jumelet catherinearnett commited on
Commit
de923ef
·
verified ·
1 Parent(s): a7d1be0

Update README.md (#4)

Browse files

- Update README.md (b7f3b6421b4648e7f22b0ace8f86fc9bbdbec4ca)


Co-authored-by: Catherine Arnett <[email protected]>

Files changed (1) hide show
  1. README.md +131 -1
README.md CHANGED
@@ -489,12 +489,142 @@ size_categories:
489
 
490
  # MultiBLiMP
491
 
492
- MultiBLiMP is a massively Multilingual Benchmark for Linguistic Minimal Pairs.
493
 
494
  The paper can be found [here](https://arxiv.org/abs/2504.02768).
495
 
496
  We split the data set by language: each language consists of a single .tsv file. The rows contain many attributes for a particular pair, most important are the `sen` and `wrong_sen` fields, which we use for evaluating the language models.
497
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
498
  # Citation Information
499
 
500
  ```
 
489
 
490
  # MultiBLiMP
491
 
492
+ MultiBLiMP is a massively Multilingual Benchmark for Linguistic Minimal Pairs. The dataset is composed of synthetic pairs generated using [Universal Dependencies](http://universaldependencies.org/) and [UniMorph](https://unimorph.github.io/).
493
 
494
  The paper can be found [here](https://arxiv.org/abs/2504.02768).
495
 
496
  We split the data set by language: each language consists of a single .tsv file. The rows contain many attributes for a particular pair, most important are the `sen` and `wrong_sen` fields, which we use for evaluating the language models.
497
 
498
+ ## Using MultiBLiMP
499
+
500
+ To download the final datasets for a given language, you can usethe example code. Make sure to use the [ISO 639-3](https://en.wikipedia.org/wiki/ISO_639-3) code, e.g. `eng` for English (see table below).
501
+
502
+ ```
503
+ from huggingface_hub import hf_hub_download
504
+
505
+ hf_hub_download(repo_id="jumelet/multiblimp", filename=f"eng/data.tsv", repo_type="dataset", local_dir='hf_cache/')
506
+ ```
507
+
508
+ To run MultiBLiMP on a model, follow the instructions in [this repo](https://github.com/catherinearnett/multiblimp). Example evaluation code:
509
+
510
+ ```
511
+ python scripts/lm_eval/eval_model.py
512
+ --model meta-llama/Meta-Llama-3-8B
513
+ --data_dir hf_cache/eng/
514
+ --src_dir multiblimp
515
+ --results_dir multiblimp_results/
516
+ --cache_dir hf_cache
517
+ ```
518
+
519
+ ## Languages
520
+
521
+ This table contains the languages covered in MultiBLiMP and the number of items for each language.
522
+
523
+ | ISO Code | Language | n |
524
+ |:--------:|:------------------:|:----:|
525
+ | abk | Abkhazian | 40 |
526
+ | aqz | Akuntsu | 14 |
527
+ | sqi | Albanian | 243 |
528
+ | amh | Amharic | 112 |
529
+ | grc | Ancient Greek | 3695 |
530
+ | hbo | Ancient Hebrew | 983 |
531
+ | apu | Apurinã | 28 |
532
+ | hye | Armenian | 1415 |
533
+ | eus | Basque | 273 |
534
+ | bel | Belarusian | 2570 |
535
+ | ben | Bengali | 21 |
536
