File size: 196,922 Bytes
6fa4bc9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
{
    "paper_id": "2022",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T06:07:11.629978Z"
    },
    "title": "Assessment of Massively Multilingual Sentiment Classifiers",
    "authors": [
        {
            "first": "Krzysztof",
            "middle": [],
            "last": "Rajda",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Wroc\u0142aw University of Science and Technology",
                "location": {
                    "addrLine": "2 Brand24 AI"
                }
            },
            "email": "[email protected]"
        },
        {
            "first": "\u0141ukasz",
            "middle": [],
            "last": "Augustyniak",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Wroc\u0142aw University of Science and Technology",
                "location": {
                    "addrLine": "2 Brand24 AI"
                }
            },
            "email": "[email protected]"
        },
        {
            "first": "Piotr",
            "middle": [],
            "last": "Gramacki",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Wroc\u0142aw University of Science and Technology",
                "location": {
                    "addrLine": "2 Brand24 AI"
                }
            },
            "email": "[email protected]"
        },
        {
            "first": "Marcin",
            "middle": [],
            "last": "Gruza",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Wroc\u0142aw University of Science and Technology",
                "location": {
                    "addrLine": "2 Brand24 AI"
                }
            },
            "email": "[email protected]"
        },
        {
            "first": "Szymon",
            "middle": [],
            "last": "Wo\u017aniak",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Wroc\u0142aw University of Science and Technology",
                "location": {
                    "addrLine": "2 Brand24 AI"
                }
            },
            "email": "[email protected]"
        },
        {
            "first": "Tomasz",
            "middle": [],
            "last": "Kajdanowicz",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Wroc\u0142aw University of Science and Technology",
                "location": {
                    "addrLine": "2 Brand24 AI"
                }
            },
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Models are increasing in size and complexity in the hunt for SOTA. But what if those 2% increase in performance does not make a difference in a production use case? Maybe benefits from a smaller, faster model outweigh those slight performance gains. Also, equally good performance across languages in multilingual tasks is more important than SOTA results on a single one. We present the biggest, unified, multilingual collection of sentiment analysis datasets. We use these to assess 11 models and 80 high-quality sentiment datasets (out of 342 raw datasets collected) in 27 languages and included results on the internally annotated datasets. We deeply evaluate multiple setups, including fine-tuning transformer-based models for measuring performance. We compare results in numerous dimensions addressing the imbalance in both languages coverage and dataset sizes. Finally, we present some best practices for working with such a massive collection of datasets and models from a multilingual perspective.",
    "pdf_parse": {
        "paper_id": "2022",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Models are increasing in size and complexity in the hunt for SOTA. But what if those 2% increase in performance does not make a difference in a production use case? Maybe benefits from a smaller, faster model outweigh those slight performance gains. Also, equally good performance across languages in multilingual tasks is more important than SOTA results on a single one. We present the biggest, unified, multilingual collection of sentiment analysis datasets. We use these to assess 11 models and 80 high-quality sentiment datasets (out of 342 raw datasets collected) in 27 languages and included results on the internally annotated datasets. We deeply evaluate multiple setups, including fine-tuning transformer-based models for measuring performance. We compare results in numerous dimensions addressing the imbalance in both languages coverage and dataset sizes. Finally, we present some best practices for working with such a massive collection of datasets and models from a multilingual perspective.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "Multilingual text representations are becoming increasingly important in science as well as the business community. However how universal and versatile they truly are? Can we use them to train one, multilingual, production-ready sentiment classifier? To verify this research question, we gathered a massive collection of sentiment analysis datasets and evaluated 11 different models on them. We want to assess the performance of fine-tuning languages models as well as language models as feature extractors for simpler, even linear models.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Sentiment analysis is subjective and both domain and language-dependent, hence there is an even greater need to understand the behaviour and performance of the multilingual setup. We focused on multilingual sentiment classification because our business use cases involve the analysis of texts in multiple languages across the world. Moreover, one universal model in a production environment is much easier to deploy, maintain, monitor, remove biases or improve the model's fairness -especially in cases when the load differs between languages and could change over time. We want to compare state-of-the-art multilingual embedding methods and select the ones with the best performance across languages.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "The main objective of this article is to answer the following Research Questions: (RQ1) Are we able to create a single multilingual sentiment classifier, that performs equally well for each language? (RQ2) Does fine-tuning of transformer-based models significantly improve sentiment classification results? (RQ3) What is the relationship between model size and performance? Is bigger always better?",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Our main contribution includes 3 main points. Firstly, we perform a large scoping review of published sentiment datasets. Using a set of rigid inclusion and exclusion criteria, we filter the initial pool of 342 datasets down to 80 high-quality datasets representing 27 languages. Secondly, we evaluated how universal and versatile multilingual text representations are for the sentiment classification problem. Finally, we compared many deep learningbased approaches with fine-tuning and without it for multilingual sentiment classification.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "The remainder of this paper is organized as follows: Section 2 presents a literature review on the topic of multilingual sentiment analysis; Section 3 describes the language models, datasets, and our evaluation methodology; Section 4 describes the conducted experiments and summarizes the results; Section 5 discusses the results in terms of research questions; Section 6 presents conclusions and describes further works.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Multilingual Text Representations. Initially, multilingual text representations were obtained using multilingual word embeddings (Ruder et al., 2019) . These were created using various training techniques, parallel corpora, and dictionaries, for example by aligning the monolingual Word2Vec (Mikolov et al., 2013a) vector spaces with linear transformations using small parallel dictionaries (Mikolov et al., 2013b) . To better represent longer texts, modern approaches use more complex contextual language models like BiLSTM (Artetxe and Schwenk, 2019) and Transformers (Feng et al., 2020; Conneau et al., 2020; Devlin et al., 2019; Xue et al., 2021; . Their multilingual capabilities result from pretraining on multilingual objective tasks like machine translation (Artetxe and Schwenk, 2019) , translation language modelling (TLM) (Conneau et al., 2020; Conneau and Lample, 2019) or translation ranking (Feng et al., 2020; Yang et al., 2019) . Details of the models used in our experiments are described in Section 3.1.",
                "cite_spans": [
                    {
                        "start": 129,
                        "end": 149,
                        "text": "(Ruder et al., 2019)",
                        "ref_id": "BIBREF53"
                    },
                    {
                        "start": 291,
                        "end": 314,
                        "text": "(Mikolov et al., 2013a)",
                        "ref_id": "BIBREF40"
                    },
                    {
                        "start": 391,
                        "end": 414,
                        "text": "(Mikolov et al., 2013b)",
                        "ref_id": "BIBREF41"
                    },
                    {
                        "start": 525,
                        "end": 552,
                        "text": "(Artetxe and Schwenk, 2019)",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 570,
                        "end": 589,
                        "text": "(Feng et al., 2020;",
                        "ref_id": "BIBREF25"
                    },
                    {
                        "start": 590,
                        "end": 611,
                        "text": "Conneau et al., 2020;",
                        "ref_id": null
                    },
                    {
                        "start": 612,
                        "end": 632,
                        "text": "Devlin et al., 2019;",
                        "ref_id": "BIBREF22"
                    },
                    {
                        "start": 633,
                        "end": 650,
                        "text": "Xue et al., 2021;",
                        "ref_id": null
                    },
                    {
                        "start": 766,
                        "end": 793,
                        "text": "(Artetxe and Schwenk, 2019)",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 833,
                        "end": 855,
                        "text": "(Conneau et al., 2020;",
                        "ref_id": null
                    },
                    {
                        "start": 856,
                        "end": 881,
                        "text": "Conneau and Lample, 2019)",
                        "ref_id": "BIBREF17"
                    },
                    {
                        "start": 905,
                        "end": 924,
                        "text": "(Feng et al., 2020;",
                        "ref_id": "BIBREF25"
                    },
                    {
                        "start": 925,
                        "end": 943,
                        "text": "Yang et al., 2019)",
                        "ref_id": "BIBREF77"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "The quality of multilingual text representations is usually evaluated with cross-and multilingual tasks like cross-lingual natural language inference (Conneau et al., 2018) , question answering (Lewis et al., 2020) , named entity recognition (Tjong Kim Sang, 2002; Tjong Kim Sang and De Meulder, 2003) or parallel text extraction (Zweigenbaum et al., 2017; Ziemski et al., 2016 ). Another important benchmark is XTREME (Hu et al., 2020) , which is designed for testing the abilities of cross-lingual transfer across 40 languages and 9 tasks. Despite its massive character, XTREME lacks benchmarking task of sentiment analysis, also only mBERT, XLM, XLM-R, and MMTE are used as baseline models. We try to fill this gap with our work. K et al. (2020) performed extensive research on the cross-lingual ability of mBERT. Wu and Dredze (2020) compared mBERT with monolingual models and found that it under-performs on low-resource languages. Liu et al. (2020) analyzed a cross-lingual ability of mBERT considering a contextual aspect of mBERT and dataset size. There is a significant lack of detailed analysis of characteristics of other language models, despite mBERT.",
                "cite_spans": [
                    {
                        "start": 150,
                        "end": 172,
                        "text": "(Conneau et al., 2018)",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 194,
                        "end": 214,
                        "text": "(Lewis et al., 2020)",
                        "ref_id": "BIBREF36"
                    },
                    {
                        "start": 242,
                        "end": 264,
                        "text": "(Tjong Kim Sang, 2002;",
                        "ref_id": "BIBREF70"
                    },
                    {
                        "start": 265,
                        "end": 301,
                        "text": "Tjong Kim Sang and De Meulder, 2003)",
                        "ref_id": "BIBREF72"
                    },
                    {
                        "start": 330,
                        "end": 356,
                        "text": "(Zweigenbaum et al., 2017;",
                        "ref_id": "BIBREF79"
                    },
                    {
                        "start": 357,
                        "end": 377,
                        "text": "Ziemski et al., 2016",
                        "ref_id": "BIBREF78"
                    },
                    {
                        "start": 419,
                        "end": 436,
                        "text": "(Hu et al., 2020)",
                        "ref_id": "BIBREF28"
                    },
                    {
                        "start": 937,
                        "end": 954,
                        "text": "Liu et al. (2020)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "Multilingual Sentiment Analysis. In literature, there are several examples of reviews, which focus on traditional sentiment analysis methods (e.g., lexicon-based, lexical features engineering, shallow models), while not mentioning any embeddingbased methods (Dashtipour et al., 2016; Sagnika et al., 2020) . They are a valuable source of information about sentiment datasets. However, modern NLP applications often utilize deep learning techniques, which were not covered there. An example of a deep learning-based approach was presented by Attia et al. (2018) , who trained a convolutional neural network (CNN) on word-level embeddings of texts in English, German and Arabic, a separate model for each language. This approach requires many resources and computations as one has to create a separate embedding dictionary for each language. An alternative approach is to use characterlevel embeddings. Wehrmann et al. (2017) trained such a model to classify tweets written in English, German, Portuguese, and Spanish as either positive or negative. This approach requires fewer parameters than word embedding models.",
                "cite_spans": [
                    {
                        "start": 258,
                        "end": 283,
                        "text": "(Dashtipour et al., 2016;",
                        "ref_id": "BIBREF20"
                    },
                    {
                        "start": 284,
                        "end": 305,
                        "text": "Sagnika et al., 2020)",
                        "ref_id": "BIBREF55"
                    },
                    {
                        "start": 541,
                        "end": 560,
                        "text": "Attia et al. (2018)",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 901,
                        "end": 923,
                        "text": "Wehrmann et al. (2017)",
                        "ref_id": "BIBREF73"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "Newer approaches to multilingual sentiment analysis use deep models and machine translation e.g. Can et al. (2018) trained a Recurrent Neural Network (RNN) on English reviews and evaluated it on machine-translated reviews in Russian, Spanish, Turkish and Dutch. They used the Google Translation API and pre-trained GloVe embeddings for English. Kanclerz et al. (2020) used LASER sentence embeddings to learn a sentiment classifier on Polish reviews and used this classifier to predict sentiment on reviews translated into other languages. As we can see most of the research covers only a couple of languages for sentiment analysis. Hence, we decided to gather a massive collection of 342 datasets in 27 languages.",
                "cite_spans": [
                    {
                        "start": 345,
                        "end": 367,
                        "text": "Kanclerz et al. (2020)",
                        "ref_id": "BIBREF32"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "We conducted several experiments to answer if there is a truly universal multilingual text representation model (Table 1 ). We tested their performance based on the largest sentiment analysis dataset in the literature.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 112,
                        "end": 120,
                        "text": "(Table 1",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Evaluation Methodology",
                "sec_num": "3"
            },
            {
                "text": "We used multiple language models as text representation methods (Table 1) . We aimed to select models varied in terms of architecture, size, and type of data used in pre-training. We selected two models which do not use transformer architecture (CNN and BiLSTM) as a baseline. We also used models, based on multiple different transformer architec- tures (T5, BERT, RoBERTa). We also included models' trained with multilingual knowledge distillation (Reimers and Gurevych, 2020) ",
                "cite_spans": [
                    {
                        "start": 449,
                        "end": 477,
                        "text": "(Reimers and Gurevych, 2020)",
                        "ref_id": "BIBREF50"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 64,
                        "end": 73,
                        "text": "(Table 1)",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Multilingual Language Models",
                "sec_num": "3.1"
            },
            {
                "text": "such as paraphrase-xlm-r-multilingual-v1 (XLM-R-dist), distiluse-base-multilingual-cased-v2 (mUSE-dist), paraphrase-multilingual-mpnet-base-v2 (MPNet).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Multilingual Language Models",
                "sec_num": "3.1"
            },
            {
                "text": "We also included models trained on multilingual corpus like Wikipedia (Wiki) or Common Crawl (CC) as well as models trained with the use of parallel datasets. Selected models differ in size -from LASER with 52M parameters to LaBSE with 470M. They also differ regarding covered languages, from 16 up to more than a hundred. By a number of languages, we mean how many were used to create a specific model, not all languages supported by the model (an example is MPNet, trained using 53 languages, but as it is based on XLM-R, it supports 100). We also compared inference time which was calculated as a mean of inference times of 500 randomly selected texts samples from all datasets. The hardware used is described in Section A.1. We searched for models comparison in similar tasks in literature but failed to find any, which compares more than 2 or 3 models. All models used are characterized in Table 1 .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 895,
                        "end": 902,
                        "text": "Table 1",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Multilingual Language Models",
                "sec_num": "3.1"
            },
            {
