RE: Parsing efficiency, CLIMB-style rule selection

A random afterthought from @Dan’s suggestion during the parsing efficiency SIG about removing rules stochastically.

I wonder if it would be useful at all to organize the grammar using declarative CLIMB style rules (in my basic understanding of turning on/off sections of declarations) such that a tool (e.g. pydelphin or even a bash script) could check for the presence of a lexeme that would trigger those rules before selecting a version of a grammar (and lazily imaging) to parse with.

So, in the wh-word example case, the entire class of wh-fronting rules could be out of the grammar by default, and if a sentence is encountered with a wh word, then CLIMB would be invoked to generate the appropriate TDL, etc.

I also wanted to say earlier today that at first I thought: oh, but would that still be a grammar of English, etc., but then I thought: actually, the concept of having some rules more readily accessible than other rules doesn’t seem all that strange at all, in processing terms?

1 Like

In the sense of determining grammaticality, it certainly seems more like a grammar than a PCFG :slight_smile: