A two-column PDF comes out with the columns interleaved
You select the body text of a paper or a journal set in two columns, copy it, paste it into Word — and one line of the left column is followed by one line of the right, over and over. Two separate arguments end up braided together, and you can no longer tell where a sentence ends.
The cause is that a PDF holds no structure for a column. Its contents are instructions: "draw this glyph at these coordinates". Nothing in the file records where one column begins and the other ends. Copying and text extraction walk the page top to bottom, line by line, in drawing position, so two columns sitting at the same height are picked up one line at a time, alternating.
So the columns have to be read back out of the layout. PDFIntact locates the blocks on the page — body text, headings, tables, figures — and puts them back in reading order before writing them out to Word, Markdown and JSON. The columns themselves are not reproduced: what you get is a single-column document in reading order.
Try it now
Check it on your own PDF
The check runs entirely inside your browser. Your file is never uploaded, and no account is needed. See how many pages, and how many tables and figures, can be extracted before you pay.
Example
行の拾われ方が変わります
左右に並んだ3行ずつの段が、どの順で並ぶかを示した図です。実際の文面ではなく、拾われる順序だけを表しています。
コピーしたとき（描かれた位置の順）
左の段 1行目 右の段 1行目 左の段 2行目 右の段 2行目 左の段 3行目 右の段 3行目
PDFIntact を通したとき（読み順）
左の段 1行目 左の段 2行目 左の段 3行目 右の段 1行目 右の段 2行目 右の段 3行目
形式ごとに、どう入るか
同じ読み順でも、書き出す形式によって入り方が違います。
Word（.docx）
読み順のまま、1段の段落として並びます。段組は再現しません。
Markdown（.md）
同じ範囲を、ページごとの見出しの下に読み順で置きます。
JSON
ブロックの種別・ページ番号・ページ内の座標つきで入るので、どの段のどこから来た文章かを後から辿れます。
Excel（.xlsx）
本文は入りません。Excelに書き出すのは表とグラフだけです。
脚注・キャプション・ページ番号
これらは本文とは別のブロックとして、読み順のその位置に段落で入ります。 2段組の論文では図のキャプションが本文と分けて取れること、 書籍のスキャンでは脚注の区切り線より下が別ブロックになりページ番号が本文に混ざらないことを、 それぞれ確認しています。Wordの脚注機能には入れません。段落として入ります。
Frequently asked questions
- Why do the columns interleave when I copy a two-column PDF?
- Because a PDF does not store a column as a column. The file only says "draw this glyph at these coordinates", with nothing to mark where one column begins and the other ends. Copying walks the page top to bottom, line by line, so two columns at the same height are picked up alternately, one line at a time.
- How does the reading order get fixed?
- The blocks on the page — body text, headings, tables, figures — are located first, then put back in reading order before anything is written out. The two-column layout itself is not reproduced: Word and Markdown receive a single-column document in reading order.
- What happens to footnotes and figure captions?
- They come out as their own blocks, as paragraphs at their place in the reading order. In our testing, figure captions were kept apart from the body text, everything below a footnote rule became its own block, and page numbers were not mixed into the body. They do not become Word footnotes; they arrive as paragraphs.
- Which formats can the text be exported to?
- Word, Markdown and JSON. JSON also carries the block kind, the page number and the position on the page, so you can trace which column and which spot a passage came from. Body text does not go into Excel: an Excel export contains tables and charts only.
- If the reading order is right, does that mean the characters are right?
- Those are two different things. The reading order can hold while the character reading breaks down, and we have seen a specimen where exactly that happened. That is why every block carries a confidence band — read, estimated, or unreadable. Check the output against the original PDF.
Check any PDF against all symptoms／ What each format contains