Guide Catalog operations
Multilingual product content: what to translate and what to keep typed.
Translate the words a shopper reads and keep identifiers, numbers and units typed. Write one source language, store standard language tags on every text and review each language on its own sample.
Published by Youzu · Editorial policy
- Published
- Reading time
- 8 min read
Key points
- Keep identifiers, numbers and units typed, and translate only the words a shopper reads.
- Use one source language, written once, so every other language has a single reference.
- Check each language on its own sample instead of trusting an average over all of them.
- Use standard language tags, and store the tag next to every piece of text.
In this guide
What to translate and what to keep typed
A product record mixes 2 kinds of content. Some of it is language: the title, the description, the bullet points and the labels of the options. The rest is data that means the same in every market: the identifier, the brand name, the number of items in a pack and the measured values with their units. Translating the second kind is the usual way multilingual catalogs go wrong.
Keep the typed values typed. A typed attribute (a typed attribute stores a value in a defined type, such as a number with a unit, a yes/no value or a choice from an allowed list) such as a width of 180 centimetres is one number with one unit, and each language only changes how the label and the unit are written for the reader. Store the value once and format it for the language, rather than storing '180 cm' in one language and '1,8 m' in another.
| Part of the record | Translate? | Why |
|---|---|---|
| Title and description | Yes | Shoppers read them in their own language |
| Option labels such as 'Olive' | Yes | The label is a word; the option value stays one code |
| Brand and model names | No | They are names, and shoppers search for them as written |
| Identifier such as a GTIN (a globally unique product identifier of 8, 12, 13 or 14 digits issued under GS1 standards and printed as a barcode) | No | It identifies the item in every market |
| Measured values and units | Format only | The number is shared; the display format follows the language |
One source language, written once
Choose one language as the reference for each product and write it carefully. Every other language is then a rendering of that reference, and a reviewer who reads 2 languages can compare them. Without a reference, 3 languages drift apart: one title mentions the material, another the use, and a buyer in each market is looking at a different product.
Write the source text for people who will not see the photo. Name the product type, the main distinguishing facts and the values that matter, in plain sentences. Short, concrete source text translates better than marketing phrasing, because idioms and wordplay rarely survive the move to another language.[3]
- Say what the product is in the first words of the title.
- Put distinguishing facts such as color, size and material in fields, not only in the sentence.
- Avoid abbreviations that only one market reads, and spell units the same way each time.
Language tags, scripts and direction
Store a language tag with every piece of text. The tags in common use follow BCP 47: a primary language such as 'de', optionally a script and a region, as in 'de-AT'.[1] A tag tells a storefront which text to show, tells a screen reader how to pronounce it and tells a search engine which audience the page is for. A field that holds 'German' as a word, or a flag, cannot do any of that.
Use the shortest tag that is accurate. A region matters when spelling, units or legal wording differ between regions, and it should not be added just to look specific. Scripts and direction matter for languages such as Arabic, where the whole layout of a product card runs right to left; make sure the text field, the card and the number formatting all follow the language rather than the page's default.[2]
Format numbers, dates and lists with the rules of the language, not by hand. Shared locale data describes decimal separators, digit groups and the order of day and month for each language, so a record can hold one value and a storefront can display it correctly in every market.[2]
A worked example: one jacket in 2 languages
The olive field jacket in size M has one identifier, one color code, one size code and one measured value for the sleeve length. Its title and description exist in English and in German. The table shows the same record in both languages: the words change, and the typed values do not.
Worked example
One jacket record in English and German
| Field | English (en) | German (de) |
|---|---|---|
| Title | Olive field jacket | Olivgrüne Feldjacke |
| Color code | olive | olive |
| Size code | M | M |
| Sleeve length | 64 cm | 64 cm |
| Identifier | 0000000000000 | 0000000000000 |
Read the 2 columns as a reviewer would. The identifier, the size code and the sleeve length match exactly, so they are safe. The titles differ, as they should, and the reviewer checks that each says the same thing: a product type, a color and a use. If the German title dropped the color or added a feature that the English one lacks, the 2 languages describe different products.
Search, filters and fallbacks
Shoppers search in the language they read, so a storefront needs text in that language to match their words. When a product has no text in a language yet, decide what to show instead: the source language with a clear marker, or nothing. A silent mix of languages on one product page confuses buyers and weakens the match with their search.
Facets and filters should work from the typed values, not from the translated labels. If 'Olive' is stored once as a color code and displayed in each language, one filter serves every market and a change to a label does not break it. If the filter is built from the words themselves, each language needs its own filter and its own maintenance.
Plan the fallback chain on paper: the requested language, then a related one if that makes sense for your market, then the source language. Test it with a product that is missing a translation, since that is the case that shows what shoppers really see.
