How do you sort translated names, titles, dictionary entries and search results correctly? Use language-aware collation rather than comparing raw Unicode code points or simply alphabetising the English translation. Unicode CLDR collation provides locale-specific rules for ordering, grouping and comparing text. A reliable multilingual sorting workflow identifies the audience’s locale, the intended comparison task, the applicable collation settings, and the exact output that readers should see.
A multilingual address book, school directory or vocabulary list can contain perfectly translated labels and still be confusing when a Swedish name, Turkish dotted-I name, Chinese title or accented word is positioned unexpectedly. People looking for Unicode CLDR collation rules, locale-aware string sorting, multilingual search matching, accent-sensitive comparisons, alphabetical order and internationalisation testing are usually trying to solve a usability problem, not merely to configure an algorithm. This guide shows how to preserve readable names while sorting and searching them predictably.
Quick route:
- Separate translation of a word from ordering that word in a list.
- Select the user’s language and region deliberately instead of assuming English order.
- Differentiate display sorting, search matching, grouping and stable database identity.
- Test accents, case, digraphs, scripts, punctuation, numerals and fallback behaviour.
- Keep the original spelling and stable record identifier even when a comparison key changes.
The visible problem: a correct translation can be placed in the wrong position
Imagine a directory with thirty translated school names. A parent may understand every name yet fail to locate the right school because the list appears to place one accented letter in an unpredictable place. Another reader searches for a title without diacritics and receives no result; someone else enters the precise original spelling and sees several results that have been merged too aggressively. The translation itself may be accurate. The flaw belongs to the operation performed on that translation.
Sorting, comparing, finding, filtering and presenting text are not interchangeable tasks. A sorted list imposes order on the whole collection. Search matching asks which records correspond to a query. A filter can ignore some differences that a display list must preserve. An index groups items under headings; a database primary key distinguishes identities. If a system uses one approximate normalised string for every purpose, it will eventually sacrifice correctness or discoverability.
This distinction matters in classrooms as well as software. Children learning vocabulary may assume that a foreign-language word should sit exactly where its English equivalent would sit. That assumption fails because alphabetical conventions are part of a writing system and its language practices, not a universal arrangement of Latin characters. Readers should learn the rule used by the resource, rather than mistake an unusual order for bad spelling.
First principles: what CLDR collation actually controls
Unicode provides a repertoire of encoded characters and algorithms; CLDR supplies locale information and tailorings used by many internationalised systems. A collation tailoring adapts a root comparison order to the needs of a language or usage. The relevant formal reference is Unicode Technical Standard #35, Part 5, on collation. This does not mean there is one universal human-approved ordering for every directory. It means the application has a principled basis for selecting an order and testing that selection.
A locale is an operational configuration such as a language with regional or other refinements. It should not be confused with nationality, a person’s identity, or the language from which a sentence was translated. A user reading a Spanish directory might need Spanish ordering while viewing the labels in a bilingual interface. A Chinese list might legitimately offer alternative orderings, for example a pinyin-oriented view or a stroke-oriented view, if the audience and application support them.
Collation strength describes which differences affect comparison. At some strengths, base-letter distinctions decide order before accents, then case or other differences are considered. The precise result depends on locale and options. It is unsafe to declare that every accented character is equal to its unaccented form or that upper-case always sorts ahead of lower-case. Write tests for the deployed implementation rather than rely on intuition from a different language.
Tailoring allows a locale to treat a sequence or character according to its own conventions. A multicharacter sequence such as Czech ch can behave differently from naive character-by-character comparison. In some languages an accented letter has a special alphabetical position; in others an accent can be a later-level distinction. These are language-specific conventions, not defects to remove in preprocessing.
Search collation and display collation can require different equivalences. A search interface may permit forgiving matches, yet a strict official-name or citation lookup must not silently equate distinct names. The CLDR specification describes a search-oriented collation type; it is not simply a command to remove every accent and compare lowercase strings. A good service documents when approximate matching is used and offers an exact alternative.
Normalization is another distinct concept. Some visually equivalent sequences can be represented by different Unicode code-point sequences. Normalization can make storage and comparison more consistent, but it does not replace a locale-aware collator, nor does it authorize altering a name’s displayed spelling. Keep the original display text, normalize only according to an explicit processing policy, and test equivalence at the right layer.
Stable sorting ensures that items regarded as equal at the chosen comparison level do not jump unpredictably. A predictable secondary rule, such as a stable stored identifier after the selected comparison key, is useful for deterministic results. Never treat the ordering result as a persistent user identifier: a locale change, library update or new collation option may reorder the same records without changing what they are.
