-
Notifications
You must be signed in to change notification settings - Fork 11
/
Copy pathanalysis-SuB-paper.R
621 lines (516 loc) · 27 KB
/
analysis-SuB-paper.R
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
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
# Analysis file for the figures and models reported in
# de Marneffe et al (2019, Sinn und Bedeutung proceedings)
# written by Marie-Catherine de Marneffe, modified by Judith Tonhauser after publication
# this file assumes that the following three directories occur at the same level:
# data (raw data and output data frames created by this analysis script)
# graphs (figures created by this analysis script)
# rscripts (the location of this file as well as helpers.R)
# set working directory to directory of script
this.dir <- dirname(rstudioapi::getSourceEditorContext()$path)
setwd(this.dir)
source("helpers.R")
# load packages
library(tidyverse)
library(dplyr)
library(xtable)
library(ordinal)
library(rcompanion)
# b/w background in figures
theme_set(theme_bw())
#load the data
d = read.csv("../data/CommitmentBank-All.csv", header=T, comment.char="")
nrow(d) #11545
length(unique(d$uID)) #1200 unique discourses
# Table 1 ----
# aggregate data
agrD = d %>%
select(uID, Verb, Embedding, Target, Prompt, Answer) %>%
group_by(uID, Verb, Embedding, Target, Prompt) %>%
summarize(Mean = mean(Answer))
agrD = as.data.frame(agrD)
nrow(agrD)
# verb by embedding
xtabs(~ Verb + Embedding, agrD)
#drop the non-epistemic modals
dataM = droplevels(subset(d, ModalType != "AB" & ModalType != "CI" & ModalType != "DE"))
length(unique(dataM$uID)) #982 unique discourses
table(dataM$ModalType) # only empty annotations & EP
nrow(dataM) #9599
# Figure 2 ----
names(dataM)
#aggregate data
aData = dataM %>%
select(uID, Verb, Embedding, Target, Prompt, Answer) %>%
group_by(uID, Verb, Embedding, Target, Prompt) %>%
summarize(Mean = mean(Answer), low.ci = ci.low(Answer), high.ci = ci.high(Answer), SD = sd(Answer), C = n())
aData = as.data.frame(aData)
nrow(aData) #982 discourses
# paste predicate and number of discourses per predicate
verbNum = as.data.frame(xtabs(~ Verb, aData))
verbNum #number of discourses per verb
dataM = merge(dataM, verbNum, by="Verb")
head(dataM)
dataM$VerbNum = paste(dataM$Verb, paste0("(",dataM$Freq,")"), sep=" ")
mean_proj = aggregate(Answer~VerbNum, data=dataM, FUN="mean")
mean_proj$YMin = mean_proj$Answer - aggregate(Answer~VerbNum, data=dataM, FUN="ci.low")$Answer
mean_proj$YMax = mean_proj$Answer + aggregate(Answer~VerbNum, data=dataM, FUN="ci.high")$Answer
mean_proj
names(mean_proj)
head(mean_proj)
nrow(mean_proj) #45 verbs
dataM$VerbNum <-factor(dataM$VerbNum, levels=mean_proj[order(mean_proj$Answer), "VerbNum"])
head(dataM)
table(dataM$VerbNum)
cols = data.frame(V=levels(dataM$VerbNum))
cols$VeridicalityGroup = as.factor(
ifelse(cols$V %in% c("learn (6)","forget (13)", "notice (32)", "understand (7)", "recognize (1)", "bother (1)","remember (5)","realize (29)","know (122)","find (17)","see (12)"), "F", "NF"))
