Content understanding: Failed to analyze the file file.html. Invalid Request. (An unknown error was encountered

Question

We're getting this error when processing some HTML files in Azure Content Understanding using the prebuilt layout analyzer.

I think it has to do with tables having a different amount of columns per row.

i.e. First row has 2 columns while next row has 4 columns.

<table>

	<tr>

		<td>                                            &#160;</td>

		<td>                                        &#160;</td>

	</tr>





	<tr>

		<td class="emptyTDSTag" colspan="1" rowspan="1"> &#160;</td>

		<td colspan="1" rowspan="1"><div class="t s3_134" id="tl_134" style="left:527px;bottom:768px;letter-spacing:0.21px;display:inline;">December 31, </div></td>

		<td colspan="1" rowspan="1"><div class="t s3_134" id="tn_134" style="left:614px;bottom:768px;letter-spacing:0.14px;word-spacing:0.03px;display:inline;">Adoption of IFRS 9 </div></td>

		<td class="emptyTDSTag" colspan="1" rowspan="1"> &#160;</td>

	</tr>

	<tr>

		<td colspan="1" rowspan="1"><div class="t s2_134" id="tk_134" style="left:108px;bottom:757px;letter-spacing:0.2px;word-spacing:0.03px;display:inline;">Opening balance sheet reconciliation </div></td>

		<td colspan="1" rowspan="1"><div class="t s3_134" id="tm_134" style="left:570px;bottom:757px;letter-spacing:0.27px;display:inline;">2022 </div></td>

		<td colspan="1" rowspan="1"><div class="t s3_134" id="to_134" style="left:646px;bottom:757px;letter-spacing:0.2px;word-spacing:0.01px;display:inline;">and IFRS 17 </div></td>

		<td colspan="1" rowspan="1"><div class="t s3_134" id="tp_134" style="left:733px;bottom:757px;letter-spacing:0.21px;display:inline;">January 1, 2023 </div></td>

	</tr>  

Changing the content is not really an option for us, but I would expect the content understanding to just handle this correctly.

Answer 1

Hi,

this is very unlikely to be caused simply by different numbers of columns per row. Html tables with uneven rows are valid and the layout model can usually handle them, especially when colspan and rowspan are used correctly.

whats stands out in ur snippet is the absolute positioning like left and bottom in inline styles. That pattern is typical of pdf to html exports. In such cases the table structure is often only visual, while the real layout is driven by absolutely positioned div elements. The service then tries to reconstruct a logical reading order and table grid from something that is essentially a positioned canvas, and this can trigger an invalid request instead of a clean parsing error. Common cause is very complex or inconsistent dom structure. Empty cells with nbsp only content, nested divs inside td elements, mixed spans and positioning can produce a grid that cannot be resolved into a consistent internal table model. The layout analyzer expects a coherent structure, not a rendering hack.

The request is truly sending valid html with the correct content type such as text html, and that the file is properly encoded in utf 8. If this html was generated from a pdf, u will almost certainly get more stable results by sending the original pdf to the layout model instead of the converted html.

To narrow it down, try validating the html strictly, then test a simplified version of the failing page without absolute positioning styles. If that works, the issue is structural rather than column count. It is also worth calling the rest api directly and capturing the full raw response body, as studio often hides additional error detail.

In practice ))) thats looks much more like a parsing limitation with pdf style html than a problem with uneven table columns.

rgds,

Alex