Datasets:
Gholamreza
commited on
Commit
•
a621773
1
Parent(s):
9b5ace6
Update README.md
Browse files
README.md
CHANGED
@@ -108,7 +108,7 @@ PQuAD is stored in the JSON format and consists of passages where each passage i
|
|
108 |
set of questions. Answer(s) of the questions is specified with answer's span (start and end
|
109 |
point of answer in paragraph). Also, the unanswerable questions are marked as unanswerable.
|
110 |
|
111 |
-
##Results
|
112 |
The estimated human performance on the test set is 88.3% for F1 and 80.3% for EM. We have
|
113 |
evaluated PQuAD using two pre-trained transformer-based language models, namely ParsBERT
|
114 |
(Farahani et al., 2021) and XLM-RoBERTa (Conneau et al., 2020), as well as BiDAF (Levy et
|
@@ -125,7 +125,7 @@ al., 2017) which is an attention-based model proposed for MRC.
|
|
125 |
+-------------+------+------+-----------+-----------+-------------+
|
126 |
```
|
127 |
|
128 |
-
##LICENSE
|
129 |
PQuAD is developed by Mabna Intelligent Computing at Amirkabir Science and Technology Park with
|
130 |
collaboration of the NLP lab of the Amirkabir University of Technology and is supported by the
|
131 |
Vice Presidency for Scientific and Technology. By releasing this dataset, we aim to ease research
|
|
|
108 |
set of questions. Answer(s) of the questions is specified with answer's span (start and end
|
109 |
point of answer in paragraph). Also, the unanswerable questions are marked as unanswerable.
|
110 |
|
111 |
+
## Results
|
112 |
The estimated human performance on the test set is 88.3% for F1 and 80.3% for EM. We have
|
113 |
evaluated PQuAD using two pre-trained transformer-based language models, namely ParsBERT
|
114 |
(Farahani et al., 2021) and XLM-RoBERTa (Conneau et al., 2020), as well as BiDAF (Levy et
|
|
|
125 |
+-------------+------+------+-----------+-----------+-------------+
|
126 |
```
|
127 |
|
128 |
+
## LICENSE
|
129 |
PQuAD is developed by Mabna Intelligent Computing at Amirkabir Science and Technology Park with
|
130 |
collaboration of the NLP lab of the Amirkabir University of Technology and is supported by the
|
131 |
Vice Presidency for Scientific and Technology. By releasing this dataset, we aim to ease research
|