means of Natural Language Processing Viorel Morari Chair of Web - - PowerPoint PPT Presentation
Mining Rhetorical Devices by means of Natural Language Processing Viorel Morari Chair of Web Technology and Information Systems Master Thesis Defense January 23 rd , 2018 viorel.morari@uni-weimar.de Prof. Dr. Benno Stein Advisor: Khalid
Chair of Web Technology and Information Systems
Viorel Morari viorel.morari@uni-weimar.de Master Thesis Defense January 23rd , 2018 Advisor: Khalid Al-Khatib
Bob
2
PERSUASION
Bob
3
PERSUASION
Bob
4
PERSUASION
…40% whiter than the other leading brands This is a must-see movie …the biggest, the brightest and the best. If you buy it now…
5
Rhetoric Invention Arrangement Style Memory Delivery
6
Rhetoric Invention Arrangement Style Memory Delivery
Happiness Sad Drink Bottle Feelings Open happiness. Grab a bottle. Don’t be sad!
Bob
7
Rhetoric Invention Arrangement Style Memory Delivery
Don’t be sad! Grab a bottle. Open happiness. Open happiness. Grab a bottle. Don’t be sad!
Bob
8
Rhetoric Invention Arrangement Style Memory Delivery
Don’t be sad! Grab a bottle. Open happiness. Feeling down? Open a bottle, open happiness! Rhetorical question Repetition, Balance
Bob
9
Rhetoric Invention Arrangement Style Memory Delivery
Feeling down? Open a bottle, open happiness! Feeling down? Open a bottle, open happiness! Feeling down? Open a bottle, open happiness!
Bob
10
Rhetoric Invention Arrangement Style Memory Delivery
Feeling down Open a bottle open happiness
Bob
11
Rhetoric Invention Arrangement Style Memory Delivery
12
Rhetoric Invention Arrangement Style Rhetorical Devices
Schemes
(syntax)
Tropes
(semantic)
Figures of thought Memory Delivery
13
Rhetoric Invention Arrangement Style Rhetorical Devices
Schemes
(syntax)
Tropes
(semantic)
Figures of thought Memory Delivery
14
Rhetoric-based NLG system
15
Rhetorical style suggestion system
16
17
18
19
Genres Topics
Authors
20
21
input
22
23
24
flexible domain specific language for defining patterns of annotations (Klügl et al. [2016]).
25
… . This is a sample sentence. …
linguistic analysis.
▪ Stanford Parser ▪ Stanford Dependencies
26
27
28
But I must explain to you how all this mistaken idea of denouncing pleasure and praising pain was born and I will give you a complete account of the system, and expound the actual teachings of the great explorer of the truth, the master- builder of human happiness. No
pleasure itself, because it is pleasure.
Device A Device B Device C
29
Balance schemes
Interplay between equivalent ideas Control the rhythm of thought
Custom schemes Omission schemes Repetition schemes
30
Omission schemes Balance schemes
Interplay between equivalent ideas Control the rhythm of thought Deliberate omission of intuitive words Cause incompleteness
Custom schemes Repetition schemes
31
Repetition schemes Omission schemes Balance schemes
Interplay between equivalent ideas Control the rhythm of thought Deliberate omission of intuitive words Cause incompleteness Repetition of key words/ideas Used for emphasis or amplification Key to persuasion (according to Aristotle)
Custom schemes
32
Custom schemes Omission schemes Balance schemes
Interplay between equivalent ideas Control the rhythm of thought Deliberate omission of intuitive words Cause incompleteness
Repetition schemes
Repetition of key words/ideas Used for emphasis or amplification Key to persuasion (according to Aristotle) Informal rhetorical devices Strong emotional effect Includes causality, comparatives and voice
33
Custom schemes Omission schemes Balance schemes Repetition schemes
Adjectives/Adverbs
Adjectives/Adverbs
34
Custom schemes Omission schemes Balance schemes Repetition schemes
Adjectives/Adverbs
Adjectives/Adverbs
35
Old farmer had a pig, a dog, a cow and a horse.