Numeric ordering is a choice, not an automatic consequence of translation. A classroom list containing Lesson 2 and Lesson 10 is often more usable in numeric sequence, but a literary catalogue might need the exact title-sort convention. Test numeric options separately from punctuation, leading zeros and the decision to preserve titles as written.
Script order needs an explicit audience decision when a list mixes Latin, Greek, Arabic, Cyrillic, Han and other writing systems. A universal alphabetic list across scripts is not obvious. Options include a locale-defined order, separate script sections or parallel transliterated browse views. Explain the grouping convention near the directory rather than quietly promising that every reader will infer it.
A repeatable twelve-stage translation and collation workflow
Stage 1: Define the reader’s job
Decide whether the list is a school directory, dictionary, catalogue, search result, name picker or examination index. Write down the question the reader will ask first. A glossary used to learn vocabulary may need headword order; a product finder may need a separate relevance ranking; a historical index may need established editorial conventions. A clear purpose prevents the easiest programming default from quietly becoming an inappropriate language rule.
At this stage, note the owner of the decision, the evidence used, and the effect on readers. Re-run the local fixture after changing this setting instead of relying on a visual impression from a single name. A reader should be able to explain the selected behaviour in plain language, and the team should be able to reproduce it using the original data rather than a screenshot alone.
Stage 2: Record source and display spelling
Store each label exactly as editorially approved, including accents, case, script, punctuation and spacing. Preserve the source text, target-language form, editorial provenance and a durable identifier in separate fields. If later processing removes a diacritic from a comparison key, the original name must still appear intact in the interface.
At this stage, note the owner of the decision, the evidence used, and the effect on readers. Re-run the local fixture after changing this setting instead of relying on a visual impression from a single name. A reader should be able to explain the selected behaviour in plain language, and the team should be able to reproduce it using the original data rather than a screenshot alone.
Stage 3: Choose the display locale
Identify the language or locale that should govern ordering for the selected view. Do not infer it solely from the user’s IP address or from a translated paragraph elsewhere on the page. Readers may switch language independently of country, and bilingual collections can require clearly labelled alternate sort views.
At this stage, note the owner of the decision, the evidence used, and the effect on readers. Re-run the local fixture after changing this setting instead of relying on a visual impression from a single name. A reader should be able to explain the selected behaviour in plain language, and the team should be able to reproduce it using the original data rather than a screenshot alone.
Stage 4: Select the operation
Specify whether the component needs sorting, exact lookup, flexible search, grouping, prefix matching, pagination or autocomplete. The same list can use more than one comparator legitimately. A comparison optimised for search discovery is not automatically correct for an official alphabetical register.
At this stage, note the owner of the decision, the evidence used, and the effect on readers. Re-run the local fixture after changing this setting instead of relying on a visual impression from a single name. A reader should be able to explain the selected behaviour in plain language, and the team should be able to reproduce it using the original data rather than a screenshot alone.
Stage 5: Choose collation type and settings
Use a deployed library that implements relevant Unicode/CLDR support. Record the locale, collation type, strength and any options such as numeric comparison or case handling. Avoid an invented home-grown comparator based on replacing a few accented letters; such shortcuts rarely generalise.
At this stage, note the owner of the decision, the evidence used, and the effect on readers. Re-run the local fixture after changing this setting instead of relying on a visual impression from a single name. A reader should be able to explain the selected behaviour in plain language, and the team should be able to reproduce it using the original data rather than a screenshot alone.
Stage 6: Design a fixture set
Assemble real-looking but non-sensitive test examples covering accents, case pairs, digraphs, numerals, punctuation, multiple scripts and names with shared prefixes. Keep the expected outcome documented by locale. Tests should include pairs that differ in only one feature, because those pairs reveal which setting controls the order.
At this stage, note the owner of the decision, the evidence used, and the effect on readers. Re-run the local fixture after changing this setting instead of relying on a visual impression from a single name. A reader should be able to explain the selected behaviour in plain language, and the team should be able to reproduce it using the original data rather than a screenshot alone.
Stage 7: Run and explain sort tests
Execute the comparator over the fixture list, capture the observed order, and compare it with language-reference expectations. Record the library and data versions used. A screenshot alone does not explain whether a difference is intentional; retain a small written explanation of the intended convention.
At this stage, note the owner of the decision, the evidence used, and the effect on readers. Re-run the local fixture after changing this setting instead of relying on a visual impression from a single name. A reader should be able to explain the selected behaviour in plain language, and the team should be able to reproduce it using the original data rather than a screenshot alone.
Stage 8: Separate search matching
Define whether search is strict, accent-tolerant, case-tolerant or backed by synonyms and transliterations. Show enough of the matched original name that readers can judge the result. When names are legally or academically distinct, forgiving search can suggest alternatives without merging identities.