                "text": "We gathered 342 sentiment analysis datasets containing texts from multiple languages, data sources and domains to check our research questions. We searched for datasets in various sources, like Google Scholar, GitHub repositories, and the Hug-gingFace datasets library. Such a large number of datasets allows us to estimate the quality of lan-guage models in various conditions with greater certainty. To the best of our knowledge, this is the largest sentiment analysis datasets collection currently gathered and researched in literature. After preliminary analysis, we selected 80 datasets of reasonable quality based on 5 criteria. 1We rejected datasets containing weak annotations (e.g., datasets with labels based on emoji occurrence or generated automatically through classification by machine learning models), as our analysis showed that they may contain too much noise (Northcutt et al., 2021) . 2We reject datasets without sufficient information about the annotation procedure (e.g., whether annotation was manual or automatic, number of annotators) because it is always a questionable decision to merge datasets created with different annotation guidelines. 3We accepted reviews datasets and mapped their rating labels to sentiment values. The mapping rules are described in section 3.2.1. 4We rejected 2-class only datasets (positive/negative without neutral), as our analysis showed their low quality in terms of 3-class usage. (5) Some datasets contain samples in multiple languages -we split them and treated each language as a separate dataset.",
                "cite_spans": [
                    {
                        "start": 878,
                        "end": 902,
                        "text": "(Northcutt et al., 2021)",
                        "ref_id": "BIBREF47"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Datasets",
                "sec_num": "3.2"
            },
            {
                "text": "Working with many datasets means that they could contain different types of text, various artefacts such as URL or HTML tags, or just different sentiment classes mappings. We applied a couple of preprocessing steps to each dataset to unify all datasets. We dropped duplicated texts. We removed URLs, Twitter mentions, HTML tags, and emails. During an exploratory analysis, we spotted that reviewbased datasets often contain many repeated texts with contradictory sentiment scores. We deduplicated such cases and applied a majority voting to choose a sentiment label. Finally, we unified labels from all datasets into 3-class (negative, neutral, positive). In the case of datasets containing user ratings (on a scale of 1-5) along with their review texts, we mapped the ratings to sentiment as follows: the middle value (3) of the rating scale was treated as a neutral sentiment, ratings below the middle as negative sentiment, and ratings above the middle as positive sentiment.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Data Preprocessing",
                "sec_num": "3.2.1"
            },
            {
                "text": "Presenting statistics of 80 datasets across 27 languages could be challenging. We checked different aggregating and sorting of datasets to make their statistics as readable as possible and easily usable for results discussion. We decided to group datasets by their language and next sorted them based on the number of examples in every aggregate - Table 2 . In total, we selected 80 datasets containing 6,164,942 text samples. Most of the texts in the datasets are in English (2,330,486 samples across 17 datasets), Arabic (932,075 samples across 9 datasets), and Spanish (418,892 samples across 5 datasets). The datasets contain text from various categories: social media (44 datasets), reviews (24 datasets), news (5 datasets), and others (7 datasets). They also differ in the mean number of words and characters in examples. See the detailed information of datasets used is in Tables 5 and 6. We also selected around 60k samples for training and validation and another 60k for testing. This is enough for training a small classifier on top of a frozen embedding or fine-tuning a transformerbased model (see Section 3.3). This was also done due to computation resources limitations.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 348,
                        "end": 355,
                        "text": "Table 2",
                        "ref_id": "TABREF1"
                    },
                    {
                        "start": 880,
                        "end": 895,
                        "text": "Tables 5 and 6.",
                        "ref_id": "TABREF4"
                    }
                ],
                "eq_spans": [],
                "section": "Data Preprocessing",
                "sec_num": "3.2.1"
            },
            {
                "text": "We have also used an internal dataset that was manually annotated. It is multi-domain and consists of texts from various Internet sources in Polish and English. It includes texts from social media, news sites, blogs and forums. We used this dataset as a gold standard. We need it because we do not know exact annotation guidelines from literature datasets and we assume that those guidelines differed between datasets. In our gold dataset, each text was annotated by 3 annotators with majority label selection. The annotators achieved a high agreement measured by Cohen's kappa: 0.665 and Krippendorff's alpha: 0.666. Statistics of this dataset are presented in Table 3 . All samples were trimmed to the length of 350 chars (mean length of 145 chars).",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 662,
                        "end": 669,
                        "text": "Table 3",
                        "ref_id": "TABREF2"
                    }
                ],
                "eq_spans": [],
                "section": "Internal Dataset",
                "sec_num": "3.2.2"
            },
            {
                "text": "We wanted to compare multilingual models in different use cases. Firstly, we wanted to see how much information is stored in pre-trained embedding. In this scenario, we used each of the text representations models listed in Section 3.1 as a feature extractor and coupled them with only a small linear classification head. We used an average from a final layer as a text representation. We will refer to this scenario Just Head -Linear. In the second scenario, we replaced a linear classifier with a BiL-STM classifier, still using the text representation model as a feature extractor. We fed BiLSTM layer with outputs from the last layer of the feature extractor (Just Head -BiLSTM). LASER and mUSE do not provide per-token embeddings and therefore, were not included in this scenario. Since most of our models are transformer-based, we decided to test them in a fine-tuning setup. This last scenario evaluated the fine-tuning of all transformer-based models (referred to as fine-tuning), with an exception made for mUSE-transformer because it was not possible to do with our implementation in Py-Torch with Huggingface models.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Scenarios",
                "sec_num": "3.3"
            },
            {
                "text": "For each scenario, we prepared 3 test metrics, which we refer to as a whole test, average by dataset and internal. Each of them separately measures model performance but all of them are based on a macro F1-score. The whole test is calculated on all samples from datasets described in 3.2 combined. It is meant to reflect the real-life performance of a model because our real-world applications often deal with an imbalance in languages distribution (with English being the most popular language used on the Internet). On average by dataset, we first calculate the macro F1-score on each dataset and then calculate the average of those scores. This is meant to show whether the model was not too over-fitted for the majority of languages or the biggest datasets. Finally, in the internal scenario, we assess them on our internal dataset (described in 3.2.2) to measure performance in our domain-specific examples.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Scenarios",
                "sec_num": "3.3"
            },
            {
                "text": "To show how each model performs in a bird's eye view, we prepared Nemenyi diagrams (Nemenyi, 1963) for all three experimental setups. Nemeneyi post-hoc statistical test finds groups of models that differ. It was used on the top of multiple comparisons Friedman test (Dem\u0161ar, 2006) . The Nemeneyi test makes a pair-wise comparison of all model's ranks. We used alpha equal to 5%. The Nemeneyi test provides critical distance for compared groups that are not significantly different from each other.",
                "cite_spans": [
                    {
                        "start": 83,
                        "end": 98,
                        "text": "(Nemenyi, 1963)",
                        "ref_id": "BIBREF45"
                    },
                    {
                        "start": 266,
                        "end": 280,
                        "text": "(Dem\u0161ar, 2006)",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation Procedure",
                "sec_num": "3.4"
            },
            {
                "text": "For each scenario, we adjusted hyperparameters. The hidden size was set to LM's embedding size for linear and fine-tuning and 32 for BiLSTM. By hidden size, we mean middle linear layer size, or in the case of BiLSTM -its hidden size parameter. BiLSTM uses a smaller hidden size because our experiments showed that it does not hurt performance but increases efficiency. The learning rate was initially the same for all scenarios, at the well-established value of 1e-3. We then modified it for each version by decreasing it for fine-tuning (to 1e-5) and slightly increasing it for BiLSTM based model (5e-3). The batch size was determined by our GPU's memory size. We used 200 for linear and BiLSTM and 6 for fine-tuning. We used dropout in classification head -0.5 for BiLSTM and 0.2 for other scenarios. We trained our models for 5 epochs in the fine-tuning scenario and 15 in two others, as those were the max number of epochs before the models started overfitting. We tested with the best F1-score on a validation dataset.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Models Setup",
                "sec_num": "3.5"
            },
            {
                "text": "We divided our results into three layers. Firstly, we show a general bird's eye view of all models -it helps to spot the best and the worst models. Then, we provide detailed results for each model aggregated per dataset. Finally, to dig deeper into the model's performance, we show numerical results for each model for each language.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Results",
                "sec_num": "4"
            },
            {
                "text": "There is no significantly best embedding model in any of the tested scenarios based on the Nemenyi diagrams - Figure 1 . However, we can see that the MPNet proved to be the best (for the linear scenario) and not significantly worse than the best -XLR-M model -in the other two scenarios. It is also worth mentioning that mBERT-based models (mBERT and DistillmBERT) proved to be the worst language models for our tasks.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 110,
                        "end": 118,
                        "text": "Figure 1",
                        "ref_id": "FIGREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Bird's Eye View",
                "sec_num": "4.1"
            },
            {
                "text": "All models achieve better results with fine-tuning (up to 0.7 F1-score) than with extraction of vectors from text and then applying linear (up to 0.61) or BiLSTM (up to 0.64) layers, shows Table 4. The performance gains are higher when fine-tuning models pretrained on MLM and TLM tasks (like mBERT or XLM-R) compared to mod- els, which were trained with sentence classification tasks, sentence similarity tasks or similar (like LaBSE). For example, mBERT had gains of 9, 11, and 14 percentage points (pp) on whole test, average by dataset and internal test cases, Distilm-BERT -9, 13 and 20pp, XLM-R -6, 10, and 15pp. At the same time, LaBSE had only 6, 8, and 7pp and MPNet -4, 7, 4pp. Still, those models achieve better overall performance. Fine-tuning reduces inequalities in the results between models (0.55 vs 0.43 for best and worst models in Just head -Linear setup, and 0.62 vs 0.56 after Fine-tuning for average by dataset metric). Those results were meant to show a general comparison between fine-tuned models against training just classification head.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Aggregated by Dataset",
                "sec_num": "4.2"
            },
            {
                "text": "The additional BiLSTM layer on top of transformer token embeddings improves the results of the model with only a linear layer in most cases. The differences are most clear in the case of the results for our internal dataset, where the result improved even by 13pp. (from 50% to 63%) for the mT5 model.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Aggregated by Dataset",
                "sec_num": "4.2"
            },
            {
                "text": "Those results show, that three models are the most promising choices: XLM-R, LaBSE and MP-Net. They achieve comparable performance in all scenarios and test cases. Furthermore, they are better than other models in almost all test cases. XLM-R-dist was very close to those, but analysis with Nemenyi diagrams shows that it is slightly worse than those three. Legend: lang -averaged by all languages, ds -averaged by dataset, ar -Arabic, bg -Bulgarian, bs -Bosnian, cs -Czech, de -German, en -English, es -Spanish, fa -Persian, fr -French, he -Hebrew, hi -Hindi, hr -Croatian, hu -Hungarian, it -Italian, ja -Japanese, lv -Latvian, pl -Polish, pt -Portuguese, ru -Russian, sk -Slovak, sl -Slovenian, sq -Albanian, sr -Serbian, sv -Swedish, th -Thai, ur -Urdu, zh -Chinese.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Aggregated by Dataset",
                "sec_num": "4.2"
            },
            {
                "text": "We assessed the performance of each model in each experimental scenario concerning the language. The texts were sub-sampled with stratification by language and class label so that language distribution in the test dataset reflects this in the whole dataset. It means that some languages are underrepresented. We also include the total Macro F1 score value in column \"all\". Results are presented in Figure 2 for fine-tuning scenario and in Figure  5 for others. Those results confirm conclusions from the previous section about the advantage of XLM-R, LaBSE and MPNet. They have the best performance in most languages and together with XLM-R-dist, there are no big differences between them.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 398,
                        "end": 406,
                        "text": "Figure 2",
                        "ref_id": "FIGREF1"
                    },
                    {
                        "start": 439,
                        "end": 448,
                        "text": "Figure  5",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Every Model for Every Language",
                "sec_num": "4.3"
            },
            {
                "text": "RQ1: Are we able to create a single multilingual sentiment classifier, performing equally well for each language? When considering only the best models (XLM-R, LaBSE, MPNet) in the fine-tuning setup, we observed that they achieve best or close to best results in every language - Figure 5 . In some languages, results are significantly worse than in others, but this is also true for other models evaluated as it may be caused by differences in the number of samples, quality, and difficulty of samples in those languages. Therefore, we can say that one model can work exceptionally well in all languages. On the other hand, statistical analysis which is presented in the form of Nemenyi diagrams in Figures 1a, 1b and 1c showed that there is no statistical difference between top models in fine-tuning setup, so it is not possible to state which of those is the best one. We can rather state which group of models proved to be significantly better than others.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 280,
                        "end": 288,
                        "text": "Figure 5",
                        "ref_id": null
                    },
                    {
                        "start": 700,
                        "end": 721,
                        "text": "Figures 1a, 1b and 1c",
                        "ref_id": "FIGREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Discussion",
                "sec_num": "5"
            },
            {
                "text": "RQ2: Does fine-tuning of transformer-based models significantly improve sentiment classification results? All models worked better when fine-tuned, but the performance gain varied from one to another. It was between 4 (mUSE-dist) and 9 (mBERT and DistilmBERT) pp. F1 on the benchmark test dataset, and between 0 (mUSE-dist) and 20 pp. (DistilmBERT) on our internal dataset. The 17, 15, and 14 pp. gain of mT5, XLM-R, and DistilmBERT on the internal dataset is also worth noting. In general, the most significant gain can be observed in models trained on language modelling only (MLM or TLM), such as XLM-R and mBERT.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Discussion",
                "sec_num": "5"
            },
            {
                "text": "The results of our experiments showed that there exists a correlation between the classification result of the language model with its number of parameters. Figure 3 shows that, for all scenarios and test dataset types, bigger models achieve better performance in most cases. However, there are some counterexamples, e.g., mUSE-dist is smaller than mBERT but achieves better performance in Just head -Linear setup, for all dataset types. This indicates that the size of the model is an important factor in its performance, but other factors, like the domain and the type of pretraining task, may also affect the results. Moreover, we observed that this correlation is weaker after fine-tuning. We can often find the model with similar performance to the best one but significantly smaller and faster for the production environment.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 157,
                        "end": 165,
                        "text": "Figure 3",
                        "ref_id": "FIGREF2"
                    }
                ],
                "eq_spans": [],
                "section": "RQ3: What is the relationship between model size and performance? Is bigger always better?",
                "sec_num": null
            },
            {
                "text": "Your Dataset Splits Matter To determine which model works best, we repeated fine-tuning five times to remove a right/wrong random seed factor for each model and dataset subsampling. Due to computation resources limitations, we selected eight models available in Huggingface for finetuning. Interestingly, we can see that one of the samples looks like the outlier - Figure 4 for almost all the evaluated models. The F1-score for this sample is even 4 percentage points worse than other samples' scores. We investigated this anomaly and spotted that it is always the same sample (the same seed for sample generation). As a reminder, since we collected a massive dataset and had limited computational resources, we sub-sampled texts for each of the five runs. Sub-samples between different models stay the same. It looks like the mentioned sample was more difficult than others or had distinctive characteristics. It is hard to tell why without in-depth analysis, hence we intend to conduct further research on the topic of data quality in sentiment analysis tasks using techniques like noise ratio (Northcutt et al., 2021) or data cartography (Swayamdipta et al., 2020) . Here, we see an outstanding example of how vital the dataset's preparation could be regarding split for train/dev/test sets.",
                "cite_spans": [
                    {
                        "start": 1096,
                        "end": 1120,
                        "text": "(Northcutt et al., 2021)",
                        "ref_id": "BIBREF47"
                    },
                    {
                        "start": 1141,
                        "end": 1167,
                        "text": "(Swayamdipta et al., 2020)",
                        "ref_id": "BIBREF67"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 365,
                        "end": 373,
                        "text": "Figure 4",
                        "ref_id": "FIGREF3"
                    }
                ],
                "eq_spans": [],
                "section": "RQ3: What is the relationship between model size and performance? Is bigger always better?",
                "sec_num": null
            },
            {
                "text": "In this work we evaluated multilingual text representations for the task of sentiment classification by comparing multiple approaches, using different deep learning methods. In the process, we gathered the biggest collection of multi-lingual sentiment datasets -80 datasets for 27 languages. We evaluated 11 models (language models and text vectorization techniques) in 3 different scenarios. We found out that it is possible to create one model which achieves the best or most competitive results in all languages in our collected dataset, but there is no statistical difference between top-performing models. We found out that there is a significant benefit from fine-tuning transformer-based language models and that a model size is correlated with performance.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions and Further Works",
                "sec_num": "6"
            },
            {
                "text": "While conducting experiments we identified further issues which we find worth addressing. Dataset quality assessment is in our opinion the most important one and we are planning to address it in further works. Meanwhile, we used datasets with a literature background and trust that they were carefully prepared and have decent quality annotations. We also found out that it is difficult to propose a coherent experiments methodology with such imbalance in languages and datasets sizes. Moreover, analyzing results is difficult, when one must address dimensions of datasets, languages, data sources, models, and experiments scenarios.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions and Further Works",
                "sec_num": "6"
            },
            {
                "text": "Finally, we found out that when sub-sampling a dataset for experiments, seeds play a significant role (see results in Figure 4 ). To analyze this phenomenon, we intend to launch further research and use noise ratio (Northcutt et al., 2021) and data cartography (Swayamdipta et al., 2020) to understand how this split differs from the others. This will be, in our opinion, a good start to a comprehensive analysis of datasets quality for the multi-lingual sentiment classification task which we intend to perform. (Nabil et al., 2015) ar SM Yes 3224 50.9/25.0/24.1 16 94 (Salameh et al., 2015) ar SM Yes 1199 48.0/10.5/41.5 11 51 (Salameh et al., 2015) ar SM Yes 1998 67.5/10.1/22.4 20 107 (Habernal et al., 2013) cs R No 91140 32.4/33.7/33.9 50 311 (Habernal et al., 2013) cs R No 92758 7.9/23.4/68.7 20 131 (Habernal et al., 2013) cs SM Yes 9752 20.4/53.1/26.5 10 59 (Habernal et al., 2013) cs Figure 5: Detailed results of models' comparison. Legend: lang -averaged by all languages, ds -averaged by dataset, ar -Arabic, bg -Bulgarian, bs -Bosnian, cs -Czech, de -German, en -English, es -Spanish, fa -Persian, fr -French, he -Hebrew, hi -Hindi, hr -Croatian, hu -Hungarian, it -Italian, ja -Japanese, lv -Latvian, pl -Polish, pt -Portuguese, ru -Russian, sk -Slovak, sl -Slovenian, sq -Albanian, sr -Serbian, sv -Swedish, th -Thai, ur -Urdu, zh -Chinese.",
                "cite_spans": [
                    {
                        "start": 215,
                        "end": 239,
                        "text": "(Northcutt et al., 2021)",
                        "ref_id": "BIBREF47"
                    },
                    {
                        "start": 261,
                        "end": 287,
                        "text": "(Swayamdipta et al., 2020)",
                        "ref_id": "BIBREF67"
                    },
                    {
                        "start": 513,
                        "end": 533,
                        "text": "(Nabil et al., 2015)",
                        "ref_id": "BIBREF43"
                    },
                    {
                        "start": 570,
                        "end": 592,
                        "text": "(Salameh et al., 2015)",
                        "ref_id": "BIBREF56"
                    },
                    {
                        "start": 629,
                        "end": 651,
                        "text": "(Salameh et al., 2015)",
                        "ref_id": "BIBREF56"
                    },
                    {
                        "start": 689,
                        "end": 712,
                        "text": "(Habernal et al., 2013)",
                        "ref_id": "BIBREF26"
                    },
                    {
                        "start": 749,
                        "end": 772,
                        "text": "(Habernal et al., 2013)",
                        "ref_id": "BIBREF26"
                    },
                    {
                        "start": 808,
                        "end": 831,
                        "text": "(Habernal et al., 2013)",
                        "ref_id": "BIBREF26"
                    },
                    {
                        "start": 868,
                        "end": 891,
                        "text": "(Habernal et al., 2013)",
                        "ref_id": "BIBREF26"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 118,
                        "end": 126,
                        "text": "Figure 4",
                        "ref_id": "FIGREF3"
                    }
                ],
                "eq_spans": [],
                "section": "Conclusions and Further Works",
                "sec_num": "6"
            }
        ],
        "back_matter": [
            {
                "text": "The work was partially supported by the Department of Artificial Intelligence at Wroclaw University of Science and Technology, and by European Regional Development Fund (ERDF) in RPO WD 2014-2020 (project no. RPDS.01.02.02-02-0065/20). We want to thank Miko\u0142aj Morzy for an initial review and feedback. We want to thank our annotators team -Barbara Or\u0142owska, Daria Sza\u0142amacha, Konrad Gajewski and Pawe\u0142 Odrow\u0105\u017c-Sypniewski.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Acknowledgements",
                "sec_num": null
            },
            {
                "text": "A.1 Hardware and SoftwareWe performed our experiments using Python 3.9 and PyTorch (1.8.1) (and Tensorflow (2.3.0) for original mUSE). Our experimental setup consists of Intel(R) Xeon(R) CPU E5-2630 v4 @ 2.20GHz and Nvidia Tesla V100 16GB.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Appendices",
                "sec_num": null
            },
            {
                "text": "We present detailed lists of datasets included in our research in Tables 5 and 6 . They include language, category, dataset size, class balance and basic texts characteristics.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 66,
                        "end": 80,
                        "text": "Tables 5 and 6",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "A.2 Detailed Datasets Information",
                "sec_num": null
            },
            {
                "text": "We include full results of our experiments with results for each language in Figure 5 . Part with finetuning results was presented earlier in Figure 2 .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 77,
                        "end": 85,
                        "text": "Figure 5",
                        "ref_id": null
                    },
                    {
                        "start": 142,
                        "end": 150,
                        "text": "Figure 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "A.3 Full Results for Languages",
                "sec_num": null
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Sentiment classifier: Logistic regression for arabic services' reviews in lebanon",
                "authors": [
                    {
                        "first": "Moustafa",
                        "middle": [],
                        "last": "Marwan Al Omari",
                        "suffix": ""
                    },
                    {
                        "first": "Nacereddine",
                        "middle": [],
                        "last": "Al-Hajj",
                        "suffix": ""
                    },
                    {
                        "first": "Amani",
                        "middle": [],
                        "last": "Hammami",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Sabra",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "2019 International Conference on Computer and Information Sciences (ICCIS)",
                "volume": "",
                "issue": "",
                "pages": "1--5",
                "other_ids": {
                    "DOI": [
                        "10.1109/ICCISci.2019.8716394"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Marwan Al Omari, Moustafa Al-Hajj, Nacereddine Hammami, and Amani Sabra. 2019. Sentiment classi- fier: Logistic regression for arabic services' reviews in lebanon. In 2019 International Conference on Computer and Information Sciences (ICCIS), pages 1-5.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "LABR: A large scale Arabic book reviews dataset",
                "authors": [
                    {
                        "first": "Mohamed",
                        "middle": [],
                        "last": "Aly",
                        "suffix": ""
                    },
                    {
                        "first": "Amir",
                        "middle": [],
                        "last": "Atiya",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics",
                "volume": "2",
                "issue": "",
                "pages": "494--498",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Mohamed Aly and Amir Atiya. 2013. LABR: A large scale Arabic book reviews dataset. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 494-498, Sofia, Bulgaria. Association for Com- putational Linguistics.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Representations and architectures in neural sentiment analysis for morphologically rich languages: A case study from Modern Hebrew",
                "authors": [
                    {
                        "first": "Adam",
                        "middle": [],
                        "last": "Amram",
                        "suffix": ""
                    },
                    {
                        "first": "Anat",
                        "middle": [],
                        "last": "Ben David",
                        "suffix": ""
                    },
                    {
                        "first": "Reut",
                        "middle": [],
                        "last": "Tsarfaty",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of the 27th International Conference on Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "2242--2252",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Adam Amram, Anat Ben David, and Reut Tsarfaty. 2018. Representations and architectures in neu- ral sentiment analysis for morphologically rich lan- guages: A case study from Modern Hebrew. In Pro- ceedings of the 27th International Conference on Computational Linguistics, pages 2242-2252, Santa Fe, New Mexico, USA. Association for Computa- tional Linguistics.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Massively Multilingual Sentence Embeddings for Zero-Shot Cross-Lingual Transfer and Beyond",
                "authors": [
                    {
                        "first": "Mikel",
                        "middle": [],
                        "last": "Artetxe",
                        "suffix": ""
                    },
                    {
                        "first": "Holger",
                        "middle": [],
                        "last": "Schwenk",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Transactions of the Association for Computational Linguistics",
                "volume": "7",
                "issue": "",
                "pages": "597--610",
                "other_ids": {
                    "DOI": [
                        "10.1162/tacl_a_00288"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Mikel Artetxe and Holger Schwenk. 2019. Massively Multilingual Sentence Embeddings for Zero-Shot Cross-Lingual Transfer and Beyond. Transactions of the Association for Computational Linguistics, 7:597- 610.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Multilingual multi-class sentiment classification using convolutional neural networks",
                "authors": [
                    {
                        "first": "Mohammed",
                        "middle": [],
                        "last": "Attia",
                        "suffix": ""
                    },
                    {
                        "first": "Younes",
                        "middle": [],
                        "last": "Samih",
                        "suffix": ""
                    },
                    {
                        "first": "Ali",
                        "middle": [],
                        "last": "Elkahky",
                        "suffix": ""
                    },
                    {
                        "first": "Laura",
                        "middle": [],
                        "last": "Kallmeyer",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Mohammed Attia, Younes Samih, Ali Elkahky, and Laura Kallmeyer. 2018. Multilingual multi-class sen- timent classification using convolutional neural net- works. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA).",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "ArSentD-LEV: A Multi-Topic Corpus for Target-based Sentiment Analysis in Arabic Levantine Tweets",
                "authors": [
                    {
                        "first": "Ramy",
                        "middle": [],
                        "last": "Baly",
                        "suffix": ""
                    },
                    {
                        "first": "Alaa",
                        "middle": [],
                        "last": "Khaddaj",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Hazem",
                        "suffix": ""
                    },
                    {
                        "first": "Wassim",
                        "middle": [],
                        "last": "Hajj",
                        "suffix": ""
                    },
                    {
                        "first": "Khaled",
                        "middle": [],
                        "last": "El-Hajj",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Bashir Shaban",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ramy Baly, Alaa Khaddaj, Hazem M. Hajj, Wassim El-Hajj, and Khaled Bashir Shaban. 2018. ArSentD- LEV: A Multi-Topic Corpus for Target-based Senti- ment Analysis in Arabic Levantine Tweets. In Pro- ceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Paris, France. European Language Resources Associ- ation (ELRA).",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Overview of the Evalita 2016 SENTIment POLarity Classification Task",
                "authors": [
                    {
                        "first": "Francesco",
                        "middle": [],
                        "last": "Barbieri",
                        "suffix": ""
                    },
                    {
                        "first": "Valerio",
                        "middle": [],
                        "last": "Basile",
                        "suffix": ""
                    },
                    {
                        "first": "Danilo",
                        "middle": [],
                        "last": "Croce",
                        "suffix": ""
                    },
                    {
                        "first": "Malvina",
                        "middle": [],
                        "last": "Nissim",
                        "suffix": ""
                    },
                    {
                        "first": "Nicole",
                        "middle": [],
                        "last": "Novielli",
                        "suffix": ""
                    },
                    {
                        "first": "Viviana",
                        "middle": [],
                        "last": "Patti",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proceedings of Third Italian Conference on Computational Linguistics (CLiC-it 2016) & Fifth Evaluation Campaign of Natural Language Processing and Speech Tools for Italian. Final Workshop",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Francesco Barbieri, Valerio Basile, Danilo Croce, Malv- ina Nissim, Nicole Novielli, and Viviana Patti. 2016. Overview of the Evalita 2016 SENTIment POLarity Classification Task. In Proceedings of Third Italian Conference on Computational Linguistics (CLiC-it 2016) & Fifth Evaluation Campaign of Natural Lan- guage Processing and Speech Tools for Italian. Final Workshop (EVALITA 2016), Naples, Italy.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Author's sentiment prediction",
                "authors": [
                    {
                        "first": "Mohaddeseh",
                        "middle": [],
                        "last": "Bastan",
                        "suffix": ""
                    },
                    {
                        "first": "Mahnaz",
                        "middle": [],
                        "last": "Koupaee",
                        "suffix": ""
                    },
                    {
                        "first": "Youngseo",
                        "middle": [],
                        "last": "Son",
                        "suffix": ""
                    },
                    {
                        "first": "Richard",
                        "middle": [],
                        "last": "Sicoli",
                        "suffix": ""
                    },
                    {
                        "first": "Niranjan",
                        "middle": [],
                        "last": "Balasubramanian",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the 28th International Conference on Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "604--615",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/2020.coling-main.52"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Mohaddeseh Bastan, Mahnaz Koupaee, Youngseo Son, Richard Sicoli, and Niranjan Balasubramanian. 2020. Author's sentiment prediction. In Proceedings of the 28th International Conference on Computational Lin- guistics, pages 604-615, Barcelona, Spain (Online). International Committee on Computational Linguis- tics.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Reliable baselines for sentiment analysis in resource-limited languages: The Serbian movie review dataset",
                "authors": [
                    {
                        "first": "Vuk",
                        "middle": [],
                        "last": "Batanovi\u0107",
                        "suffix": ""
                    },
                    {
                        "first": "Bo\u0161ko",
                        "middle": [],
                        "last": "Nikoli\u0107",
                        "suffix": ""
                    },
                    {
                        "first": "Milan",
                        "middle": [],
                        "last": "Milosavljevi\u0107",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)",
                "volume": "",
                "issue": "",
                "pages": "2688--2696",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Vuk Batanovi\u0107, Bo\u0161ko Nikoli\u0107, and Milan Milosavlje- vi\u0107. 2016. Reliable baselines for sentiment analysis in resource-limited languages: The Serbian movie review dataset. In Proceedings of the Tenth Inter- national Conference on Language Resources and Evaluation (LREC'16), pages 2688-2696, Portoro\u017e, Slovenia. European Language Resources Association (ELRA).",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "A versatile framework for resource-limited sentiment articulation, annotation, and analysis of short texts",
                "authors": [
                    {
                        "first": "Vuk",
                        "middle": [],
                        "last": "Batanovi\u0107",
                        "suffix": ""
                    },
                    {
                        "first": "Milo\u0161",
                        "middle": [],
                        "last": "Cvetanovi\u0107",
                        "suffix": ""
                    },
                    {
                        "first": "Bo\u0161ko",
                        "middle": [],
