To translate patent application numbers, publication numbers, priority numbers and IPC or CPC classification codes accurately, a translator must separate invention language from intellectual-property identity data. Patent specifications, search reports, legal opinions, licensing files and litigation records contain claims and technical prose that may need careful translation, but they also contain identifiers whose function is to point to a particular filing, publication, priority event or classification record. A fluent translation can still become legally and technically confusing if one identifier is altered or attached to the wrong document.
This guide explains how to translate patent and intellectual-property documents without changing PCT application numbers, WO publication numbers, national application numbers, priority references, kind codes, IPC symbols or CPC symbols. It solves one distinct high-intent search problem: which elements are ordinary language, which are official identifiers, how do application and publication numbers differ, and how can multilingual patent records stay traceable to the same filing and published document?
The working rule is translate the invention and legal prose, preserve official IP identifiers, identify the issuing system, and verify the relationship between every code and document. WIPO distinguishes the international application number from the international publication number; under the PCT, an international application number follows a structured PCT format, while the publication number begins with WO and identifies the published international application. IPC and CPC symbols classify technical subject matter rather than identify one filing. This specialist branch connects to Master Art of Translation, the protected Vocabulary Learning Hub, and How English Works without replacing the existing patent-claims or broader IP owners.
One invention can have several numbers
A single invention can generate first filings, priority applications, PCT records, publications, national phases and grants. Calling all of them one patent number erases procedural identity. Build a record map and name the field before translating it. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Identifiers are not vocabulary
Application and publication strings are governed bibliographic data, even when they contain readable letters. Translating a prefix or separator can create a different or invalid identifier. Translate labels and prose while copying official values exactly. Exact characters are only the first test. The target also needs the correct office, date, role or procedural stage beside the code. Record where the value was verified—source document, PATENTSCOPE, national register or client database—so another reviewer can reproduce the decision instead of trusting an unexplained correction.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
PCT application number
The international application number identifies the PCT filing record. It is easily confused with the later WO publication number because both appear on one front page. Keep separate application and publication fields. The practical test is retrieval fidelity. A reviewer who receives only the target should still be able to reach the same filing, publication or classification and understand why that identifier appears there. If retrieval or relationship is uncertain, flag the record instead of guessing or normalizing it into a more familiar pattern.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
PCT numbering structure
WIPO Section 307 uses PCT, receiving-office code, year and serial structure. Office codes and slashes can be damaged by localization or smart punctuation. Treat the complete string as a protected token. Software should store this field as text and exclude it from locale formatting, spelling correction and machine-generated rewriting. Human-language changes can then proceed around a stable bibliographic anchor. This separation is especially important when multiple family members share similar titles, applicants or serial digits.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
WO publication number
WIPO Section 404 assigns a separate WO number to the published international application. Replacing a filing number with a WO number changes the bibliographic object. Preserve both and label their different functions. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Filing year versus publication year
A filing can occur in one year and publication in the next. Editors sometimes force years to match because the difference looks suspicious. Preserve chronology and use precise verbs such as filed and published. Exact characters are only the first test. The target also needs the correct office, date, role or procedural stage beside the code. Record where the value was verified—source document, PATENTSCOPE, national register or client database—so another reviewer can reproduce the decision instead of trusting an unexplained correction.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Kind codes
A1, A2, A3 and related codes describe publication circumstances. Dropping the kind code can make a citation less precise or point to another event. Keep the code and translate only its explanation. The practical test is retrieval fidelity. A reviewer who receives only the target should still be able to reach the same filing, publication or classification and understand why that identifier appears there. If retrieval or relationship is uncertain, flag the record instead of guessing or normalizing it into a more familiar pattern.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
National numbering systems
Patent offices use different number formats, punctuation and type codes. Forcing every national number into a PCT-style pattern destroys official identity. Use the issuing office’s own representation. Software should store this field as text and exclude it from locale formatting, spelling correction and machine-generated rewriting. Human-language changes can then proceed around a stable bibliographic anchor. This separation is especially important when multiple family members share similar titles, applicants or serial digits.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Application versus publication fields
Some systems reuse serial components across different record types. Visual similarity encourages accidental deduplication. Maintain separate columns until the office-defined relationship is known. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Grant numbers
A granted right can have an identifier different from its application and publication records. Using the filing number where a granted patent number is required can confuse legal scope and status. Translate Patent No. and Application No. distinctly. Exact characters are only the first test. The target also needs the correct office, date, role or procedural stage beside the code. Record where the value was verified—source document, PATENTSCOPE, national register or client database—so another reviewer can reproduce the decision instead of trusting an unexplained correction.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Priority identity
Priority records connect a later application to earlier filings. A correct number attached to the wrong date can rewrite the legal chronology. Protect number, date and office as one tuple. The practical test is retrieval fidelity. A reviewer who receives only the target should still be able to reach the same filing, publication or classification and understand why that identifier appears there. If retrieval or relationship is uncertain, flag the record instead of guessing or normalizing it into a more familiar pattern.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Multiple priorities
One application may claim several priority filings. Sorting or row shifts can re-pair dates and numbers incorrectly. Preserve source order and verify every tuple. Software should store this field as text and exclude it from locale formatting, spelling correction and machine-generated rewriting. Human-language changes can then proceed around a stable bibliographic anchor. This separation is especially important when multiple family members share similar titles, applicants or serial digits.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Earliest priority date
Priority date and filing date answer different procedural questions. Translators can mislabel the earlier date as the current application’s filing date. Use explicit procedural labels. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Patent families
Related national and regional records share disclosure but not necessarily numbers. Collapsing family members into one canonical number loses jurisdiction-specific records. Keep office, number and status per member. Exact characters are only the first test. The target also needs the correct office, date, role or procedural stage beside the code. Record where the value was verified—source document, PATENTSCOPE, national register or client database—so another reviewer can reproduce the decision instead of trusting an unexplained correction.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Patent titles
Titles are human-language metadata and may be translated. A translated title can become hard to search if the source form disappears. Retain official source title when research traceability matters. The practical test is retrieval fidelity. A reviewer who receives only the target should still be able to reach the same filing, publication or classification and understand why that identifier appears there. If retrieval or relationship is uncertain, flag the record instead of guessing or normalizing it into a more familiar pattern.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Inventor and applicant roles
Inventor, applicant, assignee and owner are different legal roles. A row shift or loose synonym can misstate who created or owns the filing. Translate role labels precisely and preserve identity. Software should store this field as text and exclude it from locale formatting, spelling correction and machine-generated rewriting. Human-language changes can then proceed around a stable bibliographic anchor. This separation is especially important when multiple family members share similar titles, applicants or serial digits.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
INID codes
INID codes standardize bibliographic field meanings on patent documents. Renumbering them as if they were section numbers breaks cross-office readability. Preserve INID numbers and translate the field descriptions. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
IPC purpose
IPC classifies technical subject matter rather than identifying one document. Treating an IPC symbol as a patent number confuses classification with record identity. Label classification separately from publication identifiers. Exact characters are only the first test. The target also needs the correct office, date, role or procedural stage beside the code. Record where the value was verified—source document, PATENTSCOPE, national register or client database—so another reviewer can reproduce the decision instead of trusting an unexplained correction.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
IPC symbols
Letters, digits and slashes in IPC symbols are governed notation. Localizing the letters or punctuation produces a different or invalid symbol. Translate the official title, not the symbol. The practical test is retrieval fidelity. A reviewer who receives only the target should still be able to reach the same filing, publication or classification and understand why that identifier appears there. If retrieval or relationship is uncertain, flag the record instead of guessing or normalizing it into a more familiar pattern.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
IPC revisions