# ifelse(cols$V %in% c("tell (18)", "guess (16)", "bet (1)", "suspect (18)", "take (1)","decide (11)","hear (9)","fear (3)","say (67)","hypothesize (1)",
# "figure (1)","assume (5)","imagine (15)","mean (50)","insist (3)","demand (2)","hope (8)","feel (29)",
# "believe (46)","think (378)","suggest (17)","pretend (4)","expect (4)","seem (2)","suppose (5)","occur (1)"),"NF","VNF")))
cols$Colors = ifelse(cols$VeridicalityGroup == "F", "#D55E00", "#999999")
# ifelse(cols$VeridicalityGroup == "NF", "brown","black"))
head(cols$Colors)
#means with confidence intervals -- used for SuB poster
mean_proj <- mutate(mean_proj,VerbNum = reorder(VerbNum, Answer, mean)) # put mean_proj in sorted order
ggplot(mean_proj, aes(x=VerbNum, y=Answer)) +
geom_point() +
geom_errorbar(aes(ymin=YMin,ymax=YMax),color="gray50",alpha=.5) +
geom_hline(yintercept=0, linetype="dashed", color = "red") +
theme(text = element_text(size=12)) +
theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=0.4, color=cols$Colors)) +
theme(panel.background = element_blank(), plot.background = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()) +
theme(panel.grid.major.y = element_line(colour="grey90", size=0.5)) +
ylab("Mean certainty rating") +
xlab("Predicate (with number of discourses)")
ggsave(f="../graphs/Figure2.pdf",height=3,width=6)
# Figure 3 ----
# "know"
know <- droplevels(subset(dataM,dataM$Verb == "know"))
head(know)
length(unique(know$uID)) # 122 discourses
number_of_ratings = table(know$uID)
number_of_ratings <- as.data.frame(number_of_ratings)
mean(number_of_ratings$Freq) #9.3
mean_projK = aggregate(Answer~uID, data=know, FUN="mean")
mean_projK$YMin = mean_projK$Answer - aggregate(Answer~uID, data=know, FUN="ci.low")$Answer
mean_projK$YMax = mean_projK$Answer + aggregate(Answer~uID, data=know, FUN="ci.high")$Answer
mean_projK <- mutate(mean_projK,uID = reorder(uID, Answer, mean))
ggplot(mean_projK, aes(x=uID,y=Answer)) +
geom_point() +
geom_errorbar(aes(ymin=YMin,ymax=YMax),color="gray50",alpha=.5) +
#facet_wrap(~VerbNum,scales="free_x", ncol=4) +
geom_hline(yintercept=0, linetype="dashed", color = "red") +
theme(axis.text.x=element_blank(), axis.ticks.x=element_blank()) +
theme(panel.background = element_blank(), plot.background = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()) +
theme(panel.grid.major.y = element_line(colour="grey90", size=0.5)) +
ylab("Mean certainty rating") +
xlab("Discourses with \"know\" (122)")
ggsave(f="../graphs/Figure3a.pdf",height=2.5,width=6)
# "believe"
believe <- droplevels(subset(dataM,dataM$Verb == "believe"))
head(believe)
length(unique(believe$uID)) # 46 discourses
mean_projB = aggregate(Answer~uID, data=believe, FUN="mean")
mean_projB$YMin = mean_projB$Answer - aggregate(Answer~uID, data=believe, FUN="ci.low")$Answer
mean_projB$YMax = mean_projB$Answer + aggregate(Answer~uID, data=believe, FUN="ci.high")$Answer
mean_projB <- mutate(mean_projB,uID = reorder(uID, Answer, mean))
ggplot(mean_projB, aes(x=uID,y=Answer)) +
geom_point() +
geom_errorbar(aes(ymin=YMin,ymax=YMax),color="gray50",alpha=.5) +
geom_hline(yintercept=0, linetype="dashed", color = "red") +
theme(axis.text.x=element_blank(), axis.ticks.x=element_blank()) +
theme(panel.background = element_blank(), plot.background = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()) +
