If you are searching for how to translate NAICS codes, how to translate NACE codes, how to translate ISIC industry classifications, or how to move economic-activity data from one language into another without changing the classification itself, the first rule is simple: the code is reference data, while the wording is language. Translate the description and explanatory notes; do not invent a new code because the target-language industry name sounds different.
This matters in economic statistics, company databases, market research, business registration, investment analysis, labour-market data, procurement, tax administration, policy research, trade analysis and international reporting. A translation can be grammatically excellent and still be wrong if it maps a company to the wrong industry, confuses an activity with a product, mixes classification revisions, or treats NAICS, NACE and ISIC as interchangeable code systems.
This guide explains how to translate industry classifications safely across NAICS, NACE and ISIC. It focuses on preserving hierarchy, revision, activity scope and code identity while making the human-readable descriptions understandable in the target language. It also shows how to build a translation workflow that can connect to broader technical-translation practice without turning this page into another general translation hub.
1. Industry codes classify economic activity, not words
Industry classifications are designed to group establishments or economic activities according to defined rules. The description attached to a code is important, but the code’s meaning comes from the classification system and its explanatory notes, not from the surface wording alone. Two descriptions that look similar in English can belong to different categories because the underlying activity test differs.
Translation control: preserve the source code and revision first, then translate the official description. Quality assurance: verify the target wording against the scope and exclusions of the source category, not merely against a bilingual dictionary.
2. NAICS is a hierarchical North American classification
The North American Industry Classification System uses a hierarchical coded structure to distinguish sectors, subsectors, industry groups, industries and national industries. The United States Census Bureau currently works with the 2022 NAICS edition across its economic programs, with transition timing varying by program. A translator should therefore record the edition when a code is operational or statistical.
Translation control: keep the numeric code unchanged and translate the title and explanatory text. Quality assurance: do not assume a familiar older NAICS description still belongs to the same code in a newer revision.
3. NACE is the European classification of economic activities
NACE is the European Union classification of economic activities. NACE Rev. 2.1 is now the current revision used for the new generation of European statistics, with Eurostat providing structure, explanatory notes and correspondence material. Its labels may be published in multiple official languages, but the code structure remains a governed statistical reference.
Translation control: use official multilingual terminology where available instead of creating ad hoc equivalents. Quality assurance: record whether the source uses NACE Rev. 2 or NACE Rev. 2.1 before comparing codes.
4. ISIC is the United Nations international reference classification
The International Standard Industrial Classification of All Economic Activities is maintained through the United Nations statistical system and is used internationally for economic and social statistics. ISIC Revision 5 was endorsed by the United Nations Statistical Commission in 2023 to reflect changes in the world economy and improve alignment with regional classifications.
Translation control: preserve ISIC code identity and revision. Quality assurance: use the UN structure and explanatory notes when a translated label is ambiguous or when a national classification maps imperfectly to ISIC.
5. NAICS, NACE and ISIC are related but not interchangeable
These systems share the broad goal of classifying economic activity, but they are developed for different statistical communities and can differ in detail, hierarchy and coding. A translation project must not replace a NAICS code with a NACE or ISIC code merely because a translated industry label looks equivalent. That operation is classification mapping, not translation.
Translation control: keep the source namespace explicit. Quality assurance: if a crosswalk is required, use an official correspondence table or a documented analytical mapping rather than a language-model guess.
6. A code without its revision can be incomplete evidence
Industry classifications evolve. New activities appear, old categories split or merge, and wording can change. A bare code copied into a multilingual report may therefore be ambiguous if the revision is not recorded. This is especially important for longitudinal statistics, company-history datasets and regulatory archives.
Translation control: carry edition or revision metadata alongside the code wherever the source system supports it. Quality assurance: never modernize a historical code silently just because a newer revision exists.
7. Hierarchy must survive translation
Industry classifications are useful because broad sectors contain narrower categories. If translated labels are exported as a flat list without parent-child relationships, users can mistake a narrow industry for a broad sector or vice versa. The translation should preserve the structural depth as well as the words.
Translation control: maintain parent IDs, level numbers or indentation metadata. Quality assurance: reconstruct a branch from sector to detailed industry and confirm that every translated node has the same parent as the source.
8. Code length often signals level, but do not rely on appearance alone
Many classification systems use progressively longer codes for more detailed levels. That visual pattern is helpful, but it should not replace the official schema because different systems format levels differently. A translation that labels every short code as “sector” can be wrong outside the expected namespace.
