How do you translate a web archive without changing the web pages that were captured? A professional WARC 1.1 localization workflow distinguishes the original historical evidence from the multilingual descriptions, access interfaces and explanatory material that help people use it. Web archiving institutions preserve records of responses, requests, capture dates and other information precisely because a website may later change or disappear. Rewriting the captured page to make it read naturally in another language can erase the difference between what a site actually said and what an interpreter says it meant.
If you are looking for WARC translation, ISO 28500 web archive localization, WARC/1.1 record headers, WARC-Target-URI rules, WARC-Date, WARC-Record-ID, HTTP payload translation, revisit records or multilingual web archive replay, the key principle is simple: keep the captured bytes and identifiers intact, and place translated language in an explicitly separate descriptive layer. This guide explains which text can be translated, which values must remain exact, and how to check that a readable translation has not damaged a preservation record.
This is a practical guide for archivists, historians, librarians, public institutions, digital-humanities teams, translators and anyone publishing archived websites to international audiences. We will move from the structure of a WARC record to the harder judgments about dates, redirects, forms, multilingual pages, incomplete captures and ethical presentation. The aim is not merely to produce fluent prose. It is to help a reader understand a preserved event while still being able to identify, retrieve and question the evidence behind every explanation.
What a WARC file actually preserves
A WARC file is a container of records, not a translated edition of a website. A record starts with the version declaration and named header fields, followed by a blank line and a body whose byte length is declared. A response record commonly contains an HTTP response message; a request record may preserve what a crawler asked for; a metadata record may hold information about the capture. The surrounding record makes the content traceable. A multilingual catalogue or exhibit can describe the capture, but that description is not a replacement for the recorded exchange.
Three things that look like ordinary words have different jobs. A WARC field name identifies structure, a header value such as a timestamp or URI identifies a machine-readable fact, and a caption in a discovery interface communicates with a human. Translate the caption as appropriate. Do not casually translate the field name or change the value. When someone encounters WARC-Type: response, the token response is a defined record type. It is not the curator’s free-form description of how the historical website responded to a question.
The eight record types and their boundaries
WARC 1.1 defines warcinfo, response, resource, request, metadata, revisit, conversion and continuation. Those values are technical classifications and should not be translated within the WARC-Type field. Their explanations, on the other hand, can be translated in a visitor-facing guide. A response records retrieved protocol information, whereas a resource record can hold an otherwise captured resource. A revisit record can point to previously captured content when content reuse is appropriate. A continuation record plays a role when long records are segmented.
A useful translation glossary therefore has two columns: immutable token and human explanation. The row for revisit might preserve the machine token while giving a clear sentence explaining that a later capture may refer back to an earlier payload. The row for conversion can explain an archival transformation without implying that the historical author performed it. These distinctions stop readers from mistaking archival processing for a statement made by the archived website.
Record identity, target identity and capture identity
WARC-Record-ID identifies a particular archive record. WARC-Target-URI identifies the resource being captured when applicable. WARC-Date records the archival record time in a standardized form. They answer different questions and none is a title that a translator should embellish. A public catalogue may give an accessible translated title such as ‘Municipal flood advice, captured in spring’, but it must not place that title in the WARC-Target-URI or suggest it was the resource’s original address.
The safest public display shows a stable record citation and separately states the original target address, capture time and the language of the accompanying interpretation. A researcher should be able to move from a translated summary back to the archived record without reverse-translating any value. This matters when the same page has been captured many times, when a domain was sold, or when a public notice changed between two versions.
Preserve bytes before making language readable
An HTTP response body may contain HTML, CSS, JavaScript, an image, a PDF or compressed material. The Content-Length in the WARC record describes the stored block. Editing a captured sentence inside the block can change byte counts, digests and the evidential meaning of the archive. Even when an editor carefully updates every checksum, the modified content is no longer the same captured object. A translated presentation is better stored and linked as a derivative, with its own description and provenance.
A side-by-side viewer can place a translated explanation beside an unchanged replay. The user may switch between views without the translation overwriting the original. If an accessibility need requires an adapted reading layer, say that it is an adaptation, identify the source capture and document omissions. The distinction between faithful access and altered evidence is more important than whether a particular text-processing tool can technically rewrite a file.
