[pet] cheap and unknown words

Francis Bond fcbond at gmail.com
Wed Jun 27 07:40:28 CEST 2007


G'day,

I thought that cheap had the ability to fill in the CARG of unknown
words with the surface string if it thought they were names: so
something like

"Kim Novak"

would contain an EP like
 [ named_rel
    LBL: h10
    ARG0: x6
    CARG: "Novak"] >

However, it currently becomes:
          [ named_rel
            LBL: h10
            ARG0: x6
            CARG: MRS::*TOP* ]

Is that the expected behaviour?  It could be that the CARG is being
replaced by post-processing in the HOG, and I never noticed...

Note that I haven't filed a bug, as I am not sure what is the correct behaviour.

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

P.S. Tested with:

cheap -mrs -tok=yy -default-les logon/lingo/erg/english

(0, 0, 1, 1, "Kim" "Kim", 0, "null", "NNP" 1.00)(1, 1, 2, 1, "Novak"
"Novak", 0, "null", "NNP" 1.00)

Cheap version: 0.99.14svn
ERG version:  "LinGO (18-Jul-06)"

Full output:
(1) `kim novak' [0] --- 2 (0.01|0.01s) <2:19> (116.6K) [0.0s]
derivation[1] (0.277):kim novak

 [ LTOP: h1
   INDEX: e2 [ e --TPC: LUK E.MOOD: MOOD E.ASPECT.PROGR: LUK
E.ASPECT.PERF: LUK E.TENSE: BASIC_TENSE ]
   RELS: <
          [ prop-or-ques_m_rel
            LBL: h1
            ARG0: e2
            MARG: h3
            PSV: u4
            TPC: u5 ]
          [ unknown_rel
            LBL: h1
            ARG0: e2
            ARG: x6 [ x PNG.GEN: REAL_GENDER PNG.PN: 3SG DIV: - --TPC:
LUK PRONTYPE: PRONTYPE ] ]
          [ proper_q_rel
            LBL: h7
            ARG0: x6
            RSTR: h8
            BODY: h9 ]
          [ compound_name_rel
            LBL: h10
            ARG0: e12 [ e --TPC: LUK E.MOOD: MOOD E.TENSE: NO_TENSE
E.ASPECT.PROGR: - E.ASPECT.PERF: - ]
            ARG1: x11 [ x --TPC: LUK DIV: - PNG.GEN: REAL_GENDER PNG.PN: 3SG ]
            ARG2: x6 ]
          [ udef_q_rel
            LBL: h10
            ARG0: x11
            RSTR: h13
            BODY: h14 ]
          [ named_rel
            LBL: h15
            ARG0: x11
            CARG: "kim" ]
          [ named_rel
            LBL: h10
            ARG0: x6
            CARG: MRS::*TOP* ] >
   HCONS: < h3 qeq h1 h8 qeq h10 h13 qeq h15 > ]


derivation[2] (-0.2286):kim novak

 [ LTOP: h1
   INDEX: e2 [ e --TPC: LUK E.MOOD: MOOD E.ASPECT.PROGR: LUK
E.ASPECT.PERF: LUK E.TENSE: BASIC_TENSE ]
   RELS: <
          [ prop-or-ques_m_rel
            LBL: h1
            ARG0: e2
            MARG: h3
            PSV: u4
            TPC: u5 ]
          [ unknown_rel
            LBL: h1
            ARG0: e2
            ARG: x6 [ x PNG.GEN: REAL_GENDER PNG.PN: 3SG DIV: - --TPC:
LUK PRONTYPE: PRONTYPE ] ]
          [ appos_rel
            LBL: h7
            ARG0: e9 [ e --TPC: LUK E.MOOD: MOOD E.TENSE: NO_TENSE
E.ASPECT.PROGR: - E.ASPECT.PERF: - ]
            ARG1: x6
            ARG2: x8 [ x PRONTYPE: PRONTYPE --TPC: LUK DIV: - PNG.GEN:
REAL_GENDER PNG.PN: 3SG ] ]
          [ proper_q_rel
            LBL: h10
            ARG0: x6
            RSTR: h11
            BODY: h12 ]
          [ named_rel
            LBL: h7
            ARG0: x6
            CARG: "kim" ]
          [ proper_q_rel
            LBL: h13
            ARG0: x8
            RSTR: h14
            BODY: h15 ]
          [ named_rel
            LBL: h16
            ARG0: x8
            CARG: MRS::*TOP* ] >
   HCONS: < h3 qeq h1 h11 qeq h7 h14 qeq h16 > ]



More information about the pet mailing list