lautel commited on
Commit
bf3d9eb
1 Parent(s): 282ad86

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +13 -6
README.md CHANGED
@@ -56,12 +56,14 @@ marking 'supporting evidence' for the label, following how the task is defined b
56
  all the words, in the sentence, they think shows
57
  evidence for their chosen label.
58
 
59
- #### Our annotations
60
  negative 1555 |
61
  positive 1435 |
62
  no sentiment 470
63
  Total 3460
64
 
 
 
65
 
66
  ### SST2
67
 
@@ -81,12 +83,14 @@ sentiment that they do not see.
81
  *The Zuco data contains eye-tracking data for 400 instances from SST. By annotating some of these with rationales,
82
  we add an extra layer of information for future research.
83
 
84
- #### Our annotations
85
  positive 1027 |
86
  negative 900 |
87
  no sentiment 163
88
  Total 2090
89
 
 
 
90
 
91
  ### CoS-E
92
 
@@ -105,9 +109,11 @@ think that removing it will decrease your
105
  confidence toward your chosen label,
106
  please mark it.’
107
 
108
- #### Our annotations
109
  Total 3760
110
 
 
 
111
 
112
  ### Dataset Sources
113
 
@@ -116,15 +122,16 @@ Total 3760
116
  - **Repository:** https://github.com/terne/Being_Right_for_Whose_Right_Reasons
117
  - **Paper:** [Being Right for Whose Right Reasons?](https://aclanthology.org/2023.acl-long.59/)
118
 
119
- ## Uses
120
 
121
  <!-- Address questions around how the dataset is intended to be used. -->
122
  In our paper, we present a collection of three
123
  existing datasets (SST2, DynaSent and Cos-E) with demographics-augmented annotations to enable profiling of models, i.e., quantifying their alignment (or agreement) with rationales provided
124
  by different socio-demographic groups. Such profiling enables us to ask whose right reasons models are being right for and fosters future research on performance equality/robustness.
125
 
126
- For each dataset, we provide the data under a unique **'test'** split, as its original itended used was to test quality & alignment of post-hoc explainability methods.
127
- If you use it following a different split, please clarify it to ease reproducibility of your work.
 
128
 
129
 
130
  ## Dataset Structure
 
56
  all the words, in the sentence, they think shows
57
  evidence for their chosen label.
58
 
59
+ #### >Our annotations:
60
  negative 1555 |
61
  positive 1435 |
62
  no sentiment 470
63
  Total 3460
64
 
65
+ Note that all the data is uploaded under a single 'train' split (read [## Uses](uses) for further details).
66
+
67
 
68
  ### SST2
69
 
 
83
  *The Zuco data contains eye-tracking data for 400 instances from SST. By annotating some of these with rationales,
84
  we add an extra layer of information for future research.
85
 
86
+ #### >Our annotations:
87
  positive 1027 |
88
  negative 900 |
89
  no sentiment 163
90
  Total 2090
91
 
92
+ Note that all the data is uploaded under a single 'train' split (read [## Uses](uses) for further details).
93
+
94
 
95
  ### CoS-E
96
 
 
109
  confidence toward your chosen label,
110
  please mark it.’
111
 
112
+ #### >Our annotations:
113
  Total 3760
114
 
115
+ Note that all the data is uploaded under a single 'train' split (read [## Uses](uses) for further details).
116
+
117
 
118
  ### Dataset Sources
119
 
 
122
  - **Repository:** https://github.com/terne/Being_Right_for_Whose_Right_Reasons
123
  - **Paper:** [Being Right for Whose Right Reasons?](https://aclanthology.org/2023.acl-long.59/)
124
 
125
+ <a id="uses">## Uses</a>
126
 
127
  <!-- Address questions around how the dataset is intended to be used. -->
128
  In our paper, we present a collection of three
129
  existing datasets (SST2, DynaSent and Cos-E) with demographics-augmented annotations to enable profiling of models, i.e., quantifying their alignment (or agreement) with rationales provided
130
  by different socio-demographic groups. Such profiling enables us to ask whose right reasons models are being right for and fosters future research on performance equality/robustness.
131
 
132
+ For each dataset, we provide the data under a unique **'train'** split due to the current limitation of not being possible to upload a dataset with a single *'test'* split.
133
+ Note, however, that the original itended used of these collection of datasets was to **test** the quality & alignment of post-hoc explainability methods.
134
+ If you use it following different splits, please clarify it to ease reproducibility of your work.
135
 
136
 
137
  ## Dataset Structure