What is the difference between translation and localization?

Translation converts text from one language into another. Localization adapts an entire product or content set to a target market, including formats, units, currencies, imagery, legal notices and sometimes functionality. A software interface can be translated word for word and still be unusable in the target market if it has not been localized.

Is MTPE the same as machine translation?

No. Machine translation is raw software output with no human involvement. MTPE is that output corrected by a qualified human linguist, working to a defined standard. The difference matters commercially: some providers sell lightly checked machine output as human translation, so ask explicitly which service is being quoted.

What does ISO 17100 actually guarantee?

It guarantees that a defined process was followed: qualified translators, revision by a second qualified person, documented project management and client agreements. It does not guarantee that any specific translation is good, and it is not a government accreditation. It is a credible baseline, not a quality score.

Why does my translated document not fit the original layout?

Because text length changes on translation. English into French or German typically expands by 15–25 %, so a brochure that fit exactly in English overflows in French. This is handled by desktop publishing after translation, which is why DTP is quoted as a separate line item.

Do I need a translation memory?

If you translate the same or similar content more than once — product documentation, annual reports, contract templates, release notes — yes. A translation memory reduces cost on repeat content, keeps terminology stable across years, and remains your asset rather than the agency’s.

Glossary

Translation and Localization Glossary: 40 Terms Explained

This glossary defines the 40 terms most often used in translation and localization procurement, from TEP and MTPE to transcreation, back-translation and ISO 17100. Each definition is written for buyers rather than linguists, and explains what the term means in practice when you are reading a quote or a scope of work.

Maintained by Asiatis Singapore · Last updated 24 August 2026

Process and service levels

TEP (Translation, Editing, Proofreading)#: The standard three-stage professional workflow. One linguist translates, a second edits the translation against the source, and a final proofreading pass checks the text in isolation for typographical and formatting errors. TEP is the baseline for any document with legal, financial or commercial consequence.
Revision#: The comparison of a translation against its source document by a second qualified linguist, checking accuracy, completeness and terminology. Revision by a second person is the defining requirement of ISO 17100 and the main thing that separates a professional service from a single-translator service.
Review#: Assessment of a translation for suitability of purpose, usually by a subject-matter expert or the client's in-country team, without necessarily comparing it to the source. Review is about whether the text works for its audience; revision is about whether it is faithful to the original.
Proofreading#: A final check of the translated text alone, looking for typographical errors, punctuation, spacing, hyphenation and layout problems. Proofreading does not verify meaning against the source.
MTPE (Machine Translation Post-Editing)#: The correction of machine-translation output by a qualified human linguist. Light post-editing fixes only errors that affect meaning; full post-editing also aligns terminology and style with human-translation quality. MTPE is defined by ISO 18587 and is appropriate for high-volume documentation, not for contracts or regulatory submissions.
Transcreation#: The creative rewriting of a message for a different market, prioritising the intended effect over fidelity to the original wording. Used for advertising, taglines and brand messaging, where a literal translation would be accurate and useless. Transcreation is normally quoted per hour or per concept rather than per word.
Back-translation#: Translating a translated text back into the original language, by a linguist who has not seen the source, in order to verify meaning. Standard practice for clinical trial informed consent forms and patient-reported outcome measures, where regulators require documented evidence of conceptual equivalence.
Reconciliation#: The step following back-translation in which discrepancies between the original and the back-translated text are examined and the forward translation is amended where needed. The reconciliation report is often part of the regulatory submission.
Localization (l10n)#: Adapting a product or content to a specific market, going beyond text to cover date and number formats, currencies, addresses, units, images, colours, legal notices and functionality. Translating a website is a task; localizing it is a project.
Internationalization (i18n)#: Designing software or content so it can be localized without re-engineering: externalised strings, Unicode support, no hard-coded formats, layout that tolerates text expansion. Internationalization happens before localization and determines how expensive localization will be.
Locale#: A language plus a region, expressed as a code such as en-SG, zh-Hans, fr-CA or pt-BR. Locale, not language, is the correct unit for scoping a project: Simplified Chinese for Singapore and Simplified Chinese for mainland China differ in terminology and conventions.