                        "last": "Nikoli\u0107",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "PLOS ONE",
                "volume": "15",
                "issue": "11",
                "pages": "1--30",
                "other_ids": {
                    "DOI": [
                        "10.1371/journal.pone.0242050"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Vuk Batanovi\u0107, Milo\u0161 Cvetanovi\u0107, and Bo\u0161ko Nikoli\u0107. 2020. A versatile framework for resource-limited sentiment articulation, annotation, and analysis of short texts. PLOS ONE, 15(11):1-30.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Building a sentiment corpus of tweets in Brazilian Portuguese",
                "authors": [
                    {
                        "first": "Henrico",
                        "middle": [],
                        "last": "Brum",
                        "suffix": ""
                    },
                    {
                        "first": "Maria",
                        "middle": [],
                        "last": "Das Gra\u00e7as Volpe",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Nunes",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Henrico Brum and Maria das Gra\u00e7as Volpe Nunes. 2018. Building a sentiment corpus of tweets in Brazilian Portuguese. In Proceedings of the Eleventh Inter- national Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA).",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Annotated news corpora and a lexicon for sentiment analysis in slovene",
                "authors": [
                    {
                        "first": "Jo\u017ee",
                        "middle": [],
                        "last": "Bu\u010dar",
                        "suffix": ""
                    },
                    {
                        "first": "Martin",
                        "middle": [],
                        "last": "\u017dnidar\u0161i\u010d",
                        "suffix": ""
                    },
                    {
                        "first": "Janez",
                        "middle": [],
                        "last": "Povh",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Language Resources and Evaluation",
                "volume": "52",
                "issue": "3",
                "pages": "895--919",
                "other_ids": {
                    "DOI": [
                        "10.1007/s10579-018-9413-3"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Jo\u017ee Bu\u010dar, Martin \u017dnidar\u0161i\u010d, and Janez Povh. 2018. Annotated news corpora and a lexicon for sentiment analysis in slovene. Language Resources and Evalu- ation, 52(3):895-919.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "Multilingual sentiment analysis: An RNN-based framework for limited data",
                "authors": [
                    {
                        "first": "F",
                        "middle": [],
                        "last": "Ethem",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Can",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Computing Research Repository",
                "volume": "1",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1806.04511"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Ethem F. Can, Aysu Ezen-Can, and Fazli Can. 2018. Multilingual sentiment analysis: An RNN-based framework for limited data. Computing Research Repository, arXiv:1806.04511. Version 1.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "Hierarchical pre-training for sequence labelling in spoken dialog",
                "authors": [
                    {
                        "first": "Emile",
                        "middle": [],
                        "last": "Chapuis",
                        "suffix": ""
                    },
                    {
                        "first": "Pierre",
                        "middle": [],
                        "last": "Colombo",
                        "suffix": ""
                    },
                    {
                        "first": "Matteo",
                        "middle": [],
                        "last": "Manica",
                        "suffix": ""
                    },
                    {
                        "first": "Matthieu",
                        "middle": [],
                        "last": "Labeau",
                        "suffix": ""
                    },
                    {
                        "first": "Chlo\u00e9",
                        "middle": [],
                        "last": "Clavel",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Findings of the Association for Computational Linguistics: EMNLP 2020",
                "volume": "",
                "issue": "",
                "pages": "2636--2648",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/2020.findings-emnlp.239"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Emile Chapuis, Pierre Colombo, Matteo Manica, Matthieu Labeau, and Chlo\u00e9 Clavel. 2020. Hier- archical pre-training for sequence labelling in spoken dialog. In Findings of the Association for Computa- tional Linguistics: EMNLP 2020, pages 2636-2648, Online. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "A Twitter corpus and benchmark resources for German sentiment analysis",
                "authors": [
                    {
                        "first": "Mark",
                        "middle": [],
                        "last": "Cieliebak",
                        "suffix": ""
                    },
                    {
                        "first": "Jan",
                        "middle": [
                            "Milan"
                        ],
                        "last": "Deriu",
                        "suffix": ""
                    },
                    {
                        "first": "Dominic",
                        "middle": [],
                        "last": "Egger",
                        "suffix": ""
                    },
                    {
                        "first": "Fatih",
                        "middle": [],
                        "last": "Uzdilli",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "Proceedings of the Fifth International Workshop on Natural Language Processing for Social Media",
                "volume": "",
                "issue": "",
                "pages": "45--51",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/W17-1106"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Mark Cieliebak, Jan Milan Deriu, Dominic Egger, and Fatih Uzdilli. 2017. A Twitter corpus and benchmark resources for German sentiment analysis. In Proceed- ings of the Fifth International Workshop on Natural Language Processing for Social Media, pages 45- 51, Valencia, Spain. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "Unsupervised cross-lingual representation learning at scale",
                "authors": [
                    {
                        "first": "Edouard",
                        "middle": [],
                        "last": "Guzm\u00e1n",
                        "suffix": ""
                    },
                    {
                        "first": "Myle",
                        "middle": [],
                        "last": "Grave",
                        "suffix": ""
                    },
                    {
                        "first": "Luke",
                        "middle": [],
                        "last": "Ott",
                        "suffix": ""
                    },
                    {
                        "first": "Veselin",
                        "middle": [],
                        "last": "Zettlemoyer",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Stoyanov",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "8440--8451",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/2020.acl-main.747"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Guzm\u00e1n, Edouard Grave, Myle Ott, Luke Zettle- moyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. In Pro- ceedings of the 58th Annual Meeting of the Asso- ciation for Computational Linguistics, pages 8440- 8451, Online. Association for Computational Lin- guistics.",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "Crosslingual language model pretraining",
                "authors": [
                    {
                        "first": "Alexis",
                        "middle": [],
                        "last": "Conneau",
                        "suffix": ""
                    },
                    {
                        "first": "Guillaume",
                        "middle": [],
                        "last": "Lample",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of the 33rd International Conference on Neural Information Processing Systems",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "DOI": [
                        "https://dl.acm.org/doi/10.5555/3454287.3454921"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Alexis Conneau and Guillaume Lample. 2019. Cross- lingual language model pretraining. In Proceedings of the 33rd International Conference on Neural In- formation Processing Systems, Red Hook, NY, USA. Curran Associates Inc.",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "XNLI: Evaluating crosslingual sentence representations",
                "authors": [
                    {
                        "first": "Alexis",
                        "middle": [],
                        "last": "Conneau",
                        "suffix": ""
                    },
                    {
                        "first": "Ruty",
                        "middle": [],
                        "last": "Rinott",
                        "suffix": ""
                    },
                    {
                        "first": "Guillaume",
                        "middle": [],
                        "last": "Lample",
                        "suffix": ""
                    },
                    {
                        "first": "Adina",
                        "middle": [],
                        "last": "Williams",
                        "suffix": ""
                    },
                    {
                        "first": "Samuel",
                        "middle": [],
                        "last": "Bowman",
                        "suffix": ""
                    },
                    {
                        "first": "Holger",
                        "middle": [],
                        "last": "Schwenk",
                        "suffix": ""
                    },
                    {
                        "first": "Veselin",
                        "middle": [],
                        "last": "Stoyanov",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing",
                "volume": "",
                "issue": "",
                "pages": "2475--2485",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/D18-1269"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Alexis Conneau, Ruty Rinott, Guillaume Lample, Adina Williams, Samuel Bowman, Holger Schwenk, and Veselin Stoyanov. 2018. XNLI: Evaluating cross- lingual sentence representations. In Proceedings of the 2018 Conference on Empirical Methods in Nat- ural Language Processing, pages 2475-2485, Brus- sels, Belgium. Association for Computational Lin- guistics.",
                "links": null
            },
            "BIBREF19": {
                "ref_id": "b19",
                "title": "Experiments in sentiment classification of movie reviews in spanish",
                "authors": [
                    {
                        "first": "L",
                        "middle": [],
                        "last": "Fermin",
                        "suffix": ""
                    },
                    {
                        "first": "Jose",
                        "middle": [
                            "A"
                        ],
                        "last": "Cruz",
                        "suffix": ""
                    },
                    {
                        "first": "Fernando",
                        "middle": [],
                        "last": "Troyano",
                        "suffix": ""
                    },
                    {
                        "first": "Javier",
                        "middle": [],
                        "last": "Enriquez",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Ortega",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Procesamiento del Lenguaje Natural",
                "volume": "41",
                "issue": "",
                "pages": "73--80",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Fermin L Cruz, Jose A Troyano, Fernando Enriquez, and Javier Ortega. 2008. Experiments in sentiment classification of movie reviews in spanish. Proce- samiento del Lenguaje Natural, 41:73-80.",
                "links": null
            },
            "BIBREF20": {
                "ref_id": "b20",
                "title": "Multilingual sentiment analysis: state of the art and independent comparison of techniques",
                "authors": [
                    {
                        "first": "Kia",
                        "middle": [],
                        "last": "Dashtipour",
                        "suffix": ""
                    },
                    {
                        "first": "Soujanya",
                        "middle": [],
                        "last": "Poria",
                        "suffix": ""
                    },
                    {
                        "first": "Amir",
                        "middle": [],
                        "last": "Hussain",
                        "suffix": ""
                    },
                    {
                        "first": "Erik",
                        "middle": [],
                        "last": "Cambria",
                        "suffix": ""
                    },
                    {
                        "first": "Y",
                        "middle": [
                            "A"
                        ],
                        "last": "Ahmad",
                        "suffix": ""
                    },
                    {
                        "first": "Alexander",
                        "middle": [],
                        "last": "Hawalah",
                        "suffix": ""
                    },
                    {
                        "first": "Qiang",
                        "middle": [],
                        "last": "Gelbukh",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Zhou",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Cognitive computation",
                "volume": "8",
                "issue": "4",
                "pages": "757--771",
                "other_ids": {
                    "DOI": [
                        "10.1007/s12559-016-9415-7"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Kia Dashtipour, Soujanya Poria, Amir Hussain, Erik Cambria, Ahmad YA Hawalah, Alexander Gelbukh, and Qiang Zhou. 2016. Multilingual sentiment anal- ysis: state of the art and independent comparison of techniques. Cognitive computation, 8(4):757-771.",
                "links": null
            },
            "BIBREF21": {
                "ref_id": "b21",
                "title": "Statistical comparisons of classifiers over multiple data sets",
                "authors": [
                    {
                        "first": "Janez",
                        "middle": [],
                        "last": "Dem\u0161ar",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Journal of Machine Learning Research",
                "volume": "7",
                "issue": "",
                "pages": "1--30",
                "other_ids": {
                    "DOI": [
                        "https://dl.acm.org/doi/10.5555/1248547.1248548"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Janez Dem\u0161ar. 2006. Statistical comparisons of clas- sifiers over multiple data sets. Journal of Machine Learning Research, 7:1-30.",
                "links": null
            },
            "BIBREF22": {
                "ref_id": "b22",
                "title": "BERT: Pre-training of deep bidirectional transformers for language understanding",
                "authors": [
                    {
                        "first": "Jacob",
                        "middle": [],
                        "last": "Devlin",
                        "suffix": ""
                    },
                    {
                        "first": "Ming-Wei",
                        "middle": [],
                        "last": "Chang",
                        "suffix": ""
                    },
                    {
                        "first": "Kenton",
                        "middle": [],
                        "last": "Lee",
                        "suffix": ""
                    },
                    {
                        "first": "Kristina",
                        "middle": [],
                        "last": "Toutanova",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
                "volume": "1",
                "issue": "",
                "pages": "4171--4186",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/N19-1423"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language under- standing. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Tech- nologies, Volume 1 (Long and Short Papers), pages 4171-4186, Minneapolis, Minnesota. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF23": {
                "ref_id": "b23",
                "title": "BRAD 1.0: Book reviews in arabic dataset",
                "authors": [
                    {
                        "first": "Ashraf",
                        "middle": [],
                        "last": "Elnagar",
                        "suffix": ""
                    },
                    {
                        "first": "Omar",
                        "middle": [],
                        "last": "Einea",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "2016 IEEE/ACS 13th International Conference of Computer Systems and Applications (AICCSA)",
                "volume": "",
                "issue": "",
                "pages": "1--8",
                "other_ids": {
                    "DOI": [
                        "10.1109/AICCSA.2016.7945800"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Ashraf Elnagar and Omar Einea. 2016. BRAD 1.0: Book reviews in arabic dataset. In 2016 IEEE/ACS 13th International Conference of Computer Systems and Applications (AICCSA), pages 1-8.",
                "links": null
            },
            "BIBREF24": {
                "ref_id": "b24",
                "title": "Hotel Arabic-Reviews Dataset Construction for Sentiment Analysis Applications",
                "authors": [
                    {
                        "first": "Ashraf",
                        "middle": [],
                        "last": "Elnagar",
                        "suffix": ""
                    },
                    {
                        "first": "Yasmin",
                        "middle": [
                            "S"
                        ],
                        "last": "Khalifa",
                        "suffix": ""
                    },
                    {
                        "first": "Anas",
                        "middle": [],
                        "last": "Einea",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "DOI": [
                        "10.1007/978-3-319-67056-0_3"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Ashraf Elnagar, Yasmin S. Khalifa, and Anas Einea. 2018. Hotel Arabic-Reviews Dataset Construction for Sentiment Analysis Applications. Springer Inter- national Publishing, Cham.",
                "links": null
            },
            "BIBREF25": {
                "ref_id": "b25",
                "title": "Language-agnostic BERT Sentence Embedding",
                "authors": [
                    {
                        "first": "Fangxiaoyu",
                        "middle": [],
                        "last": "Feng",
                        "suffix": ""
                    },
                    {
                        "first": "Yinfei",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    },
                    {
                        "first": "Daniel",
                        "middle": [],
                        "last": "Cer",
                        "suffix": ""
                    },
                    {
                        "first": "Naveen",
                        "middle": [],
                        "last": "Arivazhagan",
                        "suffix": ""
