XML Translator: Translate XML Files With Tags, Attributes and CDATA Intact
An XML translator translates the human-readable text between the tags while preserving the markup itself — so the file that comes back is well-formed XML your system can ingest, in any of 280+ languages.
You tell Smartcat's Software Localization Agent which elements and attributes hold translatable text — by XPath, once — and everything outside those paths is left alone.
We quickly saw that Smartcat offered exactly what we needed: a translation memory, a central hub for our workflow, and no more endless emails or file juggling. Everything—and everyone—now works together in one place.
”Explore Case Study →
How Does Smartcat Translate an XML File?
XML translation means translating an XML file's text content while leaving its markup machine-readable. Smartcat's Software Localization Agent uses a configurable XML parser rather than guessing.
- Point at what counts as text — you supply an XPath to the segment-forming elements, the nodes you want to appear in the translation editor, and those become your translatable segments.
- Decide attribute by attribute — a segment-forming node can be an element, an element's attribute, or text, so an attribute that genuinely holds copy (label, title, alt) is translated only because you pointed at it, while id and enum values are never touched.
- Translate and write back — each segment is translated with your glossary and translation memory applied, then written back into the original structure.
One structural requirement: the file must contain repeating elements, since each repetition becomes a source segment.
Set the Rules Once, Then Download Well-Formed XML
Upload one .xml file or several at once, including recurring exports.
- Set the rules once — target languages, glossary, and the XPaths that define what is translatable, untranslatable, an ID, a comment or length-capped. The same rules apply to every future file of that shape.
- Review in context — open the split-screen editor if a reviewer should see strings in context, and assign team members or vetted reviewers from the Smartcat Marketplace.
- Export whole projects in one click — select them, click Download, and everything comes back as a single ZIP in the original formats, sorted into folders.
Why Does XML Break in Ordinary Translators?
XML carries meaning in places a general-purpose translator cannot tell apart from prose. Four things break:
- Tags and attributes get translated — <button label="Submit"> comes back as <bouton étiquette="Envoyer">, and your parser has never heard of <bouton>.
- Entities get "fixed" — & becomes a bare ampersand and the file stops parsing.
- Encoding declarations drift — the declaration says one thing while the content says another, and the mismatch surfaces as mojibake three systems downstream.
- CDATA gets mangled — a <![CDATA[ … ]]> block holding embedded HTML is exactly the thing naive tools mangle first.
The failure mode is not a bad translation — it is a file your import rejects at 5 p.m. on release day.
On entity and encoding behavior we would rather test a sample file with you than print a promise. The parser configuration handles the two that are configurable — tag and attribute translation, and CDATA.
Rules You Set Once, by XPath
Point the parser at the segment-forming elements in your file shape, plus separate paths for untranslatable nodes, string IDs, comments and per-segment length limits.
Configure your product feed, CMS dump or help-content export once — every future file of that shape follows the same rules.
Update Segments by ID Without Cancelling Assignments
XML is one of only two format families — with PO/Serge — where a newer version can update segments by ID without cancelling task assignments.
For every other format, replacing a document recovers translations from the translation memory but cancels the tasks, so linguists have to be reassigned. Both updating and re-uploading consume Smartwords for retranslation.
103% Key-ID Translation Memory Matching
String IDs are mapped by their own path and used for 103% key-ID matching, where standard document formats cap at 102% context matching.
Comments can be mapped the same way and surfaced as segment context for reviewers.
280+ Languages, Well-Formed Output
Structure is never edited, so what comes back parses. You choose the target languages; the markup stays as authored.
What's Preserved, Element by Element
Element names and attribute names are never translated. Everything else is opt-in by design:
Attribute values — not translated unless you name that attribute as segment-forming, so you point at label and never at id or an enum value.
Text nodes — translated where they are segment-forming, and checked against your translation memory and glossary. Untranslatable inline nodes can be excluded by their own path.
CDATA sections — configurable. With Protect HTML tags in CDATA sections on, HTML inside CDATA is substituted by Smartcat tags; with it off, those tags appear as editable text.
