Isn't the recognizer just the piece that (for example) takes the pointer location and figures out what object it correlates to? In which case CLIM already contains a recognizer. Well, I haven't read the paper so I'm probably barking up the wrong tree.
Semi-related, I have wondered many times about the relationship between the modern web and CLIM and how hard it would be to implement a CLIM-like programming interface for making web apps. That is, to get the much fancier interfaces that the web offers together with the high-level programmer interface provided by CLIM. Has this been done already?
I've attached a screenshot from the paper. Recognizers work ~like this:
Say that you have a rectangle, and inside a few output records with text put one below another.
A recognizer could suggest that the top-level rectangle is a frame, then infer that text one below another are paragraphs of a single body of text. According to the paper this process would be interactive where the user can modify system inferences, so that would be like a dialog.
Regarding CLIM on the web, I'm experimenting with porting #McCLIM directly to a browser using #ecl (that works as an webassembly program) treating <div> elements of certain class as sheets etc.