                    },
                    {
                        "first": "Wei",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Computing Research Repository",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2007.01852"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Fangxiaoyu Feng, Yinfei Yang, Daniel Cer, Naveen Ari- vazhagan, and Wei Wang. 2020. Language-agnostic BERT Sentence Embedding. Computing Research Repository, arXiv:2007.01852. Version 2.",
                "links": null
            },
            "BIBREF26": {
                "ref_id": "b26",
                "title": "Sentiment analysis in Czech social media using supervised machine learning",
                "authors": [
                    {
                        "first": "Ivan",
                        "middle": [],
                        "last": "Habernal",
                        "suffix": ""
                    },
                    {
                        "first": "Tom\u00e1\u0161",
                        "middle": [],
                        "last": "Pt\u00e1\u010dek",
                        "suffix": ""
                    },
                    {
                        "first": "Josef",
                        "middle": [],
                        "last": "Steinberger",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Proceedings of the 4th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis",
                "volume": "",
                "issue": "",
                "pages": "65--74",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ivan Habernal, Tom\u00e1\u0161 Pt\u00e1\u010dek, and Josef Steinberger. 2013. Sentiment analysis in Czech social media us- ing supervised machine learning. In Proceedings of the 4th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analy- sis, pages 65-74, Atlanta, Georgia. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF27": {
                "ref_id": "b27",
                "title": "Mansoureh Anvari, and Seyed Abolghasem Mirroshandel",
                "authors": [
                    {
                        "first": "Pedram",
                        "middle": [],
                        "last": "Hosseini",
                        "suffix": ""
                    },
                    {
                        "first": "Ali",
                        "middle": [
                            "Ahmadian"
                        ],
                        "last": "Ramaki",
                        "suffix": ""
                    },
                    {
                        "first": "Hassan",
                        "middle": [],
                        "last": "Maleki",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Computing Research Repository",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1801.07737"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Pedram Hosseini, Ali Ahmadian Ramaki, Hassan Maleki, Mansoureh Anvari, and Seyed Abolghasem Mirroshandel. 2018. SentiPers: A sentiment analysis corpus for persian. Computing Research Repository, arXiv:1801.07737. Version 2.",
                "links": null
            },
            "BIBREF28": {
                "ref_id": "b28",
                "title": "XTREME: A massively multilingual multitask benchmark for evaluating cross-lingual generalisation",
                "authors": [
                    {
                        "first": "Junjie",
                        "middle": [],
                        "last": "Hu",
                        "suffix": ""
                    },
                    {
                        "first": "Sebastian",
                        "middle": [],
                        "last": "Ruder",
                        "suffix": ""
                    },
                    {
                        "first": "Aditya",
                        "middle": [],
                        "last": "Siddhant",
                        "suffix": ""
                    },
                    {
                        "first": "Graham",
                        "middle": [],
                        "last": "Neubig",
                        "suffix": ""
                    },
                    {
                        "first": "Orhan",
                        "middle": [],
                        "last": "Firat",
                        "suffix": ""
                    },
                    {
                        "first": "Melvin",
                        "middle": [],
                        "last": "Johnson",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the 37th International Conference on Machine Learning",
                "volume": "119",
                "issue": "",
                "pages": "4411--4421",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Junjie Hu, Sebastian Ruder, Aditya Siddhant, Gra- ham Neubig, Orhan Firat, and Melvin Johnson. 2020. XTREME: A massively multilingual multi- task benchmark for evaluating cross-lingual gener- alisation. In Proceedings of the 37th International Conference on Machine Learning, volume 119 of Proceedings of Machine Learning Research, pages 4411-4421. PMLR.",
                "links": null
            },
            "BIBREF29": {
                "ref_id": "b29",
                "title": "VADER: A parsimonious rule-based model for sentiment analysis of social media text",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Clayton",
                        "suffix": ""
                    },
                    {
                        "first": "Eric",
                        "middle": [],
                        "last": "Hutto",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Gilbert",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Proceedings of the International AAAI Conference on Web and Social Media",
                "volume": "8",
                "issue": "",
                "pages": "216--225",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Clayton J. Hutto and Eric Gilbert. 2014. VADER: A par- simonious rule-based model for sentiment analysis of social media text. In Proceedings of the Interna- tional AAAI Conference on Web and Social Media, volume 8, pages 216-225.",
                "links": null
            },
            "BIBREF30": {
                "ref_id": "b30",
                "title": "Twitter us airline sentiment",
                "authors": [
                    {
                        "first": "Crowdflower",
                        "middle": [],
                        "last": "Inc",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Crowdflower Inc. 2015. Twitter us airline sentiment.",
                "links": null
            },
            "BIBREF31": {
                "ref_id": "b31",
                "title": "Cross-lingual ability of multilingual bert: An empirical study",
                "authors": [
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Karthikeyan",
                        "suffix": ""
                    },
                    {
                        "first": "Zihan",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    },
                    {
                        "first": "Stephen",
                        "middle": [],
                        "last": "Mayhew",
                        "suffix": ""
                    },
                    {
                        "first": "Dan",
                        "middle": [],
                        "last": "Roth",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "International Conference on Learning Representations",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Karthikeyan K, Zihan Wang, Stephen Mayhew, and Dan Roth. 2020. Cross-lingual ability of multilingual bert: An empirical study. In International Conference on Learning Representations.",
                "links": null
            },
            "BIBREF32": {
                "ref_id": "b32",
                "title": "Cross-lingual deep neural transfer learning in sentiment analysis",
                "authors": [
                    {
                        "first": "Kamil",
                        "middle": [],
                        "last": "Kanclerz",
                        "suffix": ""
                    },
                    {
                        "first": "Piotr",
                        "middle": [],
                        "last": "Mi\u0142kowski",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Knowledge-Based and Intelligent Information & Engineering Systems: Proceedings of the 24th International Conference KES2020",
                "volume": "176",
                "issue": "",
                "pages": "128--137",
                "other_ids": {
                    "DOI": [
                        "10.1016/j.procs.2020.08.014"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Kamil Kanclerz, Piotr Mi\u0142kowski, and Jan Koco\u0144. 2020. Cross-lingual deep neural transfer learning in senti- ment analysis. Procedia Computer Science, 176:128- 137. Knowledge-Based and Intelligent Information & Engineering Systems: Proceedings of the 24th International Conference KES2020.",
                "links": null
            },
            "BIBREF33": {
                "ref_id": "b33",
                "title": "Sentiment analysis and opinion mining applied to scientific paper reviews",
                "authors": [
                    {
                        "first": "Brian",
                        "middle": [],
                        "last": "Keith Norambuena",
                        "suffix": ""
                    },
                    {
                        "first": "Exequiel",
                        "middle": [],
                        "last": "Lettura",
                        "suffix": ""
                    },
                    {
                        "first": "Claudio",
                        "middle": [],
                        "last": "Villegas",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "",
                "volume": "23",
                "issue": "",
                "pages": "191--214",
                "other_ids": {
                    "DOI": [
                        "10.3233/IDA-173807"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Brian Keith Norambuena, Exequiel Lettura, and Clau- dio Villegas. 2019. Sentiment analysis and opinion mining applied to scientific paper reviews. Intelligent Data Analysis, 23:191-214.",
                "links": null
            },
            "BIBREF34": {
                "ref_id": "b34",
                "title": "The multilingual Amazon reviews corpus",
                "authors": [
                    {
                        "first": "Phillip",
                        "middle": [],
                        "last": "Keung",
                        "suffix": ""
                    },
                    {
                        "first": "Yichao",
                        "middle": [],
                        "last": "Lu",
                        "suffix": ""
                    },
                    {
                        "first": "Gy\u00f6rgy",
                        "middle": [],
                        "last": "Szarvas",
                        "suffix": ""
                    },
                    {
                        "first": "Noah",
                        "middle": [
                            "A"
                        ],
                        "last": "Smith",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)",
                "volume": "",
                "issue": "",
                "pages": "4563--4568",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/2020.emnlp-main.369"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Phillip Keung, Yichao Lu, Gy\u00f6rgy Szarvas, and Noah A. Smith. 2020. The multilingual Amazon reviews cor- pus. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 4563-4568, Online. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF35": {
                "ref_id": "b35",
                "title": "Multi-level sentiment analysis of PolEmo 2.0: Extended corpus of multi-domain consumer reviews",
                "authors": [
                    {
                        "first": "Jan",
                        "middle": [],
                        "last": "Koco\u0144",
                        "suffix": ""
                    },
                    {
                        "first": "Piotr",
                        "middle": [],
                        "last": "Mi\u0142kowski",
                        "suffix": ""
                    },
                    {
                        "first": "Monika",
                        "middle": [],
                        "last": "Za\u015bko-Zieli\u0144ska",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL)",
                "volume": "",
                "issue": "",
                "pages": "980--991",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/K19-1092"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Jan Koco\u0144, Piotr Mi\u0142kowski, and Monika Za\u015bko- Zieli\u0144ska. 2019. Multi-level sentiment analysis of PolEmo 2.0: Extended corpus of multi-domain con- sumer reviews. In Proceedings of the 23rd Confer- ence on Computational Natural Language Learning (CoNLL), pages 980-991, Hong Kong, China. Asso- ciation for Computational Linguistics.",
                "links": null
            },
            "BIBREF36": {
                "ref_id": "b36",
                "title": "MLQA: Evaluating cross-lingual extractive question answering",
                "authors": [
                    {
                        "first": "Patrick",
                        "middle": [],
                        "last": "Lewis",
                        "suffix": ""
                    },
                    {
                        "first": "Barlas",
                        "middle": [],
                        "last": "Oguz",
                        "suffix": ""
                    },
                    {
                        "first": "Ruty",
                        "middle": [],
                        "last": "Rinott",
                        "suffix": ""
                    },
                    {
                        "first": "Sebastian",
                        "middle": [],
                        "last": "Riedel",
                        "suffix": ""
                    },
                    {
                        "first": "Holger",
                        "middle": [],
                        "last": "Schwenk",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "7315--7330",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/2020.acl-main.653"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Patrick Lewis, Barlas Oguz, Ruty Rinott, Sebastian Riedel, and Holger Schwenk. 2020. MLQA: Evalu- ating cross-lingual extractive question answering. In Proceedings of the 58th Annual Meeting of the Asso- ciation for Computational Linguistics, pages 7315- 7330, Online. Association for Computational Lin- guistics.",
                "links": null
            },
            "BIBREF37": {
                "ref_id": "b37",
                "title": "An empirical study on sentiment classification of Chinese review using word embedding",
                "authors": [
                    {
                        "first": "Yiou",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    },
                    {
                        "first": "Hang",
                        "middle": [],
                        "last": "Lei",
                        "suffix": ""
                    },
                    {
                        "first": "Jia",
                        "middle": [],
                        "last": "Wu",
                        "suffix": ""
                    },
                    {
                        "first": "Xiaoyu",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "Proceedings of the 29th Pacific Asia Conference on Language, Information and Computation: Posters",
                "volume": "",
                "issue": "",
                "pages": "258--266",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yiou Lin, Hang Lei, Jia Wu, and Xiaoyu Li. 2015. An empirical study on sentiment classification of Chi- nese review using word embedding. In Proceedings of the 29th Pacific Asia Conference on Language, In- formation and Computation: Posters, pages 258-266, Shanghai, China.",
                "links": null
            },
            "BIBREF38": {
                "ref_id": "b38",
                "title": "Yung-Sung Chuang, and Hung yi Lee. 2020. What makes multilingual bert multilingual? Computing Research Repository",
                "authors": [
                    {
                        "first": "Chi-Liang",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "Tsung-Yuan",
                        "middle": [],
                        "last": "Hsu",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "",
                "volume": "1",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2010.10938"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Chi-Liang Liu, Tsung-Yuan Hsu, Yung-Sung Chuang, and Hung yi Lee. 2020. What makes multilingual bert multilingual? Computing Research Repository, arXiv:2010.10938. Version 1.",
                "links": null
            },
            "BIBREF39": {
                "ref_id": "b39",
                "title": "Good debt or bad debt: Detecting semantic orientations in economic texts",
                "authors": [
                    {
                        "first": "Pekka",
                        "middle": [],
                        "last": "Malo",
                        "suffix": ""
                    },
                    {
                        "first": "Ankur",
                        "middle": [],
                        "last": "Sinha",
                        "suffix": ""
                    },
                    {
                        "first": "Pekka",
                        "middle": [],
                        "last": "Korhonen",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Journal of the Association for Information Science and Technology",
                "volume": "65",
                "issue": "4",
                "pages": "782--796",
                "other_ids": {
                    "DOI": [
                        "10.1002/asi.23062"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Pekka Malo, Ankur Sinha, Pekka Korhonen, Jyrki Wal- lenius, and Pyry Takala. 2014. Good debt or bad debt: Detecting semantic orientations in economic texts. Journal of the Association for Information Science and Technology, 65(4):782-796.",
                "links": null
            },
            "BIBREF40": {
                "ref_id": "b40",
                "title": "Efficient estimation of word representations in vector space",
                "authors": [
                    {
                        "first": "Tom\u00e1s",
                        "middle": [],
                        "last": "Mikolov",
                        "suffix": ""
                    },
                    {
                        "first": "Kai",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "Greg",
                        "middle": [],
                        "last": "Corrado",
                        "suffix": ""
                    },
                    {
                        "first": "Jeffrey",
                        "middle": [],
                        "last": "Dean",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "1st International Conference on Learning Representations",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Tom\u00e1s Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013a. Efficient estimation of word representa- tions in vector space. In 1st International Conference on Learning Representations, ICLR 2013, Scottsdale, Arizona, USA, May 2-4, 2013, Workshop Track Pro- ceedings.",
                "links": null
            },
            "BIBREF41": {
                "ref_id": "b41",
                "title": "Exploiting similarities among languages for machine translation",
                "authors": [
                    {
                        "first": "Tomas",
                        "middle": [],
                        "last": "Mikolov",
                        "suffix": ""
                    },
                    {
                        "first": "V",
                        "middle": [],
                        "last": "Quoc",
                        "suffix": ""
                    },
                    {
                        "first": "Ilya",
                        "middle": [],
                        "last": "Le",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Sutskever",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Computing Research Repository",
