I’m encountering an issue where a screen reader (VoiceOver on macOS) does not correctly interpret the semantics of a table in a PDF generated using PrinceXML. Although the HTML table appears correctly structured and is accessible when using the screen reader in the browser, the generated PDF fails to maintain these semantics, leading to incorrect announcements of table data while still meeting the requirements of PDF/UA-1.

Despite these adjustments, when generating the PDF using PrinceXML, the screen reader misinterprets the table structure when viewing in Preview and Safari. For example, a cell containing “Data 1” is read as “Column 2 (2 cells), Column 2.2, Data 1” instead of “Column 1, Data 1.” which is what I find in HTML.

GitHub minimal reproduction: https://github.com/peteqian/table-accessibility-princexml
PDF: https://github.com/peteqian/table-accessibility-princexml/blob/main/output.pdf

Additional Information:
- PrinceXML Version: 15.4.1
- Screen Reader: VoiceOver

This does seem to be giving confusing results, although it's difficult to tell if that's specific to VoiceOver as it seems to be working fine in Acrobat. We will investigate some simpler examples.