Translation control: derive level names from the classification metadata. Quality assurance: validate code length and hierarchy against the correct system rather than a generic rule.
9. The classification target is usually the primary activity
Businesses often perform several activities. Statistical systems generally classify an establishment or enterprise according to defined rules about its principal or primary economic activity. A company name may emphasize one service while most economic activity occurs elsewhere. Translators should not infer the code from branding language alone.
Translation control: translate the registered activity description faithfully. Quality assurance: preserve the assigned code unless the project explicitly includes reclassification by a qualified data owner.
10. Establishment and enterprise are not always the same statistical unit
Economic datasets can classify establishments, enterprises, local units or other statistical units. A multinational company can contain establishments performing different activities. Translating every record as “company industry” can hide the unit of observation and distort interpretation.
Translation control: retain the source term for the statistical unit at the correct level of specificity. Quality assurance: check whether the code belongs to the enterprise as a whole or to an establishment or local unit.
11. Industry is not occupation
An industry classification describes what the establishment or economic unit does. An occupation classification describes what a worker does. A software engineer may work in a bank, hospital, university or software company. Translating an occupation label into an industry code is therefore a category error.
Translation control: keep industry and occupation fields separate. Quality assurance: inspect datasets that contain both and make sure no translated column heading causes one to overwrite the other.
12. Industry is not product classification
Economic classifications may be linked to product or commodity classifications, but they answer different questions. An industry code classifies the activity of the producer; a product classification categorizes goods or services. A factory can make several products while remaining classified under one principal industry.
Translation control: preserve the distinction between activity and output. Quality assurance: do not map product descriptions directly to NAICS, NACE or ISIC without a documented classification method.
13. Activity descriptions need verb-level precision
Industry titles often distinguish manufacturing, wholesale, retail, repair, rental, installation, operation, publishing, hosting or consulting. These verbs carry classification meaning. A translation that replaces several of them with one broad word such as “services” can collapse meaningful boundaries.
Translation control: build terminology around activity verbs as well as nouns. Quality assurance: compare translated verbs with inclusions and exclusions in the explanatory notes.
14. Manufacturing and assembly are not always equivalent
A source category may distinguish manufacturing from simple assembly, processing, finishing or installation. Languages differ in how these processes are named. Translators should not choose the most familiar industrial verb if it changes the economic activity described.
Translation control: preserve the process stage. Quality assurance: test examples at category boundaries where the wrong verb would move a record to another code.
15. Wholesale and retail must remain distinct
Wholesale and retail trade are often separate statistical categories because their customers, channels and economic functions differ. In languages where everyday vocabulary uses one broad term for selling, the target must still communicate the formal distinction.
Translation control: use established statistical or commercial terminology for wholesale and retail. Quality assurance: review company descriptions that mention both channels and confirm the assigned code remains unchanged.
16. Repair, maintenance and installation can classify differently
The same physical equipment can appear in manufacturing, installation, maintenance and repair categories. A translator who repeats one generic technical-service term can erase the difference. This matters for machinery, construction systems, vehicles, computers and many other sectors.
Translation control: translate the service action precisely. Quality assurance: compare the source code’s inclusions with the target verb and object.
17. Publishing, broadcasting and digital platforms require careful scope
Modern classifications distinguish activities that can sound similar in everyday language: publishing content, broadcasting it, hosting it, operating a platform or providing telecommunications infrastructure. Digital transformation has made these boundaries more important, not less.
Translation control: preserve the economic function described by the source. Quality assurance: avoid translating every digital activity as “online services.”
18. Cloud, hosting and software activities can sit in different categories
A technology firm may write software, operate data infrastructure, provide hosting, manage platforms or deliver consulting. Marketing copy often bundles these together. Industry classification requires sharper distinctions than promotional language usually provides.
Translation control: translate the assigned statistical description, not the company slogan. Quality assurance: check technical verbs such as develop, host, process, manage and consult.
19. Education categories depend on the activity, level and provider context
Education classifications can distinguish school levels, tertiary education, training, support services and other learning activities. A translation that renders all categories simply as “education services” removes the detail that makes the code useful.
Translation control: preserve educational level and institutional function. Quality assurance: compare target terminology with the classification’s level structure.
20. Health and social-care categories need careful boundaries
Hospitals, physicians, residential care, social assistance and allied services can belong to different categories. Translators should avoid medicalizing social services or, conversely, generalizing clinical activity into social support language.
Translation control: use domain-appropriate health and care terminology. Quality assurance: test whether the translated description still distinguishes clinical care, residential support and social assistance.
