If you search for CSV delimiter QA, TSV translation file errors, or quoted CSV fields breaking after localization, the core problem is structural: commas, tabs, quotes and line breaks can be both data and delimiters. A translation can be linguistically perfect while one extra comma, missing quote or embedded newline shifts every later column.
A fast localization QA workflow therefore parses CSV and TSV as tabular formats instead of treating them as plain text. RFC 4180 documents common CSV conventions such as comma-separated fields, CRLF records, double-quoted fields for commas, quotes and line breaks, and doubled quotes inside quoted fields. Real-world tools vary, so the project still needs an explicit dialect and import/export contract.
This guide explains how people translate quickly by validating delimiters and quoting before spreadsheet-style localization files are delivered. It covers CSV, TSV, quoted fields, embedded commas, tabs, newlines, headers, field counts, Excel, locale-specific separators, formulas, BOM, UTF-8, line endings, CAT round trips, parser choice, CI and safe remediation.
The owner job is narrower than Spreadsheet Safeguards. Spreadsheet Safeguards protects formulas, IDs and sheet structure broadly. Delimiter & Quoting QA owns the serialization grammar that keeps each translated value in the correct row and column.
1. CSV Is Structured Text, Not Plain Prose
A comma can separate fields or appear inside a field depending on quoting. A line break can end a record or belong inside a quoted multiline value.
That ambiguity is resolved by CSV grammar, not by visual inspection. Always parse before transforming. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
2. RFC 4180 Is a Useful Baseline
RFC 4180 documents a widely used CSV model with comma delimiters, CRLF records and double-quoted fields. It is informational rather than a universal mandate, but it provides a strong interoperability reference.
Projects can use another dialect if they document it. QA needs an explicit dialect, not assumptions. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
3. Field Counts Should Stay Stable
Each record should normally contain the expected number of fields. A translated comma that is not quoted can create an extra column.
A missing delimiter can merge two columns. Count parsed fields per row rather than raw comma characters. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
4. Headers Need the Same Field Structure
A header row should align with the data columns it names. If translation or export changes header quoting, downstream column mapping can fail.
Validate header presence, field count and names before importing target rows. Schema starts at row one. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
5. Quoted Fields Protect Embedded Commas
A field containing a comma should be enclosed in double quotes under the common CSV convention. This is especially important because target-language punctuation can introduce commas that source text did not contain.
Do not forbid commas in translation merely to protect structure. Use correct quoting instead. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
6. Embedded Double Quotes Need Doubling
Inside a double-quoted CSV field, a literal double quote is commonly escaped by doubling it. Replacing smart quotes, straight quotes or speech punctuation can change whether escaping is required.
Use a serializer rather than manual quote counting where possible. Human language should not have to avoid quotation marks. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
7. Embedded Newlines Are Valid in Quoted CSV
A translated cell can contain a paragraph or address with a line break. When the field is quoted correctly, the newline belongs to the cell rather than ending the record.
Raw-line QA that assumes one row per physical line will misdiagnose valid files. Parse logical records. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
8. Unquoted Newlines Break Rows
If a newline appears inside an unquoted field, many parsers treat it as a record boundary. One accidental hard return can shift every following value.
QA should detect multiline content whose quoting is inconsistent. Target text editors are common sources of accidental line breaks. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
9. TSV Uses Tabs as Delimiters
TSV typically separates fields with tab characters rather than commas. That makes commas in translated prose harmless but turns literal tabs into structural hazards.
Do not use space-based visual alignment as proof of column integrity. Parse tabs explicitly. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
10. Tabs Inside TSV Values Need a Dialect Rule
Some TSV implementations do not support quoting tabs inside fields cleanly; others use CSV-like conventions or escapes. The project should define how literal tabs are represented.
Translators should not insert raw tabs unless the format supports them. Use a safer interchange format if values require complex tabular content. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
11. CSV and TSV Are Not Interchangeable Labels
A file named .csv can use semicolons or tabs; a .tsv can have custom quoting. Extensions are useful hints, not complete specifications.
Document delimiter, quote, newline and charset together. Dialect metadata prevents fragile guessing. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
12. Locale Can Affect Spreadsheet Delimiters
Some spreadsheet applications use semicolons instead of commas depending on locale settings because comma is used as a decimal separator. A file that opens correctly on one reviewer’s machine can split incorrectly on another.