+ | bho | Bhojpuri | 34 |
537
+ | bor | Borôro | 241 |
538
+ | bre | Breton | 260 |
539
+ | bul | Bulgarian | 2458 |
540
+ | bua | Buriat | 103 |
541
+ | cat | Catalan | 2284 |
542
+ | chu | Church Slavonic | 4166 |
543
+ | xcl | Classical Armenian | 1623 |
544
+ | ces | Czech | 4256 |
545
+ | dan | Danish | 50 |
546
+ | nld | Dutch | 2331 |
547
+ | egy | Egyptian (Ancient) | 22 |
548
+ | eng | English | 770 |
549
+ | myv | Erzya | 464 |
550
+ | est | Estonian | 2575 |
551
+ | fao | Faroese | 232 |
552
+ | fin | Finnish | 2570 |
553
+ | fra | French | 2548 |
554
+ | glg | Galician | 753 |
555
+ | kat | Georgian | 204 |
556
+ | deu | German | 2298 |
557
+ | aln | Gheg Albanian | 677 |
558
+ | got | Gothic | 1579 |
559
+ | guj | Gujarati | 7 |
560
+ | heb | Hebrew | 2330 |
561
+ | azz | H-P Nahuatl | 207 |
562
+ | hin | Hindi | 1447 |
563
+ | hit | Hittite | 50 |
564
+ | hun | Hungarian | 845 |
565
+ | isl | Icelandic | 2801 |
566
+ | gle | Irish | 28 |
567
+ | ita | Italian | 2999 |
568
+ | quc | K'iche' | 131 |
569
+ | xnr | Kangri | 86 |
570
+ | krl | Karelian | 260 |
571
+ | kxh | Karo (Ethiopia) | 120 |
572
+ | kaz | Kazakh | 173 |
573
+ | kir | Kirghiz | 185 |
574
+ | koi | Komi-Permyak | 43 |
575
+ | kpv | Komi-Zyrian | 320 |
576
+ | lat | Latin | 3149 |
577
+ | lav | Latvian | 3032 |
578
+ | lij | Ligurian | 254 |
579
+ | lit | Lithuanian | 1180 |
580
+ | olo | Livvi | 190 |
581
+ | nds | Low German | 1774 |
582
+ | mkd | Macedonian | 39 |
583
+ | mar | Marathi | 460 |
584
+ | frm | Middle French | 294 |
585
+ | ell | Modern Greek | 1096 |
586
+ | mdf | Moksha | 82 |
587
+ | yrl | Nhengatu | 720 |
588
+ | pcm | Nigerian Pidgin | 26 |
589
+ | kmr | Northern Kurdish | 544 |
590
+ | sme | Northern Sami | 2536 |
591
+ | fro | Old French | 1976 |
592
+ | orv | Old Russian | 4615 |
593
+ | ota | Ottoman Turkish | 99 |
594
+ | fas | Persian | 2553 |
595
+ | xpg | Phrygian | 50 |
596
+ | pol | Polish | 3272 |
597
+ | por | Portuguese | 3048 |
598
+ | ron | Romanian | 2056 |
599
+ | rus | Russian | 3832 |
600
+ | san | Sanskrit | 4442 |
601
+ | gla | Scottish Gaelic | 66 |
602
+ | hbs | Serbo-Croatian | 3286 |
603
+ | sms | Skolt Sami | 263 |
604
+ | slk | Slovak | 4145 |
605
+ | slv | Slovenian | 4483 |
606
+ | spa | Spanish | 2541 |
607
+ | arb | Standard Arabic | 1215 |
608
+ | swe | Swedish | 201 |
609
+ | tam | Tamil | 382 |
610
+ | ttc | Tektiteko | 69 |
611
+ | tpn | Tupinambá | 9 |
612
+ | tur | Turkish | 1742 |
613
+ | uig | Uighur | 758 |
614
+ | ukr | Ukrainian | 2744 |
615
+ | hsb | Upper Sorbian | 186 |
616
+ | urd | Urdu | 550 |
617
+ | urb | Urubú-Kaapor | 13 |
618
+ | uzb | Uzbek | 50 |
619
+ | vep | Veps | 187 |
620
+ | wbp | Warlpiri | 12 |
621
+ | cym | Welsh | 1120 |
622
+ | hyw | Western Armenian | 1153 |
623
+ | wol | Wolof | 705 |
624
+ | sah | Yakut | 144 |
625
+ | nhi | Tenango Nahuatl | 38 |
626
+
627
+
628
  # Citation Information
629
 
630
  ```