At this stage, note the owner of the decision, the evidence used, and the effect on readers. Re-run the local fixture after changing this setting instead of relying on a visual impression from a single name. A reader should be able to explain the selected behaviour in plain language, and the team should be able to reproduce it using the original data rather than a screenshot alone.
Stage 9: Build appropriate browse headings
Decide whether list headings should reflect the locale alphabet, a script grouping or a neutral collection section. Test what happens at heading boundaries and for symbols. A correct full-list comparator cannot rescue a navigation bar that sends a reader to the wrong letter group.
At this stage, note the owner of the decision, the evidence used, and the effect on readers. Re-run the local fixture after changing this setting instead of relying on a visual impression from a single name. A reader should be able to explain the selected behaviour in plain language, and the team should be able to reproduce it using the original data rather than a screenshot alone.
Stage 10: Verify user interaction
Test keyboard navigation, screen readers, pagination, copied names, search inputs, highlighting and result count messages in the translated interface. The sorting policy should remain understandable when someone cannot rely on the visual layout of an alphabet rail.
At this stage, note the owner of the decision, the evidence used, and the effect on readers. Re-run the local fixture after changing this setting instead of relying on a visual impression from a single name. A reader should be able to explain the selected behaviour in plain language, and the team should be able to reproduce it using the original data rather than a screenshot alone.
Stage 11: Check update and cache effects
Consider cached results, cursor-based pagination and stable sorting when the site’s locale setting or collation library changes. Document a deterministic tie-breaker. A saved search must not silently omit an item because adjacent result pages were generated using different comparators.
At this stage, note the owner of the decision, the evidence used, and the effect on readers. Re-run the local fixture after changing this setting instead of relying on a visual impression from a single name. A reader should be able to explain the selected behaviour in plain language, and the team should be able to reproduce it using the original data rather than a screenshot alone.
Stage 12: Review with a language-aware editor
Ask a reviewer familiar with the audience’s conventions to inspect examples and explain surprising differences. Publish the rule succinctly and set a review date after material library changes. A system test proves what software did; editorial review examines whether that behaviour helps actual readers.
At this stage, note the owner of the decision, the evidence used, and the effect on readers. Re-run the local fixture after changing this setting instead of relying on a visual impression from a single name. A reader should be able to explain the selected behaviour in plain language, and the team should be able to reproduce it using the original data rather than a screenshot alone.
Worked examples: 22 different places where collation changes the reading experience
The examples below are instructional scenarios. They show how to diagnose an ordering or matching choice, not a claimed list of outputs from a particular installed library version. Expected exact ordering must be checked against the actual locale settings and deployment.
Worked example 1: Swedish surname directory
Source collection: Åberg, Anders, Östlund and Ängström appear together. Reader’s problem: The interface promises Swedish alphabetical order; the reader needs to find surnames quickly.
First-principles resolution: Test the locale’s ordering of å, ä and ö instead of grouping all three mechanically with plain a and o. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Ask a Swedish-reading reviewer to sort a controlled fixture and to explain the position of each name. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 2: German contact book
Source collection: Adler, Ärger, Auer and Äpfel. Reader’s problem: The application offers a dictionary-style list and a separate phonebook-style list.
First-principles resolution: These use cases can employ different collation options; do not assume one universal treatment of umlauts for every German directory. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Expose the selected view and test both before applying either to a live contact search. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 3: Spanish regional school list
Source collection: Nacional, Ñandú and Nube. Reader’s problem: A Spanish-language parent searches an alphabet bar for a name beginning with ñ.
First-principles resolution: Treat ñ according to Spanish collation conventions rather than silently changing the display name to n. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Check the letter-heading route and not just the complete sort output. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 4: Turkish personal names
Source collection: Işık, İnci, İpek and Irmak. Reader’s problem: A translator sees two kinds of I in Turkish and is tempted to lowercase using English rules.
First-principles resolution: Turkish casing and collation require locale awareness; plain English lowercase conversion can lose distinctions. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Test uppercase and lowercase lookup separately while preserving each person’s exact chosen spelling. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 5: Czech glossary
Source collection: chata, cesta and hotel. Reader’s problem: The resource is a Czech-language word index that learners browse alphabetically.
First-principles resolution: Czech ch may be handled as a collation contraction; counting characters one at a time is not a sufficient implementation. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Use a locale-specific fixture with neighbours around the ch sequence to verify real ordering. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 6: Danish town selector
Source collection: Aalborg, Ærø, Østerbro and Århus. Reader’s problem: The user expects the interface’s declared Danish convention to govern the list.
