What are the characteristics of Labelese?

[-JKP-]

I am inclined to the view that there is an overestimate of the useful of TTR in this thread. As already stated it would be expected on the face of it that there would be a higher proportion of unique and less common words amongst labels than amongst sentence text, so the TTR figure definitely has some use. However the TTR figure is a very blunt tool to tell us anything about labels, but we shouldn't be too reliant on this figure.

I know Koen can speak for himself, but I see TTR as part of a long tradition of computational attacks, and computational attacks rarely provide answers on their own (nor are intended to). They typically are used to build up a picture that gets filled in, over time, like a jigsaw puzzle.

TTR has value as one of the puzzle pieces. I doubt if Koen ever expected it to be the whole puzzle.