HTTP messages are evidence, not prose fields
HTTP status codes, response headers, request methods and locations form part of what the archive recorded. A 301 redirect is not a completed visit to the redirected page. A 404 capture is not proof that a site never existed. A redirect destination may be another archived record, a live address or a broken target. Translating ‘Not Found’ in a curator’s explanatory note is fine; silently changing an archived status line or Location header is not.
Different layers may also disagree. A page’s visible heading can say ‘Report published on Monday’ while the capture metadata shows Tuesday. A server Date header can reflect server time rather than a human publication date. Treat these as separate observations. Describe each with its source instead of selecting the convenient one and rewriting the others to match.
Multilingual titles, captions and summaries
Multilingual access is most useful where the archive offers a human-facing title, search summary, collection description, subject term or exhibit introduction. A translation can convey the purpose of a page and define obscure local institutions without claiming to be the original language. Store the language, translator or editorial responsibility, date and source record together. If the historical document already contains several languages, record that fact rather than assigning a single language because the first paragraph appears in English.
Do not make a paraphrase look like a verbatim quotation. An exact translation of a quoted passage should still be marked as translated and cited to the original capture. If the original word is ambiguous, keep a short note explaining the possible readings. A smoother sentence is not always the more faithful one, particularly in legal notices, medical claims, political statements or rapidly changing emergency instructions.
Crawl scope, missing resources and reconstructed pages
A web crawl is a collection shaped by the crawler, its permissions, configuration and timing. It may miss images, scripts, embedded services, login-only pages or resources hosted on other domains. A replay can appear visually complete while silently substituting a live asset, or appear broken because a required resource was never captured. A translated exhibit must not confidently say that the historical page ‘contained no chart’ solely because the replay shows a blank region.
Describe what was captured, not everything that once existed. When a record is missing, label the omission. When curators supply a reconstruction, use language such as ‘reconstructed from available captures’ and distinguish it from a native replay. Translation has a role here: readers deserve understandable explanations of uncertainty, not a fabricated continuity that hides archival gaps.
Rights, privacy and ethical access
Archiving a page does not automatically grant permission to republish every personal detail in every language. Local legal rules, institutional policy, copyright, personal-data duties and risk to named people may affect what can be offered. The translation layer can increase discoverability of sensitive details even when the original language created a practical barrier. That possibility deserves separate review rather than being treated as a purely linguistic task.
Where access restrictions are justified, explain the decision without exposing the withheld information. Keep the underlying preservation history according to lawful institutional procedure, but do not publish a translated summary that defeats a redaction. Rights statements are not a universal ‘safe to translate’ flag. They describe conditions to interpret in the relevant jurisdiction and use context.
A trustworthy WARC localization workflow
Start with an inventory: identify the WARC files, record identifiers, capture dates, target URIs, record types and available replay paths. Then identify fields intended for human display. Map each editable description to a provenance record and its language. Translate a small pilot sample covering a normal page, a redirect, an error response, a revisit and an incomplete capture. Review these before scaling up, because exceptional records reveal design mistakes sooner than straightforward homepages.
Next, separate translation review from preservation validation. Linguists check wording, scope, names and uncertainty. Archivists check that record references and data relationships still resolve. Technical staff compare file integrity, test replay and inspect indexing. Finally, let a reader switch from the translation back to the source and understand why the two views differ. That reversible path is the clearest sign that accessibility and evidential honesty are working together.
Worked cases: reading actual archival problems without inventing content
A city emergency page captured twice
A town posts a storm warning at 08:00 and revises it at 11:00. The archive contains two response records with different capture dates. A translator preparing a multilingual exhibition is tempted to combine both sets of advice into one smooth summary because the later warning seems more complete.
Combining them would invent a text that existed at neither capture point. Instead, write two dated captions and explain what changed. Preserve both WARC-Record-ID references, and do not convert a time difference into a narrative claim about what officials knew at an earlier moment.
