[logon] uio/erg.mtr

Stephan Oepen oe at ifi.uio.no
Wed Nov 19 20:09:24 CET 2008


hi again,

> my own version of conj_prop_ef etc. fails to redefine the transfer
> matrix one.

yes, although the MTR files look like regular TDL, there is no way of
re-defining rules just by giving them the same identifier.  i.e. there
is nothing stopping one from having multiple rules with the same name.

this is because, given the ordered nature of the rewrite formalism, it
might be desirable to use the (conceptually) same rule in more than one
place (i thought at the time).

> For now, I have just disabled loading of the matrix. 
> Any ideas how I can just replace the matrix version (which does not work
> for the German input). 

i still suspect that those robustness rules in the Transfer Matrix are
correct, as long as the assumptions they make (about SF) are met.  but
i still have not taken the time to look at your problematic items ... 

either way, i really will try to encourage use of the Transfer Matrix,
partly because i believe it already contains useful features, partly to
have it tested and generalized more (really, the `Transfer Matrix' was
an informed guess at which of the original LOGON rules might be useful
for others).

so, if an override mechanism is what it takes to encourage my users, i
just implemented that :-).

once i build new binaries, add (to `lkb/mt.lisp') something like:

  (setf *transfer-ignore*
    (list
     (mrs:vsym "conj_ques_ef")
     ...
     (mrs:vsym "conj_fut_ef")))

this will make the transfer engine ignore (skip over) all rules with
those names.  hence, your own versions need to use different names.

                                                      cheers  -  oe

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+++ Universitetet i Oslo (IFI); Boks 1080 Blindern; 0316 Oslo; (+47) 2284 0125
+++     CSLI Stanford; Ventura Hall; Stanford, CA 94305; (+1 650) 723 0515
+++       --- oe at ifi.uio.no; oe at csli.stanford.edu; stephan at oepen.net ---
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++



More information about the logon mailing list