                "volume": "1",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1309.4168"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Tomas Mikolov, Quoc V. Le, and Ilya Sutskever. 2013b. Exploiting similarities among languages for ma- chine translation. Computing Research Repository, arXiv:1309.4168. Version 1.",
                "links": null
            },
            "BIBREF42": {
                "ref_id": "b42",
                "title": "Multilingual twitter sentiment classification: The role of human annotators",
                "authors": [
                    {
                        "first": "Igor",
                        "middle": [],
                        "last": "Mozeti\u010d",
                        "suffix": ""
                    },
                    {
                        "first": "Miha",
                        "middle": [],
                        "last": "Gr\u010dar",
                        "suffix": ""
                    },
                    {
                        "first": "Jasmina",
                        "middle": [],
                        "last": "Smailovi\u0107",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "PLOS ONE",
                "volume": "11",
                "issue": "5",
                "pages": "1--26",
                "other_ids": {
                    "DOI": [
                        "10.1371/journal.pone.0155036"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Igor Mozeti\u010d, Miha Gr\u010dar, and Jasmina Smailovi\u0107. 2016. Multilingual twitter sentiment classification: The role of human annotators. PLOS ONE, 11(5):1-26.",
                "links": null
            },
            "BIBREF43": {
                "ref_id": "b43",
                "title": "ASTD: Arabic sentiment tweets dataset",
                "authors": [
                    {
                        "first": "Mahmoud",
                        "middle": [],
                        "last": "Nabil",
                        "suffix": ""
                    },
                    {
                        "first": "Mohamed",
                        "middle": [],
                        "last": "Aly",
                        "suffix": ""
                    },
                    {
                        "first": "Amir",
                        "middle": [],
                        "last": "Atiya",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing",
                "volume": "",
                "issue": "",
                "pages": "2515--2519",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/D15-1299"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Mahmoud Nabil, Mohamed Aly, and Amir Atiya. 2015. ASTD: Arabic sentiment tweets dataset. In Proceed- ings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 2515-2519, Lisbon, Portugal. Association for Computational Lin- guistics.",
                "links": null
            },
            "BIBREF44": {
                "ref_id": "b44",
                "title": "Language-independent twitter sentiment analysis",
                "authors": [
                    {
                        "first": "Sascha",
                        "middle": [],
                        "last": "Narr",
                        "suffix": ""
                    },
                    {
                        "first": "Michael",
                        "middle": [],
                        "last": "H\u00fclfenhaus",
                        "suffix": ""
                    },
                    {
                        "first": "Sahin",
                        "middle": [],
                        "last": "Albayrak",
                        "suffix": ""
                    }
                ],
                "year": 2012,
                "venue": "Workshop on Knowledge Discovery, Data Mining and Machine Learning",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Sascha Narr, Michael H\u00fclfenhaus, and Sahin Albayrak. 2012. Language-independent twitter sentiment anal- ysis. In Workshop on Knowledge Discovery, Data Mining and Machine Learning (KDML-2012).",
                "links": null
            },
            "BIBREF45": {
                "ref_id": "b45",
                "title": "Distribution-free Multiple Comparisons",
                "authors": [
                    {
                        "first": "Peter",
                        "middle": [],
                        "last": "Nemenyi",
                        "suffix": ""
                    }
                ],
                "year": 1963,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Peter Nemenyi. 1963. Distribution-free Multiple Com- parisons. Princeton University.",
                "links": null
            },
            "BIBREF46": {
                "ref_id": "b46",
                "title": "Justifying recommendations using distantly-labeled reviews and fine-grained aspects",
                "authors": [
                    {
                        "first": "Jianmo",
                        "middle": [],
                        "last": "Ni",
                        "suffix": ""
                    },
                    {
                        "first": "Jiacheng",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    },
                    {
                        "first": "Julian",
                        "middle": [],
                        "last": "Mcauley",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)",
                "volume": "",
                "issue": "",
                "pages": "188--197",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/D19-1018"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Jianmo Ni, Jiacheng Li, and Julian McAuley. 2019. Justifying recommendations using distantly-labeled reviews and fine-grained aspects. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th Interna- tional Joint Conference on Natural Language Pro- cessing (EMNLP-IJCNLP), pages 188-197, Hong Kong, China. Association for Computational Lin- guistics.",
                "links": null
            },
            "BIBREF47": {
                "ref_id": "b47",
                "title": "Confident learning: Estimating uncertainty in dataset labels",
                "authors": [
                    {
                        "first": "Curtis",
                        "middle": [],
                        "last": "Northcutt",
                        "suffix": ""
                    },
                    {
                        "first": "Lu",
                        "middle": [],
                        "last": "Jiang",
                        "suffix": ""
                    },
                    {
                        "first": "Isaac",
                        "middle": [],
                        "last": "Chuang",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "Journal of Artificial Intelligence Research",
                "volume": "70",
                "issue": "",
                "pages": "1373--1411",
                "other_ids": {
                    "DOI": [
                        "10.1613/jair.1.12125"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Curtis Northcutt, Lu Jiang, and Isaac Chuang. 2021. Confident learning: Estimating uncertainty in dataset labels. Journal of Artificial Intelligence Research, 70:1373-1411.",
                "links": null
            },
            "BIBREF48": {
                "ref_id": "b48",
                "title": "SemEval-2020 task 9: Overview of sentiment analysis of code-mixed tweets",
                "authors": [
                    {
                        "first": "Parth",
                        "middle": [],
                        "last": "Patwa",
                        "suffix": ""
                    },
                    {
                        "first": "Gustavo",
                        "middle": [],
                        "last": "Aguilar",
                        "suffix": ""
                    },
                    {
                        "first": "Sudipta",
                        "middle": [],
                        "last": "Kar",
                        "suffix": ""
                    },
                    {
                        "first": "Suraj",
                        "middle": [],
                        "last": "Pandey",
                        "suffix": ""
                    },
                    {
                        "first": "Pykl",
                        "middle": [],
                        "last": "Srinivas",
                        "suffix": ""
                    },
                    {
                        "first": "Bj\u00f6rn",
                        "middle": [],
                        "last": "Gamb\u00e4ck",
                        "suffix": ""
                    },
                    {
                        "first": "Tanmoy",
                        "middle": [],
                        "last": "Chakraborty",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the Fourteenth Workshop on Semantic Evaluation",
                "volume": "",
                "issue": "",
                "pages": "774--790",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/2020.semeval-1.100"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Parth Patwa, Gustavo Aguilar, Sudipta Kar, Suraj Pandey, Srinivas PYKL, Bj\u00f6rn Gamb\u00e4ck, Tanmoy Chakraborty, Thamar Solorio, and Amitava Das. 2020. SemEval-2020 task 9: Overview of sentiment analysis of code-mixed tweets. In Proceedings of the Fourteenth Workshop on Semantic Evaluation, pages 774-790, Barcelona (online). International Commit- tee for Computational Linguistics.",
                "links": null
            },
            "BIBREF49": {
                "ref_id": "b49",
                "title": "Zero-shot learning for cross-lingual news sentiment classification",
                "authors": [
                    {
                        "first": "Andra\u017e",
                        "middle": [],
                        "last": "Pelicon",
                        "suffix": ""
                    },
                    {
                        "first": "Marko",
                        "middle": [],
                        "last": "Pranji\u0107",
                        "suffix": ""
                    },
                    {
                        "first": "Dragana",
                        "middle": [],
                        "last": "Miljkovi\u0107",
                        "suffix": ""
                    },
                    {
                        "first": "Bla\u017e",
                        "middle": [],
                        "last": "\u0160krlj",
                        "suffix": ""
                    },
                    {
                        "first": "Senja",
                        "middle": [],
                        "last": "Pollak",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Applied Sciences",
                "volume": "",
                "issue": "17",
                "pages": "",
                "other_ids": {
                    "DOI": [
                        "10.3390/app10175993"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Andra\u017e Pelicon, Marko Pranji\u0107, Dragana Miljkovi\u0107, Bla\u017e \u0160krlj, and Senja Pollak. 2020. Zero-shot learn- ing for cross-lingual news sentiment classification. Applied Sciences, 10(17).",
                "links": null
            },
            "BIBREF50": {
                "ref_id": "b50",
                "title": "Making monolingual sentence embeddings multilingual using knowledge distillation",
                "authors": [
                    {
                        "first": "Nils",
                        "middle": [],
                        "last": "Reimers",
                        "suffix": ""
                    },
                    {
                        "first": "Iryna",
                        "middle": [],
                        "last": "Gurevych",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)",
                "volume": "",
                "issue": "",
                "pages": "4512--4525",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/2020.emnlp-main.365"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Nils Reimers and Iryna Gurevych. 2020. Making monolingual sentence embeddings multilingual us- ing knowledge distillation. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 4512-4525, Online. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF51": {
                "ref_id": "b51",
                "title": "RuSentiment: An enriched sentiment analysis dataset for social media in Russian",
                "authors": [
                    {
                        "first": "Anna",
                        "middle": [],
                        "last": "Rogers",
                        "suffix": ""
                    },
                    {
                        "first": "Alexey",
                        "middle": [],
                        "last": "Romanov",
                        "suffix": ""
                    },
                    {
                        "first": "Anna",
                        "middle": [],
                        "last": "Rumshisky",
                        "suffix": ""
                    },
                    {
                        "first": "Svitlana",
                        "middle": [],
                        "last": "Volkova",
                        "suffix": ""
                    },
                    {
                        "first": "Mikhail",
                        "middle": [],
                        "last": "Gronas",
                        "suffix": ""
                    },
                    {
                        "first": "Alex",
                        "middle": [],
                        "last": "Gribov",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of the 27th International Conference on Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "755--763",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Anna Rogers, Alexey Romanov, Anna Rumshisky, Svit- lana Volkova, Mikhail Gronas, and Alex Gribov. 2018. RuSentiment: An enriched sentiment analysis dataset for social media in Russian. In Proceedings of the 27th International Conference on Computational Linguistics, pages 755-763, Santa Fe, New Mexico, USA. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF52": {
                "ref_id": "b52",
                "title": "SemEval-2017 task 4: Sentiment analysis in Twitter",
                "authors": [
                    {
                        "first": "Sara",
                        "middle": [],
                        "last": "Rosenthal",
                        "suffix": ""
                    },
                    {
                        "first": "Noura",
                        "middle": [],
                        "last": "Farra",
                        "suffix": ""
                    },
                    {
                        "first": "Preslav",
                        "middle": [],
                        "last": "Nakov",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017)",
                "volume": "",
                "issue": "",
                "pages": "502--518",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/S17-2088"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Sara Rosenthal, Noura Farra, and Preslav Nakov. 2017. SemEval-2017 task 4: Sentiment analysis in Twitter. In Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017), pages 502- 518, Vancouver, Canada. Association for Computa- tional Linguistics.",
                "links": null
            },
            "BIBREF53": {
                "ref_id": "b53",
                "title": "A survey of cross-lingual word embedding models",
                "authors": [
                    {
                        "first": "Sebastian",
                        "middle": [],
                        "last": "Ruder",
                        "suffix": ""
                    },
                    {
                        "first": "Ivan",
                        "middle": [],
                        "last": "Vuli\u0107",
                        "suffix": ""
                    },
                    {
                        "first": "Anders",
                        "middle": [],
                        "last": "S\u00f8gaard",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Journal of Artificial Intelligence Research",
                "volume": "65",
                "issue": "",
                "pages": "569--631",
                "other_ids": {
                    "DOI": [
                        "10.1613/jair.1.11640"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Sebastian Ruder, Ivan Vuli\u0107, and Anders S\u00f8gaard. 2019. A survey of cross-lingual word embedding mod- els. Journal of Artificial Intelligence Research, 65:569-631.",
                "links": null
            },
            "BIBREF54": {
                "ref_id": "b54",
                "title": "KLEJ: Comprehensive benchmark for Polish language understanding",
                "authors": [
                    {
                        "first": "Piotr",
                        "middle": [],
                        "last": "Rybak",
                        "suffix": ""
                    },
                    {
                        "first": "Robert",
                        "middle": [],
                        "last": "Mroczkowski",
                        "suffix": ""
                    },
                    {
                        "first": "Janusz",
                        "middle": [],
                        "last": "Tracz",
                        "suffix": ""
                    },
                    {
                        "first": "Ireneusz",
                        "middle": [],
                        "last": "Gawlik",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "1191--1201",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/2020.acl-main.111"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Piotr Rybak, Robert Mroczkowski, Janusz Tracz, and Ireneusz Gawlik. 2020. KLEJ: Comprehensive benchmark for Polish language understanding. In Proceedings of the 58th Annual Meeting of the Asso- ciation for Computational Linguistics, pages 1191- 1201, Online. Association for Computational Linguis- tics.",
                "links": null
            },
            "BIBREF55": {
                "ref_id": "b55",
                "title": "A review on multi-lingual sentiment analysis by machine learning methods",
                "authors": [
                    {
                        "first": "Santwana",
                        "middle": [],
                        "last": "Sagnika",
                        "suffix": ""
                    },
                    {
                        "first": "Anshuman",
                        "middle": [],
                        "last": "Pattanaik",
                        "suffix": ""
                    },
                    {
                        "first": "Bhabani",
                        "middle": [],
                        "last": "Shankar Prasad Mishra",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Saroj",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Meher",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Journal of Engineering Science & Technology Review",
                "volume": "13",
                "issue": "2",
                "pages": "154--166",
                "other_ids": {
                    "DOI": [
                        "10.25103/jestr.132.19"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Santwana Sagnika, Anshuman Pattanaik, Bhabani Shankar Prasad Mishra, and Saroj K Meher. 2020. A review on multi-lingual sentiment analysis by ma- chine learning methods. Journal of Engineering Sci- ence & Technology Review, 13(2):154-166.",
                "links": null
            },
            "BIBREF56": {
                "ref_id": "b56",
                "title": "Sentiment after translation: A case-study on Arabic social media posts",
                "authors": [
                    {
                        "first": "Mohammad",
                        "middle": [],
                        "last": "Salameh",
                        "suffix": ""
                    },
                    {
                        "first": "Saif",
                        "middle": [],
                        "last": "Mohammad",
                        "suffix": ""
                    },
                    {
                        "first": "Svetlana",
                        "middle": [],