String IDs and comments — each mapped by their own path.
Length limits — settable per segment where a downstream schema caps a field.
Decide the CDATA checkbox deliberately, because off means a translator can edit your markup.
The one structural requirement is repeating elements: each repetition becomes a segment, so a one-off document with no repeating structure is not a fit for the custom parser.
See the XML Parser Configured on Your Own File
Bring a sample export to a 1:1 with our team.
- XPaths set live on your file shape.
- CDATA handling decided.
- Length limits mapped to your schema.
- Entity and encoding behavior tested on your own file rather than promised in a table.
Find Expert Reviewers to Edit and Train Your AI Agents
The Smartcat Marketplace connects you with experienced reviewers in your industry or target market. Use our AI-powered matching or explore profiles by specialization and ratings.
XML, XLIFF, JSON and YAML in the Same Project
XLIFF and TMX are XML too, and both are supported. XLIFF has its own workflow and its own page, the XLIFF translation editor — and note that Smartcat imports XLIFF 1.2 and 2.0 while export is 1.2 only.
Also localizing JSON or YAML? Same project, same translation memory — linguistic assets attach at project level and apply to every document in it, with one writable TM per language pair.
If you drive this from CI, the API reaches the same 50+ formats the platform does:
XML, Android XML, RESX, LOCJSON, JSON, TJSON, YML, INC, INX, STRINGS and PROPERTIES among them.
The software-localization API additionally exports json, structured-json, yaml, structured-yaml, ios-strings and android-xml.
Executables and scripts are blocked for security.
Send the Whole Export, Not a Sample
There is no cap on how many files go into one project, so a full locale export uploads as a single job sharing one glossary and one translation memory.
Splitting is only ever about speed on a very large single file, never about how many files you send.
Re-Export Every Sprint Without Losing the Work
If you ship the same XML export every release, the difference that matters is what happens on the second upload.
Translations carry over and linguists stay assigned instead of being re-invited each sprint.
Creation, translation, review and delivery stay in one platform, with repetitive steps automated and Smartwords consumed only for retranslation.
Send Well-Formed XML to Every Market
Text translated, markup untouched, your own rules about what counts as text — in 280+ languages, with AI agents doing the parse-level work.
Free for 15 days with 15,000 Smartwords and full access to translation capabilities, no credit card.
Frequently Asked Questions
What is Smartcat?
Smartcat is an AI-powered global content platform built for businesses that need to scale multilingual content across markets.
It unifies AI agents, translation memories and human collaboration into a single system, helping teams translate XML files and other content faster and more accurately.
What is an XML file?
An XML (Extensible Markup Language) file is a flexible, text-based format that organizes and stores data using custom tags.
It is designed for both humans and machines to read, which makes it useful for structuring everything from app settings to training materials.
Where are XML files commonly used?
XML files are everywhere — from app development and website localization to content management systems (CMS) and eLearning platforms.
They are the workhorse for structuring content and exchanging data across software, platforms and systems.
Why are XML files important, and who uses them?
XML keeps complex data organized and portable, which is why it shows up in app configuration, CMS exports, product feeds, eLearning packages and SCORM content.
The teams translating it are usually engineering, product content and localization teams, working from a file that some other system generates on a schedule.
What are AI Agents, and how do they support XML translation?
AI agents are task-specific, workflow-aware automation tools.
Smartcat's agents, like the Software Localization Agent, are built to handle structured formats like XML — preserving tag integrity and applying your glossaries, so localization scales without developer support.
They also learn from human feedback, which drives productivity and quality improvements even in complex enterprise workflows.[1]
Can I control which attributes get translated?
Yes, and control is the mechanism rather than a setting bolted on top. In the XML parsing configuration you give an XPath for the segment-forming elements, and a segment-forming node can be an element, an attribute or text.
An attribute is translated only if you point at it — so label becomes a segment while id and enum values never enter the editor. You can also give separate paths for untranslatable nodes, string IDs, comments and length limits.