21. Agriculture, forestry and fishing contain production-stage distinctions
Primary industries often distinguish growing, breeding, support activities, harvesting, logging, fishing and aquaculture. A broad translation such as “agriculture” may be understandable to a general reader but too coarse for classification data.
Translation control: preserve the production stage and resource type. Quality assurance: use examples from the explanatory notes to validate ambiguous terms.
22. Mining and quarrying vocabulary can be highly technical
Extraction categories depend on material, method and supporting activity. Geological and industrial terms may have several target-language equivalents, some common and some regulatory. The translated title should fit the statistical meaning rather than merely sound natural.
Translation control: use a controlled extractive-industry glossary. Quality assurance: check mineral or resource terms against the code scope.
23. Construction classifications depend on project type and activity
Building construction, civil engineering, specialty trades, installation and finishing can be separate activities. A contractor’s website may list all of them, while the statistical record assigns one primary code. Translation must not widen the official record unintentionally.
Translation control: keep the classification description narrower than marketing copy when the source is narrow. Quality assurance: compare the translated title with the coded level.
24. Transport, warehousing and postal activities are not one generic logistics bucket
Modern supply chains contain road, rail, air, water, warehousing, courier, postal and support activities. Industry codes distinguish these functions because their economics differ. A target-language term such as “logistics” can be too broad when the source code is specific.
Translation control: preserve mode and function. Quality assurance: inspect the surrounding record for transport mode, warehousing or support clues.
25. Financial services require role-specific wording
Banking, insurance, investment management, securities dealing and auxiliary services can occupy separate categories. Translators should not use one broad term such as “finance company” for every financial activity.
Translation control: preserve the economic role of the institution. Quality assurance: distinguish principal financial activity from support or brokerage functions.
26. Real-estate activity and construction are different classifications
A company can develop, build, sell, rent or manage property. Those verbs describe different economic activities. Translation is particularly risky when one language uses a broad real-estate term covering several of them.
Translation control: keep development, construction, brokerage, rental and management separate. Quality assurance: verify the source code rather than classifying from the word “property” alone.
27. Professional services need discipline-specific terminology
Legal, accounting, architecture, engineering, scientific research, advertising and consulting services may sit in distinct classes. The source title often names the professional function precisely. A vague target such as “business services” defeats that precision.
Translation control: translate the profession and service function explicitly. Quality assurance: ensure the target does not imply licensure or scope that the source does not contain.
28. Administrative support is not the same as professional consulting
Back-office support, staffing, travel services, security, cleaning and other administrative activities can be separated from higher-level professional services. Business language sometimes calls both “consulting” or “outsourcing,” which can confuse classification.
Translation control: preserve whether the activity provides expertise, labor, operations or support. Quality assurance: test borderline service descriptions against the official notes.
29. Public administration has its own statistical logic
Government functions may be classified separately from the same practical activity performed by a private enterprise. Translators should therefore pay attention to the statistical unit and institutional role, not just the service described.
Translation control: keep public-administration terminology distinct from commercial service terminology. Quality assurance: confirm the source system’s treatment of government units.
30. Cultural, arts and recreation categories can overlap in everyday language
Museums, performing arts, sports, amusement and recreation can sound broadly similar when translated for general readers. Statistical classifications usually distinguish them. The target should remain readable while retaining those distinctions.
Translation control: use sector-specific cultural and recreation terms. Quality assurance: check venue, performance, participation and operation wording.
31. Correspondence tables are mappings, not translations
When a classification is revised or mapped to another system, official correspondence tables can show one-to-one, one-to-many or many-to-one relationships. Translators should not hide this complexity by pretending every source code has one obvious target code.
Translation control: label correspondence results as mappings. Quality assurance: preserve multiple candidate relationships when the official table does not provide a single exact equivalent.
32. One-to-many mappings require business rules
If one old code maps to several new codes, language alone cannot determine the correct destination. Additional activity details are needed. An automated translation pipeline should stop and request classification evidence rather than choose whichever description looks most similar.
Translation control: treat ambiguous mappings as data exceptions. Quality assurance: require a documented decision rule or human classifier for one-to-many cases.
33. Many-to-one mappings can lose analytical detail
A newer or broader classification may combine activities previously separated, or a cross-system mapping may collapse detail. The translation should not imply that the source and target classifications have identical granularity.
Translation control: keep the original source code available when analytical lineage matters. Quality assurance: document any loss of detail in the mapping layer.
34. Historical time series need classification lineage
Economic statistics often span years in which classifications changed. Analysts may restate historical data, maintain dual-coded series or document breaks in series. A translation should preserve those methodological notes because they explain why the same industry appears under different codes across time.