A reader should see which advice belonged to each captured version, whether the page itself stated an update time, and which timestamps come solely from the archive. The test is historical rather than stylistic: can the reader reconstruct the sequence without trusting the translator’s blended prose?
A redirect to a renamed government department
A request for an older agency address produces a redirect response to a new departmental website. The destination has a different institutional name and may even use a different language from the original site. A catalogue editor wants to replace the old name everywhere so modern visitors immediately recognize the authority.
The source address and the HTTP Location header are evidence of a transition. Retain both exact addresses, and use the translated catalogue note to connect the former and current institution. Do not imply that the new website’s text was captured merely because the archive contains a redirect pointing toward it.
Verify whether the destination was actually captured as a separate response. A good translation states ‘redirected to’ rather than ‘contains’, and uses dates to prevent a present-day department name from being projected into an older archival event.
A multilingual landing page with mixed scripts
A public library page has English navigation, an Arabic heading, French programme descriptions and a small language switcher. The archive displays only the English button labels prominently. Automated language detection assigns the entire capture to English, creating a poor international search record.
Describe the actual language distribution at the level supported by the archive and catalogue. Translate the visitor-facing summary into the target language, but do not replace the historical mix with a monolingual fiction. A transliterated title can aid discovery when labelled as such; it should not erase the title in its original script.
Test search against original-script phrases and translated keywords. If the captured switcher leads to pages outside the crawl, state that those language versions were linked, not necessarily preserved. The difference matters for researchers studying multilingual public access.
An image omitted by the crawler
A historical newspaper page appears in replay with a missing banner. Its HTML contains an image reference, but the image was served from an external host not included in the crawl. A translator sees nearby text describing a protest and proposes a modern photograph as a visual substitute.
A replacement image could make the archived page appear to contain evidence it never displayed. If an illustrative image is essential, place it outside the replay and clearly credit it as a later addition. Describe the missing original asset and record the unresolved resource address separately.
Check the capture manifest, WARC target records and replay requests before concluding the image is absent. A display-layer translation should help the reader distinguish lost layout from documented historical content, rather than conceal the absence.
A 404 page whose message changed over time
A university website returns a 404 for an old course description. The 404 body contains a message in the local language and a link to the current catalogue. An editor proposes labelling the course as ‘cancelled’ because the old URL no longer resolves.
A 404 response proves only what the server returned at the recorded request. The programme might have moved, been renamed or been restricted. Translate the error message as an error message and separately explain the uncertainty. Preserve the HTTP status and target URI even if the visitor interface uses a friendly label.
Validation asks whether the translated page distinguishes ‘not found at this address and time’ from ‘never offered’. The reader should be able to follow the archive citation and inspect the status rather than inheriting a speculative institutional history.
A revisit record pointing to an earlier payload
A repeated crawl finds unchanged bytes and stores a revisit record referring to previously archived content. A curator wants to display it as a complete independent translation of a later version. The record’s timing is later, but the preserved payload may be shared through a reference.
Explain the later capture event without inventing new textual content. Follow the revisit relationship, verify the referent and expose the correct source of the replayed payload. Keep WARC-Refers-To information unchanged, and do not translate a digest or reference identifier as a prose label.
The user should understand the difference between a later observation and a newly stored body. For quality assurance, confirm that the replay system can resolve the reference and that the caption accurately states whether content was reused.
A legal notice with an ambiguous term
A historical consumer notice uses a local legal term for a protective order. The target language offers a general word that sounds familiar but carries a different legal consequence. A fluent literal translation could mislead readers into believing an injunction or judgment was issued.
Preserve the original term beside a carefully qualified explanation, especially when institutions differ across jurisdictions. The translation can describe the function the notice appears to serve, but it should not manufacture a legal equivalence. Record the translator’s interpretation separately from the captured wording.
Where possible, consult period-appropriate legal references rather than importing today’s terminology. The archived record establishes what was written, not automatically what legal outcome followed. That boundary belongs in the caption, footnote and searchable summary.
