[developers] Problem with transfer rules referencing ARG0 type or features

[Woodley Packard]

Dear Berthold,

test2_mtr stipulates [ SF pres ] while the MRS you showed exhibits [ SF 
prop ].  Do prop and pres stand in a subsumption relation in your 
grammar (I guess that would be unusual)?  Or did you perhaps mean to 
write [ SF prop ] on the transfer rule?

Woodley

On 07/02/2013 06:21 AM, Berthold Crysmann wrote:
> Dear Stephan and all,
>
> I am currently starting a new transfer grammar (frde) that builds on 
> an older baby-MT pair (ende). For some strange reason, I cannot get a 
> rule  to match that specifies an event type for ARG0 , or, more 
> interesting a variable property like TENSE or SF.
> Matching against (string) predicates, however, tends to work. Also, 
> matching on x or variable properties like [NUM sg] poses no problem.
>
> Here's a minimal pair:
>
>
> test1_mtr :=
>           monotonic_mtr &
>           [INPUT.RELS  <[ARG0 x & [NUM sg ]]>,
>            OUTPUT.RELS < +copy+ & [ARG0 [NUM pl]] >].
>
> test2_mtr :=
>           monotonic_mtr &
>           [INPUT.RELS  <[ARG0 [SF pres ]]>,
>            OUTPUT.RELS < +copy+ & [ARG0 [SF past]] >].
>
>
> And here the MRS after application of test1_mtr:
>
>  [ LTOP: h76
>    INDEX: u77
>    RELS: <
>           [ "_machen_v_n-a-haben_rel"
>             LBL: h40
>             ARG0: e41 [ e SF: PROP TENSE: PRES MOOD: IND PERF: + ]
>             ARG1: x81 [ x PERS: 3 NUM: PL GEND: !     ]
>             ARG2: x68 [ x PERS: 3 NUM: PL GEND: !     ] ]
>           [ "def_q_rel"
>             LBL: h47
>             ARG0: x68
>             RSTR: h49
>             BODY: h50 ]
>           [ "_named_n_rel"
>             LBL: h61
>             ARG0: x81
>             CARG: "tex" ]
>           [ "_pizza_n_rel"
>             LBL: h67
>             ARG0: x68 ]
>           [ "udef_q_rel"
>             LBL: h80
>             ARG0: x81
>             RSTR: h56
>             BODY: h57 ] >
>    HCONS: < h49 QEQ h67 h56 QEQ h61 > ]
>
> As far as I can tell from the LKB, the types and features are in place!
>
>
>
>
> Please let me know whether you need any further details and which.
>
> Berthold
>
> PS: Why does Logon not delete the GEND feature?
>