UIMA Ruta
Old farmer had a pig, a dog, a cow and a horse.
36
Enumeration - a rhetorical device used to list a series of details, words or phrases. (literarydevices.net)
Old farmer had a pig, a dog, a cow and a horse.
UIMA Ruta
Old farmer had a pig, a dog, a cow and a horse.
37
Enumeration - a rhetorical device used to list a series of details, words or phrases. (literarydevices.net)
Old farmer had a pig, a dog, a cow and a horse.
UIMA Ruta
Old farmer had a pig, a dog, a cow and a horse.
38
Enumeration - a rhetorical device used to list a series of details, words or phrases. (literarydevices.net)
Old farmer had a pig, a dog, a cow and a horse.
UIMA Ruta
Old farmer had a pig, a dog, a cow and a horse.
39
Enumeration - a rhetorical device used to list a series of details, words or phrases. (literarydevices.net)
Old farmer had a pig, a dog, a cow and a horse.
UIMA Ruta
Old farmer had a pig, a dog, a cow and a horse.
40
Enumeration - a rhetorical device used to list a series of details, words or phrases. (literarydevices.net)
Old farmer had a pig, a dog, a cow and a horse.
UIMA Ruta
Old farmer had a pig, a dog, a cow and a horse.
41
Enumeration - a rhetorical device used to list a series of details, words or phrases. (literarydevices.net)
Old farmer had a pig, a dog, a cow and a horse.
UIMA Ruta
Old farmer had a pig, a dog, a cow and a horse.
42
Enumeration - a rhetorical device used to list a series of details, words or phrases. (literarydevices.net)
Old farmer had a pig, a dog, a cow and a horse.
UIMA Ruta
Old farmer had a pig, a dog, a cow and a horse.
43
Enumeration - a rhetorical device used to list a series of details, words or phrases. (literarydevices.net)
Old farmer had a pig, a dog, a cow and a horse.
UIMA Ruta
Old farmer had a pig, a dog, a cow and a horse.
44
Enumeration - a rhetorical device used to list a series of details, words or phrases. (literarydevices.net)
Old farmer had a pig, a dog, a cow and a horse.
UIMA Ruta
Old farmer had a pig, a dog, a cow and a horse.
45
Enumeration - a rhetorical device used to list a series of details, words or phrases. (literarydevices.net)
Old farmer had a pig, a dog, a cow and a horse.
UIMA Ruta
Old farmer had a pig, a dog, a cow and a horse.
46
Enumeration - a rhetorical device used to list a series of details, words or phrases. (literarydevices.net)
Old farmer had a pig, a dog, a cow and a horse.
UIMA Ruta
Old farmer had a pig, a dog, a cow and a horse.
47
Old farmer had a pig, a dog, a cow and a horse.
UIMA Ruta
Old farmer had a pig, a dog, a cow and a horse.