Quality and standards

ISO 17100#: The international standard for translation services. It specifies the qualifications required of translators and revisers, mandates revision by a second person, and defines requirements for resources, client agreements and process management. It certifies the process, not the quality of any individual translation.
ISO 18587#: The international standard for full post-editing of machine translation output, defining the competences required of post-editors and the expected quality of the result.
ISO 27001#: The international standard for information security management. Relevant when confidential or personal data passes through a translation supply chain.
LQA (Linguistic Quality Assurance)#: A structured evaluation of a translation against a defined error typology — accuracy, terminology, language, style, locale conventions — with errors weighted by severity to produce a score. LQA turns "the translation is bad" into a measurable, actionable finding.
Error typology#: The classification framework used in LQA. Widely used models include MQM (Multidimensional Quality Metrics) and DQF. Agreeing an error typology before a project starts is the only way to arbitrate quality disputes objectively.
Style guide#: A client-specific document setting out tone, register, capitalisation, treatment of product names, formatting of numbers and dates, and forms of address in each target language. Without one, every linguist makes reasonable but inconsistent choices.

Technology

CAT tool (Computer-Assisted Translation)#: Software in which a human translator works, segmenting the source text and providing access to translation memory, terminology bases and quality checks. A CAT tool does not translate; it supports a translator. Common examples are Trados Studio, memoQ, Phrase and XTM.
Translation memory (TM)#: A database of previously translated segments paired with their source. When a new document contains identical or similar segments, the stored translation is proposed. Translation memory is why the second project for a client costs less than the first, and why terminology stays consistent across years.
Fuzzy match#: A segment that partially matches one already in the translation memory, expressed as a percentage. Matches above roughly 75 % are usually discounted on a sliding scale, since the linguist edits rather than translates from scratch.
Termbase (terminology database)#: A structured glossary of approved terms with their equivalents in each target language, often with definitions, context and forbidden alternatives. A termbase is enforced automatically inside the CAT tool.
Machine translation (MT)#: Automatic translation by software, today almost always neural or LLM-based. Fluent output is not the same as accurate output: modern MT produces plausible text that can be confidently wrong, which is precisely why post-editing by a qualified linguist is required for anything consequential.
Neural machine translation (NMT)#: Machine translation using neural networks trained on parallel corpora. NMT produces markedly more fluent output than earlier statistical systems, which makes its errors harder to spot without comparing against the source.
XLIFF#: An XML-based interchange format for localization data. Delivering content as XLIFF rather than as a formatted document keeps translation separate from layout and avoids re-formatting after every update.

Formats and delivery

DTP (Desktop Publishing)#: Rebuilding the layout of a document after translation so the target-language version matches the original design. Necessary because translated text changes length and because scripts such as Chinese, Japanese, Arabic and Thai have different line-breaking, font and directionality requirements.
Text expansion and contraction#: The change in text length caused by translation. English into French, Spanish or German typically expands by 15–25 %; English into Chinese or Japanese contracts substantially in character count. Layouts, user interfaces and packaging must be designed to tolerate this.
Source word count#: The number of words in the original document, used as the basis for pricing. Quoting on the source count means the price is fixed before work begins, unlike quoting on the target count, which is unknown until delivery.
EU official languages#: The 24 languages with official status in the European Union. Product and regulatory documentation must generally be supplied in the official language or languages of each Member State where the product is placed on the market, which is determined nationally rather than centrally. This is why market entry into five EU countries usually means five complete language versions, not one European version.

