Kanishka Misra
commited on
Commit
·
95e4d2d
1
Parent(s):
ff68fa6
fixed readme
Browse files
README.md
CHANGED
@@ -17,4 +17,26 @@ source_datasets:
|
|
17 |
|
18 |
# Dataset Card for "COMPS"
|
19 |
|
20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
|
18 |
# Dataset Card for "COMPS"
|
19 |
|
20 |
+
## Dataset Description
|
21 |
+
|
22 |
+
COMPS is a dataset of minimal pair sentences in English that enables the
|
23 |
+
testing knowledge of concepts and their properties in language models (LMs).
|
24 |
+
Specifically, it tests the ability of LMs to attribute properties to everyday
|
25 |
+
concepts, and demonstrate reasoning compatible with property inheritance, where
|
26 |
+
subordinate concepts inherit the properties of their superordinate (hypernyms).
|
27 |
+
|
28 |
+
- **Homepage:** [https://github.com/kanishkamisra/comps/](https://github.com/kanishkamisra/comps/)
|
29 |
+
- **Repository:** [https://github.com/kanishkamisra/comps/](https://github.com/kanishkamisra/comps/)
|
30 |
+
- **Paper:** [arxiv](https://arxiv.org/abs/2210.01963)
|
31 |
+
- **Point of Contact:** [Kanishka Misra] (https://kanishka.website)
|
32 |
+
|
33 |
+
### Citation Information
|
34 |
+
|
35 |
+
```
|
36 |
+
@article{misra2022comps,
|
37 |
+
title={COMPS: Conceptual Minimal Pair Sentences for testing Property Knowledge and Inheritance in Pre-trained Language Models},
|
38 |
+
author={Misra, Kanishka and Rayz, Julia Taylor and Ettinger, Allyson},
|
39 |
+
journal={arXiv preprint arXiv:2210.01963},
|
40 |
+
year={2022}
|
41 |
+
}
|
42 |
+
```
|
comps.py
CHANGED
@@ -32,9 +32,10 @@ _CITATION = """
|
|
32 |
|
33 |
_DESCRIPTION = """
|
34 |
COMPS is a dataset of minimal pair sentences in English that enables the
|
35 |
-
testing
|
36 |
-
|
37 |
-
concepts
|
|
|
38 |
"""
|
39 |
|
40 |
_PROJECT_URL = "https://github.com/kanishkamisra/comps"
|
|
|
32 |
|
33 |
_DESCRIPTION = """
|
34 |
COMPS is a dataset of minimal pair sentences in English that enables the
|
35 |
+
testing knowledge of concepts and their properties in language models (LMs).
|
36 |
+
Specifically, it tests the ability of LMs to attribute properties to everyday
|
37 |
+
concepts, and demonstrate reasoning compatible with property inheritance, where
|
38 |
+
subordinate concepts inherit the properties of their superordinate (hypernyms).
|
39 |
"""
|
40 |
|
41 |
_PROJECT_URL = "https://github.com/kanishkamisra/comps"
|