-
Notifications
You must be signed in to change notification settings - Fork 0
/
lab9-bwoo.txt
266 lines (178 loc) · 7.48 KB
/
lab9-bwoo.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
LAB 9
Brent Woo
==========
=====
Fixes
=====
I needed to stop generating declarative sentences for "do cats chase dogs?", like
caun ka khwei kou lai te
cat SUBJ dog OBJ chase RLS
`Cats chase dogs.'
target:
caun ka khwei kou lai la
te was previously prop-or-ques, so generated for "do cats chase dogs". Stopped this by requiring it SF to be prop:
rls-nonfut-aux-lex := sentence-final-particle &
[ SYNSEM.LOCAL [ CONT.HOOK.INDEX [ E [ MOOD rls,
TENSE nonfuture ],
SF prop ],
----
Adverbs are now _a_rel
khountamin := adverb-lex &
[ STEM < "khountamin" >,
SYNSEM.LKEYS.KEYREL.PRED "_quickly_a_rel" ].
----
Adverbs shouldn't attach to S or V: added the following VAL constraints, so:
khwei ka khountamin sa te
dog SUBJ quickly eat RLS
`Dogs eat quickly.'
* khountamin khwei ka sa te
* khwei ka sa khountamin te
adverb-lex := basic-adverb-lex & intersective-mod-lex &
[ SYNSEM [ LOCAL [ CAT [ HEAD.MOD < [ LOCAL.CAT [ HEAD verb &
[ FORM finite ],
VAL [ COMPS < >,
SUBJ cons ]
----
the locative-verb "nei" was letting negative SFP through, because it was unspecified for polarity [POL bool], so I made locative-verbs by default POL +. Only the one verb inherits from this type:
* khwei ka panjan hma nei hpu
dog park in be RLS.neg
target:
khwei ka panjan hma nei te
locative-verb-lex := verb-lex-supertype & trans-first-arg-raising-lex-item-1 &
[ SYNSEM.LOCAL.CAT.HEAD.POL +,
----
Also, the locative-verb "nei" wasn't quite playing right with SFP, in fact sentences without SFP would parse
khwei panjan hma nei te
dog park in be RLS
* khwei panjan hma nei
dog park in be
adding this constraint knocked that out:
[ SYNSEM.LOCAL.CAT.HEAD [
FORM nonfinite ],
----
Also, the locative-verb was ALSO letting in ACC-case (object) PP's as subjects, so it would generate
khwei ka panjan hma nei te
dog SUB park in be RLS
* khwei kou panjan hma nei te
dog OBJ park in be RLS
even properly linking up the "OBJ" argument with the ARG1 of be+located. adding
[ LOCAL.CAT.HEAD.CASE sub ] to the first thing on its ARG-ST list fixed that.
-----
Constraining VP-coord.
"cats chase dogs and sleep" generated a huge number of realizations.
VP-T needs to copy up FORM from its daughters, otherwise sentences like the following would generate without an SFP, because the coordinated VP-T was form bool.
* caun ka khwei kou lai ei
cat SUBJ dog OBJ chase sleep
moreover, needed to have both daughters identify in form, otherwise one conjunct might have a SFP and the other wouldn't
* caun ka khwei kou lai te ei
cat SUBJ dog OBJ chase RLS sleep
To fix this, also needed to add in the VP-B rule to copy its lone daughter's FORM. So these rules ended up looking like this:
vp5-top-coord-rule := basic-vp-top-coord-rule & apoly-top-coord-rule &
[ SYNSEM.LOCAL [ CAT.HEAD.FORM #form,
COORD-STRAT "5"],
LCOORD-DTR.SYNSEM.LOCAL.CAT.HEAD.FORM #form,
RCOORD-DTR.SYNSEM.LOCAL.CAT.HEAD.FORM #form ].
vp5-bottom-coord-rule := unary-bottom-coord-rule & vp-bottom-coord-phrase &
[ SYNSEM.LOCAL [ CAT.HEAD.FORM #form,
COORD-STRAT "5",
COORD-REL.PRED "_and_coord_rel" ],
NONCONJ-DTR.SYNSEM.LOCAL.CAT.HEAD.FORM #form ].
and we get the correct generation:
caun ka khwei kou lai te ei te
cat SUBJ dog OBJ chase RLS sleep RLS
`Cats chase dogs and sleep.'
------
As pointed out, these rules undesirably neutralize all instances of plural. So I removed them, instead I made the cmnn-lex type unspecified for NUM and left the expected singular <> sg / plural <>pl rules
PNG.NUM : NUM
...
sg <> *
sg << pl
Also, my semi.vpm wasn't quite formatted correctly, just needed to add the PNG's... :
PNG.NUM : PNG.NUM
PNG.PER : PNG.PER
PNG.FML : PNG.FML
This stopped "I chase you" from generating all 64 combinations of all pronouns in my grammar...
----
Following Alec's Uyghur system I put in a formality feature for the 2nd person pronouns.
pron2-noun-lex := pron-noun-lex & no-spr-noun-lex &
[ SYNSEM.LOCAL.CONT.HOOK.INDEX.PNG.PER 2nd ].
pron2F-noun-lex := pron2-noun-lex &
[ SYNSEM.LOCAL.CONT.HOOK.INDEX.PNG.FML formal ].
pron2I-noun-lex := pron2-noun-lex &
[ SYNSEM.LOCAL.CONT.HOOK.INDEX.PNG.FML informal ].
nin := pron2I-noun-lex & ...
khinbya := pron2F-noun-lex & ...
png :+ [ PER person,
NUM number,
FML formality ].
;;; Formality
formality := *top*.
formal := formality.
informal := formality.
-----
On rephrasing "I think that you know that dogs chase cars", thankfully it worked but it generated about 1200 results. To start cutting down on these...
My clause-embed-verb-lex type needs to restrict its SUBJ case to be sub, otherwise sentences with a [CASE obj] PP would go through.
[ ARG-ST < [ LOCAL.CAT.HEAD +np & [ CASE sub ] ], ...
"lou" the complementizer was found at the very end of sentences, pretending to be a MC complementizer like the question particle (e.g., "That the dog sleeps." would generate). I had the complementizer be an [MC -] lex item, and made it so that the comp-head-phrase copies up the MC value of its head daughter:
decl-comp-lex-item := complementizer-lex-item &
[ SYNSEM.LOCAL [ CAT.MC -,
comp-head-phrase := basic-head-1st-comp-phrase & head-final &
[ SYNSEM.LOCAL.CAT [ HEAD.INIT -,
MC #mc ],
HEAD-DTR.SYNSEM.LOCAL.CAT.MC #mc ].
My clause-embedding verbs needed to be nonfinite, to rule out sentences without SFP:
clause-embed-verb-lex := ....
[ SYNSEM.LOCAL.CAT [ HEAD.FORM nonfinite,
...
========
Coverage
========
This is the coverage I have
+ = rephrases into mya, expected alternate realizations
++ = rephrases, but certainly too many alternate realizations
X = does not rephrase
+ Dogs sleep.
+ Dogs chase cars.
+ I chase you.
+ These dogs sleep.
+ Dogs eat.
X I can eat glass.
X It doesnt hurt me.
+ The dogs chase cars.
++ I think that you know that dogs chase cars.
++ I ask whether you know that dogs chase cars.
+ Cats and dogs chase cars.
X Dogs chase cars and cats chase dogs.
+ Cats chase dogs and sleep.
+ Do cats chase dogs?
+ Hungry dogs eat.
+ Dogs eat quickly.
+ The dogs are hungry.
+ The dogs are in the park.
++ The dogs are the cats.
+ hunen liip
+ hün-en driuwe wein-en
+ ik driuwe di
+ dijir hün-en sliip
+ hün-en iit
X ik ken glees iit
X hat dö-t mi siir ek
+ di hün-en driuwe wein-en
++ ik achtsje dat dü weet-st dat hün-en driuwe wein-en
++ ik fraag di weder dü weet-st dat hün-en driuwe wein-en
+ kat-en en hün-en driuwe wein-en
X hün-en driuwe wein-en en kat-en driuwe wein-en
+ kat-en driuwe hün-en en sliip
+ driuwe kat-en hün-en
+ hol hün-en iit
+ hün-en iit hurd
+ di hün-en wiis hol
+ di hün-en wiis ön guart (INVALID PRED if rephrase wrong tree)
++ di hün-en wiis di kat-en
Every sentence generates a few sentences because of:
- SUB / OBJ drop
- SUB / OBJ case markers are optional (ka = SUB, kou = OBJ)
- 2nd-person pronoun is ambiguous (nin = informal, khinbya = informal
As far as I could tell the ones marked with '+' I have verified to all contain only sentences accounted for by the above optionality parameters.
The glass sentences do not parse because of a stack overflow. The S-coordination sentence does not parse because I did not add S-coordination. I did not make any changes to the transfer grammar.
The long, embedded sentences all have multiple realizations with all combinations of pro-drop firing. I was able to locate the "full sentence" in Burmese with no dropped arguments and all case-endings, and I'm assuming that all the rest are all permutations of dropping case endings and such, but I could not verify this.