theme(panel.grid.major.y = element_line(colour="grey90", size=0.5)) +
ylab("Mean certainty rating") +
xlab("Discourses with \"believe\" (46)")
ggsave(f="../graphs/Figure3b.pdf",height=2.5,width=6)
# "tell"
tell <- droplevels(subset(dataM,dataM$Verb == "tell"))
head(tell)
length(unique(tell$uID)) # 18 discourses
mean_projT = aggregate(Answer~uID, data=tell, FUN="mean")
mean_projT$YMin = mean_projT$Answer - aggregate(Answer~uID, data=tell, FUN="ci.low")$Answer
mean_projT$YMax = mean_projT$Answer + aggregate(Answer~uID, data=tell, FUN="ci.high")$Answer
mean_projT <- mutate(mean_projT,uID = reorder(uID, Answer, mean))
ggplot(mean_projT, aes(x=uID,y=Answer)) +
geom_point() +
geom_errorbar(aes(ymin=YMin,ymax=YMax),color="gray50",alpha=.5) +
#facet_wrap(~VerbNum,scales="free_x", ncol=4) +
geom_hline(yintercept=0, linetype="dashed", color = "red") +
theme(axis.text.x=element_blank(), axis.ticks.x=element_blank()) +
theme(panel.background = element_blank(), plot.background = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()) +
theme(panel.grid.major.y = element_line(colour="grey90", size=0.5)) +
ylab("Mean certainty rating") +
xlab("Discourses with \"tell\" (18)")
ggsave(f="../graphs/Figure3c.pdf",height=2.5,width=6)
# Figure 4 ----
factive = droplevels(subset(dataM,dataM$factive == "yes"))
# person for factive predicates
mean_Fperson = aggregate(Answer~Verb+MatSubjPer, data=factive, FUN="mean")
mean_Fperson$YMin = mean_Fperson$Answer - aggregate(Answer~Verb+MatSubjPer, data=factive, FUN="ci.low")$Answer
mean_Fperson$YMax = mean_Fperson$Answer + aggregate(Answer~Verb+MatSubjPer, data=factive, FUN="ci.high")$Answer
# add number of discourses
mean_Fperson = merge(mean_Fperson, verbNum, by="Verb")
mean_Fperson$VerbNum = paste(mean_Fperson$Verb, paste0("(",mean_Fperson$Freq,")"), sep=" ")
ggplot(mean_Fperson, aes(x=MatSubjPer,y=Answer)) +
geom_point() +
geom_errorbar(aes(ymin=YMin,ymax=YMax),color="gray50",alpha=.5) +
facet_wrap(~VerbNum) +
geom_hline(yintercept=0, linetype="dashed", color = "red") +
#theme(axis.text.x=element_blank(), axis.ticks.x=element_blank()) +
theme(panel.background = element_blank(), plot.background = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()) +
theme(panel.grid.major.y = element_line(colour="grey90", size=0.5)) +
ylab("Mean certainty rating") +
xlab("Person of predicate subject")
ggsave(f="../graphs/Figure4.pdf",height=4,width=6)
# Figure 5 ----
# tense x subject for factive
mean_Ftenseperson = aggregate(Answer~MatTense+MatSubjPer, data=factive, FUN="mean")
mean_Ftenseperson$YMin = mean_Ftenseperson$Answer - aggregate(Answer~MatTense+MatSubjPer, data=factive, FUN="ci.low")$Answer
mean_Ftenseperson$YMax = mean_Ftenseperson$Answer + aggregate(Answer~MatTense+MatSubjPer, data=factive, FUN="ci.high")$Answer
ggplot(mean_Ftenseperson, aes(x=MatSubjPer,y=Answer)) +
geom_point() +
geom_errorbar(aes(ymin=YMin,ymax=YMax),color="gray50",alpha=.5) +
facet_wrap(~MatTense) +
geom_hline(yintercept=0, linetype="dashed", color = "red") +
#theme(axis.text.x=element_blank(), axis.ticks.x=element_blank()) +
theme(panel.background = element_blank(), plot.background = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()) +