Review and measure each language apart
Translation quality is not one number. A method that does well in languages with abundant text can do badly in a language with less, so report each language on its own sample, with your own labels and your own reviewers. A figure averaged across languages hides the one market where shoppers see mistakes.
- Draw a sample of about 100 products for each language, including the awkward categories.
- Have a reader of that language mark each title and description as correct, needs edit or wrong.
- Count critical mistakes apart: a wrong size, a missing warning or a claim the product does not support.
- Repeat after any change to the source text, the method or the category mapping.
Keep the review cheap enough to repeat. A short sheet per language, with the same columns each time, shows whether quality is moving and where. When a language has a high share of edits, fix the source text first: unclear source text is the most common cause, and it costs the least to repair.
Common mistakes
- Translating the brand name or the model name, so a shopper's search for the product as written finds nothing.
- Storing units inside translated text, so a filter by width works in one language and not in another.
- Using a country flag as a language marker, which says nothing about the text that follows.
- Letting each market write its own source text, so no language is the reference for the others.
- Judging all languages by one overall score.
Pair this guide with the one on product taxonomy mapping, since category names are among the labels that need a careful rendering in each language, and with the catalog glossary for the terms used here.
Boundaries
What Youzu does not do
- Youzu publishes no accuracy figures. Measure accuracy on your own sample against your own labels, with critical errors reported separately.
- Youzu does not claim to work on every category out of the box. Supported categories are scoped with you.
- Youzu does not replace your ERP, search or storefront; those systems stay authoritative for their own data. It can replace your PIM, or work alongside it.
Checklist
Multilingual product content checklist
10 checks across source text, typed values, language tags and per-language review, with an owner column and an empty status column.
A CSV file of 10 rows, named youzu-multilingual-product-content-checklist.csv. No form to fill in first.
| Area | Check | Owner | Status |
|---|---|---|---|
| Source | One source language is named for each product | Catalog operations | |
| Source | Source text names the product type first | Content editors | |
| Typed values | Identifiers and measured values are not translated | Catalog operations | |
| Typed values | Units are stored apart from the text | Catalog operations | |
| Tags | Every text carries a BCP 47 language tag | Engineering | |
| Tags | Right-to-left text is checked on the product card | Front-end | |
| Review | Each language has its own sample and reader | Language reviewers | |
| Review | Critical mistakes are counted on their own line | Language reviewers | |
| Review | The review repeats after a change to the source text | Catalog operations | |
| Brand | Brand and model names are kept as written | Brand owner |
FAQ
Questions buyers ask
Short answers to the questions teams ask about this topic.
Titles and descriptions can be generated and localized; a catalog supports 1 to 5 of 24 language codes, and English is required when AI processing is on. Check the documentation for the current list of codes before you plan your markets.
English is required when AI processing is on, and a catalog supports 1 to 5 of 24 language codes. If your own source language is another one, plan for English text to exist alongside it, and check the documentation for the details of the language codes.
No. Youzu maps products to your category tree and extracts the attributes your category defines; numbers and yes/no values stay typed, and values the evidence does not support are left empty. A typed number keeps its value across languages, and only its display format follows the language.
Yes. Quality differs between languages, so each one needs its own sample and a reader who knows it. Count critical mistakes apart, such as a wrong size or a missing warning, and keep the sheet the same each time so that you can see whether quality is improving.
Translation moves the words into another language. Localization also adapts what depends on the market: units, date and number formats, spelling, and sometimes the example or the legal wording. A product record usually needs both, with the typed values shared and the text and the display format adapted.
Last reviewed
Sources
RFC 5646: Tags for Identifying Languages (BCP 47), IETF. rfc-editor.org/rfc/rfc5646. Accessed . Back
Unicode Common Locale Data Repository (CLDR), Unicode Consortium. cldr.unicode.org. Accessed . Back
Choosing a Language Tag, W3C Internationalization. w3.org/International/questions/qa-choosing-language-tags. Accessed . Back
About this guide
Published by Youzu · Editorial policy
- Published
Every guide defines its terms, works through one example and cites the public sources it relies on, with the date each source was checked.
Guides explain methods. They do not rank Youzu above other approaches, and an example is never presented as a customer result.
More in all guides and the catalog glossary.
Related
Keep reading
Product
AI product data enrichment from the text and images you already have
Youzu fills the attributes your categories define from text and images, localizes titles and descriptions and leaves a field empty when evidence is thin.
Glossary
What is a typed attribute?
Typed attribute defined in plain words, with a worked example, common mistakes and links to the guides and Youzu products that apply it.
Guide
Product taxonomy mapping: how to do it and how to measure it
How to map seller and supplier categories to your own taxonomy: standards, rule-based and AI methods, edge cases and how to measure accuracy.