48
Enumeration - a rhetorical device used to list a series of details, words or phrases. (literarydevices.net)
A rooster, a prince and a lion walk into a bar…
Hypozeugma - placing last, in a construction containing several words or phrases of equal value, the word or words on which all of them depend. (Silva Rhetoricae)
49
A rooster, a prince and a lion walk into a bar…
Hypozeugma - placing last, in a construction containing several words or phrases of equal value, the word or words on which all of them depend. (Silva Rhetoricae)
50
A rooster, a prince and a lion walk into a bar…
Stanford Dependencies Hypozeugma - placing last, in a construction containing several words or phrases of equal value, the word or words on which all of them depend. (Silva Rhetoricae)
A rooster, a prince and a lion walk into a bar…
51
A rooster, a prince and a lion walk into a bar…
Stanford Dependencies Hypozeugma - placing last, in a construction containing several words or phrases of equal value, the word or words on which all of them depend. (Silva Rhetoricae)
A rooster, a prince and a lion walk into a bar…
governor-dependent relation
52
A rooster, a prince and a lion walk into a bar…
Stanford Dependencies Hypozeugma - placing last, in a construction containing several words or phrases of equal value, the word or words on which all of them depend. (Silva Rhetoricae)
A rooster, a prince and a lion walk into a bar… rooster, a prince and a lion walk
UIMA Ruta
53
A rooster, a prince and a lion walk into a bar…
Stanford Dependencies Hypozeugma - placing last, in a construction containing several words or phrases of equal value, the word or words on which all of them depend. (Silva Rhetoricae)
A rooster, a prince and a lion walk into a bar… rooster, a prince and a lion walk
UIMA Ruta
54
A rooster, a prince and a lion walk into a bar…
Stanford Dependencies Hypozeugma - placing last, in a construction containing several words or phrases of equal value, the word or words on which all of them depend. (Silva Rhetoricae)
A rooster, a prince and a lion walk into a bar… rooster, a prince and a lion walk
UIMA Ruta
55
A rooster, a prince and a lion walk into a bar…
Stanford Dependencies Hypozeugma - placing last, in a construction containing several words or phrases of equal value, the word or words on which all of them depend. (Silva Rhetoricae)
A rooster, a prince and a lion walk into a bar… rooster, a prince and a lion walk
UIMA Ruta
56
A rooster, a prince and a lion walk into a bar…
Stanford Dependencies Hypozeugma - placing last, in a construction containing several words or phrases of equal value, the word or words on which all of them depend. (Silva Rhetoricae)
A rooster, a prince and a lion walk into a bar… rooster, a prince and a lion walk
UIMA Ruta
57
A rooster, a prince and a lion walk into a bar…
Stanford Dependencies Hypozeugma - placing last, in a construction containing several words or phrases of equal value, the word or words on which all of them depend. (Silva Rhetoricae)
A rooster, a prince and a lion walk into a bar… rooster, a prince and a lion walk
UIMA Ruta
58
A rooster, a prince and a lion walk into a bar…
Stanford Dependencies Hypozeugma - placing last, in a construction containing several words or phrases of equal value, the word or words on which all of them depend. (Silva Rhetoricae)
A rooster, a prince and a lion walk into a bar… rooster, a prince and a lion walk
UIMA Ruta
59
Our eyes saw it, but we could not believe our eyes.
Epanalepsis - repeats the beginning word of a sentence at the end.
60
Our eyes saw it, but we could not believe our eyes.
Epanalepsis - repeats the beginning word of a sentence at the end.
61
Our eyes saw it, but we could not believe our eyes.
Epanalepsis - repeats the beginning word of a sentence at the end.
Our eyes saw believe our eyes.
62
Our eyes saw it, but we could not believe our eyes.
Epanalepsis - repeats the beginning word of a sentence at the end.
Our eyes saw believe our eyes.
63
Our eyes saw it, but we could not believe our eyes.
Epanalepsis - repeats the beginning word of a sentence at the end.
Our eyes saw believe our eyes.
64
Our eyes saw it, but we could not believe our eyes.
Epanalepsis - repeats the beginning word of a sentence at the end.
Our eyes saw believe our eyes.
65
Our eyes saw it, but we could not believe our eyes.
Epanalepsis - repeats the beginning word of a sentence at the end.
Our eyes saw believe our eyes.
66
Our eyes saw it, but we could not believe our eyes.
Epanalepsis - repeats the beginning word of a sentence at the end.
Our eyes saw believe our eyes.
67
Our eyes saw it, but we could not believe our eyes.
Epanalepsis - repeats the beginning word of a sentence at the end.
Our eyes saw believe our eyes.
68
If I were president, I would cut taxes.
If-conditional 2 - expresses consequences that are totally unrealistic or will not likely happen in the future.
69
If I were president, I would cut taxes.
If-conditional 2 - expresses consequences that are totally unrealistic or will not likely happen in the future.
If I were president, I would cut taxes.
Stanford Dependencies
70
If I were president, I would cut taxes.
If-conditional 2 - expresses consequences that are totally unrealistic or will not likely happen in the future.
If I were president, I would cut taxes.
Stanford Dependencies
governor-dependent relations
71
If I were president, I would cut taxes.
If-conditional 2 - expresses consequences that are totally unrealistic or will not likely happen in the future.
If I were president, I would cut taxes.
Stanford Dependencies
72
If I were president, I would cut taxes.
If-conditional 2 - expresses consequences that are totally unrealistic or will not likely happen in the future.
If I were president, I would cut taxes.
Stanford Dependencies
P-clause Q-clause
73
If I were president, I would cut taxes.
If-conditional 2 - expresses consequences that are totally unrealistic or will not likely happen in the future. Stanford Dependencies
If I were president
UIMA Ruta
If I were president, I would cut taxes. I would cut
74
If I were president, I would cut taxes.
If-conditional 2 - expresses consequences that are totally unrealistic or will not likely happen in the future. Stanford Dependencies UIMA Ruta
If I were president, I would cut taxes. If I were president I would cut
75
If I were president, I would cut taxes.
If-conditional 2 - expresses consequences that are totally unrealistic or will not likely happen in the future. Stanford Dependencies UIMA Ruta
If I were president, I would cut taxes. I would cut If I were president
76
If I were president, I would cut taxes.
If-conditional 2 - expresses consequences that are totally unrealistic or will not likely happen in the future. Stanford Dependencies UIMA Ruta
If I were president, I would cut taxes. If I were president I would cut
77
If I were president, I would cut taxes.
If-conditional 2 - expresses consequences that are totally unrealistic or will not likely happen in the future. Stanford Dependencies UIMA Ruta
If I were president, I would cut taxes. I would cut If I were president
78
If I were president, I would cut taxes.
If-conditional 2 - expresses consequences that are totally unrealistic or will not likely happen in the future. Stanford Dependencies UIMA Ruta
If I were president, I would cut taxes. I would cut If I were president
79
80
𝑄𝑠𝑓𝑑𝑗𝑡𝑗𝑝𝑜 = 𝑢𝑞 𝑢𝑞 + 𝑔𝑞 𝑆𝑓𝑑𝑏𝑚𝑚 = 𝑢𝑞 𝑢𝑞 + 𝑔𝑜 𝐺1 𝑡𝑑𝑝𝑠𝑓 = 2 ∙ 𝑞𝑠𝑓𝑑𝑗𝑡𝑗𝑝𝑜 ∙ 𝑠𝑓𝑑𝑏𝑚𝑚 𝑞𝑠𝑓𝑑𝑗𝑡𝑗𝑝𝑜 + 𝑠𝑓𝑑𝑏𝑚𝑚
Evaluation measures
81
0.4 0.69 0.69
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Omission schemes
Asynd. Epizeugma Hypozeugma
0.84 0.68 1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Balance schemes
Enumeration Isocolon Pysma
82
0.4 0.69 0.69
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Omission schemes
Asynd. Epizeugma Hypozeugma
0.84 0.68 1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Balance schemes
Enumeration Isocolon Pysma
0.95 0.25
Precision Recall
83
0.4 0.69 0.69
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Omission schemes
Asynd. Epizeugma Hypozeugma
0.84 0.68 1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Balance schemes
Enumeration Isocolon Pysma
0.95 0.25
Precision Recall
Old farmer had a pig, a dog, a cow and a horse.
