Basically, a screenreader won't even acknowledge the presence of these elements. This be bad. We need some form of 'alt text' for the image, or we can put alternative text inside the <canvas> element itself I think.
Also inquired upstream, to see if there has been any discussion about it already that we wouldn't know about.
https://github.com/trifacta/vega/issues/268
Implemented: alt='...' attribute takes its value from the title attribute:
<graph title='...'>...</graph>
becomes
<img alt='...' href='...' />