Skip to content
Case study · Technical capabilities

Localization engineering: the technology behind every translation

Before a single word is translated, someone has to open the file, protect what must not change and prepare it for linguists. When no tool can do that, our engineers build one. This is how our technical team works, told through the projects that tested it.

.numbers JSON SRT IDML XLIFF </> QA CHECKS Tags Numbers Terms Length RTL
Team
Localization engineers, linguists and testers
Formats
Office, design, web, software, subtitles and more
Featured project
A custom tool for Apple Numbers files
Standards
ISO 17100, ISO 18587, ISO 9001, Cyber Essentials
The idea

The file is part of the translation

Most translation problems do not start with words. They start with files: formats that translation tools cannot open, text mixed with code or formulas, layouts that break when a sentence gets longer, and scripts that change direction. Handle the file badly and even a perfect translation arrives broken.

That is why every project at Cube passes through a technical stage before and after translation. Our engineers analyze the files, prepare them for linguists, protect everything that must not change, and rebuild the result in its original format.

  • Any format in, the same format out
  • Code, formulas and numbers protected
  • Automated checks on every file
  • Tested where people will read it
A perfect translation in a broken file is still a broken file.Cube Localization engineering

What our technical team does

Nine capabilities that sit behind our linguists on every project.

Custom file tools

Parsers and converters for formats no CAT tool supports, built in-house when a project needs them.

CAT tools and memory

Translation memories and termbases for every client, in all major CAT tools.

Automated QA

Checks for tags, variables, numbers, terminology and length limits before any delivery.

Machine translation

Engine selection and post-editing workflows that follow ISO 18587.

OCR and DTP

Scanned pages rebuilt as editable files, and layouts rebuilt in the target language, including right to left.

Subtitle and audio engineering

Timing, subtitle formats, burned-in subtitles and voice mixing for any video format.

Software and game localization

String extraction, placeholders, plural forms and linguistic testing inside the build.

Continuous localization

Connections to repositories and content systems, so new strings reach linguists automatically.

Data security

Access limited to the project team, and systems certified under Cyber Essentials.

Featured project

The spreadsheets no translation tool could open

A client sent us a large set of workbooks in Apple Numbers format: many sheets and tables mixing text with figures and formulas. Professional translation tools reject these files, and converting them to Excel breaks formulas, number formats and styles.

Inside, a Numbers file is a package of IWA archives, Snappy-compressed Protocol Buffers, with cell text kept in shared string tables, and there is no public specification. So our engineers built a dedicated program: it maps every text cell to its exact address, sends it to translators with its context, locks formulas and numbers, and writes the translation back into a new Numbers file that opens exactly like the original.

  • Every text cell extracted with its address
  • Context sent with each string
  • Formulas and numbers never exported
  • Rebuilt, then checked on a Mac
data.numbers data.numbers </> SourceOur toolTranslated

From file to delivery: our engineering pipeline

The same six stages protect every project, whatever the format.

  1. 01

    File analysis

    We check every file: format, encoding, what is text, what is code, and what could break.

  2. 02

    Preparation

    Text is extracted into a standard bilingual format, with code, tags and formulas locked.

  3. 03

    Context and references

    Screenshots, glossaries and notes are attached, so linguists see where each string lives.

  4. 04

    Translation environment

    Linguists work with translation memory, termbase and real-time checks.

  5. 05

    Automated QA

    Scripts check tags, variables, numbers, terminology and length limits.

  6. 06

    Rebuild and test

    Files are rebuilt in their original format and checked in context: in the document, the app or the game.

TagsTags and placeholders intact
NumbersFigures match the source
TermsGlossary applied everywhere
LengthWithin interface limits
ConsistencySame source, same translation
UntranslatedNo text left behind
Right to left

Arabic that displays correctly, not just translates correctly

Arabic is where many products break: letters must join, text runs right to left while numbers and Latin names run left to right, punctuation drifts to the wrong side, and a counted noun can take six forms depending on the number.

As an Arabic-first team, we test for all of it: mirrored layouts, fonts that support Arabic shaping, mixed-direction text with Latin words and numbers, and plural variants for every counted string.

ShapingLetters joined, never isolated
DirectionLayouts mirrored for right to left
Mixed textLatin words and numbers in place
PluralsSix forms, one per number range
FontsArabic-ready fonts at every size
Toolbox

We work in the tools your content lives in

Our team works in the environments your content already uses, from translation tools and content systems to video editors, so nothing has to be converted just for us.

For subtitles, that includes Aegisub, Subtitle Edit, Subtitle Workshop, EZTitles, CaptionMaker, Adobe Premiere Pro, Final Cut Pro, DaVinci Resolve and YouTube Studio. For documents and software, every common format, from Word and InDesign to JSON, XLIFF and PO.

.numbersXLSXDOCXIDMLPDFJSONXMLXLIFFPORESXSRTVTTASSTTMLSCC

Our engineering principles

Five rules that decide how we handle every file.

PrincipleIn practice
FormatStay in the original formatWe build a tool before we accept a lossy conversion
ProtectionTranslators never touch codeFormulas, tags and variables are locked or hidden
ChecksAutomate what machines do betterScripts check numbers, tags and terms; people check meaning
ContextTest where people will read itDocuments, apps and games are reviewed on screen
ReuseBuild once, use againEvery tool stays in our toolkit for the next project

What clients get

Any file, the same file back

Deliveries in the original format, ready to publish, print or ship.

Errors caught early

Automated checks stop broken tags, numbers and terms before delivery.

Faster projects

Reusable tools and translation memory remove repeated work.

Arabic that works

Right-to-left text that displays correctly in documents, apps and games.

What we learned

The lessons we now apply to every project of this kind.

01

The file is part of the translation

Quality is decided as much by how a file is handled as by the words inside it.

02

Build tools, don't work around them

A tool built once is safer than a manual workaround repeated a thousand times.

03

Automate checks, not judgement

Machines are perfect at counting tags; people are needed to judge meaning.