Comment Extraction

The Problem

Comments are the collaboration layer of Office documents. They carry the context that matters most: who raised a concern, when they raised it, what specific text they were responding to, and the threaded conversation that followed.

PDF has no concept of comments. Converting to PDF silently erases the entire review history.

Comments are stored in dedicated XML parts (word/comments.xml, xl/comments1.xml, ppt/comments/comment1.xml) with range markers linking them to the exact content they annotate. No PDF-based parser can access them because they never survive the conversion.

No other parser extracts Office comments with full attribution. See the comparison table below.

How It Works

AILANG Parse reads the Office XML directly — no intermediate conversion, no rendering, no PDF step. It opens the OOXML ZIP archive and extracts comment data from the native XML parts.

DOCX (Word)

Word stores comments in word/comments.xml as w:comment elements, each with an w:id, w:author, and w:date attribute. In the document body (word/document.xml), w:commentRangeStart and w:commentRangeEnd markers bracket the exact text the comment is attached to. AILANG Parse correlates these ranges with the comment content to produce fully anchored output.

Threaded replies are not expressed in comments.xml at all — they live in word/commentsExtended.xml, which maps each comment's final paragraph id (w14:paraId) to its parent's via w15:paraIdParent, along with a w15:done resolved flag. AILANG Parse resolves that chain back to comment ids, and a reply inherits its parent's anchor.

Where an anchor cannot be resolved with certainty — a comment body with no matching range, a range that is never closed, an id present in one part but not the other — the comment is returned with anchored: false and an empty anchor rather than being attached to nearby text. A comment confidently pinned to the wrong clause is worse than one with no target, because nothing signals that it was a guess.

PPTX (PowerPoint)

PowerPoint has two comment formats and AILANG Parse reads both. Legacy decks use ppt/comments/comment1.xml with authors resolved from ppt/commentAuthors.xml; current PowerPoint writes ppt/comments/modernComment_*.xml with GUID author ids resolved from ppt/authors.xml. Supporting only the legacy shape means silently dropping every comment made in a recent version of PowerPoint.

Which comment belongs to which slide is taken from the slide's own relationships, not from matching comment1.xml to slide1.xml by number — that coincidence holds for simple decks and mis-attributes on the rest. A PowerPoint comment is pinned to an x/y point on a slide rather than to a run of text, so the anchor we report is the slide and its opening line; we do not claim a text range the format does not carry.

XLSX (Excel)

Excel never separates a comment from its anchor: each comment carries a ref attribute (e.g. B5) naming its cell directly, so no correlation pass is needed. Legacy notes live in xl/comments1.xml with an authorId into the authors list. Modern review threads live in xl/threadedComments/, with display names in xl/persons/ and parentId linking replies.

Modern Excel writes both, the legacy part being a compatibility shim whose author is a placeholder like tc={guid}. AILANG Parse prefers the threaded version and drops the duplicate, so a two-message thread reports as two comments and not three. Alongside the cell reference we resolve the cell's own value, so a comment arrives attached to B2: 184000 rather than to a bare coordinate.

What Gets Extracted

PPTX & XLSX Comments

Comments work differently in each Office format, but AILANG Parse normalizes them into the same block structure.

PowerPoint Comments

PowerPoint comments are positional — placed at x/y coordinates on a slide rather than anchored to a text range. They are commonly used during presentation review ("Move this chart to slide 3", "Font too small for projection"). AILANG Parse extracts the comment and associates it with the slide where it appears.

Excel Comments

Excel comments (called "Notes" in modern Excel) are anchored to specific cells. They often contain data validation notes, formula explanations, or review feedback on specific values. AILANG Parse extracts the comment text, the author, and the cell reference so you know exactly which data point the comment refers to.

Modern Excel also has "Threaded Comments" (introduced in Office 365), which are stored in a separate xl/threadedComments/ part with richer metadata including timestamps and reply chains.

Example Output

A DOCX with three review comments produces output like this:

{
  "blocks": [
    {
      "type": "Text",
      "content": "The quarterly revenue target of $4.2M was met."
    },
    {
      "type": "Text",
      "content": "[Comment by Sarah Chen on 2026-01-15T09:32:00Z] on 'quarterly revenue target of $4.2M': Should we break this down by region? The EMEA numbers look flat."
    },
    {
      "type": "Text",
      "content": "[Comment by James Park on 2026-01-15T10:14:00Z] (reply to Sarah Chen): Good call. EMEA was $1.1M vs $1.3M target. Adding a regional breakdown table."
    },
    {
      "type": "Text",
      "content": "Operating expenses remained within the approved budget."
    },
    {
      "type": "Text",
      "content": "[Comment by Legal Review on 2026-01-16T14:02:00Z] on 'approved budget': Need to clarify which board approval this refers to. Add the resolution number."
    }
  ]
}

Each comment preserves the author, timestamp, the specific text it was anchored to, and reply threading. This is the raw collaboration history that PDF conversion destroys.

XLSX Example

{
  "blocks": [
    {
      "type": "Table",
      "content": "| Region | Q1 Revenue | Q1 Target | ... |",
      "metadata": { "sheet": "Revenue" }
    },
    {
      "type": "Text",
      "content": "[Comment by Audit Team on cell D7]: This figure doesn't match the GL export. Verify against SAP report #4821."
    }
  ]
}

Use Cases

Try It

Extract comments from any Office document:

# Parse a DOCX with comments
./bin/docparse contract_reviewed.docx

# Parse via the API
curl -X POST https://docparse.ailang.sunholo.com/api/v1/parse \
  -H "Content-Type: application/json" \
  -d '{"filepath": "contract_reviewed.docx", "apiKey": "dp_YOUR_KEY"}'

# Parse an XLSX with cell comments
./bin/docparse budget_annotated.xlsx

Parse in Browser    API Reference

Format Guides

Comments are extracted from DOCX, PPTX, and XLSX files. See the full format guides for everything AILANG Parse extracts:

DOCX Parsing →  ·  PPTX Parsing →  ·  XLSX Parsing →