A form that submits to a live service
An archived application page displays input fields and a submit button. In replay, the action points to a live endpoint or to an address not captured. A multilingual interface team wants to translate the labels and let visitors try the historical process for educational purposes.
Clearly present the interface as a historical form. Disable or isolate actions that could send data to a live service, and avoid describing a successful submission when none is preserved. If a demonstration is built, label it as simulated and keep it outside the original captured body.
Test every action link and form target, especially when language-specific versions add extra parameters. The translation should communicate the purpose of the form while preventing the false impression that the institution still accepts applications through the archived interface.
Server and page dates that disagree
A captured article claims an April publication date, an HTTP Date header shows a later day, and WARC-Date records the crawl in May. A translator plans to print one simple date beside the item to keep a timeline clean.
Separate asserted publication date, server response time and archive capture time. Each answers a different question and may reflect a different clock. State the source of every date in the public display, especially where politics, disaster response or medical guidance requires precise chronology.
Review time-zone conversions and daylight-saving assumptions. The safest translation leaves machine timestamps in their standardized values and adds human-readable local equivalents only as labelled display fields. Never silently shift the archived timestamp to match a preferred time zone.
A changed domain with the same logo
A local charity moves from one domain to another but continues using the same branding. Two WARC captures share visual elements, and the discovery system incorrectly treats their addresses as identical. A translator has been asked to write one institutional history from both.
Identity cannot be established from a translated name or logo alone. Preserve original target URIs, capture times and redirects. Explain that the pages appear related while separating what the archive directly shows from what external organizational records confirm.
Check whether institutional legal identity changed as well as the domain. A cautious summary can help researchers discover both captures without merging distinct record identities, and can link to additional contemporary sources where the evidence supports the relationship.
A robotic exclusion or restricted capture
A web archive’s collection contains references to a site but no full response body because of crawl restrictions, authentication or local access rules. The translator is asked to write a full-language synopsis based on descriptions from a later mirror.
Describe the archival limitation and cite the later source as separate evidence. Do not present a reconstructed synopsis as if it came from the missing WARC payload. A useful access note distinguishes excluded content, unreachable resources and content withheld by policy.
Check the collection’s rights and access statements. The multilingual catalogue must not disclose protected information by importing text from an unauthorized copy. Sound archival practice makes gaps visible instead of hiding them beneath confident prose.
A page containing quoted speech
A historical campaign page quotes a public speaker with punctuation and emphasis that do not transfer smoothly into the target language. A translator wants to paraphrase the speech inside quotation marks for readability.
Use a faithful marked translation when quoting, and reserve paraphrase for material explicitly labelled as summary. Preserve the capture reference, speaker attribution as recorded and the fact that the quote appears on a website rather than in an authenticated transcript.
Test that the original words can be found in the archived body. If quotation marks or emphasis are editorial, explain them. An accurate translation can still be misleading if it upgrades a second-hand webpage quotation into a direct record of the speech.
A PDF attachment behind a button
A municipal website links to a PDF with meeting minutes. The HTML page was captured, but the attachment may or may not have been collected. A translator prepares a summary referring to the PDF’s resolutions as though they belong to the webpage itself.
Check for a distinct archived record of the PDF and verify its exact bytes before discussing its contents. Preserve the link target and document title, and indicate whether the PDF was archived, remains live or is unavailable. If a translated PDF is created, identify it as an access derivative.
The handoff must remain traceable: translated paragraph, referenced PDF, attachment capture date and original resource address. Without that chain, a reader cannot tell whether the explanation is based on historical evidence or a later version.
An embedded video that never replayed
A news page contains an embedded player whose script loads content from another service. The screenshot-like replay shows a black rectangle, while the page title suggests a filmed interview. A curator proposes describing the full interview from the headline alone.
The headline is evidence of a claim about an interview, not evidence of the entire video content. Translate the headline and describe the missing media cautiously. If a separate archived video is found, link it explicitly instead of silently treating it as part of the page capture.
Review network dependencies and archive indexes to determine whether any stream or poster image survived. The translated exhibit should distinguish an embedded player, a playable preserved video and an independently sourced transcript.
