anno-eval 0.10.0

Evaluation harnesses, datasets, and muxer-backed sampling for anno
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
    // =========================================================================
    // Relation Extraction Datasets
    // =========================================================================
    DocRED {
        name: "DocRED",
        description: "Document-level relation extraction. 96 relation types from Wikipedia.",
        url: "https://raw.githubusercontent.com/mainlp/CrossRE/main/crossre_data/ai-test.json",
        entity_types: ["PER", "LOC", "ORG", "TIME", "NUM"],
        language: "en",
        domain: "wikipedia",
        license: "MIT",
        citation: "Yao et al. (2019)",
        paper_url: "https://aclanthology.org/P19-1074/",
        year: 2019,
        format: "JSONL",
        size_hint: "5,053 docs, 132k entities, 56k relations",
        splits: ["train", "dev", "test"],
        tasks: ["re"],
        expected_docs: 5053,
        hf_id: "docred",
        categories: [relation_extraction],
    },
    ReTACRED {
        name: "Re-TACRED",
        description: "Large-scale relation extraction. 41 relation types + no_relation. Cleaned TACRED.",
        // NOTE: Re-TACRED is distributed as a delta/cleaned relabeling that depends on the
        // original TACRED distribution (LDC). We intentionally do not provide a direct URL
        // here to avoid accidentally pointing at unrelated files.
        url: "",
        entity_types: ["PER", "ORG", "LOC", "DATE", "NUM"],
        language: "en",
        domain: "news",
        license: "LDC",
        citation: "Stoica et al. (2021)",
        paper_url: "https://aclanthology.org/2021.acl-long.359/",
        year: 2021,
        format: "JSONL",
        size_hint: "~106k relations",
        notes: "Cleaned version of TACRED with relabeling; requires original TACRED",
        access_status: DependsOnOther,
        categories: [relation_extraction],
    },

    // =========================================================================
    // Nested/Discontinuous NER Datasets
    // =========================================================================
    ACE2004 {
        name: "ACE 2004",
        description: "Nested entity recognition benchmark. Influential early corpus for nested NER research.",
        url: "",  // Requires LDC license
        entity_types: ["PER", "ORG", "GPE", "LOC", "FAC", "WEA", "VEH"],
        language: "en",
        domain: "news",
        license: "LDC",
        citation: "Doddington et al. (2004)",
        paper_url: "https://aclanthology.org/L04-1011/",
        year: 2004,
        format: "XML",
        annotation_scheme: "Standoff",
        notes: "Requires LDC license; includes entity relations; includes some nested entities",
        access_status: Registration,
        categories: [ner, nested_ner],
    },
    CADEC {
        name: "CADEC",
        description: "Clinical Adverse Drug Events. Benchmark for discontinuous NER from AskaPatient.",
        url: "https://huggingface.co/datasets/KevinSpaghetti/cadec",
        entity_types: ["ADR", "Drug", "Disease", "Symptom"],
        language: "en",
        domain: "biomedical",
        license: "Research",
        citation: "Karimi et al. (2015)",
        paper_url: "https://pubmed.ncbi.nlm.nih.gov/25817970/",
        year: 2015,
        format: "BRAT",
        annotation_scheme: "Standoff",
        size_hint: "~1,250 posts",
        example: "'severe [pain]...in my [legs]' -> ADR spans [0:10, 20:24] (discontinuous)",
        notes: "Discontinuous spans common; requires special handling; patient-written text",
        // Note: HF export endpoints currently fail (rows=422) and repo stores parquet shards.
        // Keep as HF reference but treat as non-automatable until we add parquet support or a stable jsonl mirror.
        splits: ["train", "dev", "test"],
        tasks: ["ner", "discontinuous_ner"],
        hf_id: "KevinSpaghetti/cadec",
        categories: [ner, biomedical, discontinuous_ner],
    },

    // =========================================================================
    // Bias Evaluation Datasets
    // =========================================================================
    WinoQueer {
        name: "WinoQueer",
        description: "Anti-LGBTQ+ bias benchmark. Community-in-the-loop design for queer representation.",
        url: "https://raw.githubusercontent.com/katyfelkner/winoqueer/main/data/winoqueer_final.csv",
        entity_types: ["PER"],
        language: "en",
        domain: "evaluation",
        license: "MIT",
        citation: "Felkner et al. (2023)",
        paper_url: "https://aclanthology.org/2023.acl-long.507/",
        year: 2023,
        format: "CSV",
        size_hint: "45,540 sentence pairs, ~4.8MB",
        notes: "Community-designed; tests queer stereotypes in LLMs; Winograd-schema style",
        splits: ["all"],
        tasks: ["bias_evaluation"],
        categories: [bias_evaluation],
    },
    BBQ {
        name: "BBQ",
        description: "Bias Benchmark for QA. Tests 9 social bias categories including sexual orientation.",
        url: "https://raw.githubusercontent.com/nyu-mll/BBQ/main/data/Gender_identity.jsonl",
        entity_types: [],
        language: "en",
        domain: "evaluation",
        license: "CC-BY-4.0",
        citation: "Parrish et al. (2022)",
        paper_url: "https://aclanthology.org/2022.findings-acl.165/",
        year: 2022,
        format: "JSONL",
        size_hint: "~58k QA pairs across 11 categories",
        notes: "Hand-built ambiguous contexts; age, disability, nationality, religion, etc.",
        splits: ["all"],
        tasks: ["bias_evaluation", "qa"],
        categories: [bias_evaluation],
    },
    GICoref {
        name: "GICoref",
        description: "Gender-inclusive coreference. Written by/about trans and non-binary individuals.",
        url: "https://raw.githubusercontent.com/TristaCao/into_inclusivecoref/master/GICoref/coref.combo.conll",
        entity_types: ["PER"],
        language: "en",
        domain: "evaluation",
        license: "CC-BY-4.0",
        citation: "Cao & Daume III (2020)",
        paper_url: "https://aclanthology.org/2020.acl-main.418/",
        year: 2020,
        format: "CoNLL",
        annotation_scheme: "CoNLLCoref",
        size_hint: "95 docs, 470KB",
        notes: "Includes neopronouns (ze/hir, xe/xem); singular they; first gender-inclusive coref corpus",
        splits: ["all"],
        tasks: ["coref"],
        categories: [coref, bias_evaluation],
    },