How do I translate XML files while preserving tags?
Upload the file, then define what counts as text.
Structure is never edited by the parse model: element and attribute names are left alone, and only the nodes your XPath points at become editable segments.
The translated text is written back into the original structure, so the download is well-formed XML with the same shape as the file you sent.
Can Smartcat translate Android strings.xml files?
Yes. Android XML is a supported format, and in a software-localization project you can import and export strings with their key IDs — the same as for JSON, YAML and iOS .strings — so <string name="checkout_cta"> keeps its name and R.string.checkout_cta resolves in every locale.
Plural keys are supported at project level: a key can be created with all its plural forms, and plurals export as ICU MessageFormat-compatible strings.
What XML formats does Smartcat support?
Standard XML, Android XML, XLIFF and TMX, plus custom schemas through the XML parsing configuration.
XLIFF import accepts 1.2 and 2.0 while export is 1.2 only, and XLIFF has its own workflow on the XLIFF translation editor page.
For a custom schema, the parser is configured once per file shape rather than guessed at — and it needs repeating elements to segment on.
What happens to CDATA sections with HTML inside?
You decide, with a checkbox: Protect HTML tags in CDATA sections. Checked, the HTML tags inside the CDATA block are substituted by Smartcat tags, so translators can move them but not break them. Unchecked, those tags appear as an editable piece of text. For anything machine-consumed, leave it checked.
Can Smartcat convert JSON to XML?
No — it is a translator, not a format converter. It translates JSON and XML each in their own format; see the JSON translator for locale files.
Can I automate recurring XML translations?
Yes — see “Update Segments by ID Without Cancelling Assignments” above for what happens on the second upload.
For pipelines, the Smartcat API uploads files and retrieves translations, works asynchronously with notification callbacks so a build step does not sit and poll, and has a CLI for CI/CD. Three conditions up front:
- API access is included on the Anticipate and Autonomous plans only.
- It needs a company account rather than a freelancer account.
- It is not available during the free trial.
Every method shares a limit of 4 requests per second, and the docs are public at developers.smartcat.com.
Does Smartcat offer version control for XML translations?
Yes — every edit and revision is tracked in the Smartcat Editor. You can revert to earlier versions or compare updates across language sets, which keeps localization consistent across releases.
Is my XML content secure on Smartcat?
Smartcat is SOC 2 Type II compliant. Files are encrypted in transit and at rest, workspaces are isolated per account, and access is role-based. Details are on the security page.
Will the output validate against my DTD or XSD?
Well-formedness comes from the parse model — structure is never edited.
Schema validity is partly a property of content, and translation changes content: a field your XSD limits to 40 characters can overflow, because German routinely runs longer than English.
There is a real control for exactly that — the parser configuration lets you set length limits on segments, so translators see the cap while they work instead of after the import fails. Run your own schema validation on the first files per language pair.
How much does it cost to translate XML resource files?
Translation is metered in Smartwords, with repetitions discounted, and pricing scales with volume rather than with file format.
Every plan starts free for 15 days with 15,000 Smartwords and full access to translation capabilities, no credit card.
Not covered here? Book a demo — a 1:1 with a localization engineer, no commitment.
Can I use Smartcat as an XML translator online?
Yes. Upload and translate XML files directly from your browser with Smartcat's Software Localization Agent — no downloads, no manual setup.
How many languages does Smartcat support for XML content?
Over 280 languages, including dialects and right-to-left scripts.
How big can an XML file be, and how many can I send?
The published ceiling is global rather than XML-specific:
- 6 GB per document or project file.
- Guidance is to split anything over 1 GB into smaller parts so it processes faster.
- No segment-count limit is documented, and there is no cap at all on file count — batch uploads of hundreds of files into one project are supported, and that does not vary by plan.
The constraint is per-file size, not how many files you send.
Sources
MIT News. (2024). MIT researchers develop efficiency training for more reliable AI agents. news.mit.edu