Translation control: translate revision notes and break-in-series warnings carefully. Quality assurance: compare year, classification version and code together.
35. Company databases should separate assigned codes from translated labels
A robust company database stores the classification code, system, revision and translated display label as separate fields. This prevents a language update from altering the underlying industry assignment.
Translation control: localize display labels only. Quality assurance: switch the interface language and confirm the same company record retains exactly the same code and revision.
36. Search systems should index both code and localized wording
Users may search by code, English title, local-language title or a common business phrase. Indexing those as separate fields improves discovery without changing the classification itself.
Translation control: preserve the official localized title and keep synonyms in a separate search layer. Quality assurance: ensure synonyms do not appear as if they were official code labels.
37. Spreadsheets can damage hierarchy even when codes survive
A spreadsheet may preserve every numeric code but break parent-child structure when rows are sorted independently or filtered incorrectly. Translation teams often focus on text columns and overlook relational damage.
Translation control: use stable row IDs and full-table operations. Quality assurance: rebuild several hierarchy branches after export and confirm their order and parentage.
38. Leading zeros and punctuation require exact handling
Some classification systems or local derivatives can contain leading zeros, decimal points, letters or separators. General spreadsheet formatting can remove or transform them. Even when the three systems discussed here use familiar numeric or alphanumeric patterns, the safe rule is to store codes as controlled text.
Translation control: define code columns explicitly as text. Quality assurance: compare literal source-target code strings after every file conversion.
39. CSV headers need precise translation
A file may contain fields such as classification system, revision, code, title, level, parent code and explanatory note. Translating several of those simply as “industry” or “category” can make the dataset impossible to reconstruct.
Translation control: maintain a schema glossary before translating values. Quality assurance: reimport the localized CSV and confirm that every column maps back to the intended source field.
40. JSON and APIs should keep classification values canonical
APIs can expose a canonical code and separate localized titles. Translating JSON keys, code values or enum identifiers can break client applications. The correct place for language is usually the display layer.
Translation control: localize labels, not API contracts. Quality assurance: run schema tests in every locale and compare returned codes.
41. Translation memories can carry obsolete classification wording
Industry titles repeat across releases, so a translation memory can be useful. It can also suggest a term from an older revision whose scope has changed. High match scores are not proof that the classification meaning is still identical.
Translation control: tag terminology by system and revision. Quality assurance: review every reused label against the current explanatory notes.
42. Machine translation should not assign codes
A language model can translate an industry description, but assigning a code requires classification logic, source evidence and often institutional rules. A plausible code is not the same as a verified code.
Translation control: mask or lock codes before machine translation. Quality assurance: require deterministic lookup or an authorized classifier for code assignment.
43. Generative AI should explain uncertainty instead of hiding it
When a business activity falls between categories, an AI system can summarize what information is missing: whether the activity is manufacturing or installation, wholesale or retail, software publishing or hosting, and so on. It should not invent certainty.
Translation control: allow AI to produce questions and explanations, not final classification authority. Quality assurance: retain an exception log for unresolved records.
44. OCR can corrupt codes in scanned registers
Historical business registers and scanned statistical tables may contain small-font codes that OCR confuses with punctuation or neighboring digits. Translating the extracted label without checking the code can attach a correct sentence to the wrong category.
Translation control: verify OCR-derived codes against the classification structure. Quality assurance: reject codes that do not exist in the recorded revision.
45. Right-to-left layouts need structural testing
Arabic and Hebrew reports may display left-to-right codes inside right-to-left prose. Visual reordering can make codes, parentheses or decimal separators appear confusing even when the underlying text is correct.
Translation control: use directional isolation around code strings. Quality assurance: test display, copy-paste and search behavior in the final application.
46. Accessibility requires readable labels as well as exact codes
A screen reader can read a code character by character, but a human-readable localized title gives the user meaning. Accessible interfaces should expose both without using the translated title as a substitute for the canonical code.
Translation control: pair the stable code with a clear localized label. Quality assurance: test navigation by code and by title with assistive technology.
47. Worked example: a software company
A business describes itself as a cloud software platform, data host and technology consultant. The source database already assigns an industry code. The translator translates the official coded description, not the marketing website, and leaves the code unchanged. If analysts want to reconsider the classification, that happens in a separate data-governance workflow.
Lesson: translation follows the assigned classification. Check: compare source code, revision, official description and localized display label as four separate fields.
