[developers] `fragmented' scoping MRSs

Ann Copestake Ann.Copestake at cl.cam.ac.uk
Sat Jul 19 23:14:02 CEST 2008


minor point - if the ARG1 value of `sleep' in the eg below were typed
as a `normal' nominal variable, this would give a free variable error
in the scoping code.

> hi emily,
> 
> > [...] a few examples which seem to fall into an odd category:  The
> > "Scoped MRS" option happily returns two fully scoped MRSs, but the
> > "Compare Parses" window claims that the underspecified MRS is
> > fragmented.
> 
> these are actually two independent tests.  the `fragmented' flag that
> you (can) see in the tree comparison window, strictly speaking, is not
> a report on the MRS, but on its elementary dependency simplification.
> this test is unrelated to scopal properties and rather aims to confirm
> that the dependency graph is connected, i.e. that each predicate (node)
> can be reached via some argument link (arc) from the top node.  imagine
> an MRS like the following:
> 
>   h_0:def_q(x_0, h_1, h_2)
>   h_3:boy_n(x_0)
>   h_4:sleep_v(e_0, _)
>   h_1 =q h_3
> 
> i expect this would scope (i admit i did not test this), but it would
> trigger a fragmentation flag at the dependency level.  here, clearly,
> it would seem a grammar has failed to bind the ARG1 of `sleep_v', but
> in principle i believe it is an open question whether there could be
> `meaningful' MRS configurations that legimately map to non-connected
> dependency graphs.  for the ERG, at least, i believe dan has used the
> fragmentation test as one of several MRS wellformedness indicators.
> 
>                                                      all best  -  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 developers mailing list