What is the most optimal phonological spread possible?

[Creyeditor]

optimal for what though. no, seriously.

The most spread usage of the phonological space of human language, in such a way so that the number of discernable phonemes is maximized... but without doing something crazy like including affricates, click consonants or rare articulations that make no logical sense.

Is that intuitive?

Not at all.

[HolyKnowing]

optimal for what though. no, seriously.

The most spread usage of the phonological space of human language, in such a way so that the number of discernable phonemes is maximized... but without doing something crazy like including affricates, click consonants or rare articulations that make no logical sense.

Is that intuitive?

Not at all.

Perhaps I should explain my intention:

The goal is to create a language for knowledge storage, for the benefit of LLM's like ChatGPT. Given this, interoperability with extant human languages is not necessary. What is necessary is that the language be a self-sustaining, complete, and consistent system that the LLM and its community of human operators, developers, and engineers love to use... once they've grown into it.

Note that a sound system based on familiar phonemes (e.g. Spanish five vowels, Standard Average European consonants, etc...) is a subtype of interoperability, and that is explicitly not the goal. The goal is a self-sustaining, complete, and consistent system that the LLM & devops love to use... once they've grown into it.

Lastly, keep in mind that learning the sounds of a language Is one-and-done: you commit to it once, and you never have to do it again and are stuck with the sounds forever. How large should a phoneme inventory be with that in mind?

[Creyeditor]

Okay, I see. This is an engelang and not a (naturalistic) artlang. I won't question the LLM premise then.
I think my questions are:

• What is optimal in a spread? Maximal number of phonemes? Maximal distance between phonemes? Some compromise between the two? If you need a compromise, this is not trivial at all and I doubt that there is a single optimal solution.
• What does it mean for an articulation to "make no logical sense"? Is that a subjective judgement on your part or is it based on some objective complexity measure?
• What does it mean for an articulation to be "rare"? Is there an objective limit, e.g. less than 10% of the languages in PHOIBLE have it means an articulation is "rare".
• Why are affricates "crazy"? They are among the most frequent sounds among the languages of Earth. /ts/ for example occurs in 22% of phoneme inventories in PHOIBLE.

[HolyKnowing]

Okay, I see. This is an engelang and not a (naturalistic) artlang. I won't question the LLM premise then.
I think my questions are:

Very Good. Let us proceed.

[*]What is optimal in a spread? Maximal number of phonemes? Maximal distance between phonemes? Some compromise between the two? If you need a compromise, this is not trivial at all and I doubt that there is a single optimal solution.

The greatest number of phonemes that are strong and discernable to the human ear (which connects to your question of "maximal distance between phonemes?")

It's objectively easier to search thoroughly all the strong and discernable vowel sounds than it is to do the same for the consonants.

[*] What does it mean for an articulation to "make no logical sense"? Is that a subjective judgement on your part or is it based on some objective complexity measure?

This was intended to be more of a boundary condition, meaning don't simply include every possinle phoneme in the IPA.

[*]What does it mean for an articulation to be "rare"? Is there an objective limit, e.g. less than 10% of the languages in PHOIBLE have it means an articulation is "rare".

Rare means "no discernible pattern in the distribution of world languages; only occurs sporadically here and there, or is idiosyncratic to one group or even one language."

[*]Why are affricates "crazy"? They are among the most frequent sounds among the languages of Earth. /ts/ for example occurs in 22% of phoneme inventories in PHOIBLE.

They get crazy when it becomes time to construct the phonotactics, that is, the permissible consonant clusters. But yes, /ts/ and company very probably do belong on the list.

[Darren]

Voiceless nasals are not very distinct from each other, especially when not released into a vowel. That is unoptimal. Clicks, on the other hand, are very easily distinguishable, and occur paralinguistically with almost all languages.

[bradrn]

[*]What is optimal in a spread? Maximal number of phonemes? Maximal distance between phonemes? Some compromise between the two? If you need a compromise, this is not trivial at all and I doubt that there is a single optimal solution.

The greatest number of phonemes that are strong and discernable to the human ear (which connects to your question of "maximal distance between phonemes?")

It's objectively easier to search thoroughly all the strong and discernable vowel sounds than it is to do the same for the consonants.

Next question: what does ‘strong’ mean?

Clicks, on the other hand, are very easily distinguishable, and occur paralinguistically with almost all languages.

There’s a bit of subtlety to it. If your click inventory is something very small and simple like /ᵑǀ ᵑǁ ᵑǃ/, then yes, this is fair enough. But most click languages have inventories which are far more complicated, and rapidly get very very difficult to both produce and distinguish. Even pronouncing a tenuis click can be difficult for someone who isn’t used to it.

And academia articles are designed for avant-garde, state-of-the-art, and intellectual illegitimacy. They're not intended for discerning the Truth. Just boosting the academics clout.

A bit off-topic, but this is unfair to the point of being incorrect. The linguistics literature is fascinating and can give you answers to many of the questions you’ve been asking. (For instance: the ones on aspirated fricatives which Nort sent you, especially the last one.)

[HolyKnowing]

A bit off-topic, but this is unfair to the point of being incorrect. The linguistics literature is fascinating and can give you answers to many of the questions you’ve been asking. (For instance: the ones on aspirated fricatives which Nort sent you, especially the last one.)

That was pleasantly surprising. Thank you and thank @Nort for the good research.

[bradrn]

A bit off-topic, but this is unfair to the point of being incorrect. The linguistics literature is fascinating and can give you answers to many of the questions you’ve been asking. (For instance: the ones on aspirated fricatives which Nort sent you, especially the last one.)

That was pleasantly surprising. Thank you and thank @Nort for the good research.

You’re welcome!

[malloc]

The goal is to create a language for knowledge storage, for the benefit of LLM's like ChatGPT. Given this, interoperability with extant human languages is not necessary.

Then why bother emulating human phonology at all? Computers could easily handle something more compact and built around their native data processing like hexadecimal numbers or something.

[keenir]

And academia articles are designed for avant-garde, state-of-the-art, and intellectual illegitimacy. They're not intended for discerning the Truth. Just boosting the academics clout.

and what is The Truth??

and people who are specialists in linguistics, if they have clout, its because they are discerning the truth of things. so you're either mistaken or lying.

[bradrn]

And academia articles are designed for avant-garde, state-of-the-art, and intellectual illegitimacy. They're not intended for discerning the Truth. Just boosting the academics clout.

and what is The Truth??

and people who are specialists in linguistics, if they have clout, its because they are discerning the truth of things. so you're either mistaken or lying.

…did your quotes get mangled here, perchance? I don’t follow.

[Raphael]

The goal is to create a language for knowledge storage, for the benefit of LLM's like ChatGPT. Given this, interoperability with extant human languages is not necessary.

Then why bother emulating human phonology at all? Computers could easily handle something more compact and built around their native data processing like hexadecimal numbers or something.

For once, I kind of agree with malloc.

[Torco]

optimal for what though. no, seriously.

The most spread usage of the phonological space of human language, in such a way so that the number of discernable phonemes is maximized... but without doing something crazy like including affricates, click consonants or rare articulations that make no logical sense.
Is that intuitive?

not really, no. how do affricates 'make no logical sense' but voicing in trills does? also, why would an LLM language need a phonology? if what you want is a language for llms and humans to interact (are you sure you understand what devops means? it's a way for managers to order devs around, like scrum, not a kind of dev), wouldn't the obvious choice be the languages the llm is trained on?

as for clicks... do we know they're difficult in an absolute manner (say, xhosa kids acquire them later than spanish kids acquire erre) or do we just find them difficult cause our langs don't have 'em ?

[Ketsuban]

There’s a bit of subtlety to it. If your click inventory is something very small and simple like /ᵑǀ ᵑǁ ᵑǃ/, then yes, this is fair enough. But most click languages have inventories which are far more complicated, and rapidly get very very difficult to both produce and distinguish. Even pronouncing a tenuis click can be difficult for someone who isn’t used to it.

I feel like this is circular reasoning to some extent. The click languages we know of have large, complex inventories of click consonants, but they have large, complex inventories of everything so I don't think it's inconceivable a language could evolve click consonants and then simplify its overall phonology such that it ends up with a small but contrastive inventory of click consonants (in line with the somewhat anecdotal evidence that clicks tend to be "sticky", in that once a language acquires them there's a tendency to start clicking non-click consonants in the rest of the language).

[keenir]

and what is The Truth??

and people who are specialists in linguistics, if they have clout, its because they are discerning the truth of things. so you're either mistaken or lying.

…did your quotes get mangled here, perchance? I don’t follow.

yes, they did get mangled; sorry.

[HolyKnowing]

The goal is to create a language for knowledge storage, for the benefit of LLM's like ChatGPT. Given this, interoperability with extant human languages is not necessary.

Then why bother emulating human phonology at all? Computers could easily handle something more compact and built around their native data processing like hexadecimal numbers or something.

Because everything a computer does has to have a human interface in order for humans to receive that information. And English is a crummy interface for many reasons.

You can make it perform native data processing in hexadecimal provided that you're okay with never understanding it or being at the mercy of ad hoc human language front ends like we are today.

[bradrn]

I’d be curious to know if you’ve looked at Lojban? One of its original goals was to enable unambiguous human-computer interaction.

[HolyKnowing]

Voiceless nasals are not very distinct from each other, especially when not released into a vowel. That is unoptimal. Clicks, on the other hand, are very easily distinguishable, and occur paralinguistically with almost all languages.

There are only five fundamental clicks, and they do not interact well with the rest of human phonology. They are definitely a bad idea.

[bradrn]

There are only five fundamental clicks

Six, actually.

and they do not interact well with the rest of human phonology.

They interact just fine. Plenty of people speak languages with clicks, and they don’t seem to encounter any problems doing so.

[HolyKnowing]

I’d be curious to know if you’ve looked at Lojban? One of its original goals was to enable unambiguous human-computer interaction.

Yes. That is exactly what I am seeking to overturn.

One of the flies in the ointment of Lojban is that there are many logical distinction that can theoretically be grammaticized but practically the human mind will reject it. Second person clusivity is one such instance. Lojban by design says "let's include every logical feature we can grammaticize" resulting in something human beings refuse to process naturally.

The other fly in the ointment is that its morphology is brittle. The algorithm has no way of disambiguating words (what is self-segregating morphology?) unless each word is built according to an extremely tight and rigid briva procedure resulting in every sentence looking like "slapo noho pau" repeated a billion times.