Asyndeton Enumeration
84
0.74 0.74 0.72 0.74 0.78 0.59 0.4 0.73
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Repetition schemes
Anadiplosis Diacope Epanalepsis Epiphoza Epizeuxis Mesarchia Mesod. Polysyndeton
0.4 0.69 0.69
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Omission schemes
Asynd. Epizeugma Hypozeugma
0.73 0.78 0.78 0.74 0.85 0.87 0.56 0.61 0.67 0.56 1 0.91
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Custom schemes
If-cond 0 If-cond 1 If-cond 2 If-cond 3 If-counterFact Passive Voice
Unless-cond Whether-cond
0.84 0.68 1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Balance schemes
Enumeration Isocolon Pysma
85
0.74 0.74 0.72 0.74 0.78 0.59 0.4 0.73
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Repetition schemes
Anadiplosis Diacope Epanalepsis Epiphoza Epizeuxis Mesarchia Mesod. Polysyndeton
0.4 0.69 0.69
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Omission schemes
Asynd. Epizeugma Hypozeugma
0.73 0.78 0.78 0.74 0.85 0.87 0.56 0.61 0.67 0.56 1 0.91
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Custom schemes
If-cond 0 If-cond 1 If-cond 2 If-cond 3 If-counterFact Passive Voice
Unless-cond Whether-cond
0.84 0.68 1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Balance schemes
Enumeration Isocolon Pysma
86
0.74 0.74 0.72 0.74 0.78 0.59 0.4 0.73
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Repetition schemes
Anadiplosis Diacope Epanalepsis Epiphoza Epizeuxis Mesarchia Mesod. Polysyndeton
0.4 0.69 0.69
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Omission schemes
Asynd. Epizeugma Hypozeugma
0.73 0.78 0.78 0.74 0.85 0.87 0.56 0.61 0.67 0.56 1 0.91
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Custom schemes
If-cond 0 If-cond 1 If-cond 2 If-cond 3 If-counterFact Passive Voice
Unless-cond Whether-cond
0.84 0.68 1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Balance schemes
Enumeration Isocolon Pysma
87
88
89
90
91
Frequency Combinations
+ + + + + +
Group distribution
92
Confidence interval
n * precision (rd) n/recall (rd)
Al-Khatib et al. [2017] Device Precision Recall
93
Significance Test Effect-size Test
94
A B A B
Type Topic
95
96
97
The New York Times US Presidential Debates 2016
Ben Wiseman [2016]
98
Language English Mode Written Communication Monological Medium Newspaper Type Descriptive Genre Editorial Topic Education Author Identity Audience U.S. Presidential Debates Argumentative Review Biography Debate Science Art Politics
99
Random Article-length based Matching
1000 articles 600 articles 343 articles
NYT Dataset
100
Rhetoric Detection System Rhetoric Detection System “Random” dataset “Article-length based” dataset Articles cover multiple dimensions Hard to deduce particular styles
101
Rhetoric Detection System Rhetoric Detection System “Random” dataset “Article-length based” dataset Articles cover multiple dimensions Hard to deduce particular styles CONFOUNDING
102
Author
(independent variable)
Genre
(confounding variable)
Rhetorical style
(dependent variable) 103
Author
(independent variable)
Genre
(confounding variable)
Rhetorical style
(dependent variable)
MATCHING
104
Genre 1 Genre 2 Genre 3 Genre 4
105
Genre 1 Genre 2 Genre 3 Genre 4
106
Genre 1 Genre 2 Genre 3 Genre 4
107
Genre 1 Genre 2 Genre 3 Genre 4
108
Articles written by author X
Genre 1 Genre 2 Genre 3 Genre 4
109
Genre 1 Genre 2 Genre 3 Genre 4
110
Genre 1 Genre 2 Genre 3 Genre 4
111
Genre 1 Genre 2 Genre 3 Genre 4
112
Genre 1 Genre 2 Author 1 Author 2
113
Topic 1 Topic 2
114
6.28 17.92 6.75 3.26 4.34 6.05 8.53 6.05 0.62 13.65 0.85 0.08 0.39 0.93 0.47 5.04 4.03 28.08 15.52 0.78 4.27 2.02 10.78 0.23 0.47 0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 Anadiplosis Asyndeton comparativeAdj comparativeAdv Diacope Epanalepsis Epiphoza Epizeugma Epizeuxis Hypozeugma IfCondOne IfCondThree IfCondTwo IfCondZero IfCounterfactual Isocolon Mesarchia Mesodiplosis PassiveVoice Polysyndeton Pysma superlativeAdj superlativeAdv TripleEnumeration Unless Whether
Genres: Review distribution
5.19 16.99 10.28 4.98 3.79 2.71 10.28 7.79 0.43 15.48 1.95 0.22 0.76 1.62 0.76 6.17 3.57 24.35 14.61 0.76 0.43 6.28 1.62 10.71 0.32 0.32 0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 Anadiplosis Asyndeton comparativeAdj comparativeAdv Diacope Epanalepsis Epiphoza Epizeugma Epizeuxis Hypozeugma IfCondOne IfCondThree IfCondTwo IfCondZero IfCounterfactual Isocolon Mesarchia Mesodiplosis PassiveVoice Polysyndeton Pysma superlativeAdj superlativeAdv TripleEnumeration Unless Whether
Genres: Editorial distribution 115
Style-based frequency of rhetorical devices
116
Authors
117
! Same pattern across all articles
Authors
118
Authors
Good Job, Lewis!
119
Authors
120
Genres
Comparatives
121
author topic
Genres
122
Genres: tests’ results
123
Topics
124
author genre
Style-based frequency of rhetorical devices Characteristic style patterns within each dimension
125
Topics: tests’ results
126
Style is more author- and genre-dependent
127
Style-based frequency of rhetorical devices Characteristic style patterns within each dimension
CLINTON TRUMP REST
CLINTON
CLINTON TRUMP
REST
TRUMP
128
129
Asyndeton = clarity and rhythm
130
Acceptance Speech Analysis by Huffington Post
131
Acceptance Speech Analysis by Huffington Post
132
Acceptance Speech Analysis by Huffington Post
133
Significance Test
134
Trump doesn’t change his style
Significance Test
135
Conclusions
Vague style patterns across random and article- length based subsampling: Confounding Better style identification with Matching Rhetorical style depends more on author and genre of writings rather than their topics Debates: candidates employ different styles Debates: domain experience trains an adaptive rhetorical style
136
System for rhetorical style identification in high-quality text documents Rule-based algorithms for detection of RD
Resources
Novel framework for detecting rhetorical devices Comprehensive dataset for evaluation of rhetoric detection systems Elaborative style patterns and intriguing findings
137
Conclusions
Vague style patterns across random and article- length based subsampling: Confounding Better style identification with Matching Rhetorical style depends more on author and genre of writings rather than their topics Debates: candidates employ different styles Debates: domain experience trains an adaptive rhetorical style System for rhetorical style identification in high-quality text documents Rule-based algorithms for detection of RD
Resources
Novel framework for detecting rhetorical devices Elaborative style patterns and intriguing findings
Efficiency
1st sentence → 5.8 sec. 2nd sentence → 0.4 sec. Initialization → 1.7 sec.
138
Conclusions
Vague style patterns across random and article- length based subsampling: Confounding Better style identification with Matching Rhetorical style depends more on author and genre of writings rather than their topics Debates: candidates employ different styles Debates: domain experience trains an adaptive rhetorical style System for rhetorical style identification in high-quality text documents Rule-based algorithms for detection of RD Comprehensive dataset for evaluation of rhetoric detection systems
Resources
Novel framework for detecting rhetorical devices Elaborative style patterns and intriguing findings
Future Work
Larger dataset for analysis Focus of semantical rhetoric Analysis measures like placement and flows of rhetorical devices
Efficiency
139
Conclusions
Vague style patterns across random and article- length based subsampling: Confounding Better style identification with Matching Rhetorical style depends more on author and genre of writings rather than their topics Debates: candidates employ different styles Debates: domain experience trains an adaptive rhetorical style System for rhetorical style identification in high-quality text documents Rule-based algorithms for detection of RD Comprehensive dataset for evaluation of rhetoric detection systems 1st sentence → 5.8 sec. 2nd sentence → 0.4 sec. Initialization → 1.7 sec.
