[developers] More feedback on ERG 0902

[Francis Bond]

G'day,

we have been using the new ERG 0902 and are very happy with the new
chart mapping when used in parsing.  We have found one problem though
--- in order to use the terg+tnt cpu, it is necessary to also load the
terg in the controller. That is we do both of:
*NOTE* our cpu definition includes "-mrs", we get the semantics
directly from PET, thanks to all who made it possible.

(rsa "terg")
(tsdb::tsdb :cpu :terg+tnt :file t)

I am not sure why this should be necessary --- it always seems very
wasteful :-).
But it allows us to do the following:   (mt::parse-interactively "Frodo barks.")

The problem is that we want to use terg+tnt to parse, but load a
transfer grammar:

(tsdb::tsdb :cpu :terg+tnt :file t)
(lkb::read-script-file-aux "~/logon/uio/enen/lkb/script")

or possibly

(lkb::read-script-file-aux "~/logon/uio/enja/lkb/script")

If we do this, then we can't actually parse anything:

 (mt::parse-interactively "Frodo barks.")
[t40005] Segmentation violation.
[t40005] Top level.
[t40005] MT>
[t40005] EOF

I had a look in the terg to see if there was something in user-fns or
patches that we should be loading, but couldn't see anything obvious.
So my questions are:

(i) Is there some kind of short-term solution we could quickly apply
here?  Something that enen or enja loads for example?

(ii) Could we add "parsing with terg as a client even if terg is not
the top grammar"  as a requested feature for the Barcelona release?

Yours β-testing,

-- 
Francis Bond <http://www2.nict.go.jp/x/x161/en/member/bond/>
NICT Language Infrastructure Group