Well, let's take a look at his pitches from a recent game: last night's game, I think (I'm in Tokyo and my sense of date is all screwed up, so this might be wrong).
View attachment 1239
How many different groups do you see here? I see two, really: the fastballs, in the upper-left corner, and then ... everything else. The knuckleball has so wide a range of properties that it really just fills diagrams like this.
(digression: the method by which software tries to fit the trajectory of a ball and extract quantities such as "horizontal break" and "spin" works pretty well when a ball is spinning -- but not so well when the ball is floating. Beware the values in diagrams like this for spin-less balls)
Now, an ordinary curveball thrown by a right-handed pitcher will be slow (in the lower portion of the diagram), and break toward first base (to the right in this diagram). An automated pitch-classifying program would look for pitches in the lower-right section of the diagram, and call them "curves". Here, there ARE some pitches in that region, but they are probably just extreme knuckleballs.
A person looking at this diagram would probably say, "Oh, that's a knuckle-ball pitcher. So, I see a 'fastball' group, and the rest is all 'knuckleball'." But a program doesn't draw the same conclusion, and ends up with spurious curves.
That's my guess.