What is translation memory?

Translation memory is a persistent store of source strings and their approved translations, consulted before any new translation is requested, so text that has been translated once is reused rather than translated again.

Why it matters

An application's interface is highly repetitive. Every session of the same product shows the same buttons, headings and error messages. Translating each recording from scratch would be slow, expensive and inconsistent: the same label could come back worded three different ways.

A memory fixes all three at once. The first session pays for a string, every later session gets it instantly, and a human correction sticks. This is the same idea professional translators have used for decades, applied to the session translation of a replay.

How AnyReplay handles it

AnyReplay's translation memory is keyed on the organisation, unique on organisation, target language and source hash, because sessions of one application repeat the same interface strings. Each string is extracted from the rrweb event stream with a role such as heading, button, label or input, whitespace-normalised and hashed with case kept. After a handful of recordings almost every string is already cached and costs nothing. A correction made by a reviewer is marked reviewed and never sent to a model again. One translation job runs per session per target language; asking twice reuses the first. Glossary terms take precedence over the model's judgement. The localization teams use case shows it in practice.