Buying and scoping

Source language and target language#: The source language is the language of your original document; the target language is what it becomes. Every scoping conversation starts here, and the direction matters: a linguist translating into their native language is the professional norm, so English into German and German into English are two different resources, not one.
Native speaker principle#: The convention that a translator works into their mother tongue, not out of it. It is the reason an agency's German is only as good as its German-native linguists, regardless of how many languages it lists.
In-country linguist#: A linguist who lives in the market they translate for. Language drifts: terminology, register and regulatory vocabulary change, and a linguist who left the country fifteen years ago writes a version of the language that sounds slightly dated to a local reader.
Language variant#: A regional form of a language with its own conventions: FR-FR and FR-BE, PT-PT and PT-BR, ES-ES and ES-419, DE-DE, DE-AT and DE-CH, NL-NL and NL-BE. Choosing the wrong variant is not a small error — Brazilian Portuguese in a Portuguese product manual reads as foreign to the intended user.
Word count#: The number of words in the source document, the basis for most quotes. It excludes nothing: text in images, speaker notes, alt text, error messages and overset text all have to be translated and all count once extracted.
Turnaround#: The elapsed time from confirmation to delivery. Sustained professional capacity is roughly 2,000–2,500 words per linguist per day including revision, so turnaround is a function of volume divided by the number of linguists a project can absorb without losing consistency.

Technical and file handling

Segment#: The unit a CAT tool works in, usually a sentence. Translation memory matches, quality checks and word counts all operate at segment level, which is why source documents with consistent sentence structure translate more cheaply than documents with fragmented text.
TMX and TBX#: The open interchange formats for translation memories (TMX) and termbases (TBX). If a supplier cannot export your assets in these formats, your data is locked in their tool.
Pseudo-localization#: A testing technique that replaces source strings with artificially lengthened, accented placeholder text to expose layout breakage, truncation and hard-coded strings before real translation begins. Cheap, and it prevents the most expensive class of localization bug.
String#: A unit of text in software, usually with an identifier, sometimes with variables and plural forms. Strings translated without seeing where they appear are the main source of software localization errors — hence in-context review.
RTL (right-to-left)#: Layout direction for scripts such as Arabic and Hebrew. Not relevant to European target languages, but worth knowing when a product roadmap extends beyond Europe, because RTL support is an engineering decision made long before translation.
Glossary vs termbase#: A glossary is a list of terms and their approved translations. A termbase is that list structured with definitions, context, part of speech and forbidden alternatives, and enforced automatically inside the CAT tool. The difference in practice is whether terminology is a suggestion or a control.

Frequently asked questions

What is the difference between translation and localization?: Translation converts text from one language into another. Localization adapts an entire product or content set to a target market, including formats, units, currencies, imagery, legal notices and sometimes functionality. A software interface can be translated word for word and still be unusable in the target market if it has not been localized.
Is MTPE the same as machine translation?: No. Machine translation is raw software output with no human involvement. MTPE is that output corrected by a qualified human linguist, working to a defined standard. The difference matters commercially: some providers sell lightly checked machine output as human translation, so ask explicitly which service is being quoted.
What does ISO 17100 actually guarantee?: It guarantees that a defined process was followed: qualified translators, revision by a second qualified person, documented project management and client agreements. It does not guarantee that any specific translation is good, and it is not a government accreditation. It is a credible baseline, not a quality score.
Why does my translated document not fit the original layout?: Because text length changes on translation. English into French or German typically expands by 15–25 %, so a brochure that fit exactly in English overflows in French. This is handled by desktop publishing after translation, which is why DTP is quoted as a separate line item.
Do I need a translation memory?: If you translate the same or similar content more than once — product documentation, annual reports, contract templates, release notes — yes. A translation memory reduces cost on repeat content, keeps terminology stable across years, and remains your asset rather than the agency’s.

Text expansion by language MTPE vs human translation Pricing

Ready to start your project?

Contact our Singapore office to discuss your translation needs.

+65 6950 0209

541 Orchard Road, #09-01, Liat Towers, Singapore 238881