Add translation memory for SUMM.AI translations into Easy German

[timobrembeck]

Motivation

The translations into Easy German via the SUMM.AI API consume credits on every call, even if we translate the same sentence twice.

Proposed Solution

Add a local translation memory to keep track of the already translated texts and fetch them from the memory instead of translating them again.

Alternatives

• Just ignore the matter
• Wait for SUMM.AI to implement a translation memory

Additional Context

How exactly we store the information is open for debate. A few ideas:

• In-memory cache (fast, but probably not suitable as we would need to store large amounts of data that is accessed very infrequently)
• File-based cache (discuss file forrmat, e.g. XLIFF or the format MemoQ uses for TMs?)

[osmers]

Just checking here - is there a time limit for how long the memory should be accessed? Just asking bcs I think I remember a conversation where we said that we do not want a translation memory bcs SUMM AI gets better with every translation. What I could see is that a translation memory is created and entries are accessed for a specific time after initial creation - e.g. 4 weeks After that, we assume that the translation has gotten better and ignore the TM entry? What do you think @dkehne ?

[dkehne]

I would even go for 90 days....same with MT on DeepL/Google