Use explicit import settings or standardized export formats. Desktop spreadsheet behavior is locale-sensitive. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
13. Do Not Let Target Punctuation Define Delimiters
A French or German target can use punctuation differently from English. Serialization should remain independent of linguistic punctuation.
Quote the field according to the dialect rather than restricting natural target language. Content and structure should remain separate layers. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
14. Decimal Commas Can Look Like Delimiters
Numeric target values such as 1,5 can be mistaken for two CSV fields if not quoted or formatted under the correct dialect. Locale-number formatting and CSV structure can therefore interact.
Use parser-aware export and locale-aware numeric policy. Never repair the problem by changing numeric meaning. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
15. Thousands Separators Can Add Commas
A target locale or source data can include 1,000 or 10,000 within a field. CSV needs quoting if commas are literal content.
Number-format QA and delimiter QA should cooperate. One structural comma can corrupt an entire row. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
16. List Punctuation Can Expand After Translation
A source field without commas can become a target sentence with a comma-separated list. That is a normal linguistic change.
The serializer should quote it automatically. Manual flat-file editing is weakest when target punctuation changes. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
17. Quotation Style Does Not Replace CSV Quotes
Curly quotation marks are language punctuation, not field delimiters. A field surrounded by “smart quotes” is not necessarily quoted for CSV parsing.
Use ASCII double quote as required by the dialect while allowing smart quotes inside content. Typography and serialization are separate. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
18. Quoted Empty Strings Differ From Missing Fields
An empty quoted field “” can mean an explicit empty value, while consecutive delimiters can also represent an empty field depending on parser rules. Missing trailing fields and empty values may have different application semantics.
Schema validation should count positions, not only visible text. Blank is still structural data. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
19. Trailing Delimiters Need Policy
RFC 4180 states the last field should not be followed by another comma. Some real-world tools tolerate trailing delimiters and interpret them as an extra empty field.
Choose one rule and validate consistently. Loose acceptance creates ambiguous field counts. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
20. Leading Delimiters Mean Leading Empty Fields
A row beginning with a delimiter usually contains an empty first field. That can be legitimate or a sign that the key column disappeared.
Do not trim leading delimiters as whitespace. Column position carries meaning. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
21. Whitespace Around Delimiters Is Data
RFC 4180 notes that spaces are considered part of a field rather than automatically ignored. Importers differ in whether they trim.
Localization workflows should decide whether surrounding spaces are significant. Do not rely on spreadsheet display to reveal them. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
22. Unbalanced Quotes Can Corrupt Many Rows
One missing closing quote can cause a parser to absorb later delimiters and line breaks into the same field. The visible defect may appear dozens of lines after the actual mistake.
Parser errors and field-count anomalies should point back to the earliest unmatched quote. Structure-aware diagnostics save time. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
23. Extra Closing Quotes Can Be Just as Harmful
An unexpected quote can terminate a field early or create invalid syntax. Target-language quotation edits can accidentally interact with raw CSV editing.
Let a serializer escape content. Do not ask translators to reason about field syntax while writing prose. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
24. Manual Editing Should Use CSV-Aware Tools
A code editor can safely edit CSV only when the user understands quoting and delimiters. Spreadsheet tools provide visual columns but can also change encoding, dates and leading zeros.
Choose the editor based on which risks dominate. No tool is neutral. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
25. Excel Can Reformat Values
Spreadsheet applications may convert dates, scientific-looking IDs, leading-zero numbers or long numeric strings. This broader spreadsheet problem can combine with delimiter changes on export.
Protect identifier columns and use explicit import settings. Delimiter QA should be part of a larger spreadsheet-safe workflow. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
26. Excel Export Can Change Encoding or BOM
A CSV can return with UTF-8 BOM or another encoding after review. That can affect server imports independently of delimiter correctness.
Run BOM and encoding QA after spreadsheet round trips. Each serialization layer deserves its own check. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
27. CAT Tools Can Parse CSV Into Segments
Some CAT systems can import selected columns while preserving keys and metadata. This is usually safer than exposing delimiters to translators.