    // =========================================================================
    // Multilingual Coreference
    // =========================================================================
    CorefUD {
        name: "CorefUD",
        description: "Multilingual coreference (17 languages, 22 datasets). CRAC shared task standard.",
        url: "http://hdl.handle.net/11234/1-5987",
        entity_types: ["PER", "LOC", "ORG"],
        language: "mul",
        domain: "general",
        license: "CC-BY-NC-SA-4.0",
        citation: "Nedoluzhko et al. (2022)",
        paper_url: "https://aclanthology.org/2022.lrec-1.581/",
        year: 2022,
        format: "CoNLLU",
        annotation_scheme: "CoNLLCoref",
        notes: "CRAC shared task standard; includes zero anaphora; harmonized across treebanks",
        categories: [coref, multilingual],
    },
    TransMuCoRes {
        name: "TransMuCoRes",
        description: "Coreference in 31 South Asian languages. Silver annotations via NLLB-200 translation.",
        url: "",  // HuggingFace gated
        entity_types: ["PER", "LOC", "ORG"],
        language: "mul",
        domain: "general",
        license: "Research",
        citation: "Verma et al. (2024)",
        paper_url: "https://arxiv.org/abs/2402.13571",
        year: 2024,
        format: "JSONL",
        notes: "Silver annotations via translation; fine-tuned mBERT models available",
        access_status: ContactAuthors,
        categories: [coref, multilingual],
    },
    MGAP {
        name: "mGAP",
        description: "Multilingual Gender-Ambiguous Pronouns. 27 South Asian languages.",
        url: "",  // Research access required
        entity_types: ["PER"],
        language: "mul",
        domain: "evaluation",
        license: "Research",
        citation: "Verma et al. (2025)",
        paper_url: "https://aclanthology.org/2025.chipsal-1.10/",
        year: 2025,
        format: "TSV",
        size_hint: "8,908 pronoun-name pairs",
        notes: "Cross-attention improves results; extension of GAP to South Asian languages",
        access_status: ContactAuthors,
        categories: [coref, multilingual, bias_evaluation],
    },
    CrowSPairs {
        name: "CrowS-Pairs",
        description: "Crowdsourced stereotype pairs benchmark. 9 bias categories for language models.",
        url: "https://github.com/nyu-mll/crows-pairs",
        entity_types: [],
        language: "en",
        domain: "evaluation",
        license: "CC-BY-SA-4.0",
        citation: "Nangia et al. (2020)",
        paper_url: "https://aclanthology.org/2020.emnlp-main.154/",
        year: 2020,
        format: "CSV",
        size_hint: "~1.5k sentence pairs",
        notes: "Tests stereotypical associations; gender, race, religion, age, nationality, etc.",
        splits: ["all"],
        tasks: ["bias_evaluation"],
        categories: [bias_evaluation],
    },
    StereoSet {
        name: "StereoSet",
        description: "Measuring stereotypical bias in language models. 4 target domains.",
        url: "https://github.com/moinnadeem/StereoSet",
        entity_types: [],
        language: "en",
        domain: "evaluation",
        license: "MIT",
        citation: "Nadeem et al. (2020)",
        paper_url: "https://aclanthology.org/2021.acl-long.416/",
        year: 2020,
        format: "JSONL",
        size_hint: "~17k instances",
        notes: "Intrasentence and intersentence evaluation; gender, profession, race, religion",
        splits: ["all"],
        tasks: ["bias_evaluation"],
        categories: [bias_evaluation],
    },
    RealToxicityPrompts {
        name: "RealToxicityPrompts",
        description: "100k prompts for measuring toxicity generation in language models.",
        url: "https://huggingface.co/datasets/allenai/real-toxicity-prompts",
        entity_types: [],
        language: "en",
        domain: "evaluation",
        license: "Apache-2.0",
        citation: "Gehman et al. (2020)",
        paper_url: "https://aclanthology.org/2020.findings-emnlp.301/",
        year: 2020,
        format: "JSONL",
        size_hint: "~100k prompts",
        notes: "Tests toxicity generation; perspectives API scores; diverse prompt styles",
        splits: ["all"],
        tasks: ["bias_evaluation"],
        categories: [bias_evaluation, adversarial],
    },
    BoldBias {
        name: "BOLD",
        description: "Bias in Open-ended Language Generation Dataset. Wikipedia-based prompts.",
        url: "https://github.com/amazon-science/bold",
        entity_types: [],
        language: "en",
        domain: "evaluation",
        license: "CC-BY-4.0",
        citation: "Dhamala et al. (2021)",
        paper_url: "https://aclanthology.org/2021.findings-acl.311/",
        year: 2021,
        format: "JSONL",
        size_hint: "~23k prompts",
        notes: "Tests generation bias; profession, gender, race, religion, political ideology",
        splits: ["all"],
        tasks: ["bias_evaluation"],
        categories: [bias_evaluation],
    },

    // =========================================================================
    // Literary Coreference
    // =========================================================================
    DROC {
        name: "DROC",
        description: "German novel coreference. 90 German novels from DTA (Deutsches Textarchiv).",
        url: "",
        entity_types: ["PER"],
        language: "de",
        domain: "literature",
        license: "CC-BY-4.0",
        citation: "Krug et al. (2018)",
        paper_url: "https://aclanthology.org/L18-1045/",
        year: 2018,
        format: "JSONL",
        notes: "First public German literary coreference dataset. Prior URL is dead; needs a fresh mirror.",
        splits: ["all"],
        tasks: ["coref"],
        access_status: Deprecated,
        categories: [coref, literary],
    },
    FantasyCoref {
        name: "FantasyCoref",
        description: "Fantasy fiction coreference. Handles entity transformations. 211 Grimms' stories.",
        url: "https://github.com/emorynlp/FantasyCoref/archive/refs/heads/master.zip",
        entity_types: ["PER", "LOC", "ORG"],
        language: "en",
        domain: "literature",
        license: "Research",
        citation: "Han et al. (2021)",
        paper_url: "https://aclanthology.org/2021.emnlp-main.672/",
        year: 2021,
        format: "JSONL",
        notes: "Shape-shifting, possession, disguise - unique challenges; data on GitHub",
        tasks: ["coref"],
        access_status: Public,
        categories: [coref, literary],
    },

    // =========================================================================
    // Book-Scale / Long-Document Coreference
    // =========================================================================

