5
$\begingroup$

I have no access to GPT-4, but I wonder whether it can do the following (where ChatGPT failed).

  • Make syntactic and morphological analysis of sentences in a language like Russian, marking cases, parts of speech and sentence, conjugations of verbs, etc. This is a standard school task, so I think this will be eventually possible (ChatGPT tried, but did it wrongly).

  • Reconstructing proto-forms of modern words based on known sound change laws.

  • Writing reconstructed texts in proto-languages based on the whole set of known grammar rules for those proto-languages, such as Proto-Indo-European (while of course highlighting arguable or unreliable choices). This to include stress and syllabification patterns, etc, things which while known to our science, are often omitted in recontructions because of the complexity.

  • Suggesting reconstructed sound correspondences, forms or patterns for proto-languages in automatic mode based on language's descendants. For instance, given vocabulary for Indo-Pacific languages to suggest most likely proto-forms for their proto-language.

$\endgroup$

1 Answer 1

3
$\begingroup$

So, here are some sentences GPT-4 generated in Russian (you can translate it with Google to English) and translated to PIE:

  • Я люблю лес. Он полон жизни и красоты. H₁éǵʰōm h₂lewh₂óm. H₂ésti pl̥h₁nós ǵʰéyewtis h₂ékʷeh₂s.
  • В лесу есть много деревьев. Они дают тень и кислород. H₂n̥h₁dʰgʷih₁és h₁estí h₂lewh₂óy. H₁óyti dʰeh₁núm h₃r̥h₃ḱwomkʷe.
  • Среди деревьев живут разные животные. Некоторые из них дружелюбны, а некоторые опасны. Médhyo dʰérwom bʰuh₂ónti pr̥h₃ḱwós bʰewgós. N̥sméy gʷih₃wós ǵʰóstis, n̥sméy tréyes.
  • Я вижу зайца. Он бежит быстро по траве. Widhḗmi kʷeklosóm. H₁éǵheti spéudeti pér ǵr̥h₃nom.
  • Я слышу волка. Он выет на луну. Kléwō wl̥ḱwom. H₁ówyeti ná mḗnsm̥.

They have a lot of errors. But it is much better than ChatGPT who hardly could understand what Proto-Indo-European is. One typical error is omitting diacritics, that is using k instead of and omitting superscript, that is kw instead of . I think, it is because of negligence in the training sources.

Anyway, I think that over not that long time AI would be able to generate grammatically correct text in PIE.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .