2

enter image description here

The above is a screenshot of my conversation with ChatGPT 3.5 just now. The Chinese pinyin is wrong in multiple ways.

  • It makes a systematic error, ascribing 弦 the pinyin xiàn [4th tone] (it's ​xián [2nd tone])​. If you're not used to tonal languages, you may think this is not a big deal, but this is about as wrong as saying "jig" is pronounced "gig" (they're totally different words).

  • It makes another systematic error, adding suàn​ unnecessarily (it doesn't use the character 算 in Chinese). So it takes 2-syllable words like 余弦 ("cosine") and mistakes them for 3-syllable words like 余弦算 (which I'm not sure is a word).

In comparison to its other language skills, this is terrible.

Question: Why is ChatGPT so bad at Chinese pinyin (romanization)?

Improve this question
1
  • Chinese speaker here, it may be because you haven't specified which system to use so it's either getting confused or playing mix and match. For example, everyone knows about Mandarin and Cantonese, but there are 6 other main dialects and multiple regional ones. Even with "standard" things change, for example Peking and Beijing. The AI may be blending together multiple forms.
    – Aaargh Zombies
    Commented Nov 22, 2023 at 13:19

1 Answer 1

Reset to default
2

My guess is simply lack of training data and OpenAI focusing on English. Commonly used datasets for LLM training often don't have that much Chinese data (example -> Chinese represents 3.6% of the dataset), and even less pinyin.

Improve this answer

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.