    BookCoref {
        name: "BOOKCOREF",
        description: "Book-scale coreference. First benchmark with 200k+ tokens/doc average. Character coreference on 53 Project Gutenberg novels.",
        url: "https://huggingface.co/datasets/sapienzanlp/bookcoref",
        entity_types: ["PER"],
        language: "en",
        domain: "literature",
        license: "CC-BY-NC-SA-4.0",
        citation: "Martinelli et al. (2025)",
        paper_url: "https://aclanthology.org/2025.acl-long.1197/",
        year: 2025,
        format: "JSONL",
        annotation_scheme: "CoNLLCoref",
        size_hint: "53 books, ~10.8M tokens silver, 229k tokens gold",
        example: "doc_key: pride_and_prejudice_1342\nsentences: [[CHAPTER, I.], [It, is, a, truth, ...]]\nclusters: [[[79,80], [81,82], ...], ...]\ncharacters: [{name: Mr Bennet, cluster: [[79,80]]}]",
        notes: "Gold test set: 3 books (Animal Farm, Siddhartha, Pride & Prejudice). Silver train: 45 books. Unprecedented 73k avg mention distance. Requires HF datasets Python lib to download; export to JSONL for anno loader.",
        splits: ["train", "validation", "test"],
        tasks: ["coref"],
        expected_docs: 53,
        hf_id: "sapienzanlp/bookcoref",
        access_status: DependsOnOther,
        categories: [coref, literary, long_document],
    },
    BookCorefSplit {
        name: "BOOKCOREF (Split)",
        description: "BOOKCOREF split into 1500-token windows for comparison with standard benchmarks.",
        url: "https://huggingface.co/datasets/sapienzanlp/bookcoref",
        entity_types: ["PER"],
        language: "en",
        domain: "literature",
        license: "CC-BY-NC-SA-4.0",
        citation: "Martinelli et al. (2025)",
        paper_url: "https://aclanthology.org/2025.acl-long.1197/",
        year: 2025,
        format: "JSONL",
        annotation_scheme: "CoNLLCoref",
        size_hint: "7544 train, 398 val, 152 test windows",
        notes: "Same data as BOOKCOREF but windowed. Enables fair comparison: Maverickxl gets 82.2 CoNLL F1 on split vs 61.0 on full books. Requires HF datasets Python lib to download.",
        splits: ["train", "validation", "test"],
        tasks: ["coref"],
        hf_id: "sapienzanlp/bookcoref",
        hf_config: "split",
        access_status: DependsOnOther,
        categories: [coref, literary],
    },
    LongtoNotes {
        name: "LongtoNotes",
        description: "OntoNotes with merged coreference chains. Manually merges split OntoNotes documents back into full documents.",
        url: "https://docs.google.com/forms/d/e/1FAIpQLScoWkBOgJ1HH_phtvTJ4_hGvQw6f0W6K7kw74sUKCDTG8P2iA/viewform",
        entity_types: ["PER", "ORG", "GPE", "NORP"],
        language: "en",
        domain: "mixed",
        license: "CC-BY-4.0",
        citation: "Shridhar et al. (2023)",
        paper_url: "https://aclanthology.org/2023.findings-eacl.105/",
        year: 2023,
        format: "CoNLL",
        annotation_scheme: "CoNLLCoref",
        size_hint: "2,415 documents, up to 8x longer than OntoNotes",
        notes: "Requires OntoNotes access. Documents up to 8x OntoNotes length, 2x LitBank. Multi-genre (WSJ, broadcast, web).",
        splits: ["train", "dev", "test"],
        tasks: ["coref"],
        categories: [coref, long_document],
    },
    MovieCoref {
        name: "MovieCoref",
        description: "Screenplay coreference. Character coreference in movie screenplays with unique structural challenges.",
        url: "https://aclanthology.org/attachments/2021.findings-acl.176.OptionalSupplementaryMaterial.gz",
        entity_types: ["PER"],
        language: "en",
        domain: "literature",
        license: "Research",
        citation: "Baruah et al. (2021)",
        paper_url: "https://aclanthology.org/2021.findings-acl.176/",
        year: 2021,
        format: "CoNLL",
        annotation_scheme: "CoNLLCoref",
        size_hint: "9 screenplays (~22k tokens/doc avg), 6 full + 3 excerpts",
        notes: "Screenplay structure (scene headings, character names, parentheticals) differs significantly from prose. Focus on character coreference only.",
        splits: ["all"],
        tasks: ["coref"],
        categories: [coref, literary, long_document],
    },
    // =========================================================================
    // Dialogue / Conversational Coreference
    // =========================================================================
    TwiConv {
        name: "TwiConv",
        description: "Twitter conversational coreference.",
        url: "https://raw.githubusercontent.com/berfingit/TwiConv/main/conll_skeleton/001_940791133357199360.branch7._with_boundaries_gold_conll",
        entity_types: ["PER", "LOC", "ORG"],
        language: "en",
        domain: "social_media",
        license: "Research",
        citation: "Aktaş et al. (2020)",
        paper_url: "https://aclanthology.org/2020.lrec-1.835/",
        format: "CoNLL",
        notes: "Turn-taking dynamics; speaker grounding",
        categories: [coref, dialogue, social_media],
    },
    MuDoCo {
        name: "MuDoCo",
        description: "Multi-domain document-level coreference. Dialog-based.",
        url: "https://raw.githubusercontent.com/facebookresearch/mudoco/main/mudoco_calling.json",
        entity_types: ["PER", "LOC", "ORG"],
        language: "en",
        domain: "dialogue",
        license: "MIT",
        citation: "Raghunathan et al. (2020)",
        paper_url: "https://arxiv.org/abs/2005.00816",
        format: "JSON",
        notes: "Multi-domain dialog coreference with speaker tracking",
        categories: [coref, dialogue],
    },
    DialogRE {
        name: "DialogRE",
        description: "Dialogue-based relation extraction. Multi-turn conversations requiring entity tracking across turns.",
        url: "https://github.com/nlpdata/dialogre",
        entity_types: ["PER", "ORG", "LOC"],
        language: "en",
        domain: "dialogue",
        license: "CC-BY-NC-SA-4.0",
        citation: "Yu et al. (2020)",
        paper_url: "https://aclanthology.org/2020.acl-main.444/",
        year: 2020,
        format: "JSONL",
        size_hint: "~1.8k dialogues, 36 relation types",
        notes: "Based on Friends TV show transcripts; requires tracking entities across dialogue turns",
        splits: ["train", "dev", "test"],
        tasks: ["re", "coref"],
        categories: [relation_extraction, dialogue],
    },
    MultiWOZNER {
        name: "MultiWOZ NER",
        description: "Multi-domain task-oriented dialogue with slot/entity annotations. Multi-turn conversations.",
        url: "https://github.com/budzianowski/multiwoz",
        entity_types: ["RESTAURANT", "HOTEL", "ATTRACTION", "TAXI", "TRAIN", "HOSPITAL", "POLICE"],
        language: "en",
        domain: "dialogue",
        license: "Apache-2.0",
        citation: "Budzianowski et al. (2018)",
        paper_url: "https://aclanthology.org/D18-1547/",
        year: 2018,
        format: "JSONL",
        size_hint: "~10k dialogues, 7 domains",
        notes: "Standard benchmark for dialogue state tracking; slot values correspond to entities",
        splits: ["train", "dev", "test"],
        tasks: ["ner", "slot_filling"],
        categories: [ner, dialogue],
    },
    CoQAEntities {
        name: "CoQA",
        description: "Conversational QA across 7 domains: children's stories, literature, mid/high school exams, news, Wikipedia, science, Reddit.",
        url: "https://stanfordnlp.github.io/coqa/",
        entity_types: ["ANSWER_SPAN"],
        language: "en",
        domain: "mixed",
        license: "Research",
        citation: "Reddy et al. (2019)",
        paper_url: "https://aclanthology.org/Q19-1016/",
        year: 2019,
        format: "JSONL",
        size_hint: "~8k conversations, ~127k QA turns",
        notes: "Multi-turn QA; implicit entity tracking across conversation history; 7 diverse domains",
        splits: ["train", "dev", "test"],
        tasks: ["qa", "coref"],
        categories: [coref, dialogue],
    },

    // =========================================================================
    // Cross-Document Event Coreference
    // =========================================================================
    GVC {
        name: "Gun Violence Corpus",
        description: "Cross-document event coreference for gun violence. Tests domain transfer from ECB+.",
        url: "https://github.com/cltl/GunViolenceCorpus",
        entity_types: ["EVENT", "PARTICIPANT", "WEAPON", "LOCATION", "TIME"],
        language: "en",
        domain: "news",
        license: "Research",
        citation: "Vossen et al. (2018)",
        paper_url: "https://aclanthology.org/L18-1182/",
        year: 2018,
        format: "Custom",
        size_hint: "~500 docs, 510 mentions",
        notes: "Domain-specific CDEC; requires participant/temporal reasoning unlike lemma-driven ECB+",
        splits: ["train", "dev", "test"],
        tasks: ["coref", "event_coref", "cdcr"],
        categories: [coref, event_coref],
    },
    FCC {
        name: "Football Coreference Corpus",
        description: "Cross-document event coreference for football matches. Requires temporal reasoning.",
        url: "",  // Research access required
        entity_types: ["EVENT", "PARTICIPANT", "LOC", "TIME"],
        language: "en",
        domain: "sports",
        license: "Research",
        citation: "Bugert et al. (2021)",
        paper_url: "https://direct.mit.edu/coli/article/47/3/575/102774/",
        notes: "Requires temporal reasoning about match dates",
        access_status: ContactAuthors,
        categories: [coref, event_coref],
    },
    ECBPlusMeta {
        name: "ECB+META",
        description: "ECB+ with metaphoric paraphrases. ChatGPT-transformed sentences.",
        url: "",  // Research access required
        entity_types: ["EVENT", "TIME", "LOC", "PARTICIPANT"],
        language: "en",
        domain: "news",
        license: "Research",
        citation: "Pouran Ben Veyseh et al. (2024)",
        paper_url: "https://arxiv.org/abs/2407.11988",
        notes: "Adversarial; existing systems struggle badly",
        access_status: ContactAuthors,
        categories: [coref, event_coref, adversarial],
    },