Classification schemes change over time. Silently replacing historical IPC with current IPC rewrites the source record. Record the version or period and preserve historical assignments. Software should store this field as text and exclude it from locale formatting, spelling correction and machine-generated rewriting. Human-language changes can then proceed around a stable bibliographic anchor. This separation is especially important when multiple family members share similar titles, applicants or serial digits.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
CPC purpose
CPC extends technical classification beyond IPC in participating systems. Calling all CPC entries IPC obscures the scheme used by the source. Keep CPC and IPC labels distinct. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Multiple classifications
One application can legitimately carry several technical classifications. Deleting similar-looking codes can remove relevant search dimensions. Preserve the complete classification set. Exact characters are only the first test. The target also needs the correct office, date, role or procedural stage beside the code. Record where the value was verified—source document, PATENTSCOPE, national register or client database—so another reviewer can reproduce the decision instead of trusting an unexplained correction.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Claims versus classifications
Claims define legal scope while IPC/CPC organize technical information. Combining these tasks can encourage translators to treat classification labels as claim language. Keep identifier QA separate from claim translation. The practical test is retrieval fidelity. A reviewer who receives only the target should still be able to reach the same filing, publication or classification and understand why that identifier appears there. If retrieval or relationship is uncertain, flag the record instead of guessing or normalizing it into a more familiar pattern.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Prior-art citations
Patent citations combine office code, publication number, kind code, date and title. Translating or dropping one structured element can hinder retrieval. Translate titles while preserving citation identity. Software should store this field as text and exclude it from locale formatting, spelling correction and machine-generated rewriting. Human-language changes can then proceed around a stable bibliographic anchor. This separation is especially important when multiple family members share similar titles, applicants or serial digits.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Office and country codes
Standard office prefixes form part of many publication identities. Replacing a code with a localized country abbreviation can break searches. Translate the full office name separately. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Abbreviation ambiguity
Office codes, applicant initials and legal-form abbreviations can look alike. Text extraction can remove the layout that distinguished them. Restore field context before deciding whether to translate or preserve. Exact characters are only the first test. The target also needs the correct office, date, role or procedural stage beside the code. Record where the value was verified—source document, PATENTSCOPE, national register or client database—so another reviewer can reproduce the decision instead of trusting an unexplained correction.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Spreadsheet typing
Application numbers can be interpreted as dates, formulas or scientific notation. Automatic type conversion silently changes identifiers. Import identifier columns as text and compare after export. The practical test is retrieval fidelity. A reviewer who receives only the target should still be able to reach the same filing, publication or classification and understand why that identifier appears there. If retrieval or relationship is uncertain, flag the record instead of guessing or normalizing it into a more familiar pattern.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Slashes and punctuation
Separators can be official parts of display syntax. Smart typography can replace characters with visually similar Unicode forms. Disable substitutions inside protected fields. Software should store this field as text and exclude it from locale formatting, spelling correction and machine-generated rewriting. Human-language changes can then proceed around a stable bibliographic anchor. This separation is especially important when multiple family members share similar titles, applicants or serial digits.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
OCR risk
Scanned patents can confuse 0/O, 1/I and 5/S. A plausible OCR string can consistently retrieve the wrong or no document. Verify key numbers against PATENTSCOPE or the issuing office. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Historical numbering
Older patent records often use formats that differ from current conventions. Modernizing them can erase documentary history. Preserve historical form and add modern cross-references separately. Exact characters are only the first test. The target also needs the correct office, date, role or procedural stage beside the code. Record where the value was verified—source document, PATENTSCOPE, national register or client database—so another reviewer can reproduce the decision instead of trusting an unexplained correction.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
National phase
National-phase filings receive local identifiers while remaining related to the PCT application. Substituting the national number for the PCT number collapses separate proceedings. Show both layers when the source does. The practical test is retrieval fidelity. A reviewer who receives only the target should still be able to reach the same filing, publication or classification and understand why that identifier appears there. If retrieval or relationship is uncertain, flag the record instead of guessing or normalizing it into a more familiar pattern.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Regional phase