theme(panel.grid.major.y = element_line(colour="grey90", size=0.5)) +
scale_y_continuous(limits = c(-3,3)) +
ylab("Mean certainty rating") +
xlab("Factive predicates")
ggsave(f="../graphs/Figure5a.pdf",height=3,width=6)
# tense x subject for all predicates
tmp = droplevels(subset(dataM,dataM$MatTense != ""))
mean_tenseperson = aggregate(Answer~MatTense+MatSubjPer, data = tmp, FUN="mean")
mean_tenseperson$YMin = mean_tenseperson$Answer - aggregate(Answer~MatTense+MatSubjPer, data = tmp, FUN="ci.low")$Answer
mean_tenseperson$YMax = mean_tenseperson$Answer + aggregate(Answer~MatTense+MatSubjPer, data = tmp, FUN="ci.high")$Answer
ggplot(mean_tenseperson, aes(x=MatSubjPer,y=Answer)) +
geom_point() +
geom_errorbar(aes(ymin=YMin,ymax=YMax),color="gray50",alpha=.5) +
facet_wrap(~MatTense) +
geom_hline(yintercept=0, linetype="dashed", color = "red") +
#theme(axis.text.x=element_blank(), axis.ticks.x=element_blank()) +
theme(panel.background = element_blank(), plot.background = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()) +
theme(panel.grid.major.y = element_line(colour="grey90", size=0.5)) +
scale_y_continuous(limits = c(-3,3)) +
ylab("Mean certainty rating") +
xlab("All predicates")
ggsave(f="../graphs/Figure5b.pdf",height=3,width=6)
# Figure 6 ----
# embedding x genre for all predicates
mean_embeddinggenre = aggregate(Answer~genre+Embedding, data=dataM, FUN="mean")
mean_embeddinggenre$YMin = mean_embeddinggenre$Answer - aggregate(Answer~genre+Embedding, data=dataM, FUN="ci.low")$Answer
mean_embeddinggenre$YMax = mean_embeddinggenre$Answer + aggregate(Answer~genre+Embedding, data=dataM, FUN="ci.high")$Answer
ggplot(mean_embeddinggenre, aes(x=Embedding,y=Answer)) +
geom_point() +
geom_errorbar(aes(ymin=YMin,ymax=YMax),color="gray50",alpha=.5) +
facet_wrap(~genre) +
geom_hline(yintercept=0, linetype="dashed", color = "red") +
theme(axis.text.x = element_text(angle = 70, vjust = 0.5, hjust=0.4)) +
theme(panel.background = element_blank(), plot.background = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()) +
theme(panel.grid.major.y = element_line(colour="grey90", size=0.5)) +
scale_y_continuous(limits = c(-3,3)) +
ylab("Mean certainty rating") +
xlab("Embedding")
ggsave(f="../graphs/Figure6.pdf",height=3,width=6)
# Figure 7 ----
# know: embedding x genre
know <- droplevels(subset(dataM,dataM$Verb == "know"))
nrow(know) #1134
length(unique(know$uID)) #122
aknow = know %>%
select(uID, Verb, Embedding, genre, Target, Prompt, Answer) %>%
group_by(uID, Verb, Embedding, genre, Target, Prompt) %>%
summarize(Mean = mean(Answer), low.ci = ci.low(Answer), high.ci = ci.high(Answer), SD = sd(Answer), C = n())
aknow = as.data.frame(aknow)
xtabs(~ genre + Embedding, data = aknow)
# Embedding
# genre conditional modal negation question
# BNC 12 5 28 15
# SWBD 6 0 47 6
# WSJ 0 0 3 0
know$embNum = as.character(know$Embedding)
know[know$Embedding == "conditional" & know$genre == "BNC",]$embNum = "conditional (12)"
know[know$Embedding == "modal" & know$genre == "BNC",]$embNum = "modal (5)"
know[know$Embedding == "negation" & know$genre == "BNC",]$embNum = "negation (28)"