                        "last": "Kiritchenko",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "Proceedings of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
                "volume": "",
                "issue": "",
                "pages": "767--777",
                "other_ids": {
                    "DOI": [
                        "10.3115/v1/N15-1078"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Mohammad Salameh, Saif Mohammad, and Svetlana Kiritchenko. 2015. Sentiment after translation: A case-study on Arabic social media posts. In Pro- ceedings of the 2015 Conference of the North Amer- ican Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 767-777, Denver, Colorado. Association for Compu- tational Linguistics.",
                "links": null
            },
            "BIBREF57": {
                "ref_id": "b57",
                "title": "Sanders-Twitter Sentiment Corpus",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Niek",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Sanders",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Niek J Sanders. 2011. Sanders-Twitter Sentiment Cor- pus. Sanders Analytics LLC.",
                "links": null
            },
            "BIBREF58": {
                "ref_id": "b58",
                "title": "DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter",
                "authors": [
                    {
                        "first": "Victor",
                        "middle": [],
                        "last": "Sanh",
                        "suffix": ""
                    },
                    {
                        "first": "Lysandre",
                        "middle": [],
                        "last": "Debut",
                        "suffix": ""
                    },
                    {
                        "first": "Julien",
                        "middle": [],
                        "last": "Chaumond",
                        "suffix": ""
                    },
                    {
                        "first": "Thomas",
                        "middle": [],
                        "last": "Wolf",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Computing Research Repository",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1910.01108"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. 2020. DistilBERT, a distilled ver- sion of BERT: smaller, faster, cheaper and lighter. Computing Research Repository, arXiv:1910.01108. Version 4.",
                "links": null
            },
            "BIBREF59": {
                "ref_id": "b59",
                "title": "Academic-industrial perspective on the development and deployment of a moderation system for a newspaper website",
                "authors": [
                    {
                        "first": "Dietmar",
                        "middle": [],
                        "last": "Schabus",
                        "suffix": ""
                    },
                    {
                        "first": "Marcin",
                        "middle": [],
                        "last": "Skowron",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Dietmar Schabus and Marcin Skowron. 2018. Academic-industrial perspective on the development and deployment of a moderation system for a newspaper website. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA).",
                "links": null
            },
            "BIBREF60": {
                "ref_id": "b60",
                "title": "Performing natural language processing on roman urdu datasets",
                "authors": [
                    {
                        "first": "Zareen",
                        "middle": [],
                        "last": "Sharf",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Saif Ur Rahman",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "In International Journal of Computer Science and Network Security",
                "volume": "18",
                "issue": "",
                "pages": "141--148",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Zareen Sharf and Saif Ur Rahman. 2018. Performing natural language processing on roman urdu datasets. In International Journal of Computer Science and Network Security, volume 18, pages 141-148.",
                "links": null
            },
            "BIBREF61": {
                "ref_id": "b61",
                "title": "Investigating societal biases in a poetry composition system",
                "authors": [
                    {
                        "first": "Emily",
                        "middle": [],
                        "last": "Sheng",
                        "suffix": ""
                    },
                    {
                        "first": "David",
                        "middle": [],
                        "last": "Uthus",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the Second Workshop on Gender Bias in Natural Language Processing",
                "volume": "",
                "issue": "",
                "pages": "93--106",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Emily Sheng and David Uthus. 2020. Investigating societal biases in a poetry composition system. In Proceedings of the Second Workshop on Gender Bias in Natural Language Processing, pages 93-106, Barcelona, Spain (Online). Association for Computa- tional Linguistics.",
                "links": null
            },
            "BIBREF62": {
                "ref_id": "b62",
                "title": "Twoyear study of emotion and communication patterns in a highly polarized political discussion forum",
                "authors": [
                    {
                        "first": "Pawel",
                        "middle": [],
                        "last": "Sobkowicz",
                        "suffix": ""
                    },
                    {
                        "first": "Antoni",
                        "middle": [],
                        "last": "Sobkowicz",
                        "suffix": ""
                    }
                ],
                "year": 2012,
                "venue": "Social Science Computer Review",
                "volume": "30",
                "issue": "4",
                "pages": "448--469",
                "other_ids": {
                    "DOI": [
                        "10.1177/0894439312436512"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Pawel Sobkowicz and Antoni Sobkowicz. 2012. Two- year study of emotion and communication patterns in a highly polarized political discussion forum. Social Science Computer Review, 30(4):448-469.",
                "links": null
            },
            "BIBREF63": {
                "ref_id": "b63",
                "title": "What can we learn from almost a decade of food tweets",
                "authors": [
                    {
                        "first": "Uga",
                        "middle": [],
                        "last": "Sprogis",
                        "suffix": ""
                    },
                    {
                        "first": "Matiss",
                        "middle": [],
                        "last": "Rikters",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:2007.05194"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Uga Sprogis and Matiss Rikters. 2020. What can we learn from almost a decade of food tweets. Comput- ing Research Repository, arXiv:2007.05194. Version 2.",
                "links": null
            },
            "BIBREF64": {
                "ref_id": "b64",
                "title": "Multiemotions-it: a new dataset for opinion polarity and emotion analysis for italian",
                "authors": [
                    {
                        "first": "Rachele",
                        "middle": [],
                        "last": "Sprugnoli",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the Seventh Italian Conference on Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Rachele Sprugnoli. 2020. Multiemotions-it: a new dataset for opinion polarity and emotion analysis for italian. In Proceedings of the Seventh Italian Conference on Computational Linguistics.",
                "links": null
            },
            "BIBREF66": {
                "ref_id": "b66",
                "title": "Pythainlp/wisesight-sentiment: First release (v1.0). Zenodo",
                "authors": [],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "DOI": [
                        "10.5281/zenodo.3457447"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Pythainlp/wisesight-sentiment: First release (v1.0). Zenodo.",
                "links": null
            },
            "BIBREF67": {
                "ref_id": "b67",
                "title": "Dataset cartography: Mapping and diagnosing datasets with training dynamics",
                "authors": [
                    {
                        "first": "Swabha",
                        "middle": [],
                        "last": "Swayamdipta",
                        "suffix": ""
                    },
                    {
                        "first": "Roy",
                        "middle": [],
                        "last": "Schwartz",
                        "suffix": ""
                    },
                    {
                        "first": "Nicholas",
                        "middle": [],
                        "last": "Lourie",
                        "suffix": ""
                    },
                    {
                        "first": "Yizhong",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    },
                    {
                        "first": "Hannaneh",
                        "middle": [],
                        "last": "Hajishirzi",
                        "suffix": ""
                    },
                    {
                        "first": "Noah",
                        "middle": [
                            "A"
                        ],
                        "last": "Smith",
                        "suffix": ""
                    },
                    {
                        "first": "Yejin",
                        "middle": [],
                        "last": "Choi",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)",
                "volume": "",
                "issue": "",
                "pages": "9275--9293",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/2020.emnlp-main.746"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Swabha Swayamdipta, Roy Schwartz, Nicholas Lourie, Yizhong Wang, Hannaneh Hajishirzi, Noah A. Smith, and Yejin Choi. 2020. Dataset cartography: Mapping and diagnosing datasets with training dynamics. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 9275-9293, Online. Association for Computa- tional Linguistics.",
                "links": null
            },
            "BIBREF68": {
                "ref_id": "b68",
                "title": "Sentiment strength detection for the social web",
                "authors": [
                    {
                        "first": "Mike",
                        "middle": [],
                        "last": "Thelwall",
                        "suffix": ""
                    },
                    {
                        "first": "Kevan",
                        "middle": [],
                        "last": "Buckley",
                        "suffix": ""
                    },
                    {
                        "first": "Georgios",
                        "middle": [],
                        "last": "Paltoglou",
                        "suffix": ""
                    }
                ],
                "year": 2012,
                "venue": "J. Am. Soc. Inf. Sci. Technol",
                "volume": "63",
                "issue": "1",
                "pages": "163--173",
                "other_ids": {
                    "DOI": [
                        "10.1002/asi.21662"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Mike Thelwall, Kevan Buckley, and Georgios Paltoglou. 2012. Sentiment strength detection for the social web. J. Am. Soc. Inf. Sci. Technol., 63(1):163-173.",
                "links": null
            },
            "BIBREF70": {
                "ref_id": "b70",
                "title": "Introduction to the CoNLL-2002 shared task: Language-independent named entity recognition",
                "authors": [
                    {
                        "first": "Erik",
                        "middle": [
                            "F"
                        ],
                        "last": "",
                        "suffix": ""
                    },
                    {
                        "first": "Tjong Kim",
                        "middle": [],
                        "last": "Sang",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Erik F. Tjong Kim Sang. 2002. Introduction to the CoNLL-2002 shared task: Language-independent named entity recognition. In COLING-02: The 6th",
                "links": null
            },
            "BIBREF72": {
                "ref_id": "b72",
                "title": "Introduction to the CoNLL-2003 shared task: Language-independent named entity recognition",
                "authors": [
                    {
                        "first": "Erik",
                        "middle": [
                            "F"
                        ],
                        "last": "Tjong",
                        "suffix": ""
                    },
                    {
                        "first": "Kim",
                        "middle": [],
                        "last": "Sang",
                        "suffix": ""
                    },
                    {
                        "first": "Fien",
                        "middle": [],
                        "last": "De Meulder",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Proceedings of the Seventh Conference on Natural Language Learning at HLT-NAACL 2003",
                "volume": "",
                "issue": "",
                "pages": "142--147",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Erik F. Tjong Kim Sang and Fien De Meulder. 2003. Introduction to the CoNLL-2003 shared task: Language-independent named entity recognition. In Proceedings of the Seventh Conference on Natural Language Learning at HLT-NAACL 2003, pages 142- 147.",
                "links": null
            },
            "BIBREF73": {
                "ref_id": "b73",
                "title": "A characterbased convolutional neural network for languageagnostic twitter sentiment analysis",
                "authors": [
                    {
                        "first": "Joonatas",
                        "middle": [],
                        "last": "Wehrmann",
                        "suffix": ""
                    },
                    {
                        "first": "Willian",
                        "middle": [],
                        "last": "Becker",
                        "suffix": ""
                    },
                    {
                        "first": "E",
                        "middle": [
                            "L"
                        ],
                        "last": "Henry",
                        "suffix": ""
                    },
                    {
                        "first": "Rodrigo",
                        "middle": [
                            "C"
                        ],
                        "last": "Cagnini",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Barros",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "2017 International Joint Conference on Neural Networks (IJCNN)",
                "volume": "",
                "issue": "",
                "pages": "2384--2391",
                "other_ids": {
                    "DOI": [
                        "10.1109/IJCNN.2017.7966145"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Joonatas Wehrmann, Willian Becker, Henry E. L. Cagnini, and Rodrigo C. Barros. 2017. A character- based convolutional neural network for language- agnostic twitter sentiment analysis. In 2017 Interna- tional Joint Conference on Neural Networks (IJCNN), pages 2384-2391.",
                "links": null
            },
            "BIBREF74": {
                "ref_id": "b74",
                "title": "Are all languages created equal in multilingual BERT?",
                "authors": [
                    {
                        "first": "Shijie",
                        "middle": [],
                        "last": "Wu",
                        "suffix": ""
                    },
                    {
                        "first": "Mark",
                        "middle": [],
                        "last": "Dredze",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the 5th Workshop on Representation Learning for NLP",
                "volume": "",
                "issue": "",
                "pages": "120--130",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/2020.repl4nlp-1.16"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Shijie Wu and Mark Dredze. 2020. Are all languages created equal in multilingual BERT? In Proceedings of the 5th Workshop on Representation Learning for NLP, pages 120-130, Online. Association for Com- putational Linguistics.",
                "links": null
            },
            "BIBREF75": {
                "ref_id": "b75",
                "title": "Aditya Barua, and Colin Raffel. 2021. mT5: A massively multilingual pre-trained text-to-text transformer",
                "authors": [
                    {
                        "first": "Linting",
                        "middle": [],
                        "last": "Xue",
                        "suffix": ""
                    },
                    {
                        "first": "Noah",
                        "middle": [],
                        "last": "Constant",
                        "suffix": ""
                    },
                    {
                        "first": "Adam",
                        "middle": [],
                        "last": "Roberts",
                        "suffix": ""
                    },
                    {
                        "first": "Mihir",
                        "middle": [],
                        "last": "Kale",
                        "suffix": ""
                    },
                    {
                        "first": "Rami",
                        "middle": [],
                        "last": "Al-Rfou",
                        "suffix": ""
                    },
                    {
                        "first": "Aditya",
                        "middle": [],
                        "last": "Siddhant",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
                "volume": "",
                "issue": "",
                "pages": "483--498",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/2021.naacl-main.41"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, and Colin Raffel. 2021. mT5: A massively multilingual pre-trained text-to-text transformer. In Proceedings of the 2021 Conference of the North American Chap- ter of the Association for Computational Linguistics: Human Language Technologies, pages 483-498, On- line. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF76": {
                "ref_id": "b76",
                "title": "Multilingual universal sentence encoder for semantic retrieval",
                "authors": [
                    {
                        "first": "Yinfei",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    },
                    {
                        "first": "Daniel",
                        "middle": [],
                        "last": "Cer",
                        "suffix": ""
                    },
                    {
                        "first": "Amin",
                        "middle": [],
                        "last": "Ahmad",
                        "suffix": ""
                    },
                    {
                        "first": "Mandy",
                        "middle": [],
                        "last": "Guo",
                        "suffix": ""
                    },
                    {
                        "first": "Jax",
                        "middle": [],
                        "last": "Law",
                        "suffix": ""
                    },
                    {
                        "first": "Noah",
                        "middle": [],
                        "last": "Constant",
                        "suffix": ""
                    },
                    {
                        "first": "Gustavo",
                        "middle": [],
                        "last": "Hernandez Abrego",
                        "suffix": ""
