[developers] Regular Expressions in Transfer Rules

[Woodley Packard]

Hi Stephan,

That all makes good sense, but the wrinkle I was trying to point out is that in the past, there was an intentional namespace division between ordinary PRED values ending in _rel and more ad-hoc ones like those ending in _mark and possibly other values.  The 2016 move could be interpreted as wiping that distinction out, hence leaving legacy grammarians a short ways up a creek (albeit not exactly without a paddle).

Woodley



> On Mar 1, 2017, at 1:02 PM, Stephan Oepen <oe at ifi.uio.no> wrote:
> 
> dear all,
> 
>> Incidentally, how do you feel about "~mark$" matching "_mark_rel"?  To me,
>> it would seem hard to justify ignoring quotes without also ignoring the _rel
>> suffix.  I would be curious to know the current LOGON implementation’s
>> stance on this.
> 
> from memory, i have long operated on the assumption that the ‘_rel’
> suffix is a convention in the TFS descriptions of MRSs and should be
> stripped when reading out an MRS from a feature structure.  most
> likely, that stripping happens both in the construction of transfer
> rules and when reading an MRS as input to the transfer component.
> 
> until the 2016 move towards greater predicate normalization, in the
> sense of ignoring the distinction between TDL type names vs. TDL
> strings in all MRS manipulation, the LOGON transfer code used to not
> generally match a(n unquoted) type name against a (quoted) string, but
> the regular expression operator has always been an exception, i.e. it
> would match /_mark$/ successfully against either foo_mark or
> "foo_mark".  these subtleties fortunately become irrelevant with the
> introduction of MRS predicate normalization.
> 
> best wishes, oe