When we talk about "quality translation", we usually mean text that sounds natural in the target language. But for e-shop texts, quality is much more than just fluency. It is the accuracy of numbers, preservation of HTML formatting, correct terminology, and consistency across thousands of texts. That is why every translation in the NOBAMO pipeline goes through 12 automatic checks before it is published.
1. Format check
The translation must preserve all formatting elements of the original. If the source text contains bold, italics, bullet points, or numbered lists, the translation must contain them too. This check compares the structure of source and target text and reports any difference.
Example: If the original has 3 bullet points, the translation must also have 3 bullet points — not 2, not 4.
2. Terminology check
Every e-shop has its own terminology. Words like "cart", "checkout", "shipping", "returns" must always be translated the same way. The pipeline uses a terminology dictionary specific to the given e-shop and language. If the translation uses a different term than what is in the dictionary, the check flags it.
Example: If the dictionary defines "Add to cart" = "Pridat do kosika", the translation "Vlozit do kosika" will be flagged for review.
3. Consistency check
Consistency means that the same source text is always translated the same way. If the sentence "Free shipping on orders over 50 EUR" appears in 10 places, all 10 translations must be identical. The check compares translations of the same texts and reports deviations.
4. Number and unit check
Numbers in the translation must match numbers in the original. If the original says "38,000 products", the translation must not say "3,800" or "380,000". The check extracts all numbers from both texts and compares them. It also checks currencies (EUR, CZK), measurements (cm, kg), and percentages.
Example: "Free shipping over 50 EUR" — the number 50 and currency EUR must be preserved in the translation.
5. HTML tag check
E-shop texts often contain HTML tags — links, bold text, bullet points, images. The translation must preserve all tags in the correct order without mismatches. This check parses HTML in both texts and compares the tag structure. A missing closing tag or crossed tags can break the layout of an entire page.
6. Length check
A translation should not be significantly shorter or longer than the original. A very short translation may mean missing content. A very long translation may overflow from UI elements. The check compares relative lengths and reports deviations greater than the set threshold (typically ±40%).
7. Placeholder check
Many e-shop texts contain placeholders — variables that are replaced with dynamic values. For example: "Your order #{ORDER_ID} has been shipped". The placeholder {ORDER_ID} must be preserved exactly in the translation — without translation, without case changes, without spaces. The check extracts all placeholders and compares them.
8. Grammar check
The basic grammar check catches the most common errors — subject-verb agreement, correct cases, diacritics (if relevant for the given language), punctuation. This check uses language-specific rules and continuously improves based on feedback.
9. Context analysis
Some translations are grammatically correct but contextually wrong. For example, translating "bank" as a financial institution instead of a river bank in a nature text. The context analysis compares the translation with surrounding texts and reports suspicious inconsistencies.
For an e-shop, this is especially important with homonyms — words with multiple meanings that often appear in different contexts.
10. Duplicate check
Duplicate translations are not always an error — many UI texts are identical across pages. But when the same source text is translated in two different ways, that is a problem. The check identifies duplicate sources with different translations and flags them.
11. Tone check
E-shop texts have a specific tone — friendly, professional, concise. The translation must maintain the same tone as the original. The check analyzes sentiment and formality of the translation and compares it with the original. A formal German translation of an informal English text is just as problematic as the reverse situation.
12. Completeness check
The final check verifies that the translation covers all content of the original. No sentence was skipped, no paragraph was omitted. This is especially important for long texts (product descriptions, blog articles, legal texts), where it is easy to skip part of the content.
How the checks work in practice
All checks run automatically after each phase of the translation pipeline. If a translation does not pass any check, it is returned for correction. The correction can be automatic (e.g., adding a missing HTML tag) or manual (e.g., contextual correction).
Each check has a configurable strictness level. For e-shop product descriptions, we use the highest level — every number, every HTML tag, every placeholder must be exact. For blog articles and marketing texts, the level is slightly more relaxed, because fluency matters more than formal precision.
The checks are hierarchical — formal checks (HTML tags, numbers, placeholders) run first, as they are unambiguously correct or incorrect. Then semantic checks (terminology, consistency, context) that require deeper understanding. Finally, stylistic checks (tone, length, completeness) that evaluate overall quality.
The result is a system where:
- Error rate dropped below 0.3% (compared to 4-5% without checks).
- Terminology consistency reaches 99.7%.
- Proofreading time decreased by 85%.
- No translation with a missing HTML tag ever reaches production.
- Numbers in translations are always accurate — no "38,000" translated as "3,800".
Why exactly 12 checks
The number 12 is not arbitrary. During development, we tested various combinations of checks on a sample of 50,000 translations. We found that 12 checks cover 99.4% of all error types that occur in translations. Adding more checks would increase coverage only marginally (by 0.2%), but would significantly slow down the pipeline.
Each check was calibrated on real data from IZMAEL.eu operations. For example, the length check — the ±40% threshold was established based on analysis of 200,000 translations into 20 languages. German translations are on average 25% longer than English, Hungarian 15% shorter. These differences are normal and the check does not flag them. But when a translation is 50% shorter, it almost certainly means missing content.
Conclusion
12 automatic checks are not a luxury — they are the foundation of a quality translation process. Without them, quality is random and unpredictable. With them, you know exactly what you will get. For an e-shop where every text directly affects conversions and customer trust, this is an investment that pays back immediately.
Every translation error is a potentially lost conversion. An incorrect number in a price, a missing HTML tag that breaks the layout, inconsistent terminology that confuses the customer — all these problems are caught by our checks before the text reaches the customer.
Want to see how the NOBAMO pipeline works on your texts? Learn more about our translation services or contact us for a free demo.