An advertisement added by replay software
A web replay includes a navigation toolbar, a capture selector and an archive service banner. A visitor mistakenly believes all these labels belonged to the original historical site. Translating them in the same style as the website could deepen the confusion.
Use visual and semantic separation between the replay interface and the archived payload. Translate archive navigation for accessibility, but clearly attribute it to the archive provider. Original page text should remain identifiable even when both layers use the same target language.
A usability test can ask a new reader to point to one element supplied by the archive and one supplied by the original site. If they cannot do so, the translation or presentation has collapsed the layers and needs redesign.
A screenshot extracted for teaching
A teacher wants to crop a WARC replay to highlight one disputed sentence. The crop removes the page header, its time context and a warning that the claim was tentative. The translated caption might make the excerpt look definitive.
Retain a link to the complete capture, label the image as an excerpt and supply enough surrounding context for a fair reading. Translation should not erase qualifications or transform a partial screenshot into a whole-page claim.
Check that the crop really matches the cited record and that no live resources were introduced by the replay environment. A compact teaching artefact is useful only if students can follow the evidence chain beyond the screen image.
A page whose text changed after JavaScript execution
An archived landing page builds its main content dynamically. The raw HTML contains a placeholder, while the replay displays text supplied by archived scripts or data endpoints. The translator has access to a text extraction that differs from the visual replay.
Document the extraction source and distinguish source markup from rendered content. Do not assume that either representation alone is a complete historical record. The choice of translation source should follow the research question and the available captured dependencies.
Compare the browser display, network records and extraction method. When the page requires missing scripts, mark the reconstructed result accordingly. A reader needs to know not just the translated sentence but the route by which the sentence appeared.
A capture from a disputed or renamed place
A tourism website uses a place name whose official status and preferred spellings differ among communities. The translator wishes to standardize the name to the current official map label, but the archive documents the historical wording.
Keep the historically recorded name in the transcription or quotation. In accompanying explanatory text, use a context-appropriate current label and explain the relationship without silently editing the source. Where necessary, make search aliases available for multiple names.
Check who used each name and when. The translation should help international readers locate the place while preserving the linguistic evidence that historians or local communities may need to examine.
A record with an incorrect charset declaration
A captured webpage declares an encoding that produces corrupted characters in a modern viewer. A translator sees nonsensical symbols and assumes the author’s language is garbled. The archive may still contain bytes that can be interpreted with a different character set.
Investigate the encoding before translating content or correcting the recorded response. A display-layer repair can apply an explicitly documented decoding interpretation while the underlying WARC bytes remain unchanged. Do not rewrite the HTTP charset header merely to make a local viewer behave.
Compare decoding candidates against surviving original-language phrases and typography. State when a text was recovered through interpretation. The resulting translation should explain the intervention rather than disguise an uncertainty as a simple editorial correction.
Sensitive personal names in an old directory
A historical community directory lists residents, dates and home addresses. Translating and indexing the page into a global language could make details easier to find than they were when originally published.
Separate archival preservation from unrestricted publication. Consider applicable rights, privacy rules, harm potential and the archive’s policies before translating names, descriptions or searchable abstracts. Where necessary, provide a high-level collection description without reproducing sensitive personal data.
Document the access decision for reviewers, and make sure external snippets or schema markup do not accidentally reveal restricted content. A technically accurate translation is not automatically a responsible publication.
A crawl with duplicate identical pages
A multilingual portal serves the same HTML at two query-string variants. The archive captures both, and a translator wants to collapse them under one cleaned-up address because their visible text looks identical.
Retain the distinct target URIs and capture records. You may provide a discovery grouping for related records, but do not replace original identifiers. Query parameters can affect provenance even when rendered output happens to match.
Verify body digests, timestamps and HTTP headers before calling captures duplicates. A useful catalogue note can say that particular payloads appear identical while still maintaining the separate retrieval histories.
A partial WARC continuation sequence
A large resource was split across records, and one continuation is unavailable. A system replays only the initial segment, but a translator sees enough text to think the item is complete.
