and ideas and still keep its 'cohesion' high


What I'm primarily look for is a measure - a continuum - with which even two natural languages can be compared. not necessarily an artificial language should be involved - it is just the extreme case (i.e., I do expect to find artificial languages in one side of the continuum). I suspect the "level of cohesion" between languages is different and might reflect and give us insight about several things...

Edit 2: following a discussion in comments. I should explain that we can assume we understand the language (i.e., it's not language(?) in the Voynich manuscript, but rather Spanish, Akkadian, old-English, modern-English, etc...), but for the sake of the question, let us assume also we know nothing but the language itself (i.e., no history or even other languages)

  • This is almost certainly possible. No-one with any linguistic background would look at Lojban, Ithkuil, or even Esperanto and think they were a natural language. The relevant question is how reliably could one distinguish natlangs from naturalistic conlangs, and also how clearcut the distinction between natlang and conlang even is in the first place
    – Tristan
    Jul 16, 2021 at 15:05
  • Real languages are embedded within speech and cultural communities. There are no artificial "languages" that have them, and every natural language does, until it becomes extinct. That's how easy it is to tell.
    – jlawler
    Jul 16, 2021 at 16:45
  • 2
    @jlawler there are native esperantists, and a distinct esperanto culture, even if no full-time esperanto ccommunities. Regardless, there's no reason to think a conlang could develop a full-time speech community
    – Tristan
    Jul 16, 2021 at 16:48
  • 2
    @Tristan -Before “almost certainly” telling if the text is in an artificial or natural language, you have first to be able to tell if it's in a language. Can this be done? Looks like no, or else you can tell us whether it is an artificial or natural language that's used in the Voynich manuscript. Can you?
    – Yellow Sky
    Jul 16, 2021 at 17:12
  • 1
    In fact, it is already a hard problem to decide if something is language or not, I am thinking of the famous Voynich manuscript or the so-called Indus Script. Distinguishing artificial languages from natural ones will be an even harder problem, specially for naturalistic conlangs vs. highly regulated natlangs like French. Jul 16, 2021 at 17:43

1 Answer 1


The question has never come up (officially: maybe you are the first to suggest developing a test). The problem is that we already know the answer, at least if you're dealing with a decent-sized corpus and not just a 3-word written snippet.

You could re-frame the question as being about distinctive properties of conlangs as contrasted with those of natural languages. Also a historical divide should be made between older conlangs and contemporary post-Klingon creations, because the latter set are strongly influenced by contemporary linguistic research (not theory so much, but findings of fact). The difference between a natural language vs. a conlang is simply whether the language developed naturally as a product of various social forces and is memorized, as is, by children, versus whether the language is deliberately constructed by a small number of people who had some purpose in mind. There is a strong tendency for conlangs to be phonologically simple (e.g. relatively unmarked segment inventory, avoidance of complex morphophonemics as exists in Arabic verb inflection), because customers tend to avoid languages that they can't handle (except for a small number who thrive on a language where the word for "mother" is [ʈqfħmaʔktsp]).

There is also a tendency for conlangs to be over-regular. Natural languages tend to accumulate fixed constructions because they have been spoken for many centuries and nobody purges commonly-used contractions like "will" + "not". Conlangs are (1) a few years old and (2) not naturally acquired by a community, so keeping irregularities out is easier.

The challenge in performing a large-scale diagnostic program would be the lack of sufficiently extensive specification of many conlangs. For example, where's the grammar or dictionary of Romulan? Nobody (official) has even written a grammar of The Hero's Tongue. Also, many non-humanoid conlangs have human anatomically-impossible sounds which are a dead giveaway, or maybe they "talk" in whispy patterns of smoke.

  • Thanks for this answer. I've edited my question a little bit. I'm not interested in knowing which language are artificial and which are natural - as you say, we already know that.... I'm interested in the : how artificial languages and natural languages differ, a measurable property with which they can be discerned.
    – d_e
    Jul 16, 2021 at 16:45
  • 2
    You sound a bit too optimistic. The test for artificiality you suggest is something to be performed with a text which is supposed to be in a language. What form is the text supposed to be presented in? In IPA? But what if it's in a macabre complicated ideographic script of a Tangut kind? What if it's a modulated stream of neutrinos from another galaxy? In Lem's novel His Master's Voice the word's best minds faced the problem of telling if that is a text in a language at all or just an unknown natural phenomenon.
    – Yellow Sky
    Jul 16, 2021 at 17:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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