know[know$Embedding == "question" & know$genre == "BNC",]$embNum = "question (15)"
know[know$Embedding == "conditional" & know$genre == "SWBD",]$embNum = "conditional (6)"
know[know$Embedding == "negation" & know$genre == "SWBD",]$embNum = "negation (47)"
know[know$Embedding == "question" & know$genre == "SWBD",]$embNum = "question (6)"
know[know$Embedding == "negation" & know$genre == "WSJ",]$embNum = "negation (3)"
mean_kembeddinggenre = aggregate(Answer~genre+embNum, data=know, FUN="mean")
mean_kembeddinggenre$YMin = mean_kembeddinggenre$Answer - aggregate(Answer~genre+embNum, data=know, FUN="ci.low")$Answer
mean_kembeddinggenre$YMax = mean_kembeddinggenre$Answer + aggregate(Answer~genre+embNum, data=know, FUN="ci.high")$Answer
ggplot(mean_kembeddinggenre, aes(x=embNum,y=Answer)) +
geom_point() +
geom_errorbar(aes(ymin=YMin,ymax=YMax),color="gray50",alpha=.5) +
facet_wrap(~genre,scales="free") +
geom_hline(yintercept=0, linetype="dashed", color = "red") +
theme(axis.text.x = element_text(angle = 70, vjust = 0.5, hjust=0.4)) +
theme(panel.background = element_blank(), plot.background = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()) +
theme(panel.grid.major.y = element_line(colour="grey90", size=0.5)) +
scale_y_continuous(limits = c(-3,3)) +
ylab("Mean certainty rating") +
xlab("Embedding")
ggsave(f="../graphs/Figure7a.pdf",height=3,width=6)
know.conv2 <- droplevels(subset(know,know$MatSubjPer == "third"))
length(unique(know.conv2$uID)) #59 discourses
aknow3 = know.conv2 %>%
select(uID, Verb, Embedding, genre, Target, Prompt, Answer) %>%
group_by(uID, Verb, Embedding, genre, Target, Prompt) %>%
summarize(Mean = mean(Answer), low.ci = ci.low(Answer), high.ci = ci.high(Answer), SD = sd(Answer), C = n())
aknow = as.data.frame(aknow3)
xtabs(~ genre + Embedding, data = aknow3)
# Embedding
# genre conditional modal negation question
# BNC 10 5 22 8
# SWBD 4 0 6 1
# WSJ 0 0 3 0
aknow3$embNum = as.character(aknow3$Embedding)
aknow3[aknow3$Embedding == "conditional" & aknow3$genre == "BNC",]$embNum = "conditional (10)"
aknow3[aknow3$Embedding == "modal" & aknow3$genre == "BNC",]$embNum = "modal (5)"
aknow3[aknow3$Embedding == "negation" & aknow3$genre == "BNC",]$embNum = "negation (22)"
aknow3[aknow3$Embedding == "question" & aknow3$genre == "BNC",]$embNum = "question (8)"
aknow3[aknow3$Embedding == "conditional" & aknow3$genre == "SWBD",]$embNum = "conditional (4)"
aknow3[aknow3$Embedding == "negation" & aknow3$genre == "SWBD",]$embNum = "negation (6)"
aknow3[aknow3$Embedding == "question" & aknow3$genre == "SWBD",]$embNum = "question (1)"
aknow3[aknow3$Embedding == "negation" & aknow3$genre == "WSJ",]$embNum = "negation (3)"
mean_k3embeddinggenre = aggregate(Answer~genre+embNum, data=know.conv2, FUN="mean")
mean_k3embeddinggenre$YMin = mean_k3embeddinggenre$Answer - aggregate(Answer~genre+embNum, data=know.conv2, FUN="ci.low")$Answer
mean_k3embeddinggenre$YMax = mean_k3embeddinggenre$Answer + aggregate(Answer~genre+embNum, data=know.conv2, FUN="ci.high")$Answer
ggplot(mean_k3embeddinggenre, aes(x=embNum,y=Answer)) +
geom_point() +
geom_errorbar(aes(ymin=YMin,ymax=YMax),color="gray50",alpha=.5) +
facet_wrap(~genre,scales="free") +
