Hi,
Thanks for the feedback. We have created a work item for this document.
As a suggestion, it doesn't make a lot of sense to make large Detect requests. Better strategy is to select the relevant portion of the text and detect just that. For instance the body of the email, or the comment on the web page, or the first or second paragraph of the document.
You may break the document at natural breakpoints, for instance paragraph marks or section headers, and then determine the language for each individually.
When you send a large document, you get a single set of probabilities for the entire long document: Not that useful.
Regards,
Yutong