Test the filter configuration on a sample before production. Column mapping is part of localization engineering. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
28. CAT Export Can Requote Fields
A CAT tool may reconstruct CSV using its own serializer. The result can differ visually while remaining structurally valid.
Compare parsed records rather than raw line formatting only. Meaningful QA distinguishes harmless serialization from broken columns. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
29. Round-Trip One Sample Before Scaling
Import a representative CSV/TSV, translate one row containing comma, quote and newline edge cases, export and re-import it. If all fields return to the same columns, the filter is probably configured correctly.
Do this before handing over thousands of rows. Small tests have large leverage. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
30. Delimiter Count Alone Is Not Enough
A quoted field can contain many commas that are not delimiters. Counting raw separator characters produces false errors.
Use a real parser and compare parsed field counts. Regex is useful only for narrow supplementary checks. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
31. CSV Parsing Should Be Library-Based
Mature libraries understand quoted fields, embedded newlines and doubled quotes. Manual split(‘,’) logic fails on valid CSV.
Localization importers should use proper parsers. Simple code is not always simpler behavior. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
32. TSV Parsing Also Needs a Defined Dialect
split(‘\t’) can work for strict single-line TSV without embedded tabs. It fails once the format allows quoted multiline or escaped content.
Write the actual TSV rules down. Parser complexity should match the data complexity. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
33. Schema Can Validate Column Meaning
After parsing, verify required columns such as key, source, target, comment and status. A row with the right number of fields can still have shifted semantics if headers were mapped incorrectly.
Use names and types, not only positions. Structural correctness continues after delimiter parsing. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
34. Column Types Need Separate Validation
A key column, locale column, numeric ID and translatable text column have different rules. Do not run language transformations across every field.
Schema-aware import prevents formulas, IDs or statuses from becoming translatable content. Tabular localization works best with explicit field roles. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
35. Comments Columns Should Stay Out of Target
Reviewer notes can sit beside target text in CSV. A one-column shift can expose comments as user-facing strings.
Validate header-to-field mapping on import. Comment Leakage QA complements delimiter QA. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
36. Formula Columns Need Protection
Spreadsheet exports can include formulas or calculated fields. Translators should not alter them unless the workflow explicitly converts them to data.
Protect formula columns before localization. Broad spreadsheet safeguards still apply around delimiter integrity. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
37. IDs With Leading Zeros Need Text Handling
A resource key such as 00123 can become 123 in spreadsheet software. The row still has the correct delimiter count but identity has changed.
Mark IDs as text and validate after export. Delimiter QA cannot replace identifier preservation. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
38. Line Ending Policy Interacts With CSV
RFC 4180 documents CRLF records, while many real-world CSV files use LF. The parser should accept the project’s actual dialect.
Mixed line endings can still create tool differences. Run line-ending and CSV-structure QA separately. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
39. Multiline Cells Need Stable Line Ending Semantics
A quoted cell can contain CRLF or LF depending on producer and consumer. Do not convert embedded line breaks blindly when normalizing record endings.
Parser-level rewrite can preserve logical cell content. Raw replacement can damage data. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
40. Quotes in RTL Text Need Structural Separation
Arabic or Hebrew punctuation and bidi behavior can make quoted content visually complex. CSV delimiters remain logical code points regardless of visual order.
Use parser results rather than screen position. RTL visual QA should inspect rendered content after parsing. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
41. CJK Text Often Has Fewer ASCII Delimiters Naturally
That does not reduce structural risk because one imported comma or tab can still shift a row. Keep parser-based QA consistent across scripts.
Do not tune delimiter detection based on language appearance. Structure is language-independent. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
42. Unicode Comma-Like Characters Are Not CSV Commas
Full-width comma and other punctuation characters can look similar to ASCII comma but do not act as delimiters in standard CSV parsers. That can be linguistically correct inside target content.
Do not normalize punctuation into ASCII just to match delimiter appearance. Confusable punctuation and structural delimiter are different concepts. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
43. Tabs Can Hide in Copied Text
A translator can paste a tab from a spreadsheet or document into a TSV target value. The field then splits silently if the dialect has no quoting protection.
Scan target values for raw delimiter characters where they are forbidden. Clipboard data is a common entry route. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
44. Newlines Can Hide in Copied Text
Copying multi-line text into a single-line CSV cell can create an embedded newline. If the exporter quotes correctly, this may be fine; if not, row structure breaks.