geom_hline(yintercept=0, linetype="dashed", color = "red") +
theme(axis.text.x = element_text(angle = 70, vjust = 0.5, hjust=0.4)) +
theme(panel.background = element_blank(), plot.background = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank()) +
theme(panel.grid.major.y = element_line(colour="grey90", size=0.5)) +
scale_y_continuous(limits = c(-3,3)) +
ylab("Mean certainty rating") +
xlab("Embedding")
ggsave(f="../graphs/Figure7b.pdf",height=3,width=6)
## Models
dataM$Answer <- as.factor(dataM$Answer)
# Models: predicting projection from factivity (section 3.1.1) ----
m.factive <- clmm(Answer ~ factive + (1|WorkerID), data = dataM)
summary(m.factive) # factive predicates higher ratings than non-factive
# Models: Table 2: Nagelkerke R^2 ----
# null model
m.null = clmm(Answer ~ 1 + (1|WorkerID), data = dataM)
# factivity
nagelkerke(fit = m.factive, null = m.null) #.126
# genre (WSJ, BNC, SWBD)
table(dataM$genre)
m.genre <- clmm(Answer ~ genre + (1|WorkerID), data = dataM)
summary(m.genre) #SWBD and WSJ lower ratings than BNC
nagelkerke(fit = m.genre, null = m.null) #6.2
# predicate lemma
m.verb <- clmm(Answer ~ Verb + (1|WorkerID), data = dataM)
summary(m.verb)
nagelkerke(fit = m.verb, null = m.null) #22.8
nagelkerke(fit = m.verb, null = m.factive) #11.6
# Embedding
table(dataM$Embedding)
m.embedding <- clmm(Answer ~ Embedding + (1|WorkerID), data = dataM)
summary(m.embedding) # modal higher than conditional, negation lower than conditional, no diff to question
nagelkerke(fit = m.embedding, null = m.null) #9.1
# predicate tense
#at time of publication, there were a few items without tense annotation, these got fixed after publication -- hence some slight discrepancies with what is reported in the paper
table(dataM$MatTense)
m.matTense <- clmm(Answer ~ MatTense + (1|WorkerID), data = droplevels(subset(dataM,dataM$MatTense != "")))
summary(m.matTense) #past and present lower than future
nagelkerke(fit = m.matTense, null = m.null) #4.1 (different than what was reported in the paper)
# person of predicate subject (where 1/3sg and 1/3pl are different)
dataM$PersonSub = paste(dataM$MatSubjPer,dataM$MatSubjNum)
dataM$PersonSub = as.factor(dataM$PersonSub)
m.matPersonSubj <- clmm(Answer ~ PersonSub + (1|WorkerID), data = dataM)
nagelkerke(fit = m.matPersonSubj, null = m.null) #12.4 (this was reported for "person of predicate subject")
# not reported in the paper
# person of predicate subject (as 1, 2 or 3 person)
table(dataM$MatSubjPer)
m.matSubjPer <- clmm(Answer ~ MatSubjPer + (1|WorkerID), data = dataM)
summary(m.matSubjPer) #second and third higher than first
nagelkerke(fit = m.matSubjPer, null = m.null) #11.7
# not reported in the paper
# matrix subject number ("unknown" are "you" items, all BNC and WSJ codes whether "you" is generic or not: ImpMatSubj)
table(data$MatSubjNum)
m.matSubjNum <- clmm(Answer ~ MatSubjNum + (1|WorkerID), data = data)
summary(m.matSubjNum) # singular lower than plural, unknown higher than plural
nagelkerke(fit = m.matSubjNum, null = m.null) #4
# Models: Table 3 ----
m.FactiveR <- clmm(Answer ~ factive + (1|WorkerID) + (1|Verb), data = dataM)
summary(m.FactiveR)
