I have still never nailed down how, precisely, one does this via github.
Under 2019.1 (and it may have been occurring earlier, I haven't played with this recently) I have an MS-Word fillable form. When I initially open the form and begin traversing the text edit fields and checkboxes, NVDA keeps saying "blank" rather than reading the labels assigned to each of these items until or unless I hit F1 on any one of them to force help text to appear once, then afterward the labels are being read. However, for the checkboxes, when I hit spacebar to change state for them, I get no feedback as to "checked" or "unchecked," and I know it used to tell me the state the checkbox was in when I landed on it initially and after any key press that would toggle the state.
I can submit the .dotx file from which each document gets created and/or an instance of the .docx document itself. Something's changed, though, because I know this used to work as expected and without the need for any gyrations to get the labels for each form field to read as you traversed the form.
Brian - Windows 10 Home, 64-Bit, Version 1809, Build 17763
The Internet is so big, so powerful and pointless that for some people it is a complete substitute for life.