Regional offices create their own filing, publication and grant identities. Assuming one global numbering scheme causes field errors. Keep regional-office context with the identifier. Software should store this field as text and exclude it from locale formatting, spelling correction and machine-generated rewriting. Human-language changes can then proceed around a stable bibliographic anchor. This separation is especially important when multiple family members share similar titles, applicants or serial digits.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Priority translations
A translated priority document remains evidence of the same earlier filing. Language change does not create a new priority number. Preserve the original filing identity. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Legal-status events
Registers record filing, publication, grant, lapse, opposition and transfer. A correct number with the wrong event date gives a false status history. Verify event, date and document together. Exact characters are only the first test. The target also needs the correct office, date, role or procedural stage beside the code. Record where the value was verified—source document, PATENTSCOPE, national register or client database—so another reviewer can reproduce the decision instead of trusting an unexplained correction.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Ownership changes
Applications can be assigned after filing. Replacing historical applicant information with a current owner rewrites the original record. Distinguish source-period data from later updates. The practical test is retrieval fidelity. A reviewer who receives only the target should still be able to reach the same filing, publication or classification and understand why that identifier appears there. If retrieval or relationship is uncertain, flag the record instead of guessing or normalizing it into a more familiar pattern.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Inventor names
Names may have multiple scripts and romanizations. Semantic translation of a surname or inconsistent transliteration fragments identity. Use office spelling or an approved transliteration policy. Software should store this field as text and exclude it from locale formatting, spelling correction and machine-generated rewriting. Human-language changes can then proceed around a stable bibliographic anchor. This separation is especially important when multiple family members share similar titles, applicants or serial digits.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Technical terminology in titles
Short patent titles can contain dense domain terminology. Literal translation can change technical scope or search terms. Translate with specification context and a controlled glossary. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Machine translation protection
Automated systems can normalize punctuation and expose identifiers to generation. A single character edit can break record retrieval. Mask or lock identifiers before machine translation. Exact characters are only the first test. The target also needs the correct office, date, role or procedural stage beside the code. Record where the value was verified—source document, PATENTSCOPE, national register or client database—so another reviewer can reproduce the decision instead of trusting an unexplained correction.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Translation memory risk
Legacy TM can contain obsolete formatting or a related family member’s number. Fuzzy reuse can overwrite the current source identifier. Reuse prose but take identifiers from the current record. The practical test is retrieval fidelity. A reviewer who receives only the target should still be able to reach the same filing, publication or classification and understand why that identifier appears there. If retrieval or relationship is uncertain, flag the record instead of guessing or normalizing it into a more familiar pattern.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Database normalization
Different patent databases display spacing and punctuation differently. A formatting difference can be mistaken for a different record or vice versa. Document normalization rules and retain source form. Software should store this field as text and exclude it from locale formatting, spelling correction and machine-generated rewriting. Human-language changes can then proceed around a stable bibliographic anchor. This separation is especially important when multiple family members share similar titles, applicants or serial digits.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Search links
Visible citation text and hyperlink destination are separate layers. A publication label linked to another family member misdirects the reader. Test destination against the displayed identifier. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Family-table relationships
Every value can be valid while rows are paired incorrectly. Sorting one column independently creates false family relationships. Move whole records and validate tuples. Exact characters are only the first test. The target also needs the correct office, date, role or procedural stage beside the code. Record where the value was verified—source document, PATENTSCOPE, national register or client database—so another reviewer can reproduce the decision instead of trusting an unexplained correction.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
IP identity register
Large portfolios need a single source of truth for identifiers and relationships. Independent copying across languages multiplies drift. Maintain one protected registry used by every language team. The practical test is retrieval fidelity. A reviewer who receives only the target should still be able to reach the same filing, publication or classification and understand why that identifier appears there. If retrieval or relationship is uncertain, flag the record instead of guessing or normalizing it into a more familiar pattern.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Terminology versus identity resources
