Discussions

A question about the best input format when submitting documents through the API.

A lot of what we send for translation arrives as PDF: vendor manuals, regulatory notices, product sheets. Submitting the PDF directly works, but the segmentation comes back noisy — hard-wrapped lines become separate segments and tables get flattened, so the translator sees fragments with no context.

We started converting to Markdown first, which segments much more cleanly because headings and table rows survive as structure rather than as layout. The conversion runs in the browser with pdf2md , which matters here because a chunk of this material is under NDA and cannot be uploaded to a third-party service just to be reformatted.

Two things I could not find in the docs:

1. Does the API segmenter treat a Markdown upload differently from a plain .txt upload, or is Markdown flattened to text on ingest? If the structure is preserved, is there a list of which constructs are respected — headings, tables, lists, code fences?
   2. When a document is submitted as Markdown, does the delivered translation come back with the Markdown intact, or should we expect to re-apply the formatting on our side?

If Markdown is the recommended path we would rather standardise on it before wiring up the pipeline.