First-principles resolution: Avoid translating or romanising a town name merely to make sorting look familiar to English readers. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Test explicitly whether legacy spellings and modern spellings are treated as expected by the chosen implementation. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 7: Chinese titles by pinyin
Source collection: A group of Chinese book titles with an approved pinyin reading for each. Reader’s problem: One readership expects pinyin order, while another needs a stroke-oriented browse mode.
First-principles resolution: Collation is an editorial choice for navigation; a one-size-fits-all English transliteration is not the only option. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Label the browse convention and test whether the data and library support the chosen mode. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 8: Japanese book catalogue
Source collection: Mixed hiragana, katakana and kanji titles. Reader’s problem: The catalogue includes both written titles and optional pronunciation readings.
First-principles resolution: Readers may need a pronunciation-oriented search alongside a preserved original-script display; generic byte order cannot deliver that experience. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Keep approved readings separately and record when a title lacks authoritative reading data. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 9: French accent distinction
Source collection: cote, côte, coté and côté. Reader’s problem: A language learner expects spelling to remain visible while the list follows the chosen locale’s order.
First-principles resolution: Do not strip accents from published headwords simply because an accent-insensitive search mode is available. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Compare strict and forgiving searches and make the difference clear in the results. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 10: Greek titles
Source collection: Names with base letters and tonos accents. Reader’s problem: The collection serves Greek-reading users and allows readers to search without diacritics.
First-principles resolution: A display sort and a forgiving query can use different settings without rewriting the stored titles. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Test both accented and unaccented queries and show the original matched title. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 11: Arabic classroom headings
Source collection: Arabic headwords with optional vowel marks. Reader’s problem: The learning list contains annotated and unannotated forms of some words.
First-principles resolution: Treat the presence or absence of marks as a deliberate educational feature; a search index may be forgiving while the teaching display must be accurate. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Ask a qualified reviewer whether visually similar entries are linguistically interchangeable in that exercise. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 12: Hindi and Devanagari entries
Source collection: Learner entries include consonant clusters and distinct matras. Reader’s problem: An English-only developer attempts to alphabetise by code-point sequence.
First-principles resolution: Unicode encoding order is not a substitute for the chosen Hindi collation and curriculum conventions. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Use local-language fixtures and judge navigation from the learner’s perspective. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 13: Lesson numbers
Source collection: Lesson 2, Lesson 10, Lesson 003 and Lesson 11. Reader’s problem: A parent is scrolling a digital workbook list.
First-principles resolution: A numeric comparison option may be more usable than lexicographic order, yet leading-zero labels might have an archival significance. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Decide whether the interface sorts by lesson index or exact title and document the rule. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 14: Product sizes
Source collection: Size 2, Size 8, Size 10 and Size 10A. Reader’s problem: A translated shopping filter is mixing textual model IDs with numeric sizes.
First-principles resolution: A numeric sort may help for sizes but must not turn distinct alphanumeric models into the same record. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Test the intended parsing of units and suffixes without rewriting identifiers. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 15: Accented author search
Source collection: José Alvarez and Jose Alvarez are separate records. Reader’s problem: A reader types a name without the accent on a mobile keyboard.
First-principles resolution: Forgiving search can suggest both entries, but the system should not silently collapse two distinct people. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Display full names and other disambiguating information, then permit an exact refinement. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 16: Punctuation in titles
Source collection: The River, River, The-River and River!. Reader’s problem: A literary index chooses how initial articles and punctuation affect placement.
First-principles resolution: The choice to ignore a leading article is an editorial indexing rule, not a property of translation itself. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Make title-sort keys separate from display titles and test punctuation-sensitive and punctuation-insensitive views. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 17: Mixed-script library
Source collection: English, Arabic and Japanese books in one catalogue. Reader’s problem: Readers cannot rely on a single familiar alphabet line for all scripts.
First-principles resolution: Create labelled script groupings or a well-described locale-driven order instead of pretending all scripts share identical headings. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Test the group-switch and full-search paths with users of multiple writing systems. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 18: Bilingual school directory
Source collection: English and Mandarin school names with some official bilingual pairs. Reader’s problem: A parent switches the display language without changing which school is selected.
First-principles resolution: Stable school IDs must connect the same entity while separate localized sort keys change its position. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Check that bookmarks and detail-page links remain stable after a language switch. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 19: Autocomplete suggestions
Source collection: Search term produces many near-identical accented entries. Reader’s problem: The interface ranks probable matches and then sorts tied scores.