    // =========================================================================
    // Abstract Anaphora / Discourse
    // =========================================================================
    ARRAU {
        name: "ARRAU 3.0 (v2)",
        description: "Multi-genre anaphoric annotation: identity, bridging, discourse deixis, split antecedents.",
        url: "",  // LDC + free subsets
        entity_types: ["PER", "LOC", "ORG", "EVENT"],
        language: "en",
        domain: "mixed",
        license: "LDC + Research",
        citation: "Poesio et al. (2024)",
        paper_url: "https://aclanthology.org/2024.codi-1.12/",
        year: 2024,
        format: "MMAX2",
        notes: "Most comprehensive anaphora resource; RST/TRAINS/Pear/GENIA subsets; LDC2023T05",
        splits: ["train", "dev", "test"],
        tasks: ["coref", "bridging", "discourse_deixis"],
        access_status: Registration,
        categories: [coref, abstract_anaphora],
    },
    ISNotes {
        name: "ISNotes",
        description: "Unrestricted bridging anaphora on OntoNotes. ~660 bridging pairs.",
        url: "https://github.com/nlpAThits/ISNotes1.0/archive/refs/heads/master.zip",
        entity_types: ["PER", "LOC", "ORG"],
        language: "en",
        domain: "news",
        license: "Research",
        citation: "Hou et al. (2018)",
        paper_url: "https://direct.mit.edu/coli/article/44/2/237/1596/",
        format: "MMAX2",
        notes: "Part-whole set-member bridging; requires OntoNotes for full text",
        tasks: ["coref"],
        access_status: DependsOnOther,
        categories: [coref, abstract_anaphora],
    },
    ShellNouns {
        name: "Shell Nouns (ASN)",
        description: "Anaphoric shell noun resolution. 670 English shell nouns from Schmid taxonomy.",
        url: "",  // Research access
        entity_types: [],
        language: "en",
        domain: "academic",
        license: "Research",
        citation: "Kolhatkar & Hirst (2012)",
        paper_url: "https://aclanthology.org/D12-1036/",
        notes: "Factual, linguistic, mental, modal, eventive categories",
        access_status: ContactAuthors,
        categories: [abstract_anaphora],
    },
    PDTBv3 {
        name: "PDTB 3.0",
        description: "Penn Discourse TreeBank v3. 43 discourse relation types.",
        url: "",  // LDC required
        entity_types: [],
        language: "en",
        domain: "news",
        license: "LDC",
        citation: "Prasad et al. (2019)",
        paper_url: "https://catalog.ldc.upenn.edu/LDC2019T05",
        notes: "Shallow discourse parsing; connective-argument pairs",
        access_status: Registration,
        categories: [abstract_anaphora],
    },
    CODICRACBridging {
        name: "CODI-CRAC Bridging",
        description: "Universal Anaphora bridging annotations. One of the largest bridging datasets.",
        url: "https://github.com/UniversalAnaphora/UA-CODI-CRAC",
        entity_types: ["BRIDGING_REF"],
        language: "en",
        domain: "dialogue",
        license: "CC-BY-4.0",
        citation: "CODI-CRAC (2022)",
        paper_url: "https://aclanthology.org/2024.lrec-main.1484.pdf",
        year: 2022,
        format: "CoNLLUA",
        size_hint: "AMI, LIGHT, PERSUASION subsets",
        notes: "Dialogue-heavy; extensive bridging annotations; Universal Anaphora format",
        splits: ["train", "dev", "test"],
        tasks: ["coref", "bridging"],
        categories: [coref, abstract_anaphora, dialogue],
    },
    AnaphoraAccessibility {
        name: "Anaphora Accessibility",
        description: "Discourse anaphora accessibility evaluation. Tests non-NP antecedents.",
        url: "",
        entity_types: ["DISCOURSE_DEIXIS", "EVENT_ANAPHORA", "CLAUSAL_ANTECEDENT"],
        language: "en",
        domain: "evaluation",
        license: "Research",
        citation: "Accessibility Authors (2025)",
        paper_url: "https://arxiv.org/html/2502.14119v1",
        year: 2025,
        format: "JSONL",
        size_hint: "Controlled evaluation set",
        notes: "2025 evaluation dataset; focuses on discourse-level anaphora understanding; non-nominal antecedents",
        splits: ["test"],
        tasks: ["coref", "discourse_deixis"],
        access_status: NotYetReleased,
        categories: [coref, abstract_anaphora],
    },
    HumanVoiceAgentInteraction {
        name: "Human-Voice Agent Interaction",
        description: "Naturalistic French dialogue from human-voice agent interactions. Response tokens, aside sequences, discourse deixis.",
        url: "",  // Local dataset at testdata/human_voice_agent/
        entity_types: ["RESPONSE_TOKEN", "DISCOURSE_DEIXIS", "PROPOSITIONAL_ANAPHORA"],
        language: "fr",
        domain: "dialogue",
        license: "Research",
        citation: "Rudaz, Broth & Mlynář (2025)",
        paper_url: "https://dl.acm.org/journal/tochi",
        year: 2025,
        format: "JSONL",
        size_hint: "70 turns, 10 discourse deixis examples, 11 response tokens",
        notes: "French dialogue from Pepper robot (2022) and ChatGPT voice mode (2025). Documents 'managed omnirelevance of speech' - how VAD-based agents misinterpret response tokens and asides. Local dataset at testdata/human_voice_agent/",
        splits: ["all"],
        tasks: ["coref", "abstract_anaphora"],
        access_status: Local,
        categories: [abstract_anaphora, dialogue],
    },
    DiscoBench {
        name: "Disco-Bench",
        description: "Discourse-aware evaluation benchmark for language modeling. 9 document-level testsets covering cohesion, coherence across Chinese/English literature.",
        url: "",
        entity_types: [],
        language: "zh-en",  // Chinese-English bilingual
        domain: "literature",
        license: "CC-BY-4.0",
        citation: "Wang et al. (2023)",
        paper_url: "https://arxiv.org/abs/2307.08074",
        year: 2023,
        format: "Custom",
        size_hint: "9 document-level testsets + diagnostic suite",
        notes: "Tests intra-sentence discourse properties: cohesion, coherence, entity tracking. Evaluates LLMs on discourse phenomena that cross sentences.",
        splits: ["test"],
        tasks: ["discourse_coherence"],
        access_status: NotYetReleased,
        categories: [abstract_anaphora, multilingual],
    },
    DiscoTrack {
        name: "DiscoTrack",
        description: "Multilingual LLM benchmark for discourse tracking. 12 languages, 4 levels: salience, entity tracking, discourse relations, bridging.",
        url: "",
        entity_types: ["SALIENT_ENTITY", "TRACKED_ENTITY", "BRIDGING_REF"],
        language: "mul",
        domain: "general",
        license: "CC-BY-4.0",
        citation: "Bu, Levine & Zeldes (2025)",
        paper_url: "https://arxiv.org/abs/2510.17013",
        year: 2025,
        format: "Custom",
        size_hint: "12 languages, 4 task levels",
        notes: "Tests implicit information and pragmatic inference across documents. Four levels: salience recognition, entity tracking, discourse relations, bridging inference. State-of-the-art models still struggle.",
        splits: ["test"],
        tasks: ["coref", "bridging", "discourse_deixis"],
        access_status: NotYetReleased,
        categories: [coref, abstract_anaphora, multilingual],
    },
    LIEDER {
        name: "LIEDER",
        description: "Linguistically-Informed Evaluation for Discourse Entity Recognition. Tests existence, uniqueness, plurality, novelty.",
        url: "",
        entity_types: ["DISCOURSE_ENTITY"],
        language: "en",
        domain: "evaluation",
        license: "Research",
        citation: "Zhu & Frank (2024)",
        paper_url: "https://arxiv.org/abs/2403.06301",
        year: 2024,
        format: "Custom",
        size_hint: "Controlled evaluation set",
        notes: "Tests LLM knowledge of semantic properties governing discourse entity introduction/reference: existence, uniqueness, plurality, novelty. Models show sensitivity to all except novelty.",
        splits: ["test"],
        tasks: ["discourse_deixis"],
        access_status: ContactAuthors,
        categories: [abstract_anaphora],
    },
    GCDC {
        name: "GCDC",
        description: "Grammarly Corpus of Discourse Coherence. Real-world texts with coherence ratings across 4 domains.",
        url: "",
        entity_types: [],
        language: "en",
        domain: "mixed",
        license: "Research",
        citation: "Lai & Tetreault (2018)",
        paper_url: "https://arxiv.org/abs/1805.04993",
        year: 2018,
        format: "TSV",
        size_hint: "~2000 texts, 4 domains",
        notes: "Four domains: Clinton emails, Enron, Yahoo answers, Yelp. First large-scale discourse coherence evaluation.",
        splits: ["train", "dev", "test"],
        tasks: ["discourse_coherence"],
        access_status: DependsOnOther,
        depends_on: "Yahoo L6 corpus (request from Yahoo, then email tetreaul@gmail.com)",
        categories: [abstract_anaphora],
    },
    DISAPERE {
        name: "DISAPERE",
        description: "Discourse Structure in Peer Review Discussions. 20k sentences in 506 review-rebuttal pairs with argumentation annotation.",
        url: "https://github.com/nnkennard/DISAPERE",
        entity_types: ["REVIEW_ARG", "REBUTTAL_ARG", "STANCE"],
        language: "en",
        domain: "scientific",
        license: "CC-BY-4.0",
        citation: "Kennard et al. (2022)",
        paper_url: "https://arxiv.org/abs/2110.08520",
        year: 2022,
        format: "JSONL",
        size_hint: "20k sentences, 506 review-rebuttal pairs",
        notes: "Discourse relations between reviews and rebuttals. Fine-grained rebuttal annotation: context in review, stance toward arguments. Expert-annotated. DOWNLOADED to anno cache.",
        splits: ["train", "dev", "test"],
        tasks: ["discourse_relations"],
        access_status: Public,
        categories: [abstract_anaphora, dialogue],
    },
    // PragmEval: 11 discourse-focused datasets for pragmatics evaluation
    PragmEval {
        name: "PragmEval",
        description: "Pragmatics-centered evaluation framework: 11 datasets covering discourse relations, speech acts, stance, sarcasm, verifiability.",
        url: "https://github.com/sileod/pragmeval",
        entity_types: [],
        language: "en",
        domain: "dialogue",
        license: "Research",
        citation: "Sileo et al. (2022)",
        paper_url: "https://aclanthology.org/2022.lrec-1.255",
        year: 2022,
        format: "TSV",
        size_hint: "20 subsets, ~130k examples total",
        notes: "Compilation of PDTB, STAC, GUM (discourse relations), Emergent (stance), SarcasmV2, SwitchBoard/MRDA (speech acts), Verifiability, Persuasion, Squinky, EmoBank. DOWNLOADED to anno cache.",
        splits: ["train", "dev", "test"],
        tasks: ["discourse_relations", "speech_act_classification"],
        hf_id: "pragmeval",
        access_status: Public,
        categories: [abstract_anaphora, dialogue],
    },
    // DISRPT 2025: Unified cross-framework discourse benchmark
    DISRPT2025 {
        name: "DISRPT 2025",
        description: "Cross-formalism benchmark for discourse segmentation, connective detection, and relation classification. 39 corpora, 16 languages, 6 frameworks.",
        url: "https://github.com/disrpt/sharedtask2025",
        entity_types: ["DISCOURSE_UNIT", "CONNECTIVE", "DISCOURSE_RELATION"],
        language: "mul",
        domain: "general",
        license: "Research",
        citation: "DISRPT Organizers (2025)",
        paper_url: "https://aclanthology.org/2025.disrpt-1.1.pdf",
        year: 2025,
        format: "CoNLL",
        size_hint: "39 corpora, 5.1M tokens, 311k relations",
        notes: "Unified 17-label relation scheme mapping 353 original labels. Includes RST, eRST, SDRT, PDTB, dependency, ISO frameworks. Czech, German, English, Basque, Persian, French, Italian, Dutch, Nigerian Pidgin, Polish, Portuguese, Russian, Spanish, Thai, Turkish, Chinese. EMNLP 2025 shared task.",
        splits: ["train", "dev", "test"],
        tasks: ["discourse_segmentation", "discourse_relations"],
        access_status: Public,
        categories: [abstract_anaphora, multilingual, dialogue],
    },

