Henry Higgins did not wait for Eliza Doolittle to ask for help. He simply kept correcting her until the problem was solved. This project takes the same approach.
Running on your own hardware, Higgins will interrupt you several times a day (over a messaging app of your choice) by sending a spontaneous conversational question in your target language. This is not some dry grammar exercise, but the kind of thing a curious and friendly stranger might actually ask at a dinner party. You respond when you can, and then Higgins will reply with a thumbs-up, or an improved version of your response, if one is necessary, and with errors clearly marked.
◇◆◇◆◇◆◇◆◇◆◇◆◇◆◇◆◇◆◇
Higgins is language-agnostic, delivery-agnostic, and LLM-agnostic. It can even use your own local LLM stack if you have one.
The questions are not drawn from a static library. Instead, Higgins generates each one spontaneously by seeding an LLM with a few random nouns, a social situation, and a discourse type. These seeds are all in simple text files that you can extend or customize yourself to zero in on target areas you want to improve.
This is much more interesting than typical language learning drills, dredging up answers from your own memories that will force you into language corners you haven’t explored before.
And then wagging a finger under your nose when you get it wrong, just like a stern tutor.
We don’t promise it’s as good as letting a stranger force you to mingle at a high-society party in your new language, but don’t be surprised if you find yourself muttering: “Just you white, ‘Enry ‘Iggins, just you white!”
