So I recently started to learn Japanese, and especially when new to a language, I find reading in sentence mode with automatic translation + audio to be the most useful way to read. However, it seems every sentence is translated individually, rather than using the context of the rest of the lesson. Not only is this incredibly wasteful of compute (it seems to me like every time a user opens a sentence, a translation is newly generated), it also makes for poorer translation. From what I can tell, Japanese verbs are conjugated less for person / number than in English, and pronouns can be optional. For example, in the Eating Out (or something similar) beginners course, someone is asked how their parents are doing, and their response (a seperate sentence) is translated as āIām fineā. This is because the Japanese sentence on its own is ambiguous, it basically just means āfine is/areā.
Ideally, I would propose translations be generated lesson-wide and stored with the lesson. This would also allow us to submit corrections. If this is not feasible, generate translations once for the entire lesson, so the translation software has the context of the entire lesson.