    // =========================================================================
    // Ancient / Classical Languages
    // =========================================================================
    AncientGreekUD {
        name: "Ancient Greek UD",
        description: "Universal Dependencies for Ancient Greek. Homeric through Byzantine.",
        url: "https://raw.githubusercontent.com/UniversalDependencies/UD_Ancient_Greek-Perseus/master/grc_perseus-ud-test.conllu",
        entity_types: ["PER", "LOC", "ORG"],
        language: "grc",
        domain: "literature",
        license: "CC-BY-NC-SA-3.0",
        citation: "Celano et al. (2016)",
        paper_url: "https://aclanthology.org/L16-1158/",
        year: 2016,
        format: "CoNLLU",
        example: "# text = μῆνιν ἄειδε θεὰ Πηληϊάδεω Ἀχιλῆος\n1\tμῆνιν\tμῆνις\tNOUN\t_\t_\t2\tobj\t_\tO\n5\tἈχιλῆος\tἈχιλλεύς\tPROPN\t_\t_\t4\tnmod\t_\tB-PER",
        notes: "Perseus treebank; spans 1500+ years of Greek; Homeric to Byzantine",
        categories: [ner, ancient],
    },
    LatinUD {
        name: "Latin UD",
        description: "Universal Dependencies for Latin. Classical through Medieval.",
        url: "https://raw.githubusercontent.com/UniversalDependencies/UD_Latin-ITTB/master/la_ittb-ud-test.conllu",
        entity_types: ["PER", "LOC", "ORG"],
        language: "la",
        domain: "literature",
        license: "CC-BY-NC-SA-3.0",
        citation: "Passarotti et al. (2017)",
        paper_url: "https://aclanthology.org/W17-6526/",
        format: "CoNLLU",
        notes: "Index Thomisticus treebank; medieval scholastic",
        categories: [ner, ancient],
    },
    CopticScriptorium {
        name: "Coptic Scriptorium",
        description: "Sahidic Coptic with multi-layer annotation. ~50k tokens.",
        url: "https://data.copticscriptorium.org/",
        entity_types: ["PER", "LOC", "ORG"],
        language: "cop",
        domain: "religious",
        license: "CC-BY-4.0",
        citation: "Zeldes & Schroeder (2016)",
        paper_url: "https://aclanthology.org/L16-1313/",
        format: "CoNLLU",
        notes: "Multi-layer morphology/syntax/entities/coreference; requires ANNIS export",
        access_status: Registration,
        categories: [ner, ancient],
    },
    LT4HALA {
        name: "LT4HALA Hebrew",
        description: "Biblical Hebrew NER and coreference annotation.",
        url: "",  // Research access
        entity_types: ["PER", "LOC", "ORG", "GPE"],
        language: "hbo",
        domain: "religious",
        license: "Research",
        citation: "LREC-COLING 2024 LT4HALA Workshop",
        paper_url: "https://lt4hala2024.github.io/",
        notes: "First systematic biblical Hebrew NER+coref",
        tasks: ["ner", "coref"],
        access_status: ContactAuthors,
        categories: [ner, coref, ancient],
    },
    ORACC {
        name: "ORACC",
        description: "Open Richly Annotated Cuneiform Corpus. Sumerian, Akkadian, Urartian.",
        url: "http://oracc.museum.upenn.edu/",
        entity_types: ["PER", "LOC", "ORG", "DIVINE"],
        language: "akk",
        domain: "historical",
        license: "CC-BY-SA-3.0",
        citation: "ORACC Project",
        paper_url: "http://oracc.museum.upenn.edu/doc/about/index.html",
        format: "JSON",
        notes: "Cuneiform logographic+syllabic polyphony challenges; JSON export via API",
        access_status: Registration,
        categories: [ner, ancient],
    },

