[developers] TERG named entity regression?

Eric Nichols ericnichols79 at gmail.com
Sun May 17 16:32:18 CEST 2009 

Dan,

Francis and I are enjoying the progress you have made in TERG.
The enhanced TNT support is particularly appreciated, however,
I think I have found a reversion in the latest LOGON TERG. I
rebuilt my dictionary rules to sync with TERG's MRS rel names
and changed ja2en.lisp to use the terg CPU instead of erg. This
resulted in us losing all sentences containing named rels. Looking
in the fans, I noticed a new error message:

[09:04:15] (100) |私 は ベートーベン が 今 まで の 最高 の 作曲 家 だ と 思い ます 。| --- 5
(0.04|0.03:0.05 s) <87:1175> {1495:969} (008.3m = 8.3m).
|
|-[0.48] # 0 {20.00} --- error: unknown transfer predicates: |"ja:_ka_a_6_rel"|.
|
|-[0.52] # 1 {10.00} --- error: unknown transfer predicates: |"ja:_ka_a_6_rel"|.
|
|-[18.74] # 2 {10.00} --- 16 (17.96:0.53|0.00:0.00 s) <:958> (560m 13g
0= 13g) [4].
| |
| |-[18.80] # 0 {-427.04} --- error: invalid predicates:
|named_rel("beethoven")|.
| |
| |-[18.85] # 1 {-433.67} --- error: invalid predicates:
|named_rel("beethoven")|.
| |
| |-[18.91] # 2 {-550.33} --- error: invalid predicates:
|named_rel("beethoven")|.
| |
| |-[18.97] # 3 {-550.85} --- error: invalid predicates:
|named_rel("beethoven")|.
| |
| |-[19.02] # 4 {-557.54} --- error: invalid predicates:
|named_rel("beethoven")|.
|
|-[37.26] # 3 {10.00} --- 16 (17.99:0.53|0.00:0.00 s) <:958> (560m 13g
0= 13g) [4].
| |
| |-[37.31] # 0 {-427.04} --- error: invalid predicates:
|named_rel("beethoven")|.
| |
| |-[37.37] # 1 {-433.67} --- error: invalid predicates:
|named_rel("beethoven")|.
| |
| |-[37.43] # 2 {-550.33} --- error: invalid predicates:
|named_rel("beethoven")|.
| |
| |-[37.49] # 3 {-550.85} --- error: invalid predicates:
|named_rel("beethoven")|.
| |
| |-[37.55] # 4 {-557.54} --- error: invalid predicates:
|named_rel("beethoven")|.
|
|-[37.58] # 4 {10.00} --- error: unknown transfer predicates:
|"ja:_ka_a_6_rel"|.
|
|< |私 は ベートーベン が 今 まで の 最高 の 作曲 家 だ と 思い ます 。| (100) --- 5 x 32 x 0 = 0
|@ |I think Beethoven is as great a composer as ever lived.|
|= 9:0 of 10 {90.0+0.0}; 7:0 of 9:0 {77.8 0.0}; 3:0 of 7:0 {42.9 0.0}
@ 3 of 10 {30.0} <0.04 0.14|0.05 0.16>.

Are you trying something new with named rel generation, perhaps?
At Francis' suggestion, I reverted to the 0902 TERG release, and
my problems were solved. Below is a log comparing the MRSes of
0902 and the current LOGON trunk. Ostensibly the MRSes for a
simple sentence with a named rel are identical (they pass the diff
test), so the change must be beyond the MRS level.

I hope this helps,

Eric

$ svn info
Path: .
URL: http://svn.emmtee.net/erg/tags/0902
Repository Root: http://svn.emmtee.net
Repository UUID: 3df82f5b-d43a-0410-af33-fce91db48ec5
Revision: 6523
Node Kind: directory
Schedule: normal
Last Changed Author: oe
Last Changed Rev: 1952
Last Changed Date: 2009-04-11 04:54:49 +0900 (Sat, 11 Apr 2009)

$ cd terg.0902
$ echo I like Beethoven. | cheap --mrs english.grm
reading `pet/english.set'... including `pet/common.set'... including
`pet/global.set'... including `pet/mrs.set'... loading `english.grm'
(ERG (0902)) reading ME model `jhpstg.mem'... [913032 features]
read-vpm(): reading file `semi.vpm'.
92441 types in 9.2 s

(1) `i like beethoven.' [0] --- 1 (0.01|0.01s) <15:88> (569.8K) [0.0s]
derivation[1] (-3.811):i like beethoven.

