Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
+
pretty_name: Bar Exam QA
|
| 5 |
+
---
|
| 6 |
+
`Bar Exam QA` is a dataset for legal information retrieval and question answering.
|
| 7 |
+
|
| 8 |
+
This repo contains the public MBE subset of questions (historical exams), which we release with gold passage annotations. In our work, we also evaluate on a private test set of questions from a BarBri multistate bar exam test preparation textbook.
|
| 9 |
+
|
| 10 |
+
The dataset was created by linking multistate bar exam multiple-choice questions with supporting gold passages from primary sources (caselaw) and secondary sources (online resources, textbooks, etc.).
|
| 11 |
+
The gold passages were annotated by a law student, simulating a legal research workflow.
|
| 12 |
+
|
| 13 |
+
## qa
|
| 14 |
+
The dataset can be loaded with the `qa` configuration. Each example consists of a question, multiple-choice answers, an answer label, and a gold passage (passage ID) label.
|
| 15 |
+
The dataset contains 1,195 examples.
|
| 16 |
+
```python
|
| 17 |
+
dataset = load_dataset("reglab/barexam_qa", name="qa")
|
| 18 |
+
```
|
| 19 |
+
where
|
| 20 |
+
- `idx`: unique example ID (string)
|
| 21 |
+
- `prompt`: the question prompt (context before the question, can be shared between questions)
|
| 22 |
+
- `question`: the question
|
| 23 |
+
- `choice_a`: answer choice A text
|
| 24 |
+
- `choice_b`: answer choice B text
|
| 25 |
+
- `choice_c`: answer choice C text
|
| 26 |
+
- `choice_d`: answer choice D text
|
| 27 |
+
- `answer`: A, B, C, or D
|
| 28 |
+
- `gold_passge`: gold passage text
|
| 29 |
+
- `gold_idx`: passage ID of gold passage (matches with passage ID in `passages`)
|
| 30 |
+
|
| 31 |
+
## passages
|
| 32 |
+
The passage pool can be loaded with `passages` configuration. Each passages consists of a passage ID, the passage text, and additional metadata if applicable (e.g, case ID for passages from Caselaw Access Project).
|
| 33 |
+
The passage pool contains ~900K passages.
|
| 34 |
+
```python
|
| 35 |
+
dataset = load_dataset("reglab/barexam_qa", name="passages")
|
| 36 |
+
```
|
| 37 |
+
where
|
| 38 |
+
- `idx`: unique passage ID (string)
|
| 39 |
+
- `faiss_id`: unique Faiss ID (integer)
|
| 40 |
+
- `case_id`: case ID (from Caselaw Access Project)
|
| 41 |
+
- `opinion_id`: opinion ID (from Caselaw Access Project)
|
| 42 |
+
- `absolute_paragraph_id`: paragraph ID in case across all opinions (from Caselaw Access Project)
|
| 43 |
+
- `relative_paragraph_id`: paragraph ID in opinion (from Caselaw Access Project)
|
| 44 |
+
- `text`: passage text
|