    // =========================================================================
    // Low-Resource / African Languages
    // =========================================================================
    MasakhaNER {
        name: "MasakhaNER",
        description: "NER for 10 African languages. PER/LOC/ORG/DATE.",
        url: "https://raw.githubusercontent.com/masakhane-io/masakhane-ner/main/data/yor/test.txt",
        entity_types: ["PER", "LOC", "ORG", "DATE"],
        language: "mul",
        domain: "news",
        license: "CC-BY-4.0",
        citation: "Adelani et al. (2021)",
        paper_url: "https://aclanthology.org/2021.tacl-1.66/",
        year: 2021,
        format: "CoNLL",
        annotation_scheme: "BIO",
        example: "Olúṣẹ́gun B-PER\nObásanjọ́ I-PER\nní O\nìlú O\nAbẹ́òkúta B-LOC\n. O",
        notes: "Critically underrepresented languages; community-driven; tonal diacritics preserved",
        hf_id: "masakhane/masakhaner",
        categories: [ner, multilingual, low_resource, african_language],
    },
    MasakhaNER2 {
        name: "MasakhaNER 2.0",
        description: "Extended MasakhaNER with 20+ African languages.",
        url: "https://huggingface.co/datasets/masakhane/masakhaner2",
        entity_types: ["PER", "LOC", "ORG", "DATE"],
        language: "mul",
        domain: "news",
        license: "CC-BY-NC-4.0",
        citation: "Adelani et al. (2022)",
        paper_url: "https://aclanthology.org/2022.emnlp-main.298/",
        year: 2022,
        format: "CoNLL",
        annotation_scheme: "BIO",
        notes: "Extended to 20 languages; includes tonal languages",
        splits: ["train", "dev", "test"],
        tasks: ["ner"],
        hf_id: "masakhane/masakhaner2",
        categories: [ner, multilingual, low_resource, african_language],
    },
    AfriSenti {
        name: "AfriSenti",
        description: "Sentiment analysis for 14 African languages. 110k+ tweets. SemEval 2023 Task 12.",
        url: "https://huggingface.co/datasets/shmuhammad/AfriSenti-twitter-sentiment",
        entity_types: ["positive", "neutral", "negative"],
        language: "mul",
        domain: "social_media",
        license: "CC-BY-4.0",
        citation: "Muhammad et al. (2023)",
        paper_url: "https://aclanthology.org/2023.emnlp-main.862/",
        year: 2023,
        format: "HuggingFace",
        size_hint: "~111k tweets",
        example: "tweet: ይሄው ነው አይደል የእውቀትሽ ጥግ (Amharic)\nlabel: negative",
        notes: "Amharic, Algerian/Moroccan Arabic, Hausa, Igbo, Kinyarwanda, Oromo, Nigerian Pidgin, Mozambican Portuguese, Swahili, Tigrinya, Xitsonga, Twi, Yoruba",
        splits: ["train", "validation", "test"],
        tasks: ["sentiment"],
        hf_id: "shmuhammad/AfriSenti-twitter-sentiment",
        categories: [multilingual, low_resource, social_media, african_language],
    },
    AfriQA {
        name: "AfriQA",
        description: "Cross-lingual QA for 10 African languages. Wikipedia-based.",
        url: "https://huggingface.co/datasets/masakhane/afriqa",
        entity_types: [],  // QA dataset
        language: "mul",
        domain: "wikipedia",
        license: "CC-BY-4.0",
        citation: "Ogundepo et al. (2023)",
        paper_url: "https://aclanthology.org/2023.findings-emnlp.997/",
        year: 2023,
        format: "JSONL",
        notes: "Cross-lingual retrieval QA; questions in African languages, passages in English/target",
        splits: ["train", "dev", "test"],
        tasks: ["qa"],
        hf_id: "masakhane/afriqa",
        categories: [multilingual, low_resource, african_language],
    },
    MasakhaNEWS {
        name: "MasakhaNEWS",
        description: "News topic classification for 16 African languages.",
        url: "https://huggingface.co/datasets/masakhane/masakhanews",
        entity_types: ["business", "entertainment", "health", "politics", "religion", "sports", "technology"],
        language: "mul",
        domain: "news",
        license: "Apache-2.0",
        citation: "Adelani et al. (2023)",
        paper_url: "https://aclanthology.org/2023.acl-long.574/",
        year: 2023,
        format: "HuggingFace",
        notes: "7 topics: business, entertainment, health, politics, religion, sports, technology",
        splits: ["train", "dev", "test"],
        tasks: ["text_classification"],
        hf_id: "masakhane/masakhanews",
        categories: [multilingual, low_resource, news, african_language],
    },
    // =========================================================================
    // Text Classification (for GLiNER multi-task classification capability)
    // =========================================================================
    // Note: GLiNER multi-task supports zero-shot classification via entity_types as class labels

    AGNews {
        name: "AG News",
        description: "News article topic classification. 4 classes: World, Sports, Business, Sci/Tech.",
        url: "https://huggingface.co/datasets/fancyzhx/ag_news",
        entity_types: ["World", "Sports", "Business", "Sci/Tech"],
        language: "en",
        domain: "news",
        license: "Non-commercial",
        citation: "Zhang et al. (2015)",
        paper_url: "https://arxiv.org/abs/1509.01626",
        year: 2015,
        format: "HuggingFace",
        size_hint: "120k train, 7.6k test",
        notes: "Character-level ConvNet paper; standard text classification benchmark",
        splits: ["train", "test"],
        tasks: ["text_classification"],
        hf_id: "ag_news",
        access_status: Public,
        categories: [ner],  // News articles contain named entities; can be used for transfer
    },

    DBPedia14 {
        name: "DBPedia-14",
        description: "Wikipedia article classification. 14 non-overlapping classes from DBpedia ontology.",
        url: "https://huggingface.co/datasets/fancyzhx/dbpedia_14",
        entity_types: ["Company", "EducationalInstitution", "Artist", "Athlete", "OfficeHolder", "MeanOfTransportation", "Building", "NaturalPlace", "Village", "Animal", "Plant", "Album", "Film", "WrittenWork"],
        language: "en",
        domain: "wikipedia",
        license: "CC-BY-SA-3.0",
        citation: "Zhang et al. (2015)",
        paper_url: "https://arxiv.org/abs/1509.01626",
        year: 2015,
        format: "HuggingFace",
        size_hint: "560k train, 70k test",
        notes: "14 classes: Company, EducationalInstitution, Artist, Athlete, OfficeHolder, MeanOfTransportation, Building, NaturalPlace, Village, Animal, Plant, Album, Film, WrittenWork",
        splits: ["train", "test"],
        tasks: ["text_classification"],
        hf_id: "dbpedia_14",
        access_status: Public,
        categories: [ner, entity_linking],  // Wikipedia entities; relates to entity linking
    },