Validate both field policy and parser output. User intent and serialization need to agree. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
45. Exported Quotes Can Become Smart Quotes
Some word processors replace straight quotes with typographic quotes. This can damage CSV syntax if the structural delimiter quote is transformed.
Do not run smart-quote conversion over raw CSV files. Typography transformations belong inside field content, not serialization. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
46. Semicolon CSV Needs Its Own Contract
Many European desktop workflows use semicolon-separated values. That is a real dialect and should not be treated as malformed comma CSV.
Document delimiter and quote character explicitly. Filename extension alone cannot tell the full grammar. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
47. Pipe-Delimited Files Need Similar QA
Some localization exports use | as a field separator. Translated text may also contain vertical bars for formatting or notation.
Apply the same principles: parser, quoting or escaping, schema, field count. The concept is delimiter integrity, not comma worship. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
48. Custom Delimiters Increase Handoff Risk
Every nonstandard dialect asks vendors and tools to configure import rules correctly. Use standard, parser-supported formats when possible.
If a custom delimiter is unavoidable, include a sample and machine-readable specification. Ambiguity costs more across organizations. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
49. MIME Type and Charset Metadata Can Help
RFC 4180 registers text/csv and allows charset and header parameters. Real deployments vary, but explicit metadata can improve interoperability.
Encoding, header and dialect should be documented together. Flat files need more context than their extension provides. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
50. CSV Injection Is a Separate Security Concern
Spreadsheet programs can interpret cells beginning with =, +, – or @ as formulas depending on context. Localization teams handling user-controlled content should follow the application’s security policy.
Do not solve injection by breaking delimiter syntax. Security and localization structure are separate owners. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
51. Do Not Prefix Apostrophes Blindly
Spreadsheet users sometimes prefix values with apostrophes to force text. That apostrophe can become part of exported data depending on the tool.
Use explicit column typing or safe import settings where possible. Ad-hoc spreadsheet tricks can leak into localization values. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
52. Parser Warnings Should Include Row and Column
A message such as ‘expected 5 fields, found 7 at record 213’ is immediately actionable. Include the raw excerpt and header names if safe.
Do not report only ‘invalid CSV.’ Diagnostics are part of productive QA. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
53. Compare Parsed Tables Before and After Translation
Load source and target into a structured table and compare row counts, key columns and schema. This catches shifts that visual spreadsheet review can miss.
Do not require source and target text cells to match length. Compare structure, not language. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
54. Keys Should Remain Aligned With Targets
If a delimiter error shifts columns, a target can become attached to the wrong key without obvious text corruption. Key-target pairing is therefore a high-value structural invariant.
Verify a stable ID column whenever the format provides one. Identity makes tabular localization safer. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
55. Sort Order Should Not Be the Only Identity
Spreadsheets are easy to sort accidentally. If the import relies only on row order, targets can map to the wrong source after review.
Use stable keys rather than position. Delimiter correctness cannot compensate for weak identity mapping. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
56. Duplicate Rows Need Separate Checks
A valid CSV can contain the same key twice. That belongs to Duplicate-Key QA rather than delimiter QA.
Keep owners narrow even when one file can fail many ways. Layered checks give clearer fixes. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
57. Missing Rows Need Key-Parity QA
A CSV can be perfectly quoted and still omit source records. Missing-key or row-parity checks detect that.
Do not treat successful parsing as proof of completeness. Structure has multiple dimensions. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
58. Extra Rows Need Stale-Key QA
Old target records can remain after source deletion. Delimiter QA will accept them because the syntax is valid.
Run stale-key checks after parsing. A clean table can still contain the wrong release set. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
59. Round-Trip Reimport Is a Powerful Final Check
Export the translated file and import it again using the same production parser. Compare row count, columns and key-target associations.
This verifies the actual artifact rather than the CAT workspace. Round-trip testing closes the handoff loop. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
60. CI Can Parse Every CSV/TSV Artifact
Build pipelines can reject malformed quoting, inconsistent field counts and duplicate headers automatically. Use the same parser configuration as production where possible.
