Software Localisation Services

Software localisation puts a product into another market’s conventions rather than only its vocabulary: the interface text inside its resource files, the plural categories a sentence needs, the order a list sorts in, the way a figure and its currency are written, and the documentation a user reaches when the screen has not helped. Espresso Translations localises software from £0.10 per word over 2 to 3 working days, with native linguists, revision against the source by a second linguist under ISO 17100:2015, 150+ languages and a free quote inside the hour.

ISO 17100:2015 Certified

Native Linguists, Five Years Minimum

£0.10 Per Word

Revised Against the Source

150+ Languages

Quote Back in 1 Hour

Espresso Translations on Google
Espresso Translations on Trustpilot
Espresso Translations ISO 17100:2015 certification for translation services

Send Your String Files for a Free Quote

A project manager answers within 1 hour

Forbes logo, an Espresso Translations client
IMDB logo, an Espresso Translations client
Jaguar logo, an Espresso Translations client
Universal logo, an Espresso Translations client
National Geographic logo, an Espresso Translations client
EY logo, an Espresso Translations client
Sixt logo, an Espresso Translations client
Amazon logo, an Espresso Translations client
EDF logo, an Espresso Translations client
FREENOW logo, an Espresso Translations client

What is software localisation? Software localisation is the work of making a product behave as though it had been built for another market: translating the interface text inside its own resource files, supplying every plural category the target language grammatically requires, drawing dates, numbers and currencies from locale data instead of writing them by hand, mirroring a right-to-left layout, and localising the help centre, release notes and store listing around the product. Espresso Translations localises software from £0.10 per word over 2 to 3 working days, with a native linguist on each language, revision against the source by a second linguist under ISO 17100:2015, and a free quote within 1 hour. The correction UK product teams most often need is a legal one. The European Accessibility Act has applied since 28 June 2025 and sets accessibility requirements on listed digital products and services, but it places no obligation on you to offer the interface itself in any particular language; its language duties attach to instructions, safety information, contact details, documentation for authorities and the declaration of conformity.

What Is Software Localisation, and Where Does Translation Stop?


Translation stops at the words; localisation carries on to everything the words sit inside. In a software product that means the resource file a string lives in, the message format that decides its plural and gender branching, the locale data that renders its dates and numbers, the width of the control it has to fit, the direction the layout runs in and the sort order of the list it appears in. Espresso Translations does the linguistic half of all of that, in 150+ languages, and tells you plainly which half belongs in your code. American sources and most tooling spell it software localization with a z, and the two spellings describe the same work. Translation carries a sentence from one language into another; localisation changes everything the sentence sits in.

Internationalisation comes first, and it is engineering rather than language. It means no user-facing text left hard-coded in the source, every string reachable through a resource file under a stable key, locales identified by BCP 47 language tags, plural and gender logic expressed through a message format instead of by gluing fragments together, and dates, numbers and currencies formatted by the platform rather than assembled by hand. Only once that is true can a linguistic team work without touching your code. Where strings are still being concatenated we say so before you pay for the translation, because that is a code change and no amount of language work substitutes for it. Done in that order, a localisation workflow fits inside your software development cycle instead of running alongside it.

A product is more than its interface, and on most projects the interface is the smaller word count. A localisation project normally takes in onboarding, system notifications and transactional email, the help centre and knowledge base, error and empty-state text, release notes, the store listing, and the policy pages a user has to accept. The marketing site and the public pages around the product are a separate scope with a different audience and a different brief, covered on our website and store pages.

On the legal question the honest answer is narrower than the one this market is usually sold. Directive (EU) 2019/882, the European Accessibility Act, was to be transposed by 28 June 2022 and has applied since 28 June 2025; it covers consumer computer hardware and operating systems, consumer terminal equipment, e-readers, e-commerce and listed transport services, with a transitional period running to 28 June 2030. It imposes no requirement that a software interface be offered in any given language. Its language obligations attach to four things: product instructions and safety information in a language the Member State determines, contact details, documentation supplied to a national authority, and the EU declaration of conformity. The reason to localise is that users buy more and support queues get shorter, not that a statute compels it.

Which Resource Files Can You Send, and What Comes Back?


Send whatever your toolchain already exports. A resource file carries the identifiers, the developer comments, the placeholders and the plural categories that a spreadsheet of sentences throws away, and it is the only form in which a translation memory can tell a genuinely new string from one you have already paid for once.

Resource files we work in, and what comes back
Resource fileToolchain it comes fromWhat comes back
XLIFF 2.1 (.xlf), the XML Localization Interchange File FormatThe current OASIS Standard, approved 13 February 2018; Xcode exports and imports localisations in itThe same document with its target segments populated, ready to import where it came from
gettext .po, .pot and .moGNU gettext: xgettext writes the .pot template, each .po holds one language, the .mo is the compiled binary formOne .po per locale with every msgid untouched, so msgmerge picks up only what changed next release
strings.xml and the values foldersAn Android project under res/values, holding string, string-array and plurals resourcesA values folder per locale, with quantity strings in the categories that language grammatically requires rather than the ones English needs
String Catalogue (.xcstrings), .strings and .stringsdictXcode: the String Catalogue is Apple’s current recommendation and the property-list pair is the earlier routePlural variants populated, with the format keys and variable names left exactly where they were
RESX, JSON, YAML, .properties, Markdown and CSVA .NET solution, a web front end, a Java bundle, a documentation repository, or an export from a translation management systemIdentical structure and escaping with only the values changed, so a diff shows nothing except language
Uploading resource files for a free software localisation quote
Second linguist checking a translated string against its character limit
Project manager agreeing the product glossary before translation starts
Translated resource files returned in the format they arrived in

Locale Data: Unicode CLDR 49, and Its New Name

Locale data is where dates, times, numbers, currencies, measurement units, plural rules and collation order come from, and the reference set is Unicode CLDR, version 49, released on 22 October 2026. Unicode describes it as the key building blocks for software supporting the world’s languages, used by all major software systems including every mobile phone. Two details in that release matter, because almost every page written before late October 2026 has them wrong: CLDR now expands as Common Locale Data & Rules rather than Common Locale Data Repository, and UTS #35 has been retitled to match, so LDML is no longer the specification’s name. If your product formats a date itself instead of asking the platform, this is the data it is failing to use.

Plural Rules and ICU MessageFormat 2

Count is not plurality. English needs two forms; Lithuanian uses its singular for 1 and again for 101, which Android’s own documentation flags as the trap it is; Korean needs one form only. A label stitched together from a figure and a noun cannot express any of that, which is why the branching has to live inside the translatable string. The current instrument for that is published as UTS #35 Part 9, MessageFormat, inside CLDR 49 and dated 21 October 2026, and Unicode’s own Status section calls it a stable document, citable as a normative reference, while warning that any function inside it marked Draft may still move. For a product team that reads as a sequencing decision rather than a migration: the specification has settled, the libraries have not, so stay on what your platform supports today and write strings that will cost nothing to move later.

Language Tags: BCP 47 Is Two RFCs

BCP 47 is RFC 5646, Tags for Identifying Languages, of September 2009, together with RFC 4647, Matching of Language Tags, of September 2006. Both are Best Current Practice, and the RFC Editor records nothing obsoleting or updating either one. The second is the half people forget, and it is the half that decides which of your locales a browser asking for es-419 is actually given. For a UK team the practical point is closer to home than that: en-GB and en-US are distinct language tags, and serving one of them to both markets is a decision rather than an oversight.

Bidirectional Interfaces and UAX #9

Right-to-left markets are the ones where a build looks wrong while every string is right. Hebrew, Arabic, Persian and Urdu run from the right, and Unicode Standard Annex #9, the Unicode Bidirectional Algorithm at Revision 52 of 1 September 2026 inside Unicode 18.0.0, is what settles the position of each Latin word, figure and punctuation mark embedded in them. The symptoms are not linguistic. A company registration number or an order code loses its reading order halfway through a sentence, an amount and its symbol separate, and the chevron on a back button carries on pointing the way it did in English. Our half is the language and the report that comes with it; mirroring the layout is a change in your build.

Which Parts of a SaaS Product Need Localising?


Start where a user cannot proceed without reading, then work outwards. In order: the user interface and its resource files, the onboarding and transactional messages that arrive by email and push, the help centre, the store listing and release notes, and finally a testing pass that checks all of it in context. A SaaS product localises continuously rather than once, because the product changes every release. Which markets come first is a target market decision rather than a linguistic one, and we will price two or three candidates so it can be made on numbers.

The Product Interface and Its Resource Files

The interface is the smallest word count and the highest risk, because every label is short and short strings expand most. We translate in the resource format, supply the plural categories the language needs, leave placeholders and markup exactly where they were, and write to a character budget wherever you can give us one. Where a target language cannot say it shorter, we propose different wording instead of letting a control truncate at runtime, and we will tell you which labels we would change in English too. UI strings and software strings in general are where user experience is won or lost, because a product interface that reads as a translation is read as a worse product.

Dense screens are the hard case: tables, chart legends, filter chips, dashboard tiles and anything in a fixed column. For those we want a screenshot or a staging login, because the question is never what the phrase means but whether it fits. Your product terminology is agreed in a glossary first and applied with the same discipline as any technical content, so the interface and the documentation never drift apart. A translation management system of your own is welcome in the loop; if you have not got one, the files move by secure transfer.

Onboarding, Notifications and Transactional Email

Onboarding and transactional messages decide whether a new user ever reaches the product, and they usually live somewhere other than the interface, in a messaging platform, an email template store or a notification service. Send those too. A push notification is truncated by the operating system rather than by your design, a receipt has to carry a currency and a date in the local convention, and a verification email that reads as machine output loses the account. We localise these as copy written for the market rather than as strings matched to the English.

Two things go wrong here reliably. The first is a template assembled from fragments, where the greeting, the body and the sign-off are separate records that no linguist can see together; for those we ask for the rendered output as well as the parts. The second is a variable dropped into the middle of a sentence that changes its grammar, because a name, a plan name or a number behaves differently in German, Polish or Finnish, and the fix is usually a different sentence rather than a better translation.

The Help Centre and Knowledge Base

The help centre normally carries more words than everything else combined and is read at the worst possible moment, so it is the part of the project where agreement with the interface matters most. We work from the export format your platform produces, keep article structure, metadata and internal links intact, and apply the same glossary used on the strings, so that an article and the button it describes use the same noun. Where the knowledge base is versioned with the product, only what changed is translated at each release.

Localising the whole help centre at once is rarely the right first move. A short list of articles, the ones your support queue actually links to plus anything in the signup and billing path, covers most of the volume, and the rest can follow a release later. We price the full set and the short list separately, so the decision is made on a number rather than on a feeling. Document translation of the manuals and agreements that sit outside the help centre is quoted the same way.

Store Listings, Release Notes and In-Product Announcements

A store listing is the only localised text a prospective user reads before installing a mobile app, and it answers a different brief from the interface: it is marketing copy with a character limit and a search function behind it. We localise the title, subtitle, description, keyword field, screenshot captions and the what’s-new text for each market, keeping the product nouns identical to the ones inside the app. Release notes and in-product announcements then follow the same glossary, so a new feature carries one name in the changelog, in the interface and in the campaign that announces it.

Keep announcement copy short in English and it survives translation; write it to fill the English character limit exactly and it will not. This is the one place where we routinely come back with a rewrite rather than a translation, because a localised store listing that reads as a translation of an English listing performs like one. Multilingual content works when each market’s version reads as though it had been written there.

Locale Testing on Device Before the Release Ships

Testing is a separate pass and it belongs in the plan rather than at the end of it. Two things only appear in a running build: whether the text fits, and whether a string means in context what it meant in the file. Give us a staging environment or a build and a list of the screens that matter, and the review comes back as an ordered list of strings to change with a reason for each. A pseudo-localised build, run before any translation is ordered, finds the structural faults earlier still.

Before a single language is commissioned, run the build with its strings swapped for machine-made stand-ins: accented or other-script characters, padded out and bracketed with markers. Nothing has been translated, and the screens already say which labels clip, which containers crop, which glyphs the shipped font has not got, which text never left the source, and which sentences are being glued together at runtime. There is no specification behind the technique. The Unicode glossary has no entry for it, W3C’s internationalization materials carry no article on it, and Microsoft sets it out as a methodology without pointing at any standards body, so treat it as an afternoon’s testing rather than a compliance exercise.

What Does a Localisation Project Deliver, and What Does It Cost?


Each file comes back in the format it was sent in, translated and revised, with its keys, placeholders and escaping unchanged. Every language is written by a linguist working into their own first language with at least five years of experience, and a second linguist then revises the translation against the source. That two-person process is the largest single contributor to translation quality on a software project, it is what ISO 17100:2015 requires of a translation service, we are certified to it and we are an ATC member; what the standard does not do is govern localisation engineering, so it should never be presented as a software localisation standard. Every translator working on your account writes into their first language, and automated quality assurance checks run over the files before anyone sees them. One project manager runs the work end to end.

A glossary and a style note are agreed before anything is translated, and on a first project they are the most valuable thing you get. The glossary pins down the words only you use: what your objects are called, what your pricing tiers are called, the labels in your navigation, the imperatives on your buttons, and the brand terms that deliberately stay in English. The style note records how the product speaks to a user and which languages take a formal register. Together with the translation memory, that is what makes the fourth release read like the first.

Day to day the work runs on standard translation management tooling: a translation memory per account, a term base, and automated checks for an untranslated segment, an altered number, a missing placeholder, a broken tag or a string over its length limit. Machine translation has a place in a pipeline, for triage, for internal strings and for a first pass where you ask for one, and where it is used we say so, because ISO 17100’s own scope excludes raw machine output with post-editing and the two should not be described as the same product. Neural machine translation and the translation technology around it have improved steadily, and neither it nor any other software translation tool decides what ships. Files move under a non-disclosure agreement over secure transfer.

Around the strings we also translate the agreements and policies that come with selling in a new market, certify a translation where something has to be filed, and turn recorded product walkthroughs into text that can be translated and subtitled.

Two units, because the work has two shapes. Strings and documentation are counted in words at £0.10 per word over 2 to 3 working days, with 24-hour express on selected work; a document that has to be certified is counted in pages at £30 per page of up to 250 words, back in 24 to 48 hours. What arrives is broken out by language pair and by file, which is the only way to see whether a market or a component is driving the figure, and the segments your memory already covers appear as their own line rather than disappearing into the total. Nothing is charged for the quote and it is with you inside the hour.

Software localisation prices and delivery
ServicePriceDelivery
Interface strings, resource files and product documentationFrom £0.10 per word2 to 3 working days
Certified translation of agreements, policies and filingsFrom £30 per page of up to 250 words24 to 48 hours
Express work on a release already in flightQuoted per project24 hours for selected projects
QuoteFreeWithin 1 hour

Which Other Language Work Uses the Same Glossary?

One glossary, one memory and the same linguists also cover the public website and store, the campaign that opens a market, and the recorded demos and webinars that have to become text before anybody can translate or subtitle them. Holding all of it in one place is what stops a single feature collecting three different names.

How Does a Localisation Project Run, Release by Release?


Four steps on the first release and three on every one after it, because by then the glossary and the memory are already in place. Send the files and the quote comes back free within 1 hour.

78% PLACE ORDER

Step 1: Send the String Files and Say When the Branch Closes

Upload whatever your toolchain exports, whether that is .xlf, .po, strings.xml, an .xcstrings catalogue, RESX, JSON, YAML or Markdown, with the target languages, a note on where the strings appear, and the date the release branch closes. The developer comments inside those files are the most useful thing in the package: a key with a note saying it is a 14-character button label is translated once, and the same key without one is translated, queried and translated again.

Step 2: A Free Quote in 1 Hour, With the Glossary Agreed First

A project manager separates the genuinely new words from the segments the memory already covers, prices each language pair and each file, and returns an itemised quote within 1 hour at no cost. On a first project the glossary and style note are proposed at the same time, and nothing is translated until you have signed them off, which is also when we raise anything in the English that will not survive translation, such as a reused key or a sentence assembled from parts.

ORIGINAL EN A 文 TRANSLATED AR FR DE ES JA ZH AR CERTIFIED TRANSLATION

Step 3: Translation, Self-Check and Revision Against the Source

The batch goes to a linguist who works into their first language and knows software, and it comes back with placeholders, plural categories and markup exactly where they were; that linguist checks their own output, which ISO 17100 treats as part of translating rather than as review. A second linguist then reads the file against the source. Machine checks run across both passes, hunting for a segment nobody translated, a figure that changed on its own, a placeholder that vanished, or a string that outgrew the limit you set.

Step 4: Return the Files, Test in Context, Then the Next Release

Files come back in the structure they left in and go straight into your repository. Then we test: given a staging environment or a build, we review the screens you nominate and return an ordered list of strings to change with a reason for each, because length and context faults are invisible in a file. From the next release onwards only the changed strings come to us, which is what continuous localisation means in practice and why a later batch costs less than the first.

Why Does an English Interface Still Need Work for the United States?


Because en-GB and en-US are two locales, and a product written for one of them gets things wrong in the other that a reader notices immediately. Dates are the worst of it: 03/04/2026 is April in London and March in Chicago, and the only safe repair is a pattern taken from locale data rather than typed into the code. Then come a 24-hour clock against a 12-hour one with am and pm, a week starting on Monday against one starting on Sunday, postcode against ZIP code, county against state, mobile against cell, paper size on anything printable, and metric against US customary units in anything measured. Spelling is the part everybody remembers and the cheapest part to fix. Dates and currency are mechanical; register, idiom and cultural nuances are not, and they are the reason a target language is written by somebody who lives in it.

The practical approach is to treat en-US as a target locale with its own resource files rather than as a find-and-replace. That means a language tag of its own, number and date formatting drawn from CLDR for that locale, a glossary entry for every term that changes, and a review of forms and address fields, since a form demanding a postcode in a UK pattern will reject a valid US address. It is a small project next to a new language, and for a UK product it is usually the first one worth doing.

Which Problems Belong in the Code Rather Than the Translation?

Three belong to your developers and two to whoever writes the English. In the code: a sentence assembled at runtime out of a figure and two fragments, which cannot be branched in any language; a date, number or amount formatted by hand instead of drawn from locale data; and a control with a hard width and no headroom. In the copy: a key carrying no note about where it appears, which comes back as a noun when it was meant as a verb; and one key doing duty on two screens that mean different things, which means one of the two must read wrongly. Not one of the five is a translation problem, all five arrive as translation complaints, and an hour with a pseudo-localised build finds every one of them before a penny is committed.

Which Other Language Services Sit Alongside Localisation?


A localisation project usually arrives with something attached to it: a contract to sign in the new market, a website going out in the same release, a recording that needs a transcript first. One of these is normally the right next page.

Certified Translation Services

Opening a market brings paperwork with it, from a reseller agreement to a filing for a local entity. Explore our Certified Translation Services, charged per page of up to 250 words and returned in 24 to 48 hours.

Legal Translation Services

A licence agreement, a set of terms or a privacy notice is read by a lawyer rather than a user, and the register has to match. Visit our Legal Translation Services, with our sworn translation page for a destination that asks for one by name.

Medical Translation Services

Anything clinical, from a patient-facing app to care records, turns on the terminology being settled in advance. See our Medical Translation Services.

Technical Translation Services

Hardware manuals, integration guides and engineering documentation behave differently again, because repetition and file format set the figure. Learn more about our Technical Translation Services.

Digital Marketing Translation

A marketing site has to persuade in the market’s own idiom, which is a different brief from an interface label. Visit our Digital Marketing Translation page, with our website translation page for the public pages themselves.

Transcription Services

Anything recorded becomes text first, whether it is a webinar, a product demo or a support call. Explore our Transcription Services, and our subtitling service once the video is cut.

2,000+

Linguists

ISO 17100

Certified quality standard

1 hour

Free quote

150+

Languages Supported

What Else Do Product Teams Ask About Software Localisation?


How is localisation different from translation?

Translation is one task inside localisation, not a shorter name for it. A translator is handed a sentence; a localisation project is handed the file that sentence lives in, the rule deciding its plural form, the width of the box it has to sit in, the locale data that renders the figures inside it, the direction the page runs in, and the help article, release note and store listing that repeat it elsewhere. The difference shows in what changes hands. Ask for translation and you send a document. Ask for localisation and you send .xlf, .po, strings.xml or .xcstrings files, and they come back with every key, placeholder and plural category exactly where it was.

What is internationalisation, and when is it needed?

Before the first language is bought, rather than alongside it. Internationalisation is the engineering that makes a product capable of being localised at all, and it reduces to five properties: no user-facing sentence written into the source; every string addressable by a stable key in a resource file; locales named with BCP 47 tags; plural, gender and ordinal logic expressed through a message format instead of by concatenation; and dates, figures and amounts rendered by the platform out of locale data. Hardly any product arrives with all five, and that is not a reason to wait. We take what has been extracted, translate it, and hand back a list of the keys that need a developer rather than a linguist, so nobody is billed for working around a defect they cannot reach.

How much does software localisation cost?

Words for strings and documentation at £0.10 per word; pages for certified documents at £30 per page of up to 250 words. The rate is the part that does not vary. The total varies on three things: how much of the batch your memory already holds, since a repeated string is not a new word; how many markets open at once, because a glossary built for one language is reused by the next; and whether the strings arrive as files or as screenshots, since rebuilding a resource file by hand costs more than translating it. Urgent handling on a release that has already left the gate is quoted per project. An itemised figure reaches you free, inside the hour.

How long does it take to localise a product?

The scheduling windows are 2 to 3 working days for standard work, 24 hours on selected express work, and 24 to 48 hours for a certified document, and those are facts about our queue rather than promises about your release. Three things sit outside them on a first project: agreeing the glossary and style note, which comes before any translation; your own review of the running build, which comes after delivery; and any string that turns out to need a developer. By the second release the glossary exists and only changed strings arrive, so the cycle usually fits inside a sprint. The date for your own files is settled with the quote, an hour after they land.

Which file formats are supported?

Whatever your toolchain writes, which in practice falls into four groups. Interchange: XLIFF 2.1, the XML Localization Interchange File Format. Platform resources: strings.xml and its values folders on Android, String Catalogues and the older .strings and .stringsdict pair on Apple, RESX on .NET. Open-source tooling: GNU gettext .po, .pot and .mo. Structured data and documentation: JSON, YAML, .properties, Markdown, HTML and CSV. A bilingual table is the fallback where a product will not give its strings up any other way, and compiled or database-held strings need an extraction route agreed at the quote. What we will not ask you for is a retyped spreadsheet, because the keys, the comments and the plural categories are the half that matters.

What is continuous localisation?

It is localisation that runs at the pace of your releases instead of in one block at the end. The pattern is simple: your pipeline exports the new and changed strings when a branch is merged, the batch comes to us, the translated batch returns into the same branch, and the memory and glossary carry forward so nothing already approved is paid for twice. For a SaaS product that ships weekly or fortnightly, this is the only arrangement that works, because a quarterly handover means every release in between goes out partly untranslated. It needs two things from your side: resource files in version control, and someone who can tell us when the branch closes.

Does the European Accessibility Act require us to translate our interface?

No. Directive (EU) 2019/882, the European Accessibility Act, was to be transposed by 28 June 2022 and has applied since 28 June 2025, covering consumer computer hardware and operating systems, consumer terminal equipment with interactive computing capability, e-readers, e-commerce and listed transport services, with a transitional period to 28 June 2030. It sets accessibility requirements, and it imposes no obligation to offer the software interface in any particular language. Its language duties attach to four other things: product instructions and safety information in a language the Member State concerned determines, contact details, documentation supplied to a competent national authority, and the EU declaration of conformity. Selling the Act as a localisation mandate is the commonest error in this market.

Does the B2 plain-language rule in the Accessibility Act apply to our app?

Almost certainly not, and the detail matters because the rule is quoted constantly as though it were general. The Act’s Annex I does contain a plain-language ceiling, requiring information to be understandable without exceeding a level of complexity above level B2, upper intermediate, of the Council of Europe’s Common European Framework of Reference for Languages. It sits in Section IV of that Annex and applies to consumer banking services. It is not a general accessibility rule for software, for websites or for all services. If your product is a consumer banking service it is directly relevant; if it is not, a supplier citing it at you is reading one Annex section as though it governed the whole Directive.

Do you provide localisation testing?

Yes, as a linguistic review of the running build rather than as a QA engineering service. Give us a staging environment or a test build, plus a list of the screens that matter and any character limits, and we review the localised interface in context and return an ordered list of strings to change with the reason for each: truncated, clipped, wrong register, wrong part of speech, plural category missing, placeholder in the wrong position. Two classes of fault only exist on a screen and never in a file, and those are the two this pass is for. Running a pseudo-localised build before any translation is ordered finds the structural problems earlier and more cheaply still.

Do you use AI in localisation?

Not in place of the linguist. Each language is written by somebody working into their first language and revised against the source by a second person, and that is what the price buys. Machine output has honest uses either side of that, sorting a backlog, internal-only strings, a first pass you have asked for, and where one has been used we say which strings it touched, so AI translation is never handed over as though a person had written it. There is a formal reason as well as a commercial one: ISO 17100:2015 puts raw machine output with post-editing outside its own scope, so the two things cannot both be sold as ISO 17100 work. Placeholder, number and untranslated-segment checking is automated on every file, and a global market is not won by shipping into international markets faster than the text can carry.

How much longer does translated text get?

Longest exactly where there is least room. W3C’s internationalization guidance, citing IBM, measures translated length as a proportion of the English, and the proportion falls as the string grows: a label of ten characters or fewer averages two to three times its original length, while anything past seventy characters settles around 130 per cent. Microsoft works to plus 40 per cent in its own pseudo-localisation tooling, with outliers of 200 to 400 per cent, and notes that some translations shrink instead. The example worth carrying around is a measured one from the same W3C article: one English word on a Flickr interface came out 2.6 to 3 times as long in Portuguese, French and Italian. Whatever flat percentage a supplier quotes, buttons break before paragraphs do.

Can you localise right-to-left and Asian-language interfaces?

Yes, inside the 150+ languages we cover: Hebrew, Arabic, Persian and Urdu on the right-to-left side, and Japanese, Korean, Thai and both Simplified and Traditional Chinese among the Asian languages. Each group brings a problem that is not about words. Right to left, the position of any Latin fragment inside the sentence is decided by Unicode Standard Annex #9 at Revision 52, so a screen can be wrong while every string is correct. For the CJK and Thai languages the issues are line breaking without spaces, text that comes out far shorter than the English and leaves a container looking empty, and fonts that have to be shipped rather than assumed. All of it is reported with the delivery, and these languages are worth reviewing on a handset rather than in a file.

Which content should a SaaS company localise first?

The text a user cannot get past, in this order: signup, onboarding and billing screens; the interface strings on the screens most used after that; the transactional email and notifications that arrive outside the product; then the handful of help articles your support queue links to most; then the store listing and release notes. Leave the blog, the long-tail help centre and the marketing site until a market has proved itself. The reason to sequence it this way is that the first group decides whether anyone becomes a user at all, and it is also the smallest word count, so it is the cheapest test of whether a market responds.

Is XLIFF 2.2 the current standard?

No. XLIFF 2.1, approved as an OASIS Standard on 13 February 2018, is still the version to build against. Version 2.2 is real and is drafted to replace it, but on 2 November 2026 it had got as far as Committee Specification, dated 13 March 2025: OASIS publishes an approved standard under a directory of its own, and version 2.2 has none. The committee’s page still names 2.1 as the latest approved version. Both halves of the usual sales pitch are therefore wrong, whether somebody tells you 2.2 is today’s standard or that 2.1 is the newest thing in existence. We default to 2.1 because that is what the tools emit, and we will take a 2.2 file from anybody whose toolchain writes one.

Contact Us


Get in touch with our team for professional translation services. We're here to help you with all your translation needs.

Email
info@espressotranslations.com

Phone
+44 203 488 1841

Espresso Translations Google Reviews rating
Espresso Translations Trustpilot rating
ISO 17100 certified translation services

Request a Quote Now

We’ll get back to you with a competitive quote.