Kotoba is 言葉: words, written as the leaves of speech. Kotodama, the spirit said to live in them, is the older belief that a word spoken aloud acts on the world.
The JLPT word lists are long and flat. Learning them with a standard app mostly repeats what you already know, because the drill has no idea what is decaying.
I am working towards it myself and wanted a tool that puts the words I am currently forgetting in front of me.
Spaced repetition across the JLPT vocabulary. Every word carries an interval that shifts after each answer. Example sentences are generated rather than taken from a fixed database.
The example sentences come from a language model rather than a vetted sentence collection. A collection would have been reliable, but would only ever have covered the words somebody entered beforehand.
The price: every sentence costs an API call, quality varies, and a wrong sentence is not something I reliably catch at my own level. That dependency is the next thing to remove.
No measured values yet. Once the sentence generation runs locally on the Mac Mini in Tokyo from January, this chapter will report latency, power draw, and cost per thousand sentences.
From January the sentence generation runs locally on a Mac Mini in Tokyo instead of through someone else's API. Then the result chapter carries real numbers for the first time: latency, power draw, cost per thousand sentences.
