PDF table columns glued together? Fix the cells before you translate
Body text can look fine while tables fall apart—two columns merge into one line, or several lines inside one cell become a single blob. Often the model is not “wrong”; the table was already segmented incorrectly. Yibelin cuts structure first, then translates. For table-heavy files, preview with Table Enhanced.
Try Table Enhanced on a PDF with tablesWhat the result looks like
Click an image to zoom and check row and column alignment.
Documents this page is for
- Research result tables, inclusion/exclusion grids, multi-query keyword tables
- Product manuals and tech specs with parameter tables
- Lab reports and summary tables with many columns
- Contract annex tables with a clear grid
For two-column papers and captions, see paper PDF translation. Spec sheets mixed with TOC and callouts: manuals. Photo or scan-only tables need scanned PDF OCR first.
Why PDF tables break so easily
Most PDFs have no real table object. You see a grid; the program mostly sees text coordinates, lines, and whitespace. Reading order is not the same as cell order.
- Column sticky — two columns merge into one paragraph
- Multi-line cell glue — several lines inside one cell become one blob
- Far-apart merge — a number and a label on the same row, far apart, get joined into one sentence
- Over-split — one cell is torn into many tiny fragments
Half-drawn rules and very thin lines make multi-line cells harder. In preview, decide whether columns merged or lines inside a cell glued, then choose whether to re-run with Table Enhanced.
How Yibelin handles tables
- Locate text and table regions on the page
- Split by column and row where geometry allows
- Translate by paragraph and place text near the original spot
- Edit individual cells in the browser, then export PDF
Keep Scan Enhanced for most uploads. For papers, lab reports, and dense tables, switch to Enhanced (recommended) in Translation settings and compare a second run.
How we check quality
We evaluate segmentation on the public paper-table set SciTSR (~3000 pages). The focus is whether cells stay readable after translation—not detection-box scores alone.
- About 80% of test pages show no multi-cell sticky paragraphs (~79%)
- Heavily sticky pages drop to about the 2% range (~1.7%; earlier setup ~8%)
These are segmentation / sticky metrics, not word-level translation accuracy. Borderless, multi-page, and blurry scanned tables can still fail—check the preview before download.
A practical workflow
- Upload one or two sample pages with tables first
- In preview, watch headers, column glue, and multi-line cells
- If columns clearly stick, re-run with Table Enhanced
- Fix leftover cells in the editor instead of retranslating the whole file
Questions
Two columns become one line after translation?
That is classic column sticky. Check the preview; for table-heavy PDFs switch to Enhanced (recommended). Open with Table Enhanced
The grid looks fine—why are lines inside a cell still glued?
Lines you treat as cell borders may not all register as row separators, especially partial or thin rules. Try Table Enhanced and split leftover cells in the editor if needed.
A distant label and a number become one sentence?
Segmentation may have split them, then a later merge glued them again. If preview shows two pieces but the translation joins them, edit that cell—no need to re-run the whole document.
Do I always need Table Enhanced?
No. Stay on Scan Enhanced for ordinary files. Use Enhanced when result tables or multi-column grids clearly glue after the first pass.
Borderless or multi-page tables?
Open grids, tables that span pages, and blurry scans remain hard. Clear ruled tables with visible column gaps work more reliably—always preview first.
Try a PDF you already have open. Check headers, number columns, and whether cells stay separate.
Translate a PDF with tablesRelated pages
English home · Papers · Manuals · Scanned PDFs · Contracts