9
$\begingroup$

"An artificial or constructed language (sometimes called a conlang) is a language that has been created by a person or small group, instead of being formed naturally as part of a culture." (Source: Simply English Wikipedia)

My question is, could an AI make construct it's own natural language, with words, conjugations and grammar rules? Basically, a language that humans could use to speak to each other. (Preferably to communicate abstract, high-level concepts.)

What techniques could such an AI use? Could it be based on existing natural languages or would it have few connections to existing natural languages? Could it design a language that's easier to learn than existing languages (even Esperanto)?

$\endgroup$
  • $\begingroup$ Question seems to be predicated on the idea of the evolution a "natural" language, which presumably would require stable thought processes, individuation, and a culture, none of which have clear parameters for AI's at their current stage of development. Maybe rephrase with a more specific premise? $\endgroup$ – dynrepsys Sep 1 '16 at 15:40
  • $\begingroup$ For example HTTP is kind of a constructed language (with verbs and nouns). So if an AI constructs something similar that would be such an example? $\endgroup$ – Trilarion Sep 6 '16 at 20:02
  • $\begingroup$ @Trilarion HTTP as in Hypertext Transfer Protocol? $\endgroup$ – Kevin Sep 7 '16 at 10:24
  • $\begingroup$ Yes, see also REST. $\endgroup$ – Trilarion Sep 7 '16 at 10:58
9
$\begingroup$

could an AI make construct it's own natural language, with words, conjugations and grammar rules?

Sure. This might be helpful: Simulated Evolution of Language: a Review of the Field

Basically, a language that humans could use to speak to each other. (Preferably to communicate abstract, high-level concepts.)

I'm not sure how useful such a machine adapted language would be to human speech. I suspect not very. Perhaps it could be useful as a kind of "common byte-code format" to translate between multiple human languages... But English kind of already serves in that role. Doing so would probably be an academic exercise.

Could it be based on existing natural languages or would it have few connections to existing natural languages?

You could probably generate languages in either direction. Languages not linked to human-natural languages will probably take shapes that reflect the problem spaces they work with. For instance, if this is an ant simulation, the generated words will probably reflect states related to food, energy, other ants, etc.

Could it design a language that's easier to learn than existing languages (even Esperanto)?

Easier for a machine? Definitely. Easier for a human? Probably not. Our brains are somewhat adapted to the languages we use. And the languages we use are somewhat adapted to our brains.

What is your goal? To create a language that is easier to use for humans than existing human languages?

If your intention is to build a "universal language" that could be "the most efficient" for machines, humans and aliens - no such thing can exist. The space of all possible machines is infinite and therefor limits our ability to define communicative abstractions that have utility across all contexts.

If we make lots of assumptions about machines, like they have intentions, they exist in 3 dimensions, they differentiate between temporally linked events, they have eye balls, a need to consume foods and liquids, a need to carry the food from place to place, etc... Then yes, common communicative abstractions may have utility across the set of those kinds of machines. But then we're no longer dealing with the general case, but one much more specific.

These links also seem interesting and somewhat related:

$\endgroup$
  • $\begingroup$ As I learned yesterday, answers or comments with just links to other places are not quite appropriate to the site. The moderator that had removed my comment with links noted that we aimed to provide more comprehensive answers here rather than linking people to other places. I would simply edit the answer to amalgamate the information in the links if you choose to pose this as an answer. $\endgroup$ – Avik Mohan Sep 2 '16 at 13:20
  • $\begingroup$ Good point. I'll briefly summarize why I think the top link is relevant. Thanks. $\endgroup$ – Doxosophoi Sep 2 '16 at 13:42
  • $\begingroup$ Thanks for you answer. Normally human languages are shaped over hundreds of years or are constructed which take much work. I was curious if a machine (an AI) could do this task much quicker. For me it would be awesome to speak a (constructed natural) language that is made by an AI. Also, I think it could give more insight in how languages envolve. $\endgroup$ – Kevin Sep 2 '16 at 20:20
  • $\begingroup$ Very nice answer. $\endgroup$ – BobbyPi Sep 4 '16 at 14:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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