First-principles resolution: Relevance ranking is distinct from collation. Use collation as a documented tie-breaker, not as a replacement for relevance. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Test that the best exact match is findable even when a related name would sort earlier. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 20: Database pagination
Source collection: A long directory loads fifty results at a time. Reader’s problem: Results are sorted in the database but filtered later using different locale logic.
First-principles resolution: Comparator mismatch can create omissions and duplicates at page boundaries. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Use compatible sort keys and deterministic tie-breakers for every page of the same view. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 21: Version migration
Source collection: A Unicode or collation library is upgraded. Reader’s problem: The team sees some directory positions change after deployment.
First-principles resolution: A different resulting order is not automatically corruption, but it may disrupt saved page cursors and user expectations. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Retest fixtures, review differences and reindex if the application stores comparison keys. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Worked example 22: Translation quality review
Source collection: A glossary contains approved translations plus aliases. Reader’s problem: Reviewers argue whether a synonym should appear as a separate headword.
First-principles resolution: Editorial synonym policy and locale comparison are different layers; one must not be hidden inside the other. Before changing any label, separate the material that the translator controls from the application’s comparison settings. Confirm whether this particular view needs alphabetical browsing, exact matching, search discovery or a special grouping convention. The same entry can legitimately participate in several of those activities without adopting several spellings.
Quality check: Keep the glossary relation explicit and ask a learner to perform an actual find-it task. Record the input fixture, the audience locale, the selected operation and the comparison outcome. Then ask whether a user can find the entry without knowing the implementation detail. If a different form is needed only for searching or indexing, store it as a separate key rather than replacing the visible translation. This protects accurate spelling, makes the issue reproducible, and prevents a local repair from silently harming other languages.
Common failure modes and how to repair them
Failure 1: Sorting bytes or code points
What goes wrong: The code is consistent but users see unfamiliar groupings. Corrective move: Use a locale-aware comparison implementation for the human-facing view, and test with minimal pairs. The useful diagnostic is not merely whether the program produces a result; it is whether the result supports the reader’s intended action and is consistent across the full collection. Keep a before-and-after fixture, test the translated interface, and review adjacent entries so the repair does not introduce a second, quieter defect elsewhere in the list.
Failure 2: Equating search and alphabetical browsing
What goes wrong: Forgiving matches become an unstable or linguistically odd ordered list. Corrective move: Specify distinct search and sort comparators, and let results show original names. The useful diagnostic is not merely whether the program produces a result; it is whether the result supports the reader’s intended action and is consistent across the full collection. Keep a before-and-after fixture, test the translated interface, and review adjacent entries so the repair does not introduce a second, quieter defect elsewhere in the list.
Failure 3: Deleting accents from stored titles
What goes wrong: Two legitimately different spellings collapse into one irreversible representation. Corrective move: Preserve source and display fields; generate indexing keys separately under a documented policy. The useful diagnostic is not merely whether the program produces a result; it is whether the result supports the reader’s intended action and is consistent across the full collection. Keep a before-and-after fixture, test the translated interface, and review adjacent entries so the repair does not introduce a second, quieter defect elsewhere in the list.
Failure 4: Assuming country determines language
What goes wrong: A bilingual user gets the wrong alphabetical view. Corrective move: Allow a clear language or sort preference without converting it into a claim about the person’s identity. The useful diagnostic is not merely whether the program produces a result; it is whether the result supports the reader’s intended action and is consistent across the full collection. Keep a before-and-after fixture, test the translated interface, and review adjacent entries so the repair does not introduce a second, quieter defect elsewhere in the list.
Failure 5: Using English case folding everywhere
What goes wrong: Turkish and other locale-specific case behaviours are mishandled. Corrective move: Use language-aware casing when casing is actually needed, and avoid unnecessary destructive conversion. The useful diagnostic is not merely whether the program produces a result; it is whether the result supports the reader’s intended action and is consistent across the full collection. Keep a before-and-after fixture, test the translated interface, and review adjacent entries so the repair does not introduce a second, quieter defect elsewhere in the list.
Failure 6: Treating all punctuation as clutter
What goes wrong: Book titles, formal citations or codes cease to be distinguishable. Corrective move: Decide punctuation significance by task, keeping immutable identifiers distinct from display sort keys. The useful diagnostic is not merely whether the program produces a result; it is whether the result supports the reader’s intended action and is consistent across the full collection. Keep a before-and-after fixture, test the translated interface, and review adjacent entries so the repair does not introduce a second, quieter defect elsewhere in the list.
Failure 7: Losing stable pagination
What goes wrong: Records move between pages after an inconsistent sort change. Corrective move: Use deterministic tie-breakers, consistent comparators and planned reindexing after relevant updates. The useful diagnostic is not merely whether the program produces a result; it is whether the result supports the reader’s intended action and is consistent across the full collection. Keep a before-and-after fixture, test the translated interface, and review adjacent entries so the repair does not introduce a second, quieter defect elsewhere in the list.