ranef(m.FactiveR)$Verb
# Models (section 3.2): predicting certainty ratings from tense and subject person, and their interaction ----
# Paper: "This is true for models fitted to the factive predicates and models
# fitted to all predicates. (All models included random by-annotator intercepts.)"
# factive predicates
m.TensePersonIF <- clmm(as.factor(Answer) ~ MatTense * MatSubjPer + (1|WorkerID), data = factive)
m.TensePersonAF <- clmm(as.factor(Answer) ~ MatTense + MatSubjPer + (1|WorkerID), data = factive)
anova(m.TensePersonIF, m.TensePersonAF)
# model with interaction is better
# all predicates
dataM$MatSubjPer <- relevel(dataM$MatSubjPer, ref = "first")
dataM$MatTense <- relevel(dataM$MatTense, ref = "present")
m.TensePersonI <- clmm(Answer ~ MatTense * MatSubjPer + (1|WorkerID), data = droplevels(subset(dataM,dataM$MatTense != "")))
m.TensePersonA <- clmm(Answer ~ MatTense + MatSubjPer + (1|WorkerID), data = droplevels(subset(dataM,dataM$MatTense != "")))
anova(m.TensePersonI, m.TensePersonA)
# model with interaction is better
# Models (section 3.4): plausibility of the CC given the context ----
# mean no target rating
table(dataM$mean.noTarget)
nrow(dataM[!is.na(dataM$mean.noTarget),])
tmp = droplevels(subset(dataM,!is.na(dataM$mean.noTarget)))
nrow(tmp)
names(tmp)
m.nullS = clmm(Answer ~ 1 + (1|WorkerID), data = tmp)
m.mean.noT = clmm(Answer ~ mean.noTarget + (1|WorkerID), data = tmp)
summary(m.mean.noT) # the higher the mean no target rating, the higher the projection rating
nagelkerke(fit = m.mean.noT, null = m.nullS) #0.036
# Models (section 3.5): models with all predictors ----
#"We fitted an ordinal mixed effects model that predicts the certainty ratings from fixed effects of
# embedding and genre, and their interaction, as well as tense and person, and their interaction,
# and the predicate lemma. We again included random by-annotator intercepts."
# full data, with interaction between gender and embedding
m.null <- clmm (Answer ~ 1 + (1|WorkerID), data = dataM)
m.Verb.i <- clmm (Answer ~ Verb + genre + Embedding + genre:Embedding + MatSubjPer + MatTense + (1|WorkerID), data = dataM)
m.Factive.i <- clmm (Answer ~ factive + genre + Embedding + genre:Embedding + MatSubjPer + MatTense + (1|WorkerID), data = dataM)
nagelkerke(fit = m.Verb.i, null = m.null) #34.4
nagelkerke(fit = m.Factive.i, null = m.null) #28.9
drop1(m.Verb.i,test="Chi")
# all factors matter (interaction for genre and Embedding, no info on individual factors)
# Single term deletions
#
# Model:
# Answer ~ Verb + genre + Embedding + genre:Embedding + MatSubjPer +
# MatTense + (1 | WorkerID)
# Df AIC
# <none> 32592
# Verb 44 34542
# MatSubjPer 2 32721
# MatTense 2 32605
# genre:Embedding 6 32746
# LRT
# <none>
# Verb 2038.34
# MatSubjPer 133.32
# MatTense 17.06
# genre:Embedding 165.49
# Pr(>Chi)
# <none>
# Verb < 2.2e-16 ***
# MatSubjPer < 2.2e-16 ***
# MatTense 0.0001972 ***
# genre:Embedding < 2.2e-16 ***
# ---
# Signif. codes:
# 0 '***' 0.001 '**' 0.01
# '*' 0.05 '.' 0.1 ' ' 1
# full data, without interaction between gender and embedding
m.Verb.s <- clmm (Answer ~ Verb + genre + Embedding + MatSubjPer + MatTense + (1|WorkerID), data = dataM)
m.Factive.s <- clmm (Answer ~ factive + genre + Embedding + MatSubjPer + MatTense + (1|WorkerID), data = dataM)
nagelkerke(fit = m.Verb.s, null = m.null) #0.333
nagelkerke(fit = m.Factive.s, null = m.null) #0.275
drop1(m.Verb.s,test="Chi")
# all factors matter
# Single term deletions
#
# Model:
# Answer ~ Verb + genre + Embedding + MatSubjPer + MatTense + (1 |
# WorkerID)
# Df AIC LRT Pr(>Chi)
# <none> 32746
# Verb 44 34695 2037.29 < 2.2e-16 ***
# genre 2 32759 17.68 0.0001447 ***
# Embedding 3 32952 212.47 < 2.2e-16 ***
# MatSubjPer 2 33046 304.78 < 2.2e-16 ***
# MatTense 2 32759 17.07 0.0001965 ***
# ---
# Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#"We also fitted a variant of this model which includes the plausibility means of the CCs as a fixed effect,
