[developers] Re: pending/674: generator broken on everything except the ERG?

Stephan Oepen oe at csli.Stanford.EDU
Tue Oct 25 20:42:42 CEST 2005

> The fresh start gave me a working generator.  Before that I was
> getting no generated strings though I was getting edges.   I had
> increased *maximum-number-of-edges* because it ran out but even after
> that it still wasn't generating, though I didn't get the warning
> about going over the number of edges again. So perhaps there is some
> complexity from packing?

this all sounds strange, as none of it should be affected by restarting
afresh?  similar to the forest construction phase, unpacking could run
out of edges.  it passes :unpack as the .type. argument to next-edge(),
so the error() text will be different.  but that should really be the
only difference i would expect.  unpacking uses an independent limit:

  ;;; when unpacking, edges are a lot cheaper, as we are deterministic at this
  ;;; point (e.g. there will be very few failing unifications, if any).  hence,
  ;;; impose a separate limit for unpacking, although it could be argued that 
  ;;; we should count on top of the edges built already (like PET does).  i
  ;;; prefer independent counting, however, since active edges (in generation 
  ;;; at least) are also included in the edge count, but we could dispose of 
  ;;; them when we move on into unpacking.                      (26-nov-04; oe)
  (defparameter *unpack-edge-allowance* 25000)

--- i fear there is nothing more to do, unless you happen to re-produce 
the effect in future testing.

> I hadn't checked that yet, but it would have caused problems ...

we are about to do generation in our class at oslo this week.  so your
timing making me test on a handle-free grammar was impeccable :-).  i
have in the meantime checked in my update to the MRS test.

                                                     good night  -  oe

+++ Universitetet i Oslo (ILN); Boks 1102 Blindern; 0317 Oslo; (+47) 2285 7989
+++     CSLI Stanford; Ventura Hall; Stanford, CA 94305; (+1 650) 723 0515
+++       --- oe at csli.stanford.edu; oe at hf.uio.no; stephan at oepen.net ---

More information about the developers mailing list