Failure 8: Ignoring script switching
What goes wrong: A keyboard user cannot locate names written in another script. Corrective move: Offer script-aware groupings and search alternatives appropriate to the resource. The useful diagnostic is not merely whether the program produces a result; it is whether the result supports the reader’s intended action and is consistent across the full collection. Keep a before-and-after fixture, test the translated interface, and review adjacent entries so the repair does not introduce a second, quieter defect elsewhere in the list.
Failure 9: Omitting editorial guidance
What goes wrong: Users mistake a correct locale rule for an error. Corrective move: Name the sort convention beside the directory and explain special cases briefly. The useful diagnostic is not merely whether the program produces a result; it is whether the result supports the reader’s intended action and is consistent across the full collection. Keep a before-and-after fixture, test the translated interface, and review adjacent entries so the repair does not introduce a second, quieter defect elsewhere in the list.
Failure 10: Testing one sample name only
What goes wrong: The comparator appears correct until encountering digraphs or accented letters. Corrective move: Create locale fixtures of near-neighbours and boundary cases before sign-off. The useful diagnostic is not merely whether the program produces a result; it is whether the result supports the reader’s intended action and is consistent across the full collection. Keep a before-and-after fixture, test the translated interface, and review adjacent entries so the repair does not introduce a second, quieter defect elsewhere in the list.
Failure 11: Assuming normalisation equals collation
What goes wrong: Visually equivalent encodings are handled but language sorting still fails. Corrective move: Use normalization for its intended processing role and collation for human ordering. The useful diagnostic is not merely whether the program produces a result; it is whether the result supports the reader’s intended action and is consistent across the full collection. Keep a before-and-after fixture, test the translated interface, and review adjacent entries so the repair does not introduce a second, quieter defect elsewhere in the list.
Failure 12: Changing order without communicating
What goes wrong: Saved links, classroom handouts and cached lists differ unexpectedly. Corrective move: Publish relevant notes, invalidate stale keys and keep stable record identities. The useful diagnostic is not merely whether the program produces a result; it is whether the result supports the reader’s intended action and is consistent across the full collection. Keep a before-and-after fixture, test the translated interface, and review adjacent entries so the repair does not introduce a second, quieter defect elsewhere in the list.
Practice, transfer and self-check
Use the exercises as a workshop for students, editors or developers. For each, write down the original display text, the selected locale, the intended reader action, a suggested comparison policy, and a test that would falsify your assumption.
Practice 1: Swedish surname directory
Task: You are given åberg, anders, östlund and ängström appear together. The interface promises Swedish alphabetical order; the reader needs to find surnames quickly. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: Test the locale’s ordering of å, ä and ö instead of grouping all three mechanically with plain a and o. Ask a Swedish-reading reviewer to sort a controlled fixture and to explain the position of each name. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 2: German contact book
Task: You are given adler, ärger, auer and äpfel. The application offers a dictionary-style list and a separate phonebook-style list. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: These use cases can employ different collation options; do not assume one universal treatment of umlauts for every German directory. Expose the selected view and test both before applying either to a live contact search. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 3: Spanish regional school list
Task: You are given nacional, ñandú and nube. A Spanish-language parent searches an alphabet bar for a name beginning with ñ. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: Treat ñ according to Spanish collation conventions rather than silently changing the display name to n. Check the letter-heading route and not just the complete sort output. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 4: Turkish personal names
Task: You are given işık, i̇nci, i̇pek and irmak. A translator sees two kinds of I in Turkish and is tempted to lowercase using English rules. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: Turkish casing and collation require locale awareness; plain English lowercase conversion can lose distinctions. Test uppercase and lowercase lookup separately while preserving each person’s exact chosen spelling. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 5: Czech glossary
Task: You are given chata, cesta and hotel. The resource is a Czech-language word index that learners browse alphabetically. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: Czech ch may be handled as a collation contraction; counting characters one at a time is not a sufficient implementation. Use a locale-specific fixture with neighbours around the ch sequence to verify real ordering. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 6: Danish town selector
Task: You are given aalborg, ærø, østerbro and århus. The user expects the interface’s declared Danish convention to govern the list. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: Avoid translating or romanising a town name merely to make sorting look familiar to English readers. Test explicitly whether legacy spellings and modern spellings are treated as expected by the chosen implementation. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 7: Chinese titles by pinyin
Task: You are given a group of chinese book titles with an approved pinyin reading for each. One readership expects pinyin order, while another needs a stroke-oriented browse mode. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: Collation is an editorial choice for navigation; a one-size-fits-all English transliteration is not the only option. Label the browse convention and test whether the data and library support the chosen mode. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 8: Japanese book catalogue
Task: You are given mixed hiragana, katakana and kanji titles. The catalogue includes both written titles and optional pronunciation readings. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: Readers may need a pronunciation-oriented search alongside a preserved original-script display; generic byte order cannot deliver that experience. Keep approved readings separately and record when a title lacks authoritative reading data. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 9: French accent distinction
Task: You are given cote, côte, coté and côté. A language learner expects spelling to remain visible while the list follows the chosen locale’s order. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: Do not strip accents from published headwords simply because an accent-insensitive search mode is available. Compare strict and forgiving searches and make the difference clear in the results. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 10: Greek titles
Task: You are given names with base letters and tonos accents. The collection serves Greek-reading users and allows readers to search without diacritics. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: A display sort and a forgiving query can use different settings without rewriting the stored titles. Test both accented and unaccented queries and show the original matched title. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 11: Arabic classroom headings
Task: You are given arabic headwords with optional vowel marks. The learning list contains annotated and unannotated forms of some words. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: Treat the presence or absence of marks as a deliberate educational feature; a search index may be forgiving while the teaching display must be accurate. Ask a qualified reviewer whether visually similar entries are linguistically interchangeable in that exercise. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 12: Hindi and Devanagari entries
Task: You are given learner entries include consonant clusters and distinct matras. An English-only developer attempts to alphabetise by code-point sequence. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: Unicode encoding order is not a substitute for the chosen Hindi collation and curriculum conventions. Use local-language fixtures and judge navigation from the learner’s perspective. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 13: Lesson numbers
Task: You are given lesson 2, lesson 10, lesson 003 and lesson 11. A parent is scrolling a digital workbook list. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: A numeric comparison option may be more usable than lexicographic order, yet leading-zero labels might have an archival significance. Decide whether the interface sorts by lesson index or exact title and document the rule. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 14: Product sizes
Task: You are given size 2, size 8, size 10 and size 10a. A translated shopping filter is mixing textual model IDs with numeric sizes. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: A numeric sort may help for sizes but must not turn distinct alphanumeric models into the same record. Test the intended parsing of units and suffixes without rewriting identifiers. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 15: Accented author search
Task: You are given josé alvarez and jose alvarez are separate records. A reader types a name without the accent on a mobile keyboard. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: Forgiving search can suggest both entries, but the system should not silently collapse two distinct people. Display full names and other disambiguating information, then permit an exact refinement. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Practice 16: Punctuation in titles
Task: You are given the river, river, the-river and river!. A literary index chooses how initial articles and punctuation affect placement. Should you change the published spelling, use a dedicated search key, change the locale comparator, or revise the browse interface?
Model reasoning: The choice to ignore a leading article is an editorial indexing rule, not a property of translation itself. Make title-sort keys separate from display titles and test punctuation-sensitive and punctuation-insensitive views. This is not solved by making every string look English. First preserve identity and spelling; next choose the operation; finally verify one or two neighbouring entries under the relevant locale. A good answer explains what the reader sees, what remains stable in storage, and how the result would be tested by someone who understands the language.
Frequently asked questions about locale-aware multilingual sorting
Does CLDR translate words automatically?
No. It supplies locale data and rules, including collation data, that software can use. Human-approved or otherwise checked translations still need to be supplied. An accurate translation can be sorted badly, and a correct comparator cannot repair a mistranslated term.
For an actual project, document the exact collection, desired reader task and comparison settings before changing code or translations. A language-aware reviewer should see the test cases, because a result that works for one example can still fail across a full alphabet or across a mixed-script catalogue.
Is Unicode code-point order good enough?
It may be useful for machine-internal deterministic operations, but not as the default promise of human alphabetical order across languages. Code points encode characters; readers follow language-specific conventions. A user-facing index needs a declared rule.
For an actual project, document the exact collection, desired reader task and comparison settings before changing code or translations. A language-aware reviewer should see the test cases, because a result that works for one example can still fail across a full alphabet or across a mixed-script catalogue.
Are accents always ignored in search?
No. Some search experiences tolerate missing marks, but the relevant language may distinguish words or names importantly. A reasonable design can offer forgiving discovery while preserving exact names and an exact-match option.
For an actual project, document the exact collection, desired reader task and comparison settings before changing code or translations. A language-aware reviewer should see the test cases, because a result that works for one example can still fail across a full alphabet or across a mixed-script catalogue.
Can I remove punctuation before sorting?