Mark the incompleteness and preserve segment identifiers and ordering. Do not fill the gap from a different copy without explaining its origin. A translator may summarize the surviving portion, but should not silently reconstruct the whole.
Validate the segment chain, declared lengths and total-length information where present. The public record should distinguish a confirmed complete capture, an incomplete capture and an editorial composite.
A checksum that fails after processing
An institution exports WARC records into another repository and a digest comparison fails. A localization team has recently edited descriptions in the package and assumes this harmless change caused the mismatch.
Determine exactly which bytes the digest covers. Changes to a separate catalogue can be legitimate, but changing stored record blocks or headers may invalidate fixity evidence. Keep the preservation object immutable while recording new descriptive metadata in its own managed location.
Do not simply recalculate a checksum to make an unexplained discrepancy disappear. Diagnose whether the difference is intended, accidental or malicious, and preserve the audit trail of the investigation.
A translated archive search result
An international scholar searches for a historic slogan in their own language. The system returns a record whose original content used a local phrase. The search snippet shows only the translation and gives no indication that the match is interpretive.
Show translated and original terms in a way that identifies the language and editorial layer. Search expansion can aid discovery, but it must not create the false appearance that the translated term occurs literally in the preserved payload.
Test exact-search and translated-search behavior separately. The results page should retain a direct path to the record and make clear whether the match came from captured text, OCR, metadata or a later translation.
A disputed translation corrected after publication
An exhibit’s translation of a historical statement is challenged by a community expert. The original WARC remains unchanged, but the public descriptive record must be corrected.
Version the translation and its editorial notes rather than overwrite the history of interpretation without explanation. Record who approved the correction and preserve the previous wording where institutional policy permits. The source record’s identity does not need to change.
Readers benefit from a succinct correction notice that explains what changed. This creates accountability without suggesting that the original archived page itself was edited or that an earlier capture magically contained the new translation.
A field-by-field validation checklist
Confirm that WARC/1.1, WARC-Type, WARC-Record-ID, WARC-Date, WARC-Target-URI, Content-Length and applicable digests retain the values from the preservation source. Verify that bytes in the stored capture have not been rewritten. Check request and response relationships, redirects, reference targets, continuation chains and any known truncation. For every translated display label, record the language, source capture, responsibility and date of the translation.
Try five reader tasks: identify the original target URL, distinguish capture time from publication time, find an original quoted phrase, understand a missing asset warning, and reach the unchanged source from the translated summary. If any task requires guessing which layer is original, the interface needs revision. Also test screen-reader labels and right-to-left scripts, and confirm that no translation crosses into a machine field accidentally.
Frequently asked questions
Can the WARC file itself contain translated text? A new archival record or separately documented derivative can contain a translation, but that is not a license to replace the original response body while presenting it as the same capture. Keep identifiers and provenance explicit.
Should I translate WARC-Type values? No. The defined tokens are protocol-level values. Translate explanations displayed to users, not the machine field contents.
Does WARC-Date tell me when a webpage was written? No. It identifies the time associated with the archival record. A webpage may display a different publication or revision date, and both should be described accurately.
Can a replay screenshot prove that the complete website was preserved? No. A crawl can miss resources or depend on later rendering behavior. Review the relevant captured records and describe coverage limits.
Can I improve search with translated keywords? Yes, when they are labelled as descriptive or interpretive metadata and maintain links back to the original capture.
What if the archive contains a harmful disclosure? Apply the relevant law, rights and access review rather than assuming that translation and indexing are always appropriate.
Further reading and connected subjects
Authoritative technical starting points include the International Internet Preservation Consortium’s WARC 1.1 annotated specification and ISO 28500:2017. For related archival concerns, read the eduKateSG guides to OCFL 1.1 object preservation, ResourceSync change records, and PREMIS preservation events. These adjacent guides address storage, exchange and preservation description; this WARC guide owns the distinct task of translating access to archived web exchanges.
Read the IIPC WARC 1.1 annotated specification and the ISO 28500:2017 standard catalogue for formal definitions. A trustworthy translated web archive is not one in which every historical trace is made to sound modern. It is one in which more readers can understand the trace without losing their right to inspect it.