Deterministic structure should not depend on manual review. Automation frees reviewers for language. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
61. Vendor Acceptance Can Be Machine-Checked
A returned localization package can be parsed before human QA starts. Reject or quarantine structurally invalid files immediately.
This prevents reviewers from working inside corrupted tables. Technical acceptance should precede linguistic acceptance. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
62. Final QA Should Inspect the Shipping File
A spreadsheet can look correct while a later export changes delimiter or quoting behavior. Validate the exact CSV/TSV sent to the product or client.
Serialized structure is part of the deliverable. Do not stop at the editor state. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
63. Fix the Exporter, Not Individual Broken Rows
If every multiline target fails, the serializer configuration is wrong. Repair the export pipeline instead of manually quoting hundreds of cells.
Systemic defects need systemic fixes. Localization QA should make patterns visible. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
64. Document the Dialect Once
Store delimiter, quote, escape, header, charset, newline and multiline policy in the project technical brief. Vendors and automation can then use the same contract.
Flat-file formats become reliable when their hidden grammar is explicit. Documentation reduces repeated experimentation. The correct decision depends on the declared CSV/TSV dialect and the parser that will consume the final artifact.
A practical review step is to parse the file, verify header and field counts, compare stable key columns, inspect quoted fields containing delimiter characters or newlines, and rerun the production parser after export. Do not repair structural errors by restricting natural target punctuation.
Practical CSV/TSV Delimiter & Quoting QA Workflow
Document delimiter, quote, escape, newline, header and charset policy.
Parse source and target using a real CSV/TSV library rather than raw string splitting.
Validate record counts, header names and field counts.
Verify stable key columns before and after translation.
Inspect fields containing commas, tabs, quotes and embedded newlines.
Run encoding, BOM and line-ending checks after spreadsheet round trips.
Re-import the exact delivery artifact with the production parser.
Worked Example: One Comma Moves Every Column
A localization CSV contains columns key, source, target and comment. An English source string has no comma, but the translated Spanish target naturally adds one. A translator edits the raw file in a text editor without adding quotes around the field.
The next import reads five fields instead of four. The text after the comma shifts into the comment column, while the original comment shifts out of the expected schema. The file is still readable, but the target is attached to the wrong structure.
Delimiter QA parses the file, reports the row’s field-count mismatch and shows the expected header mapping. The team switches to a serializer-driven export so translated punctuation no longer threatens column boundaries.
The fix preserves natural Spanish and repairs the serialization layer instead of asking translators to avoid commas.
Frequently Asked Questions
When should a CSV field be quoted?
Under the common RFC 4180 convention, fields containing commas, double quotes or CRLF should be enclosed in double quotes, with internal quotes doubled.
Can a CSV cell contain a newline?
Yes, in common CSV grammar when the field is quoted correctly. Use a parser rather than assuming every physical line is one record.
Why can the same CSV open differently on different computers?
Spreadsheet locale settings can affect delimiter interpretation, encoding and number conversion. Use explicit import settings and a documented dialect.
Is TSV safer than CSV?
TSV avoids comma collisions but raw tabs become structural delimiters. It is only safer when values and tools obey the TSV contract.
Can CI validate CSV/TSV structure?
Yes. Parsing, field-count checks, header validation and key-column parity are deterministic and should be automated for recurring releases.
Current CSV Guidance
RFC 4180 documents a common CSV format in which fields are comma-separated, fields containing commas, double quotes or CRLF are quoted, and double quotes inside quoted fields are doubled. It also notes that real implementations vary. See RFC 4180.
The localization lesson is simple: treat CSV and TSV as structured tabular formats with an explicit dialect. Parser-based validation is more reliable than counting delimiters by eye or restricting target-language punctuation.
Conclusion
CSV and TSV Delimiter & Quoting QA protects the row-and-column structure that lets translation data survive handoffs.
A fast workflow parses the file, validates schema and key alignment, preserves natural punctuation through correct quoting and round-trips the final artifact through the production importer.
When flat-file grammar is explicit, translators can write natural language without becoming accidental CSV engineers.
Continue the Translation Series
Read How People Translate Quickly | Spreadsheet Safeguards.
Read How People Translate Quickly | Mixed Line-Ending QA.
Read How People Translate Quickly | Mojibake and File-Encoding QA.