                    },
                    {
                        "first": "Steve",
                        "middle": [],
                        "last": "Yuan",
                        "suffix": ""
                    },
                    {
                        "first": "Chris",
                        "middle": [],
                        "last": "Tar",
                        "suffix": ""
                    },
                    {
                        "first": "Yun-Hsuan",
                        "middle": [],
                        "last": "Sung",
                        "suffix": ""
                    },
                    {
                        "first": "Brian",
                        "middle": [],
                        "last": "Strope",
                        "suffix": ""
                    },
                    {
                        "first": "Ray",
                        "middle": [],
                        "last": "Kurzweil",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: System Demonstrations",
                "volume": "",
                "issue": "",
                "pages": "87--94",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/2020.acl-demos.12"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Yinfei Yang, Daniel Cer, Amin Ahmad, Mandy Guo, Jax Law, Noah Constant, Gustavo Hernandez Abrego, Steve Yuan, Chris Tar, Yun-hsuan Sung, Brian Strope, and Ray Kurzweil. 2020. Multilingual universal sen- tence encoder for semantic retrieval. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: System Demonstrations, pages 87-94, Online. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF77": {
                "ref_id": "b77",
                "title": "Improving multilingual sentence embedding using bidirectional dual encoder with additive margin softmax",
                "authors": [
                    {
                        "first": "Yinfei",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    },
                    {
                        "first": "Gustavo",
                        "middle": [],
                        "last": "Hernandez Abrego",
                        "suffix": ""
                    },
                    {
                        "first": "Steve",
                        "middle": [],
                        "last": "Yuan",
                        "suffix": ""
                    },
                    {
                        "first": "Mandy",
                        "middle": [],
                        "last": "Guo",
                        "suffix": ""
                    },
                    {
                        "first": "Qinlan",
                        "middle": [],
                        "last": "Shen",
                        "suffix": ""
                    },
                    {
                        "first": "Daniel",
                        "middle": [],
                        "last": "Cer",
                        "suffix": ""
                    },
                    {
                        "first": "Yun-Hsuan",
                        "middle": [],
                        "last": "Sung",
                        "suffix": ""
                    },
                    {
                        "first": "Brian",
                        "middle": [],
                        "last": "Strope",
                        "suffix": ""
                    },
                    {
                        "first": "Ray",
                        "middle": [],
                        "last": "Kurzweil",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI-19",
                "volume": "",
                "issue": "",
                "pages": "5370--5378",
                "other_ids": {
                    "DOI": [
                        "10.24963/ijcai.2019/746"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Yinfei Yang, Gustavo Hernandez Abrego, Steve Yuan, Mandy Guo, Qinlan Shen, Daniel Cer, Yun-hsuan Sung, Brian Strope, and Ray Kurzweil. 2019. Im- proving multilingual sentence embedding using bi- directional dual encoder with additive margin soft- max. In Proceedings of the Twenty-Eighth Inter- national Joint Conference on Artificial Intelligence, IJCAI-19, pages 5370-5378. International Joint Con- ferences on Artificial Intelligence Organization.",
                "links": null
            },
            "BIBREF78": {
                "ref_id": "b78",
                "title": "The United Nations parallel corpus v1.0",
                "authors": [
                    {
                        "first": "Micha\u0142",
                        "middle": [],
                        "last": "Ziemski",
                        "suffix": ""
                    },
                    {
                        "first": "Marcin",
                        "middle": [],
                        "last": "Junczys-Dowmunt",
                        "suffix": ""
                    },
                    {
                        "first": "Bruno",
                        "middle": [],
                        "last": "Pouliquen",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)",
                "volume": "",
                "issue": "",
                "pages": "3530--3534",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Micha\u0142 Ziemski, Marcin Junczys-Dowmunt, and Bruno Pouliquen. 2016. The United Nations parallel cor- pus v1.0. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16), pages 3530-3534, Portoro\u017e, Slovenia. European Language Resources Association (ELRA).",
                "links": null
            },
            "BIBREF79": {
                "ref_id": "b79",
                "title": "Overview of the second BUCC shared task: Spotting parallel sentences in comparable corpora",
                "authors": [
                    {
                        "first": "Pierre",
                        "middle": [],
                        "last": "Zweigenbaum",
                        "suffix": ""
                    },
                    {
                        "first": "Serge",
                        "middle": [],
                        "last": "Sharoff",
                        "suffix": ""
                    },
                    {
                        "first": "Reinhard",
                        "middle": [],
                        "last": "Rapp",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "Proceedings of the 10th Workshop on Building and Using Comparable Corpora",
                "volume": "",
                "issue": "",
                "pages": "60--67",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/W17-2512"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Pierre Zweigenbaum, Serge Sharoff, and Reinhard Rapp. 2017. Overview of the second BUCC shared task: Spotting parallel sentences in comparable corpora. In Proceedings of the 10th Workshop on Building and Using Comparable Corpora, pages 60-67, Vancou- ver, Canada. Association for Computational Linguis- tics.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "uris": null,
                "type_str": "figure",
                "text": "Nemenyi diagrams based on the ranking of models according to the F1-score on each dataset",
                "num": null
            },
            "FIGREF1": {
                "uris": null,
                "type_str": "figure",
                "text": "Detailed results of models' comparison.",
                "num": null
            },
            "FIGREF2": {
                "uris": null,
                "type_str": "figure",
                "text": "Results for models by their size and scenario.",
                "num": null
            },
            "FIGREF3": {
                "uris": null,
                "type_str": "figure",
                "text": "Results of multiple runs of fine-tuning experiments with different seeds.",
                "num": null
            },
            "TABREF0": {
                "html": null,
                "num": null,
                "content": "<table><tr><td>Model</td><td colspan=\"3\">Inf. time [s] #params #langs base a</td><td>data</td><td>reference</td></tr><tr><td>mT5</td><td>1.69</td><td>277M</td><td>101 T5</td><td>CC b</td><td>(Xue et al., 2021)</td></tr><tr><td>LASER</td><td>1.64</td><td>52M</td><td>93 BiLSTM</td><td>OPUS c</td><td>(Artetxe and Schwenk, 2019)</td></tr><tr><td>mBERT</td><td>1.49</td><td>177M</td><td>104 BERT</td><td>Wiki</td><td>(Devlin et al., 2019)</td></tr><tr><td>MPNet**</td><td>1.38</td><td>278M</td><td>53 XLM-R</td><td colspan=\"2\">OPUS c , MUSE d , Wikititles e (Reimers and Gurevych, 2020)</td></tr><tr><td>XLM-R-dist**</td><td>1.37</td><td>278M</td><td>53 XLM-R</td><td colspan=\"2\">OPUS c , MUSE d , Wikititles e (Reimers and Gurevych, 2020)</td></tr><tr><td>XLM-R</td><td>1.37</td><td>278M</td><td>100 XLM-R</td><td>CC</td><td>(Conneau et al., 2020)</td></tr><tr><td>LaBSE</td><td>1.36</td><td>470M</td><td>109 BERT</td><td>CC, Wiki + mined bitexts</td><td>(Feng et al., 2020)</td></tr><tr><td>DistilmBERT</td><td>0.79</td><td>134M</td><td>104 BERT</td><td>Wiki</td><td>(Sanh et al., 2020)</td></tr><tr><td>mUSE-dist**</td><td>0.79</td><td>134M</td><td colspan=\"3\">53 DistilmBERT OPUS c , MUSE d , Wikititles e (Reimers and Gurevych, 2020)</td></tr><tr><td>mUSE-transformer*</td><td>0.65</td><td>85M</td><td>16 transformer</td><td>mined QA + bitexts, SNLI</td><td>(Yang et al., 2020)</td></tr><tr><td>mUSE-cnn*</td><td>0.12</td><td>68M</td><td>16 CNN</td><td>mined QA + bitexts, SNLI</td><td>(Yang et al., 2020)</td></tr><tr><td colspan=\"6\">*mUSE models were used in TensorFlow implementation in contrast to others in torch a Base model is either monolingual version on which it was based or another multilingual model which was used and adopted b Colossal Clean Crawled Corpus in multilingual version (mC4) c multiple datasets from OPUS website (https://opus.nlpl.eu), d bilingual dictionaries from MUSE (https://github.com/facebookresearch/MUSE), e just titles from wiki articles in multiple languages</td></tr></table>",
                "text": "Models used in experiments -inference times, number of parameters, and languages used in pre-training, base model and data used in pre-training",
                "type_str": "table"
            },
            "TABREF1": {
                "html": null,
                "num": null,
                "content": "<table><tr><td/><td>Count</td><td>Category</td><td/><td/><td>Samples</td><td/><td/><td>Mean #</td></tr><tr><td/><td/><td colspan=\"2\">N O R SM</td><td>NEG</td><td>NEU</td><td colspan=\"3\">POS words characters</td></tr><tr><td>English</td><td>17</td><td>3 4 4</td><td>6</td><td colspan=\"3\">305,782 289,847 1,734,857</td><td>42</td><td>233</td></tr><tr><td>Arabic</td><td>9</td><td>0 1 4</td><td>4</td><td colspan=\"2\">139,173 192,463</td><td>600,439</td><td>28</td><td>159</td></tr><tr><td>Spanish</td><td>5</td><td>0 1 3</td><td>2</td><td colspan=\"2\">110,156 120,668</td><td>188,068</td><td>145</td><td>864</td></tr><tr><td>Chinese</td><td>2</td><td>0 0 2</td><td>0</td><td>118,023</td><td>68,953</td><td>144,726</td><td>48</td><td>-</td></tr><tr><td>German</td><td>6</td><td>0 0 1</td><td>5</td><td>105,416</td><td>99,291</td><td>111,180</td><td>19</td><td>131</td></tr><tr><td>Polish</td><td>4</td><td>0 0 2</td><td>2</td><td>78,309</td><td>61,041</td><td>97,338</td><td>39</td><td>245</td></tr><tr><td>French</td><td>3</td><td>0 0 1</td><td>2</td><td>84,324</td><td>43,097</td><td>83,210</td><td>19</td><td>108</td></tr><tr><td>Japanese</td><td>1</td><td>0 0 1</td><td>0</td><td>83,985</td><td>41,976</td><td>83,819</td><td>60</td><td>-</td></tr><tr><td>Czech</td><td>4</td><td>0 0 2</td><td>2</td><td>39,687</td><td>59,181</td><td>97,419</td><td>29</td><td>168</td></tr><tr><td>Portuguese</td><td>4</td><td>0 0 0</td><td>4</td><td>57,737</td><td>54,145</td><td>45,952</td><td>12</td><td>73</td></tr><tr><td>Slovenian</td><td>2</td><td>1 0 0</td><td>1</td><td>34,178</td><td>50,055</td><td>29,310</td><td>161</td><td>1054</td></tr><tr><td>Russian</td><td>2</td><td>0 0 0</td><td>2</td><td>32,018</td><td>47,852</td><td>31,060</td><td>11</td><td>73</td></tr><tr><td>Croatian</td><td>2</td><td>1 0 0</td><td>1</td><td>19,907</td><td>19,298</td><td>38,389</td><td>86</td><td>556</td></tr><tr><td>Serbian</td><td>3</td><td>0 0 2</td><td>1</td><td>25,580</td><td>31,762</td><td>19,026</td><td>176</td><td>1094</td></tr><tr><td>Thai</td><td>2</td><td>0 0 1</td><td>1</td><td>9,327</td><td>28,615</td><td>34,377</td><td>18</td><td>317</td></tr><tr><td>Bulgarian</td><td>1</td><td>0 0 0</td><td>1</td><td>14,040</td><td>28,543</td><td>19,567</td><td>12</td><td>85</td></tr><tr><td>Hungarian</td><td>1</td><td>0 0 0</td><td>1</td><td>9,004</td><td>17,590</td><td>30,088</td><td>11</td><td>83</td></tr><tr><td>Slovak</td><td>1</td><td>0 0 0</td><td>1</td><td>14,518</td><td>12,735</td><td>29,370</td><td>13</td><td>97</td></tr><tr><td>Albanian</td><td>1</td><td>0 0 0</td><td>1</td><td>6,958</td><td>14,675</td><td>22,651</td><td>13</td><td>90</td></tr><tr><td>Swedish</td><td>1</td><td>0 0 0</td><td>1</td><td>16,664</td><td>12,912</td><td>11,770</td><td>14</td><td>94</td></tr><tr><td>Bosnian</td><td>1</td><td>0 0 0</td><td>1</td><td>12,078</td><td>11,039</td><td>13,066</td><td>12</td><td>75</td></tr><tr><td>Urdu</td><td>1</td><td>0 1 0</td><td>0</td><td>5,244</td><td>8,580</td><td>5,836</td><td>13</td><td>69</td></tr><tr><td>Hindi</td><td>1</td><td>0 0 0</td><td>1</td><td>4,992</td><td>6,392</td><td>5,615</td><td>27</td><td>128</td></tr><tr><td>Persian</td><td>1</td><td>0 0 1</td><td>0</td><td>1,619</td><td>5,074</td><td>6,832</td><td>21</td><td>104</td></tr><tr><td>Italian</td><td>2</td><td>0 0 0</td><td>2</td><td>4,043</td><td>4,193</td><td>3,829</td><td>16</td><td>104</td></tr><tr><td>Hebrew</td><td>1</td><td>0 0 0</td><td>1</td><td>2,283</td><td>238</td><td>6,098</td><td>22</td><td>110</td></tr><tr><td>Latvian</td><td>1</td><td>0 0 0</td><td>1</td><td>1,379</td><td>2,617</td><td>1,794</td><td>20</td><td>138</td></tr></table>",
                "text": "Summary of 80 high-quality datasets selected. Categories: N -News, O -Other, R -Reviews, SM -Social Media",
                "type_str": "table"
            },
            "TABREF2": {
                "html": null,
                "num": null,
                "content": "<table><tr><td>pl</td><td colspan=\"2\">2968 14% 60% 26%</td></tr><tr><td>en</td><td>943</td><td>4% 74% 22%</td></tr></table>",
                "text": "Statistics of the internal dataset lang samples NEG NEU POS",
                "type_str": "table"
            },
            "TABREF3": {
                "html": null,
                "num": null,
                "content": "<table><tr><td>XLM-R</td><td>LaBSE</td><td>MPNet</td><td>XLM-R-dist</td><td>mT5</td><td>mBERT</td><td>DistilmBERT</td><td>mUSE-dist</td><td>LASER</td><td>mUSE-trans.</td><td>mUSE-cnn</td></tr><tr><td/><td/><td/><td colspan=\"4\">Just Head -Linear</td><td/><td/><td/><td/></tr><tr><td colspan=\"11\">W 62 62 63 60 59 56 55 59 55 55 54 A 51 54 55 51 49 45 43 50 47 47 45 I 55 61 61 56 50 43 38 60 50 49 50</td></tr><tr><td/><td/><td/><td colspan=\"5\">Just Head -BiLSTM</td><td/><td/><td/></tr><tr><td colspan=\"8\">W 66 62 63 62 65 60 59 62 A 57 55 56 54 56 49 48 54 I 64 63 64 63 63 54 48 64</td><td>---</td><td>---</td><td>---</td></tr><tr><td/><td/><td/><td/><td colspan=\"3\">Fine-tuning</td><td/><td/><td/><td/></tr><tr><td colspan=\"8\">W 68 68 67 67 66 65 64 63 A 61 62 62 62 60 56 56 56 I 70 69 65 67 67 57 58 60</td><td>---</td><td>---</td><td>---</td></tr></table>",
                "text": "Aggregated results of models (F1 score in %). The best results for each test set are highlighted. (Wwhole test, A -avg. by dataset, I -internal)",
                "type_str": "table"
            },
            "TABREF4": {
                "html": null,
                "num": null,
                "content": "<table><tr><td>(Al Omari et al., 2019)</td><td>ar</td><td>R</td><td>No</td><td>3096</td><td>13.0/10.2/76.8</td><td>9</td><td>51</td></tr><tr><td>(Elnagar et al., 2018)</td><td>ar</td><td>R</td><td>No</td><td>400101</td><td>13.0/19.9/67.1</td><td>22</td><td>127</td></tr><tr><td>(Aly and Atiya, 2013)</td><td>ar</td><td>R</td><td>No</td><td>6250</td><td>11.6/17.9/70.5</td><td>65</td><td>343</td></tr><tr><td>(Elnagar and Einea, 2016)</td><td>ar</td><td>R</td><td>No</td><td>504007</td><td>15.4/21.0/63.6</td><td>77</td><td>424</td></tr><tr><td>(Baly et al., 2018)</td><td>ar</td><td colspan=\"2\">SM Yes</td><td>2809</td><td>47.2/23.9/29.0</td><td>22</td><td>130</td></tr></table>",
                "text": "List of all monolingual datasets used in experiments. Category (Cat.): R -Reviews, SM -Social Media, C -Chats, N -News, P -Poems, M -Mixed. HL -human labeled, #Words and #Chars are mean valuesPaperLang Cat. HL Samples NEG/NEU/POS #Words #Char.",
                "type_str": "table"
            },
            "TABREF6": {
                "html": null,
                "num": null,
                "content": "<table><tr><td>Paper</td><td colspan=\"7\">Cat. Lang HL Samples (NEG/NEU/POS) #Words #Char.</td></tr><tr><td>(Narr et al., 2012)</td><td>SM</td><td>de</td><td>Yes</td><td>953</td><td>10.0/75.1/14.9</td><td>12</td><td>80</td></tr><tr><td/><td/><td>de</td><td>Yes</td><td>1781</td><td>16.9/63.3/19.8</td><td>13</td><td>81</td></tr><tr><td/><td/><td>en</td><td>Yes</td><td>7073</td><td>17.4/60.0/22.6</td><td>14</td><td>78</td></tr><tr><td/><td/><td>fr</td><td>Yes</td><td>685</td><td>23.4/53.4/23.2</td><td>14</td><td>82</td></tr><tr><td/><td/><td>fr</td><td>Yes</td><td>1786</td><td>25.0/54.3/20.8</td><td>15</td><td>83</td></tr><tr><td/><td/><td>pt</td><td>Yes</td><td>759</td><td>28.1/33.2/38.7</td><td>14</td><td>78</td></tr><tr><td/><td/><td>pt</td><td>Yes</td><td>1769</td><td>30.7/33.9/35.4</td><td>14</td><td>78</td></tr><tr><td>(Keung et al., 2020)</td><td>R</td><td>de</td><td>No</td><td>209073</td><td>40.1/20.0/39.9</td><td>33</td><td>208</td></tr><tr><td/><td/><td>en</td><td>No</td><td>209393</td><td>40.0/20.0/40.0</td><td>34</td><td>179</td></tr><tr><td/><td/><td>es</td><td>No</td><td>208127</td><td>40.2/20.0/39.8</td><td>27</td><td>152</td></tr><tr><td/><td/><td>fr</td><td>No</td><td>208160</td><td>40.2/20.1/39.7</td><td>28</td><td>160</td></tr><tr><td/><td/><td>ja</td><td>No</td><td>209780</td><td>40.0/20.0/40.0</td><td>2</td><td>101</td></tr><tr><td/><td/><td>zh</td><td>No</td><td>205977</td><td>39.8/20.1/40.1</td><td>1</td><td>50</td></tr><tr><td colspan=\"2\">(Rosenthal et al., 2017) M</td><td>ar</td><td>Yes</td><td>9391</td><td>35.5/40.6/23.9</td><td>14</td><td>105</td></tr><tr><td/><td/><td>en</td><td>Yes</td><td>65071</td><td>19.1/45.7/35.2</td><td>18</td><td>111</td></tr><tr><td>(Patwa et al., 2020)</td><td>SM</td><td>es</td><td>Yes</td><td>14920</td><td>16.8/33.1/50.0</td><td>16</td><td>86</td></tr><tr><td/><td/><td>hi</td><td>Yes</td><td>16999</td><td>29.4/37.6/33.0</td><td>27</td><td>128</td></tr><tr><td>(Mozeti\u010d et al., 2016)</td><td>SM</td><td>bg</td><td>Yes</td><td>62150</td><td>22.6/45.9/31.5</td><td>12</td><td>85</td></tr><tr><td/><td/><td>bs</td><td>Yes</td><td>36183</td><td>33.4/30.5/36.1</td><td>12</td><td>75</td></tr><tr><td/><td/><td>de</td><td>Yes</td><td>90534</td><td>19.7/52.8/27.4</td><td>12</td><td>94</td></tr><tr><td/><td/><td>en</td><td>Yes</td><td>85784</td><td>26.8/44.1/29.1</td><td>12</td><td>77</td></tr><tr><td/><td/><td>es</td><td colspan=\"2\">Yes 191412</td><td>11.8/37.9/50.3</td><td>14</td><td>92</td></tr><tr><td/><td/><td>hr</td><td>Yes</td><td>75569</td><td>25.7/23.9/50.4</td><td>12</td><td>91</td></tr><tr><td/><td/><td>hu</td><td>Yes</td><td>56682</td><td>15.9/31.0/53.1</td><td>11</td><td>83</td></tr><tr><td/><td/><td>pl</td><td colspan=\"2\">Yes 168931</td><td>30.0/26.1/43.9</td><td>11</td><td>82</td></tr><tr><td/><td/><td>pt</td><td colspan=\"2\">Yes 145197</td><td>37.2/35.0/27.8</td><td>10</td><td>61</td></tr><tr><td/><td/><td>ru</td><td>Yes</td><td>87704</td><td>32.0/40.1/27.8</td><td>10</td><td>67</td></tr><tr><td/><td/><td>sk</td><td>Yes</td><td>56623</td><td>25.6/22.5/51.9</td><td>13</td><td>97</td></tr><tr><td/><td/><td>sl</td><td colspan=\"2\">Yes 103126</td><td>29.9/43.3/26.8</td><td>13</td><td>91</td></tr><tr><td/><td/><td>sq</td><td>Yes</td><td>44284</td><td>15.7/33.1/51.1</td><td>13</td><td>90</td></tr><tr><td/><td/><td>sr</td><td>Yes</td><td>67696</td><td>34.8/42.8/22.4</td><td>13</td><td>81</td></tr><tr><td/><td/><td>sv</td><td>Yes</td><td>41346</td><td>40.3/31.2/28.5</td><td>14</td><td>94</td></tr></table>",
                "text": "List of all multilingual datasets used in experiments. Category (Cat.): R -Reviews, SM -Social Media, C -Chats, N -News, P -Poems, M -Mixed. HL -human labeled",
                "type_str": "table"
            }
        }
    }
}