Datasets:

Modalities:
Text
Formats:
parquet
Sub-tasks:
extractive-qa
Libraries:
Datasets
pandas
License:
lenglaender commited on
Commit
a9f3d48
·
verified ·
1 Parent(s): d79054c

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +101 -0
README.md CHANGED
@@ -1,4 +1,18 @@
1
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  dataset_info:
3
  - config_name: m2qa.chinese.creative_writing
4
  features:
@@ -243,3 +257,90 @@ configs:
243
  - split: validation
244
  path: m2qa.turkish.product_reviews/validation-*
245
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: cc-by-nd-4.0
3
+ language:
4
+ - de
5
+ - zh
6
+ - tr
7
+ size_categories:
8
+ - 10K<n<100K
9
+ multilinguality:
10
+ - multilingual
11
+ pretty_name: M2QA
12
+ task_categories:
13
+ - question-answering
14
+ task_ids:
15
+ - extractive-qa
16
  dataset_info:
17
  - config_name: m2qa.chinese.creative_writing
18
  features:
 
257
  - split: validation
258
  path: m2qa.turkish.product_reviews/validation-*
259
  ---
260
+
261
+ M2QA: Multi-domain Multilingual Question Answering
262
+ =====================================================
263
+
264
+ M2QA (Multi-domain Multilingual Question Answering) is an extractive question answering benchmark for evaluating joint language and domain transfer. M2QA includes 13,500 SQuAD 2.0-style question-answer instances in German, Turkish, and Chinese for the domains of product reviews, news, and creative writing.
265
+
266
+ This Hugging Face datasets repo accompanies our paper "[M2QA: Multi-domain Multilingual Question Answering](TODO_INSERT_ARXIV_LINK)". If you want an explanation and code to reproduce all our results or want to use our custom-built annotation platform, have a look at our GitHub repository: [https://github.com/adapter-hub/m2qa](https://github.com/adapter-hub/m2qa)
267
+
268
+
269
+ Loading & Decrypting the Dataset
270
+ -----------------
271
+
272
+ Following [Jacovi et al. (2023)](https://aclanthology.org/2023.emnlp-main.308/), we encrypt the validation data to prevent leakage of the dataset into LLM training datasets. But loading the dataset is still easy:
273
+
274
+ To load the dataset, you can use the following code:
275
+ ```python
276
+ from datasets import load_dataset
277
+ from cryptography.fernet import Fernet
278
+
279
+ # Load the dataset
280
+ subset = "m2qa.german.news" # Change to the subset that you want to use
281
+ dataset = load_dataset("lenglaender/m2qa", subset) # TODO change to new repo name
282
+
283
+ # Decrypt it
284
+ fernet = Fernet(b"aRY0LZZb_rPnXWDSiSJn9krCYezQMOBbGII2eGkN5jo=")
285
+
286
+ def decrypt(example):
287
+ example["question"] = fernet.decrypt(example["question"].encode()).decode()
288
+ example["context"] = fernet.decrypt(example["context"].encode()).decode()
289
+ example["answers"]["text"] = [fernet.decrypt(answer.encode()).decode() for answer in example["answers"]["text"]]
290
+ return example
291
+
292
+ dataset["validation"] = dataset["validation"].map(decrypt)
293
+ ```
294
+
295
+ Overview / Data Splits
296
+ ----------
297
+ All used text passages stem from sources with open licenses. We list the licenses here: [https://github.com/adapter-hub/m2qa/tree/main/m2qa_dataset](https://github.com/adapter-hub/m2qa/tree/main/m2qa_dataset)
298
+
299
+ We have validation data for the following domains and languages:
300
+
301
+ | Subset Name | Domain | Language | #Question-Answer instances |
302
+ | --- | --- | --- | --- |
303
+ | `m2qa.german.product_reviews` | product_reviews | German | 1500 |
304
+ | `m2qa.german.creative_writing` | creative_writing | German | 1500 |
305
+ | `m2qa.german.news` | news | German | 1500 |
306
+ | `m2qa.chinese.product_reviews` | product_reviews | Chinese | 1500 |
307
+ | `m2qa.chinese.creative_writing` | creative_writing | Chinese | 1500 |
308
+ | `m2qa.chinese.news` | news | Chinese | 1500 |
309
+ | `m2qa.turkish.product_reviews` | product_reviews | Turkish | 1500 |
310
+ | `m2qa.turkish.creative_writing` | creative_writing | Turkish | 1500 |
311
+ | `m2qa.turkish.news` | news | Turkish | 1500 |
312
+
313
+ ### Additional Training Data
314
+ We also provide training data for five domain-language pairs, consisting of 1500 question-answer instances each, totalling 7500 training examples. These are the subsets that contain training data:
315
+ - `m2qa.chinese.news`
316
+ - `m2qa.chinese.product_reviews`
317
+ - `m2qa.german.news`
318
+ - `m2qa.german.product_reviews`
319
+ - `m2qa.turkish.news`
320
+
321
+ The training data is not encrypted.
322
+
323
+ Citation
324
+ ----------
325
+
326
+ If you use this dataset, please cite our paper:
327
+ ```
328
+ @article{englaender-etal-2024-m2qa,
329
+ title="M2QA: Multi-domain Multilingual Question Answering",
330
+ author={Engl{\"a}nder, Leon and
331
+ Sterz, Hannah and
332
+ Poth, Clifton and
333
+ Pfeiffer, Jonas and
334
+ Kuznetsov, Ilia and
335
+ Gurevych, Iryna},
336
+ journal={arXiv preprint},
337
+ year="2024"
338
+ }
339
+ ```
340
+
341
+ License
342
+ -------
343
+
344
+ This dataset is distributed under the [CC-BY-ND 4.0 license](https://creativecommons.org/licenses/by-nd/4.0/legalcode).
345
+
346
+ Following [Jacovi et al. (2023)](https://aclanthology.org/2023.emnlp-main.308/), we decided to publish with a "No Derivatives" license to mitigate the risk of data contamination of crawled training datasets.