[developers] code changes to LKB and [incr tsdb()]
Stephan Oepen
oe at csli.Stanford.EDU
Mon Oct 3 03:25:47 CEST 2005
well, it turns out this cafe stays open a little longer tonight. here
is a brief summary of recent code changes in the LKB:
- allow customization of post-generation string read-out: when bound,
the new global *gen-extract-surface-hook* will be invoked on edges
returned from generation; the function must return a string which
will be recorded in the (new) `string' slot on edges. see the ERG
`user-fns.lsp' for an (relatively un-appealing) example.
- new MRS comparison functions, intended to replace the restrictive
mrs-equalp().
- eliminate unnecessary combinatorics in selectively-unpack-edges(),
i.e. only put new indices on agenda that are actually required.
- support to serialize and restore generator indices (in a CDB).
- add VCS (aka vacuous lexical entries) slot to simple MRS exchange
format; this is intended to support generation from MRSs without
having to invoke trigger rules (e.g. in ACE) and should ordinarily
not be visible.
- experimentally, allow a grammar to invoke a set of transfer rules
on generator inputs, e.g. to patch MRSs up or increase robustness.
- experimentally, generalize the top-level (empty) path in feature
structures when testing subsumption for packing; see `dag.lsp'.
- new all-records() function in CDB.
- fix bug in index-for-generator(), viz. setting *top-semantics-type*
when restoring cached indices, so that predicate underspecification
works with construction semantics too.
- bug fixes in transfer code; remove most LOGON-specific settings.
- bug fixes in LUI and LSP.
please note that some of the above remain disabled by default, i.e. are
conditionalized on the :logon feature. i plan to go through these and
eliminate unnecessary conditionals over the next few days.
regarding [incr tsdb()], there will be more to say, but note that there
are new fields in several relations, i.e. generating new profiles will
require up-to-date skeletons (otherwise, the code will refuse to write
to old-style profiles). as always, reading old profiles should remain
functional, all the way back to the 1996 data format.
best - 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