4 ms·

I wonder if a GPT or something alike could possibly produce a good conlang easy to understand and learn for people of a specific linguistic group by optimally i

I wonder if a GPT or something alike could possibly produce a good conlang easy to understand and learn for people of a specific linguistic group by optimally interpolating between their languages and filtering out idiosyncrasies.

pmontra 3y ago
Maybe, but it would take some real effort. This naive attempt of mine is an obvious non starter (it's ChatGPT 3.5). Transcript with some minimal editing to fit it into HN: User: Would you be able to develop an artificial language easy to learn and use for speakers of English and romance languages? Extremely simple grammar. ChatGPT: Sure, I'd focus on minimal grammar, consistent rules, and common vocabulary. It's doable. User: Good. Give a name to that language ChatGPT: How about "LinguaEase"? User: OK. Give me some examples starting with translations of the sentences in this chat so far. ChatGPT "LinguaEase Exempla: > 'Sure, I'd focus on minimal grammar, consistent rules, and common vocabulary.' becomes 'Yep, keep grammar simple, rules steady, vocab shared.' > 'How about "LinguaEase"?' becomes 'What about "LinguaEase"?' > 'OK. Give me some examples starting with translations of the sentences in this chat so far.' becomes 'Alright, show me samples, begin with translations of this chat's sentences.'" I went with English and romance languages because they are the languages I know. It looks all English to me. I would probably have had to be more specific. For example: only present tense. No plural. No gender. Only one set of pronouns. Aim to sentences like Beer two me drink yesterday. Beer ten me buy tomorrow. Then work on vocabulary.
- d1sxeyes 3y ago
  I agree it would be very difficult. I think that current LLMs would fail here because I don't think they would be able to make a value judgement about what is 'harder' between 'I drank two beers yesterday' and 'Beer two me drink yesterday'. In fact, I'm not sure most linguists would agree. All languages are something of a compromise between accuracy and simplicity. An LLM is currently not equipped to make decisions about the correct balance between accuracy and simplicity. For example, English eschews cases (in most situations) in favour of fixed word order. 'Beer two me drink yesterday' violates that rule (and in fact, manages to use case incorrectly in one of the few places the vestiges of a case system still exist in English). However, a sentence constructed using the same structure as yours highlights a problem. 'Me two dog bite yesterday' Did you bite two dogs or were you bitten by two dogs? How can you tell? English has the subject of the sentence (the thing/person doing the verb) first, then the verb, then the object (the thing having the verb done to it). In languages with more fluid word order, normally a case marker is added to either the determiner or the noun itself: 'Me-o two dog bite yesterday'. Let's choose '-o' as our accusative case ending. Now we can tell that it was you who was bitten by the dog. Next problem: what if instead of 'me' it's 'man'. 'Man-o two dog bite yesterday'. How do we know whether it's two men, or two dogs? Again, English keeps the number with the noun it refers to. Most languages do, but some also put a marker on the number to make it clear. 'Man-o two-o dog bite yesterday'. Now we can tell that two men were bitten by some number of dogs. Next problem is that we don't know how many dogs bit the men. Were two men bitten by the same dog, or different dogs? Were two men surrounded by a pack of dogs? Is it important? Most languages care about this enough to at least mark the plural. What if we don't know exactly when the men were bitten? We could have some generic word like 'past', to make 'man-o two-o dog bite past'. But at that point, is your conlang really any easier than English, or just different?
  - pmontra 3y ago
    Of course all of this is a game. I address this point of yours > However, a sentence constructed using the same structure as yours highlights a problem. > 'Me two dog bite yesterday' > Did you bite two dogs or were you bitten by two dogs? "Me two" could be ungrammatical in that language. "Me VERB" is grammatical. Generalizing: "SUBJECT VERB" and for subject I'd go with "NOUN modifiers", so "dog two" and not "two dog". So the sentence should be either "Me bite dog two yesterday" or "Me dog two bite yesterday" or "Dog two me bite yesterday" or "Dog two bite me yesterday" or even "yesterday" + any variation of the above. Being a native Italian speaker I'm used to free order sentences. Given that language should be a mix of English and romance languages, some free ordering should be expected. Cases should stay away from any language aiming at simplicity. We simplified Latin a lot to get Italian (and French, Spanish, etc.) We also have way more speakers now. What happens in the real world is that when adult speakers of different languages meet in a bare place and start communicating with too simple rules like "beer two me drink yesterday" (that would be a pidgin [1]) is that they eventually have children that learn the language and add grammatical layers to it. So you would end up with Yesterday two beers me drink" or whatever they feel sounding better or being more economical to speak and understand. Then they become adults and have their own children and the language stabilizes. That's a creole [2] and it can be as complicated as any of the languages it derives from. This answers your last question. An alternative is developing an artificial language for maybe a restricted community, keep it simple by enforcing the grammar with an academia, and that will stay simple forever. I don't know if that's the case of Klingon or Vulcan (are they simple? do they have a governing body?) but it might be. [1] https://en.wikipedia.org/wiki/Pidgin https://en.wikipedia.org/wiki/Pidgin [2] https://en.wikipedia.org/wiki/Creole_language https://en.wikipedia.org/wiki/Creole_language
    - d1sxeyes 3y ago
      > "Me two" could be ungrammatical in that language. Absolutely :) but then introducing a new grammar for this hypothetical language is starting to push it down the road to complexity.
    - qwerty456127 3y ago
      > ...that would be a pidgin... That's a creole... What is it called when people (e.g. an colonized nation) generally speak one language but introduce a huge lot of loanwords from another into it (e.g. use their native words while speaking the colonizer language or the opposite) and generations grow speaking this 2-language hybrid as their mother tounge?

