I also mentioned that Mandarin tones were the final boss of AI mixed language representation. If you ask an AI to reproduce Mandarin tones, the classical example being the four forms of the word "ma", it will give you identical tones for all four, and explain in english how it bent each of those tones to be so different. It's blind to its own inability. My hope was to simply have an eternally patient conversation partner. I could ask it how to say some sentence in Mandarin, it would reproduce the sentence aurally and in pinyin. Not much more structured or formal than any other immersion type language learning process. I think spanish/english mixing for such a project presents fewer obstacles, in that both use the same alphabet, and neither has tones.