    YahooAnswers {
        name: "Yahoo Answers Topic",
        description: "Question-answer topic classification. 10 classes covering diverse topics.",
        url: "https://huggingface.co/datasets/community-datasets/yahoo_answers_topics",
        entity_types: ["Society", "Science", "Health", "Education", "Computers", "Sports", "Business", "Entertainment", "Family", "Politics"],
        language: "en",
        domain: "qa",
        license: "Non-commercial",
        citation: "Zhang et al. (2015)",
        paper_url: "https://arxiv.org/abs/1509.01626",
        year: 2015,
        format: "HuggingFace",
        size_hint: "1.4M train, 60k test",
        notes: "10 classes: Society, Science, Health, Education, Computers, Sports, Business, Entertainment, Family, Politics",
        splits: ["train", "test"],
        tasks: ["text_classification"],
        hf_id: "community-datasets/yahoo_answers_topics",
        access_status: Public,
        categories: [dialogue],
    },

    TREC {
        name: "TREC Question Classification",
        description: "Question type classification. 6 coarse classes, 50 fine-grained types.",
        url: "https://huggingface.co/datasets/trec",
        entity_types: ["ABBR", "DESC", "ENTY", "HUM", "LOC", "NUM"],
        language: "en",
        domain: "qa",
        license: "CC-BY-4.0",
        citation: "Li & Roth (2002)",
        paper_url: "https://www.aclweb.org/anthology/C02-1150/",
        year: 2002,
        format: "HuggingFace",
        size_hint: "5.5k train, 500 test",
        notes: "6 coarse: Abbreviation, Entity, Description, Human, Location, Numeric; 50 fine-grained",
        splits: ["train", "test"],
        tasks: ["text_classification"],
        hf_id: "trec",
        access_status: Public,
        categories: [dialogue],
    },

    TweetTopic {
        name: "TweetTopic",
        description: "Multi-label topic classification for tweets. 6 domains, 19 topics.",
        url: "https://huggingface.co/datasets/cardiffnlp/tweet_topic_multi",
        entity_types: ["arts", "business", "daily_life", "pop_culture", "science", "sports"],
        language: "en",
        domain: "social_media",
        license: "CC-BY-4.0",
        citation: "Antypas et al. (2022)",
        paper_url: "https://aclanthology.org/2022.coling-1.299/",
        year: 2022,
        format: "HuggingFace",
        size_hint: "~11k tweets",
        notes: "Multi-label; domains: arts, business, daily_life, pop_culture, science, sports; zero-shot benchmark",
        splits: ["train", "validation", "test"],
        tasks: ["text_classification"],
        hf_id: "cardiffnlp/tweet_topic_multi",
        access_status: Public,
        categories: [social_media],
    },

    MasakhaPOS {
        name: "MasakhaPOS",
        description: "Part-of-speech tagging for 20 African languages.",
        url: "https://github.com/masakhane-io/masakhane-pos",
        entity_types: ["NOUN", "VERB", "ADJ", "ADV", "PRON", "PROPN", "ADP", "AUX", "CCONJ", "DET", "INTJ", "NUM", "PART", "PUNCT", "SCONJ", "SYM", "X"],
        language: "mul",
        domain: "general",
        license: "MIT",
        citation: "Dione et al. (2023)",
        paper_url: "https://aclanthology.org/2023.acl-long.609/",
        year: 2023,
        format: "CoNLL-U",
        annotation_scheme: "IOB2",
        notes: "Universal Dependencies tagset; includes Bambara, Ewe, Mossi, Chichewa",
        splits: ["train", "dev", "test"],
        tasks: ["pos"],
        hf_id: "masakhane/masakhane-pos",
        categories: [multilingual, low_resource, african_language],
    },
    WikiANN {
        name: "WikiANN",
        description: "Silver-standard NER from Wikipedia hyperlinks. 282 languages.",
        url: "https://huggingface.co/datasets/unimelb-nlp/wikiann",
        entity_types: ["PER", "LOC", "ORG"],
        language: "mul",
        domain: "wikipedia",
        license: "CC-BY-SA-4.0",
        citation: "Pan et al. (2017)",
        paper_url: "https://aclanthology.org/P17-1178/",
        example: "tokens: [Berlin, is, the, capital, of, Germany]\nner_tags: [B-LOC, O, O, O, O, B-LOC]",
        notes: "Silver annotations; noisy but massive coverage",
        hf_id: "unimelb-nlp/wikiann",
        hf_config: "en",
        access_status: HuggingFace,
        categories: [ner, multilingual, low_resource],
    },
    NaijaNER {
        name: "NaijaNER",
        description: "Nigerian Pidgin NER corpus.",
        url: "",  // Research access
        entity_types: ["PER", "LOC", "ORG"],
        language: "pcm",
        domain: "social_media",
        license: "Research",
        citation: "Oyewusi et al. (2021)",
        paper_url: "https://arxiv.org/abs/2102.05236",
        notes: "Nigerian Pidgin English; code-mixing common",
        access_status: ContactAuthors,
        categories: [ner, low_resource],
    },