Only as a deliberate, tested indexing policy. Punctuation may be insignificant for some browse views but meaningful in titles, identifiers, quotations and personal names. Keep the original form and distinguish comparison behaviour from editorial correction.
For an actual project, document the exact collection, desired reader task and comparison settings before changing code or translations. A language-aware reviewer should see the test cases, because a result that works for one example can still fail across a full alphabet or across a mixed-script catalogue.
What does a collation strength control?
It controls how different levels of string distinctions contribute to comparison, subject to the implementation and locale. Do not assume a single global rule about accents or case. Inspect the selected library’s documentation and verify with fixtures.
For an actual project, document the exact collection, desired reader task and comparison settings before changing code or translations. A language-aware reviewer should see the test cases, because a result that works for one example can still fail across a full alphabet or across a mixed-script catalogue.
Why are two visually identical words stored differently?
Some characters can be represented with different Unicode sequences. Normalization can help with text processing, but it should be managed explicitly. The editorial display form and the comparison policy remain separate choices.
For an actual project, document the exact collection, desired reader task and comparison settings before changing code or translations. A language-aware reviewer should see the test cases, because a result that works for one example can still fail across a full alphabet or across a mixed-script catalogue.
Should Chinese titles always use pinyin sorting?
Not always. Audience, region, catalogue purpose and available approved reading data matter. Alternative collation modes may be appropriate. A labelled pinyin view can be useful without implying that all Chinese-language browsing must use it.
For an actual project, document the exact collection, desired reader task and comparison settings before changing code or translations. A language-aware reviewer should see the test cases, because a result that works for one example can still fail across a full alphabet or across a mixed-script catalogue.
What is the best way to sort translated vocabulary lists for children?
Use the learner’s target language and the convention taught in that learning context. Include a short explanation of unusual letter groups, and make search available so students can confirm the original spelling rather than guess where an item belongs.
For an actual project, document the exact collection, desired reader task and comparison settings before changing code or translations. A language-aware reviewer should see the test cases, because a result that works for one example can still fail across a full alphabet or across a mixed-script catalogue.
Why can a search find a name but place it strangely?
Search ranking and display collation are separate stages. A query may retrieve candidates by relevance and then break ties using a locale-specific comparator. If the two operations are not coordinated, results can feel arbitrary.
For an actual project, document the exact collection, desired reader task and comparison settings before changing code or translations. A language-aware reviewer should see the test cases, because a result that works for one example can still fail across a full alphabet or across a mixed-script catalogue.
How can we test multilingual alphabetical order?
Use controlled minimal pairs, include accented and unaccented forms, test multiple scripts, compare expected and actual orders, and ask a reviewer proficient in the target language. Save fixture data and implementation version so regressions are detectable.
For an actual project, document the exact collection, desired reader task and comparison settings before changing code or translations. A language-aware reviewer should see the test cases, because a result that works for one example can still fail across a full alphabet or across a mixed-script catalogue.
Do I need one sort order for all languages?
No. A multilingual application may expose language-specific ordering or separate groupings. What matters is a consistent, understandable choice within each view, and a stable underlying identity so an entry remains the same object as the order changes.
For an actual project, document the exact collection, desired reader task and comparison settings before changing code or translations. A language-aware reviewer should see the test cases, because a result that works for one example can still fail across a full alphabet or across a mixed-script catalogue.
When should a translator escalate a sorting problem?
Escalate when a name disappears, two distinct identities merge, an official record is corrupted, a teaching convention is broken, or a user’s navigation repeatedly fails. Provide examples, current locale, expected behaviour and the observed output to the development team.
For an actual project, document the exact collection, desired reader task and comparison settings before changing code or translations. A language-aware reviewer should see the test cases, because a result that works for one example can still fail across a full alphabet or across a mixed-script catalogue.
Further reading, translation routes and sources
For the normative technical background, consult Unicode CLDR Part 5: Collation, the CLDR collation guidelines, and Unicode CLDR core locale documentation. These references describe the data model and rationale; any actual library’s version and options still require testing.
Continue to eduKateSG’s Vocabulary Learning Hub for understanding how words and forms are organised in learning, and How English Works for comprehension and language structure. For software localisation where plural forms and placeholders are the main difficulty, see GNU gettext context, plurals and placeholders. This article’s canonical job is multilingual sorting and searching; it does not replace the translation of meaning, register, idioms or terminology addressed elsewhere in the series.
Final method: translate the text faithfully, keep its identity stable, select the reader’s locale and task, test the comparator with challenging examples, and explain the chosen order. A multilingual list is successful when people can find the right item confidently, not merely when a programmer can reproduce an array order.