Scottes,
Your guess at the meaning of "instrumentalness" is correct. Glad the name was obvious enough!
The first thing to keep in mind is that instrumentalness is what we call an "alpha" attribute. That's why it was not announced. We include it in the API so we can widen the exposure and the kinds of experiments we can conduct, but we are not 100% happy with the accuracy yet. We know there are a number of cases where it doesn't line up. (There's no need to report specific examples at this time. We have plenty we are looking at!) But we felt like it was interesting enough to let you experiment.
Second, instrumentalness and speechiness are not opposites. They estimate different things. Speechiness indicates the likelihood the track contains spoken words. Sung vocals are not speech. A track that is not instrumental is not necessarily "speech". Tracks with singing have low speechiness AND low instrumentalness, whereas rap tracks could have low instrumentalness and MEDIUM speechiness. High speechiness tracks should PROBABLY have low instrumentalness, but we don't correlate the two attributes, so you should not make this assumption.
Hope this helps!
-David