What Is Translation Memory (TM) and How Does It Work?

Translation memory (TM) is a database that stores source sentences paired with their approved translations, called segments, and suggests them whenever the same or similar text appears again. That reuse cuts redundant work and translation costs, and it keeps phrasing consistent across documents. According to Smartling's 2024 State of Translation Report, translation volumes grew 30% year over year, which makes reuse essential at scale. When new text matches a stored segment exactly or closely, the translator confirms or edits the suggestion instead of translating from scratch, and vendors bill matched text at a discount. This guide will explain translation memory, how it works, and the benefits of leveraging it in your translation management system. What is translation memory? Translation memory is different from a glossary or termbase, which stores approved translations of individual terms rather than whole sentences. TM is useful for businesses handling large volumes of content, such as product manuals, websites, and legal documents. Reusing translated segments accelerates the translation process, improves consistency, and lowers costs. How does translation memory work? Translation memory works by breaking translated text into smaller sections, called segments, and storing them in a database. As a translator (or automated translation) works on new content, the system searches for matches between the text and previously translated segments. These matches are categorized as either perfect matches (100% identical to a stored segment) or fuzzy matches (similar but not exact, requiring minor edits). For example, suppose the sentence "Click Save to continue" was translated into Spanish last year. When the same sentence appears in new content, translation memory returns it as a 100% match, and the translator only confirms it. When "Click Save to proceed" appears, translation memory returns the earlier translation as a fuzzy match, and the translator changes one word instead of translating the whole sentence. Translation memory match types and how they're priced   Match type What it means Example rate (share of the full per-word price) 100% match The new segment is identical to a stored segment 10% 95%–99.9% fuzzy match Nearly identical; usually punctuation or one small word differs 30% 85%–94.9% fuzzy match Similar; a few words need editing 60% 0%–84.9% (no useful match) Translated from scratch 100%   In human translation or editing, the translator can accept, modify, or replace any detected match. New or updated translations are then saved for future use. Translators often use TM within a computer-assisted translation (CAT) tool, which helps manage and reuse translations efficiently. TM is also often leveraged in a translation management system (TMS), a platform that streamlines translation workflows for large-scale projects. What is the difference between translation memory and machine translation?   Translation memory Machine translation Where the output comes from Translations people have already approved Generated new by a neural model or LLM Consistency Very high: the same sentence always gets the same translation Can vary between runs and engines Cost Matched text is billed at a discount A small fraction of human per-word rates Best for Repeated, updated, or regulated content New content, large volumes, first drafts Used together? Yes. Most translation management systems apply translation memory first and send unmatched text to machine translation. Translation memory and machine translation (MT) both speed up the translation process, and they can be used together, but they serve different purposes. Machine translation automatically translates text without human involvement. Machine translation engines use neural machine translation, which relies on artificial intelligence and deep learning to analyze language patterns and generate translations. Machine translation provides fast content translation that communicates the general meaning of a document, but it could lack accuracy, especially with complex language or specialized terminology. Some MT solutions use translation memory to get around this issue. The user can set thresholds that dictate when the system

The post What Is Translation Memory (TM) and How Does It Work? appeared first on Smartling Blog.

For more information, please visit https://www.smartling.com/blog/what-is-translation-memory