pmontra 3y ago

Maybe, but it would take some real effort. This naive attempt of mine is an obvious non starter (it's ChatGPT 3.5). Transcript with some minimal editing to fit it into HN: User: Would you be able to develop an artificial language easy to learn and use for speakers of English and romance languages? Extremely simple grammar. ChatGPT: Sure, I'd focus on minimal grammar, consistent rules, and common vocabulary. It's doable. User: Good. Give a name to that language ChatGPT: How about "LinguaEase"? User: OK. Give me some examples starting with translations of the sentences in this chat so far. ChatGPT "LinguaEase Exempla: > 'Sure, I'd focus on minimal grammar, consistent rules, and common vocabulary.' becomes 'Yep, keep grammar simple, rules steady, vocab shared.' > 'How about "LinguaEase"?' becomes 'What about "LinguaEase"?' > 'OK. Give me some examples starting with translations of the sentences in this chat so far.' becomes 'Alright, show me samples, begin with translations of this chat's sentences.'" I went with English and romance languages because they are the languages I know. It looks all English to me. I would probably have had to be more specific. For example: only present tense. No plural. No gender. Only one set of pronouns. Aim to sentences like Beer two me drink yesterday. Beer ten me buy tomorrow. Then work on vocabulary.

d1sxeyes 3y ago
I agree it would be very difficult. I think that current LLMs would fail here because I don't think they would be able to make a value judgement about what is 'harder' between 'I drank two beers yesterday' and 'Beer two me drink yesterday'. In fact, I'm not sure most linguists would agree. All languages are something of a compromise between accuracy and simplicity. An LLM is currently not equipped to make decisions about the correct balance between accuracy and simplicity. For example, English eschews cases (in most situations) in favour of fixed word order. 'Beer two me drink yesterday' violates that rule (and in fact, manages to use case incorrectly in one of the few places the vestiges of a case system still exist in English). However, a sentence constructed using the same structure as yours highlights a problem. 'Me two dog bite yesterday' Did you bite two dogs or were you bitten by two dogs? How can you tell? English has the subject of the sentence (the thing/person doing the verb) first, then the verb, then the object (the thing having the verb done to it). In languages with more fluid word order, normally a case marker is added to either the determiner or the noun itself: 'Me-o two dog bite yesterday'. Let's choose '-o' as our accusative case ending. Now we can tell that it was you who was bitten by the dog. Next problem: what if instead of 'me' it's 'man'. 'Man-o two dog bite yesterday'. How do we know whether it's two men, or two dogs? Again, English keeps the number with the noun it refers to. Most languages do, but some also put a marker on the number to make it clear. 'Man-o two-o dog bite yesterday'. Now we can tell that two men were bitten by some number of dogs. Next problem is that we don't know how many dogs bit the men. Were two men bitten by the same dog, or different dogs? Were two men surrounded by a pack of dogs? Is it important? Most languages care about this enough to at least mark the plural. What if we don't know exactly when the men were bitten? We could have some generic word like 'past', to make 'man-o two-o dog bite past'. But at that point, is your conlang really any easier than English, or just different?
- pmontra 3y ago
  Of course all of this is a game. I address this point of yours > However, a sentence constructed using the same structure as yours highlights a problem. > 'Me two dog bite yesterday' > Did you bite two dogs or were you bitten by two dogs? "Me two" could be ungrammatical in that language. "Me VERB" is grammatical. Generalizing: "SUBJECT VERB" and for subject I'd go with "NOUN modifiers", so "dog two" and not "two dog". So the sentence should be either "Me bite dog two yesterday" or "Me dog two bite yesterday" or "Dog two me bite yesterday" or "Dog two bite me yesterday" or even "yesterday" + any variation of the above. Being a native Italian speaker I'm used to free order sentences. Given that language should be a mix of English and romance languages, some free ordering should be expected. Cases should stay away from any language aiming at simplicity. We simplified Latin a lot to get Italian (and French, Spanish, etc.) We also have way more speakers now. What happens in the real world is that when adult speakers of different languages meet in a bare place and start communicating with too simple rules like "beer two me drink yesterday" (that would be a pidgin [1]) is that they eventually have children that learn the language and add grammatical layers to it. So you would end up with Yesterday two beers me drink" or whatever they feel sounding better or being more economical to speak and understand. Then they become adults and have their own children and the language stabilizes. That's a creole [2] and it can be as complicated as any of the languages it derives from. This answers your last question. An alternative is developing an artificial language for maybe a restricted community, keep it simple by enforcing the grammar with an academia, and that will stay simple forever. I don't know if that's the case of Klingon or Vulcan (are they simple? do they have a governing body?) but it might be. [1] https://en.wikipedia.org/wiki/Pidgin https://en.wikipedia.org/wiki/Pidgin [2] https://en.wikipedia.org/wiki/Creole_language https://en.wikipedia.org/wiki/Creole_language
  - d1sxeyes 3y ago
    > "Me two" could be ungrammatical in that language. Absolutely :) but then introducing a new grammar for this hypothetical language is starting to push it down the road to complexity.
  - qwerty456127 3y ago
    > ...that would be a pidgin... That's a creole... What is it called when people (e.g. an colonized nation) generally speak one language but introduce a huge lot of loanwords from another into it (e.g. use their native words while speaking the colonizer language or the opposite) and generations grow speaking this 2-language hybrid as their mother tounge?