EditorialXML

editorial-production

From PDF to JATS XML: The Editorial Production Process

EditorialXML6 min read
  • jats
  • production
  • pdf

Most Latin American scientific journals receive manuscripts in PDF or Microsoft Word, but indexing repositories — SciELO, PubMed Central, Redalyc — require delivery in JATS XML. Closing the gap between the author's format and the archival format is the core of modern editorial production.

This article describes the complete PDF → JATS XML workflow, the editorial decisions at each stage, and the points where quality is won or lost.

Why PDF remains the starting point

Despite decades of promotion for structured formats, PDF remains the preferred delivery format for authors and peer reviewers. It is universal, preserves typographic layout, and requires no specialized tools. For editorial production, this creates a challenge: PDF is an opaque container where the article's semantic structure is implicit, not explicit.

Conversion strategies vary depending on journal maturity:

StrategyAdvantageLimitation
PDF → OCR + manual taggingWorks with any sourceSlow, error-prone
Word → automated XMLMore structure preservedQuality depends on author template
LaTeX → JATSIdeal for exact sciencesRequires specialized pipeline
Native XML editingHighest qualitySteep learning curve

Stage 1: Manuscript intake and assessment

Before converting, assess whether the manuscript is ready for production:

  • Has it completed peer review and author corrections?
  • Are figures at publication resolution (300 dpi for print, 150 dpi minimum for web)?
  • Do references follow the journal's style?
  • Are author metadata (ORCID, affiliations) complete?

An incomplete manuscript converted to XML produces incomplete XML. The golden rule: garbage in, garbage out.

Stage 2: Content extraction

From Word (.docx)

DOCX files are ZIP archives with internal XML (Office Open XML). Conversion tools can map Word styles (Heading 1, Heading 2, Normal) to JATS elements (<sec>, <title>, <p>). Quality depends critically on authors using consistent styles rather than direct formatting.

From PDF

Extraction from PDF is inherently more difficult:

  1. Native PDF (generated from Word/LaTeX): text is selectable and structure partially recoverable.
  2. Scanned PDF: requires OCR with mandatory human review.

In both cases, expect to correct manually: incorrect line breaks, corrupted special characters, fragmented tables, and unlinked figures.

From LaTeX

For mathematics and physics journals, the LaTeX → JATS pipeline preserves equations in MathML (<mml:math>) with greater fidelity than any conversion from PDF.

Stage 3: Semantic tagging

Extraction produces plain or semi-structured text. Semantic tagging assigns JATS meaning to each block:

<body>
  <sec sec-type="methods">
    <title>Materials and methods</title>
    <p>An observational study was conducted...</p>
    <sec>
      <title>Study design</title>
      <p>The design was cross-sectional...</p>
    </sec>
  </sec>
</body>

Editorial decisions at this stage:

  • How are sections named? ("Methodology" or "Materials and methods"?)
  • Do subsections require nested <sec> or a <title> within the paragraph?
  • Do equations go inline (<inline-formula>) or display (<disp-formula>)?

Document these decisions in an internal XML style guide to maintain consistency across issues.

Stage 4: Metadata and front matter

The <front> block is built in parallel with or after the body:

<front>
  <journal-meta>...</journal-meta>
  <article-meta>
    <article-id pub-id-type="doi">10.1234/rev.2026.001</article-id>
    <title-group>
      <article-title>Full article title</article-title>
    </title-group>
    <contrib-group>...</contrib-group>
    <abstract xml:lang="en">...</abstract>
    <kwd-group xml:lang="en">...</kwd-group>
  </article-meta>
</front>

Author normalization is the most labor-intensive subtask: separating surnames from given names, unifying duplicate affiliations, validating ORCID, and assigning roles (author, editor, corresponding author).

Stage 5: Bibliographic references

Converting references from PDF or Word to structured <element-citation> is frequently the bottleneck. Strategies:

  1. Automatic parsing with tools such as AnyStyle, Crossref API, or ParsCit.
  2. Cross-checking DOIs via Crossref to complete metadata.
  3. Human review mandatory for 10–20% of ambiguous references.

A correctly structured reference:

<ref id="ref1">
  <element-citation publication-type="journal">
    <person-group person-group-type="author">
      <name><surname>Silva</surname><given-names>Patricia</given-names></name>
    </person-group>
    <article-title>Study results</article-title>
    <source>Revista Brasileira</source>
    <year>2025</year>
    <volume>30</volume>
    <fpage>45</fpage>
    <lpage>52</lpage>
  </element-citation>
</ref>

Stage 6: Figures, tables, and supplementary material

Each figure requires:

  • Image file in appropriate format and resolution.
  • <fig id="fig1"> element with <label>, <caption>, and <graphic>.
  • In-text reference with <xref ref-type="fig" rid="fig1">.

Tables extracted from PDF usually need manual cell-by-cell reconstruction. Budget specific time for this work.

Stage 7: Validation and quality control

No article should advance without JATS validation. Errors detected at this stage are orders of magnitude cheaper to fix than those discovered after publication or during indexing.

Run validation in three layers:

  1. Well-formed — the XML parses without syntax errors.
  2. Valid against schema — complies with JATS Archiving.
  3. Business rules — meets SciELO, PMC, or other destination requirements.

Use the JATS validator for the first layer before applying repository-specific rules.

Stage 8: Derivative generation

Once JATS is validated, automatically generate:

  • HTML for the journal website.
  • PDF for archive and distribution.
  • ePub for mobile readers (optional).
  • Metadata XML for Crossref and DOI registration.

JATS XML becomes the canonical source format; all other formats are regenerable derivatives.

Indicative timelines and costs

For a standard research article (6,000–8,000 words, 5 figures, 40 references):

MethodEstimated timeRelative cost
Full manual conversion8–16 hoursHigh
Semi-automated + review3–6 hoursMedium
Automated pipeline + QA1–3 hoursLow (high initial investment)

Investment in automation pays off from the third or fourth annual issue onward.

When to outsource production

Outsource PDF → JATS conversion when XML is already costing you editorial time or reputation with indexers:

  • Recurring returns or reprocessing due to structure, metadata, or references.
  • Indexing deadlines with low tolerance for errors.
  • Need to comply with SciELO and/or PMC without dedicated XML staff.
  • Manuscripts in inconsistent formats and an internal workflow that does not scale (even modest volume — one or two dozen articles per year — is often enough to notice).

You do not need to "wait until you grow": a pilot issue shows real cost and time. The JATS XML production service manages the complete workflow with validation included and delivery in the formats your indexer requires.

Request a quote

To evaluate costs for your journal, request a quote indicating annual article volume, target repositories, and current manuscript format.

Conclusion

The path from PDF to JATS XML is not a simple format conversion: it is an editorial process that requires decisions about structure, metadata, and quality at every stage. Journals that treat it as a core competency — not a peripheral technical task — achieve more discoverable publications, smoother indexing, and workflows that scale with journal growth.

To get started, evaluate a pilot article with JATS XML production or request a quote tailored to your publication's volume.

Related articles