 [ LTOP: h1
   INDEX: e2 [ e SF: PROP TENSE: PRES MOOD: INDICATIVE PROG: - PERF: - ]
   RELS: <
          [ pron_rel
            LBL: h3
            ARG0: x4 [ x PERS: 1 NUM: SG PRONTYPE: STD_PRON ] ]
          [ pronoun_q_rel
            LBL: h5
            ARG0: x4
            RSTR: h7
            BODY: h6 ]
          [ "_like_v_1_rel"
            LBL: h8
            ARG0: e2
            ARG1: x4
            ARG2: x9 [ x PERS: 3 NUM: SG IND: + ] ]
          [ proper_q_rel
            LBL: h10
            ARG0: x9
            RSTR: h12
            BODY: h11 ]
          [ named_rel
            LBL: h13
            ARG0: x9
            CARG: "Beethoven" ] >
   HCONS: < h7 qeq h3 h12 qeq h13 > ]

$ cd ../terg.logon
$ svn info
Path: .
URL: http://svn.emmtee.net/erg/trunk
Repository Root: http://svn.emmtee.net
Repository UUID: 3df82f5b-d43a-0410-af33-fce91db48ec5
Revision: 6523
Node Kind: directory
Schedule: normal
Last Changed Author: oe
Last Changed Rev: 6501
Last Changed Date: 2009-04-28 21:42:06 +0900 (Tue, 28 Apr 2009)

$ flop -t english.tdl
reading `Version.lisp'...
converting `english.tdl' (ERG (trunk)) into `english.grm' ...
loading `english.tdl'... including `fundamentals.tdl'... including
`lextypes.tdl'... including `tmt.tdl'... including `tmr.tdl'...
including `tmr/ptb.tdl'... including `tmr/spelling.tdl'... including
`tmr/ne1.tdl'... including `tmr/split.tdl'... including
`tmr/ne2.tdl'... including `tmr/class.tdl'... including
`tmr/punctuation.tdl'... including `tmr/pos.tdl'... including
`tmr/case.tdl'... including `lfr.tdl'... including `syntax.tdl'...
including `lexrules.tdl'... including `auxverbs.tdl'... including
`letypes.tdl'... including `lexicon.tdl'... including
`handon-propers.tdl'... including `gle.tdl'... including
`constructions.tdl'... including `lexrinst.tdl'... including
`inflr.tdl'... including `inflr-pnct.tdl'... including
`parse-nodes.tdl'... including `roots.tdl'... postloading
`pet/qc.tdl'...

finished parsing - 0 syntax errors, 231435 lines in 1.63 s
processing type constraints (90257 types):
- expanding templates: 0 template instantiations
- type hierarchy (leaf types [87806], bitcodes, glbs [2385], recomputing)
- building dag representation
- computing appropriateness
- applying appropriateness constraints for types
- delta / full expansion for types
- partitioning hierarchy (36 partitions)
dumping grammar (symbols 2998k, hierarchy 484k, feature tables 20k,
supertypes 49k, lexicon 54k, types 16667k)
finished conversion - output generated in 13.8 s
# echo I like Beethoven. | cheap --mrs english.grm
reading `pet/english.set'... including `pet/common.set'... including
`pet/global.set'... including `pet/mrs.set'... loading `english.grm'
(ERG (trunk)) reading ME model `jhpstg.mem'... [913032 features]
read-vpm(): reading file `semi.vpm'.
92642 types in 9.3 s

(1) `i like beethoven.' [0] --- 1 (0.01|0.01s) <15:88> (559.8K) [0.0s]
derivation[1] (-3.811):i like beethoven.

 [ LTOP: h1
   INDEX: e2 [ e SF: PROP TENSE: PRES MOOD: INDICATIVE PROG: - PERF: - ]
   RELS: <
          [ pron_rel
            LBL: h3
            ARG0: x4 [ x PERS: 1 NUM: SG PRONTYPE: STD_PRON ] ]
          [ pronoun_q_rel
            LBL: h5
            ARG0: x4
            RSTR: h7
            BODY: h6 ]
          [ "_like_v_1_rel"
            LBL: h8
            ARG0: e2
            ARG1: x4
            ARG2: x9 [ x PERS: 3 NUM: SG IND: + ] ]
          [ proper_q_rel
            LBL: h10
            ARG0: x9
            RSTR: h12
            BODY: h11 ]
          [ named_rel
            LBL: h13
            ARG0: x9
            CARG: "Beethoven" ] >
   HCONS: < h7 qeq h3 h12 qeq h13 > ]

More information about the developers mailing list