A glossary answers how to express technical and legal terms; an identity register answers which record is involved. Combining them encourages accidental editing of identifiers. Maintain separate resources with different permissions. Software should store this field as text and exclude it from locale formatting, spelling correction and machine-generated rewriting. Human-language changes can then proceed around a stable bibliographic anchor. This separation is especially important when multiple family members share similar titles, applicants or serial digits.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Automated QA
Exact source-target comparison catches character changes but not wrong row pairing. A perfect string in the wrong context remains wrong. Run token equality checks followed by relationship validation. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
PCT worked example
A filing such as PCT/XX2026/000321 may later be published as WO 2027/123456 A1. Calling both strings the same patent number obscures filing and publication stages. Explain each function while preserving each token. Exact characters are only the first test. The target also needs the correct office, date, role or procedural stage beside the code. Record where the value was verified—source document, PATENTSCOPE, national register or client database—so another reviewer can reproduce the decision instead of trusting an unexplained correction.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Multiple-priority example
Two priority records can have different offices and dates. Alphabetical resorting of translated office names can change procedural order. Keep source relationships and flag discrepancies. The practical test is retrieval fidelity. A reviewer who receives only the target should still be able to reach the same filing, publication or classification and understand why that identifier appears there. If retrieval or relationship is uncertain, flag the record instead of guessing or normalizing it into a more familiar pattern.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
IPC and CPC example
A search report can list both systems for the same disclosure. Similar symbols can tempt a reviewer to merge them. Translate scheme titles and preserve every symbol. Software should store this field as text and exclude it from locale formatting, spelling correction and machine-generated rewriting. Human-language changes can then proceed around a stable bibliographic anchor. This separation is especially important when multiple family members share similar titles, applicants or serial digits.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Error clinic
Typical failures include prefix translation, kind-code loss, year harmonization, priority row shifts and classification relabeling. These errors survive fluent prose review because they are data errors. Use a dedicated identifier QA pass. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Release checklist
Application, publication, grant, priority, kind code and classification fields all need exact preservation. Checking only the body text misses bibliographic failures. Verify tokens, relationships, dates, links and scheme labels before release. Exact characters are only the first test. The target also needs the correct office, date, role or procedural stage beside the code. Record where the value was verified—source document, PATENTSCOPE, national register or client database—so another reviewer can reproduce the decision instead of trusting an unexplained correction.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Frequently asked questions
Patent application numbers, WO publication numbers and classification symbols are not translation targets. Users often ask whether one invention should have only one number. Explain that procedural stages and jurisdictions legitimately create multiple identifiers. The practical test is retrieval fidelity. A reviewer who receives only the target should still be able to reach the same filing, publication or classification and understand why that identifier appears there. If retrieval or relationship is uncertain, flag the record instead of guessing or normalizing it into a more familiar pattern.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Connection to eduKateSG architecture
This specialist owner protects patent record identity rather than claim scope. Creating another broad patent hub would cannibalize existing owners. Route substantive claim translation to the existing patent guide and use this page for identifiers. Software should store this field as text and exclude it from locale formatting, spelling correction and machine-generated rewriting. Human-language changes can then proceed around a stable bibliographic anchor. This separation is especially important when multiple family members share similar titles, applicants or serial digits.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Authoritative sources
WIPO publishes PCT numbering, kind-code and IPC guidance. Secondary databases may normalize fields differently. Verify core identity against WIPO or the responsible patent office. Mark the label as translatable, the identifier as protected, and the label-to-identifier pairing as a relationship that must be checked after export. This three-layer review catches errors that ordinary proofreading misses because the prose can remain fluent while the bibliographic record changes.
Before release, ask three questions: what changed linguistically, what had to remain exact, and what patent-office record does this field point to? A complete target lets a reviewer answer all three without reconstructing hidden assumptions from the source.
Reference verification route
For PCT application numbering, consult WIPO’s Administrative Instructions Section 307. For international publication numbers, see Section 404. For PCT kind codes, use WIPO’s Kind Codes. For classification, use the current IPC Publication and relevant CPC resources.