# for the 558 BNC and WSJ discourses for which we have such annotations.
# plausibility-annotated data, with interaction between genre and embedding
m.nullS <- clmm(Answer ~ 1 + (1|WorkerID), data = droplevels(subset(dataM,!is.na(dataM$mean.noTarget))))
m.s.AllV.i <- clmm (Answer ~ Verb + genre + Embedding + genre:Embedding + MatSubjPer + MatTense + mean.noTarget + (1|WorkerID), data = droplevels(subset(dataM,!is.na(dataM$mean.noTarget))))
m.s.AllF.i <- clmm (Answer ~ factive + genre + Embedding + genre:Embedding + MatSubjPer + MatTense + mean.noTarget + (1|WorkerID), data = droplevels(subset(dataM,!is.na(dataM$mean.noTarget))))
nagelkerke(fit = m.s.AllV.i, null = m.nullS) #30.9
nagelkerke(fit = m.s.AllF.i, null = m.nullS) #23.7
drop1(m.s.AllV.i,test="Chi")
# all factors significant except MatTense and the interaction between genre and embedding
# (different than what was reported in the paper)
# Single term deletions
#
# Model:
# Answer ~ Verb + genre + Embedding + genre:Embedding + MatSubjPer +
# MatTense + mean.noTarget + (1 | WorkerID)
# Df AIC
# <none> 19081
# Verb 35 20437
# MatSubjPer 2 19103
# MatTense 2 19082
# mean.noTarget 1 19162
# genre:Embedding 3 19080
# LRT
# <none>
# Verb 1426.81
# MatSubjPer 26.86
# MatTense 5.38
# mean.noTarget 83.93
# genre:Embedding 5.14
# Pr(>Chi)
# <none>
# Verb < 2.2e-16 ***
# MatSubjPer 1.473e-06 ***
# MatTense 0.06802 .
# mean.noTarget < 2.2e-16 ***
# genre:Embedding 0.16183
# ---
# Signif. codes:
# 0 '***' 0.001 '**' 0.01
# '*' 0.05 '.' 0.1 ' ' 1
# plausibility-annotated data, without interaction between genre and embedding
m.s.AllV.s <- clmm (Answer ~ Verb + genre + Embedding + MatSubjPer + MatTense + mean.noTarget + (1|WorkerID), data = droplevels(subset(dataM,!is.na(dataM$mean.noTarget))))
m.s.AllF.s <- clmm (Answer ~ factive + genre + Embedding + MatSubjPer + MatTense + mean.noTarget + (1|WorkerID), data = droplevels(subset(dataM,!is.na(dataM$mean.noTarget))))
nagelkerke(fit = m.s.AllV.s, null = m.nullS) #30.9
nagelkerke(fit = m.s.AllF.s, null = m.nullS) #23.4
drop1(m.s.AllV.s,test="Chi")
# all factors matter, except genre and matrix tense (the latter is marginal)
# (different than what was reported in the paper)
# Single term deletions
#
# Model:
# Answer ~ Verb + genre + Embedding + MatSubjPer + MatTense + mean.noTarget +
# (1 | WorkerID)
# Df AIC LRT Pr(>Chi)
# <none> 19080
# Verb 35 20470 1459.75 < 2.2e-16 ***
# genre 1 19079 1.63 0.20128
# Embedding 3 19172 98.27 < 2.2e-16 ***
# MatSubjPer 2 19102 26.32 1.926e-06 ***
# MatTense 2 19081 5.01 0.08147 .
# mean.noTarget 1 19163 84.94 < 2.2e-16 ***
# ---
# Signif. codes:
# 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1