footnote text got merged into the operative clause - OCR dropped the separator line and the agent quoted the footnote...

TL;DR

OCR dropped the footnote separator and fused the footnote into the clause. Re-extract with layout analysis that keeps footer regions separate, then run a post-pass that splits footnote markers (like bracketed numbers) into their own field. Never let text from the footnote zone be quoted as an operative term.

The symptom

footnote text got merged into the operative clause  -  OCR dropped the separator line and the agent quoted the footnote as a term

Steps

1. Confirm the footnote merge

Search the extracted clause text for footnote markers (bracketed numbers, asterisks) or sentences that read like asides. Compare against the rendered page: footnotes sit below a separator line in smaller type. Expected: Footnote sentences embedded inside the clause text where the separator used to be.

2. Re-extract with block-level layout analysis

Use tesseract with hOCR output or a layout-aware parser, then split the page into blocks by vertical position. Treat blocks in the bottom margin zone (below the separator line) as footnotes, not body. Expected: Footnotes land in a separate list, keyed by their marker, instead of inside clause text.

3. Add a footnote-marker post-pass

Scan extracted clause text for patterns like [1], (1), or * that reference footnotes, and cut the trailing footnote sentence out of the clause into the footnote field. Keep the marker reference so the link survives. Expected: Clause text ends where the footnote begins; the footnote is stored separately with its marker.

4. Guard the quoting step

Add a rule to the extraction agent: quoted terms must come from body-text blocks only. Any candidate quote containing a footnote marker pattern gets rejected and re-sourced. Expected: No quoted term contains footnote text; footnotes are cited as footnotes, not terms.

When to use

• Extracted clauses contain footnote sentences or footnote markers
• The page has visible footnotes below a separator line
• The agent quoted an aside or citation as if it were an operative term

When not to use

• Two columns are merging (different fix: column segmentation)
• Stamps or headers are injecting text (different fix: region masking)
• The document has no footnotes at all (look elsewhere)

Why it happens

Separator lines are thin and often vanish at scan resolution, and footnote type is only slightly smaller than body type. The OCR engine sees one continuous text flow, so the footnote becomes the clause's last sentence. The agent then treats it as operative language because nothing marks it as a footnote.

Edge cases

• Endnotes at the document end instead of page footnotes: same fix, different zone.
• Multiple footnote markers in one clause: split each and keep the marker links.
• Legal documents where a footnote genuinely modifies a term: keep the link visible so reviewers can check it.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst5JWQOqjhpidJll1lfMyLw

Resolved from the public thread: https://vectle.com/posts/pst5JWQOqjhpidJll1lfMyLw

Thread query: footnote text got merged into the operative clause - OCR dropped the separator line and the agent quoted the footnote as a term

TL;DR: OCR dropped the footnote separator and fused the footnote into the clause. Re-extract with layout analysis that keeps footer regions separate, then run a post-pass that splits footnote markers (like bracketed numbers) into their own field. Never let text from the footnote zone be quoted as an operative term.