48. Worked example: a retailer that also manufactures
A brand manufactures some products but earns most revenue from retail stores. A classification authority may assign the record according to the principal activity under its rules. The translator should not move the company into manufacturing merely because the product description dominates the website.
Lesson: business narrative and statistical classification are different evidence layers. Check: preserve the source assignment and escalate classification questions rather than “correcting” them linguistically.
49. Worked example: mapping NAICS to NACE
An analyst needs a European comparison for a North American industry. The translation team first translates the NAICS description, then uses an official or documented correspondence resource to identify possible NACE categories. If one NAICS industry corresponds to several NACE classes, the analyst receives all valid possibilities with the required activity detail, not a fabricated one-to-one answer.
Lesson: cross-system mapping is analytical work. Check: preserve both source and target namespaces and record the mapping method.
50. Worked example: historical data across revisions
A ten-year dataset contains older NACE or NAICS codes for early years and newer revision codes later. The translated report keeps each year’s recorded classification and explains the revision break. It does not rewrite the earlier data to look uniform unless an official backcast or restatement has been supplied.
Lesson: comparability requires provenance. Check: audit year, revision and code together before charting trends.
51. Search intent: “translate NAICS code”
A person using this search may want the meaning of a code in another language, a translation of an industry title, or an equivalent classification in another country. Those are three different tasks. The safe answer identifies the code and revision, translates its official description, and treats any cross-system equivalent as a mapping problem.
Best practice: show the source NAICS code unchanged, provide the localized title, and clearly label any NACE or ISIC correspondence as a separate analytical result.
52. Search intent: “translate NACE code”
European users often need NACE titles in another language or want to compare NACE with a national or international classification. Because NACE is already maintained in a multilingual European statistical environment, official terminology should be preferred where available.
Best practice: identify whether the source is Rev. 2 or Rev. 2.1, preserve the code and consult Eurostat’s structure, explanatory notes and correspondence material before creating a target description.
53. Search intent: “translate ISIC code”
ISIC users may be working with international datasets in which one code must be explained in several languages. The code should remain the same while the title and explanatory notes are translated. If a country uses a derived national classification, that national code should not be presented as if it were the ISIC code.
Best practice: state the ISIC revision explicitly and keep any national mapping separate.
54. Official reference route for NAICS
The U.S. Census Bureau provides the official NAICS structure, definitions and transition information for Census programs. Its current materials explain the hierarchical structure and the use of the 2022 NAICS edition. Technical teams should use the official source rather than copying an old industry list from a marketing database.
Reference: U.S. Census Bureau NAICS. Record the release or revision date used in the project.
55. Official reference route for NACE
Eurostat publishes NACE material, including NACE Rev. 2.1 structure, guidance and explanatory notes. Eurostat states that the revision became official in 2023 and that European statistics began moving to Rev. 2.1 from 2025 onward.
Reference: Eurostat NACE. Use official multilingual labels when they exist and keep local project synonyms separate.
56. Official reference route for ISIC
The United Nations Statistics Division maintains the ISIC reference pages and publishes the current structure, explanatory material and revision history. ISIC Revision 5 was endorsed by the UN Statistical Commission at its 54th session in 2023.
Reference: UN Statistics Division ISIC. Record the revision and any correspondence table used.
57. Connection to the master translation architecture
This page belongs beneath eduKateSG’s Master Art of Translation — Technical Translation System. That master owns the broad method for specifications, terminology, data protection, validation and change control. This article owns the narrower problem of economic-activity classification.
The vocabulary layer still matters: readers who need to build stronger technical word knowledge can use the Vocabulary Learning Hub, while this article remains focused on code identity, classification meaning and multilingual industry data.
58. Release checklist
Before release, verify the classification system, revision, code, level, parent relationship and official description. Confirm that industry has not been confused with occupation or product classification. Review activity verbs, preserve historical codes, document crosswalks, protect code fields in spreadsheets and APIs, lock identifiers in machine translation, test RTL rendering where relevant, and make sure every localized label still belongs to the same source node.
Then perform ordinary linguistic QA for clarity, naturalness, grammar and terminology. Classification integrity and language quality are complementary controls; neither one replaces the other.
59. Final rule: translate the industry description, preserve the classification evidence
Industry classifications let governments, researchers and businesses compare economic activity across millions of records. Their usefulness depends on stable code identity, clear revision history and disciplined interpretation. Translation should make the classification understandable to a new reader without changing what the source has classified. Preserve the code and its lineage, translate the official meaning, and keep cross-system mapping as a separate, auditable analytical step.