140
stops/my-debate-nightmare-a-duller-donald-trump.html
Argumentation Strategies across Topics. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing (EMNLP 17), pages 1362–1368. Association for Computational Linguistics, September 2017. URL http://aclweb.org/anthology/D17-1142.
141
142
election/presidential-debate-schedule/
143
Device Total No. Precision Recall F1-score Device Total No. Precision Recall F1-score Anadiplosis 60 0.76 0.73 0.74 If Conditional Two 60 0.82 0.75 0.78 Asyndeton 60 0.25 0.95 0.4 If Conditional Zero 60 0.71 0.76 0.73 Comparative Adjective 67 0.51 0.61 0.56 If Counterfactual 60 0.84 0.87 0.85 Comparative Adverb 71 0.6 0.62 0.61 Isocolon 180 0.57 0.83 0.68 Diacope 60 0.75 0.73 0.74 Mesarchia 20 0.45 0.85 0.59 Enumeration 60 0.76 0.93 0.84 Mesodiplosis 40 0.28 0.68 0.4 Epanalepsis 60 0.63 0.83 0.72 Passive Voice 60 0.79 0.98 0.87 Epiphoza 60 0.61 0.93 0.74 Polysyndeton 60 0.77 0.7 0.73 Epizeugma 60 0.68 0.7 0.69 Pysma 60 1 1 1 Epizeuxis 60 0.79 0.77 0.78 Superlative Adjective 70 0.62 0.73 0.67 Hypozeugma 60 0.61 0.8 0.69 Superlative Adverb 70 0.63 0.5 0.56 If Conditional One 60 0.78 0.78 0.78 Unless Conditional 60 1 1 1 If Conditional Three 60 0.86 0.65 0.74 Whether Conditional 60 1 0.83 0.91
144
Device Total No. Precision Recall F1-score Device Total No. Precision Recall F1-score Anadiplosis 60 0.76 0.73 0.74 If Conditional Two 60 0.82 0.75 0.78 Asyndeton 60 0.25 0.95 0.4 If Conditional Zero 60 0.71 0.76 0.73 Comparative Adjective 67 0.51 0.61 0.56 If Counterfactual 60 0.84 0.87 0.85 Comparative Adverb 71 0.6 0.62 0.61 Isocolon 180 0.57 0.83 0.68 Diacope 60 0.75 0.73 0.74 Mesarchia 20 0.45 0.85 0.59 Enumeration 60 0.76 0.93 0.84 Mesodiplosis 40 0.28 0.68 0.4 Epanalepsis 60 0.63 0.83 0.72 Passive Voice 60 0.79 0.98 0.87 Epiphoza 60 0.61 0.93 0.74 Polysyndeton 60 0.77 0.7 0.73 Epizeugma 60 0.68 0.7 0.69 Pysma 60 1 1 1 Epizeuxis 60 0.79 0.77 0.78 Superlative Adjective 70 0.62 0.73 0.67 Hypozeugma 60 0.61 0.8 0.69 Superlative Adverb 70 0.63 0.5 0.56 If Conditional One 60 0.78 0.78 0.78 Unless Conditional 60 1 1 1 If Conditional Three 60 0.86 0.65 0.74 Whether Conditional 60 1 0.83 0.91
145
Device Total No. Precision Recall F1-score Device Total No. Precision Recall F1-score Anadiplosis 60 0.76 0.73 0.74 If Conditional Two 60 0.82 0.75 0.78 Asyndeton 60 0.25 0.95 0.4 If Conditional Zero 60 0.71 0.76 0.73 Comparative Adjective 67 0.51 0.61 0.56 If Counterfactual 60 0.84 0.87 0.85 Comparative Adverb 71 0.6 0.62 0.61 Isocolon 180 0.57 0.83 0.68 Diacope 60 0.75 0.73 0.74 Mesarchia 20 0.45 0.85 0.59 Enumeration 60 0.76 0.93 0.84 Mesodiplosis 40 0.28 0.68 0.4 Epanalepsis 60 0.63 0.83 0.72 Passive Voice 60 0.79 0.98 0.87 Epiphoza 60 0.61 0.93 0.74 Polysyndeton 60 0.77 0.7 0.73 Epizeugma 60 0.68 0.7 0.69 Pysma 60 1 1 1 Epizeuxis 60 0.79 0.77 0.78 Superlative Adjective 70 0.62 0.73 0.67 Hypozeugma 60 0.61 0.8 0.69 Superlative Adverb 70 0.63 0.5 0.56 If Conditional One 60 0.78 0.78 0.78 Unless Conditional 60 1 1 1 If Conditional Three 60 0.86 0.65 0.74 Whether Conditional 60 1 0.83 0.91
146
Device Total No. Precision Recall F1-score Device Total No. Precision Recall F1-score Anadiplosis 60 0.76 0.73 0.74 If Conditional Two 60 0.82 0.75 0.78 Asyndeton 60 0.25 0.95 0.4 If Conditional Zero 60 0.71 0.76 0.73 Comparative Adjective 67 0.51 0.61 0.56 If Counterfactual 60 0.84 0.87 0.85 Comparative Adverb 71 0.6 0.62 0.61 Isocolon 180 0.57 0.83 0.68 Diacope 60 0.75 0.73 0.74 Mesarchia 20 0.45 0.85 0.59 Enumeration 60 0.76 0.93 0.84 Mesodiplosis 40 0.28 0.68 0.4 Epanalepsis 60 0.63 0.83 0.72 Passive Voice 60 0.79 0.98 0.87 Epiphoza 60 0.61 0.93 0.74 Polysyndeton 60 0.77 0.7 0.73 Epizeugma 60 0.68 0.7 0.69 Pysma 60 1 1 1 Epizeuxis 60 0.79 0.77 0.78 Superlative Adjective 70 0.62 0.73 0.67 Hypozeugma 60 0.61 0.8 0.69 Superlative Adverb 70 0.63 0.5 0.56 If Conditional One 60 0.78 0.78 0.78 Unless Conditional 60 1 1 1 If Conditional Three 60 0.86 0.65 0.74 Whether Conditional 60 1 0.83 0.91
147
Device Total No. Precision Recall F1-score Device Total No. Precision Recall F1-score Anadiplosis 60 0.76 0.73 0.74 If Conditional Two 60 0.82 0.75 0.78 Asyndeton 60 0.25 0.95 0.4 If Conditional Zero 60 0.71 0.76 0.73 Comparative Adjective 67 0.51 0.61 0.56 If Counterfactual 60 0.84 0.87 0.85 Comparative Adverb 71 0.6 0.62 0.61 Isocolon 180 0.57 0.83 0.68 Diacope 60 0.75 0.73 0.74 Mesarchia 20 0.45 0.85 0.59 Enumeration 60 0.76 0.93 0.84 Mesodiplosis 40 0.28 0.68 0.4 Epanalepsis 60 0.63 0.83 0.72 Passive Voice 60 0.79 0.98 0.87 Epiphoza 60 0.61 0.93 0.74 Polysyndeton 60 0.77 0.7 0.73 Epizeugma 60 0.68 0.7 0.69 Pysma 60 1 1 1 Epizeuxis 60 0.79 0.77 0.78 Superlative Adjective 70 0.62 0.73 0.67 Hypozeugma 60 0.61 0.8 0.69 Superlative Adverb 70 0.63 0.5 0.56 If Conditional One 60 0.78 0.78 0.78 Unless Conditional 60 1 1 1 If Conditional Three 60 0.86 0.65 0.74 Whether Conditional 60 1 0.83 0.91
148
Stanford Dependencies Governors
4 tokens
P-clause Q-clause If-conditional
grammatical rules
149
Identify P and Q
P → past tense
(VBN/VBD)
Q → modals
(would/could…)
150
Comparatives
151