    // =========================================================================
    // Arcane / Specialized Domains
    // =========================================================================
    WIESP2022NER {
        name: "WIESP2022-NER (DEAL)",
        description: "Astrophysics NER from NASA ADS. 31 entity types: facilities, wavelengths, telescopes, archives.",
        url: "https://huggingface.co/datasets/adsabs/WIESP2022-NER",
        entity_types: ["WAVELENGTH", "TELESCOPE", "FACILITY", "MODEL", "ARCHIVE", "DATASET", "MISSION"],
        language: "en",
        domain: "scientific",
        license: "CC-BY-4.0",
        citation: "Grezes et al. (2022)",
        paper_url: "https://aclanthology.org/2022.wiesp-1.9/",
        year: 2022,
        format: "JSONL",
        size_hint: "~3000 annotated abstracts",
        notes: "AACL-IJCNLP 2022 WIESP shared task; NASA ADS astrophysics literature",
        hf_id: "adsabs/WIESP2022-NER",
        categories: [ner, arcane_domain],
    },
    DutchArchaeology {
        name: "Dutch Archaeology NER",
        description: "Archaeological excavation reports from DANS archive. 31k annotations across 6 entity types.",
        url: "https://live.european-language-grid.eu/catalogue/corpus/13410",
        entity_types: ["ARTEFACT", "PERIOD", "MATERIAL", "LOCATION", "SPECIES", "CONTEXT"],
        language: "nl",
        domain: "archaeology",
        license: "CC-BY-4.0",
        citation: "Brandsen et al. (2020)",
        paper_url: "https://aclanthology.org/2020.lrec-1.562/",
        year: 2020,
        format: "CoNLL",
        size_hint: "~31k entity annotations, high IAA (0.95)",
        notes: "Dutch grey literature; basis for ArcheoBERTje model",
        categories: [ner, arcane_domain],
    },
    ENer {
        name: "E-NER (EDGAR-NER)",
        description: "NER for US SEC EDGAR filings. 52 documents, 400k+ tokens with legal entities.",
        url: "https://raw.githubusercontent.com/terenceau1/E-NER-Dataset/main/all.csv",
        entity_types: ["PERSON", "COURT", "BUSINESS", "GOVERNMENT", "LOCATION", "LEGISLATION"],
        language: "en",
        domain: "legal",
        license: "GPL-3.0",
        citation: "Au et al. (2022)",
        paper_url: "https://aclanthology.org/2022.nllp-1.22/",
        year: 2022,
        format: "CSV",
        size_hint: "52 SEC filings, 400k+ tokens",
        notes: "10-K, 8-K, prospectuses; CSV token,tag format with BIO scheme",
        tasks: ["ner"],
        categories: [ner, arcane_domain],
    },
    // =========================================================================
    // Niche Domain Datasets (Food, Cybersecurity, etc.)
    // =========================================================================
    FINER {
        name: "FINER (Food Ingredients NER)",
        description: "Food ingredient NER from AllRecipes. 182k sentences with ingredient phrases in IOB2 format.",
        url: "https://figshare.com/ndownloader/files/36144501",
        entity_types: ["INGREDIENT", "PRODUCT", "QUANTITY", "UNIT", "STATE"],
        language: "en",
        domain: "food",
        license: "CC-BY-4.0",
        citation: "Komariah et al. (2022)",
        paper_url: "https://doi.org/10.6084/m9.figshare.20222361",
        year: 2022,
        format: "CoNLL",
        annotation_scheme: "IOB2",
        size_hint: "~182k sentences, ingredient phrases",
        notes: "Semi-supervised multi-model construction from AllRecipes; RAR archive with CoNLL format",
        splits: ["train", "test"],
        tasks: ["ner", "slot_filling"],
        categories: [ner, arcane_domain],
    },
    AnnoCTR {
        name: "AnnoCTR (Cyber Threat Reports)",
        description: "Cyber threat intelligence NER with MITRE ATT&CK linking. 400 annotated documents from commercial CTI vendors.",
        url: "https://github.com/boschresearch/anno-ctr-lrec-coling-2024/archive/refs/heads/main.zip",
        entity_types: ["ORGANIZATION", "LOCATION", "SECTOR", "TIME", "CODE", "THREAT_ACTOR", "MALWARE", "TOOL", "TACTIC", "TECHNIQUE"],
        language: "en",
        domain: "cybersecurity",
        license: "CC-BY-SA-4.0",
        citation: "Lange et al. (2024)",
        paper_url: "https://arxiv.org/abs/2404.07765",
        year: 2024,
        format: "JSONL",
        annotation_scheme: "BIO",
        size_hint: "400 documents, multi-layer annotation",
        notes: "LREC-COLING 2024; links to Wikipedia and MITRE ATT&CK KB; includes entity linking task",
        splits: ["train", "dev", "test"],
        tasks: ["ner", "entity_linking"],
        categories: [ner, arcane_domain],
    },
    CRAFT {
        name: "CRAFT",
        description: "Colorado Richly Annotated Full-Text. 97 PubMed articles with multi-layer annotation including coreference.",
        url: "https://github.com/UCDenver-ccp/CRAFT/archive/refs/heads/master.zip",
        entity_types: ["GENE", "PROTEIN", "CHEMICAL", "CELL", "ORGANISM"],
        language: "en",
        domain: "biomedical",
        license: "CC-BY-3.0",
        citation: "Bada et al. (2012)",
        paper_url: "https://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-13-161",
        year: 2012,
        format: "XML",
        annotation_scheme: "Standoff",
        size_hint: "97 full-text articles, ~790k tokens",
        notes: "Full-text (not just abstracts); 10 ontologies used for normalization",
        categories: [coref, biomedical, arcane_domain],
    },

    // =========================================================================
    // Adversarial / Robustness Evaluation
    // =========================================================================
    WNUT16 {
        name: "WNUT-16",
        description: "Twitter NER workshop shared task. Focus on rare and emerging entities in noisy social media text.",
        // Stable raw snapshot from aritter/twitter_nlp (has train/dev/test in one folder).
        url: "https://raw.githubusercontent.com/aritter/twitter_nlp/65f3d77134c40d920db8d431c5c6faef1c051c94/data/annotated/wnut16/data/test",
        // Note: dataset uses lowercase hyphenated labels (person, geo-loc, company, facility, product, other)
        entity_types: ["person", "geo-loc", "company", "facility", "product", "other"],
        language: "en",
        domain: "social_media",
        license: "CC-BY-4.0",
        citation: "Strauss et al. (2016)",
        paper_url: "https://aclanthology.org/W16-3919/",
        year: 2016,
        format: "CoNLL",
        annotation_scheme: "BIO",
        size_hint: "3,856 test tweets, 2,394 train",
        notes: "89% unseen test entities; predecessor to WNUT-17; harder than standard benchmarks",
        splits: ["train", "dev", "test"],
        tasks: ["ner"],
        categories: [ner, social_media, adversarial],
    },

    // =========================================================================
    // More Ancient Languages
    // =========================================================================
    SanskritUD {
        name: "Sanskrit UD",
        description: "Universal Dependencies for Vedic and Classical Sanskrit. Includes Vedas and epics.",
        url: "https://raw.githubusercontent.com/UniversalDependencies/UD_Sanskrit-Vedic/master/sa_vedic-ud-test.conllu",
        entity_types: ["PER", "LOC", "ORG"],
        language: "sa",
        domain: "religious",
        license: "CC-BY-SA-4.0",
        citation: "Hellwig et al. (2020)",
        paper_url: "https://aclanthology.org/2020.lrec-1.632/",
        year: 2020,
        format: "CoNLLU",
        notes: "Oldest Indo-European language with extensive NLP resources; Devanagari script",
        categories: [ner, ancient, low_resource],
    },
    OldEnglishUD {
        name: "Old English UD",
        description: "Universal Dependencies for Old English (Anglo-Saxon). York-Toronto-Helsinki corpus.",
        url: "https://raw.githubusercontent.com/UniversalDependencies/UD_Old_English-Cairo/118f11ad906fb15d930825fadce9ef9eccca9347/ang_cairo-ud-test.conllu",
        entity_types: ["PER", "LOC", "ORG"],
        language: "ang",
        domain: "historical",
        license: "CC-BY-SA-4.0",
        citation: "Tischler & Walkden (2019)",
        paper_url: "https://aclanthology.org/W19-4214/",
        year: 2019,
        format: "CoNLLU",
        notes: "Anglo-Saxon; insular script variations; 5th-11th century CE. NOTE: UD repo naming drifts; URL pinned to a specific commit.",
        categories: [ner, ancient, historical, low_resource],
    },
    OldNorseUD {
        name: "Old Norse UD",
        description: "Universal Dependencies for Old Norse/Icelandic Sagas. PROIEL and ISWOC treebanks.",
        url: "https://raw.githubusercontent.com/UniversalDependencies/UD_Icelandic-IcePaHC/ca72a59affd87c4b7b9067ae56efa7a694a7b4c4/is_icepahc-ud-test.conllu",
        entity_types: ["PER", "LOC", "ORG"],
        language: "non",
        domain: "literature",
        license: "CC-BY-SA-4.0",
        citation: "Rögnvaldsson et al. (2012)",
        paper_url: "https://aclanthology.org/L12-1148/",
        year: 2012,
        format: "CoNLLU",
        notes: "IcePaHC treebank (historical Icelandic; Old Norse family). URL pinned to a specific commit.",
        categories: [ner, ancient, literary, low_resource],
    },

    // =========================================================================
    // Code-Switching Datasets
    // =========================================================================
    CALCS2018 {
        name: "CALCS-2018",
        description: "Code-Switching Workshop shared task. English-Spanish Twitter NER with 9 entity types.",
        url: "https://code-switching.github.io/2018/",
        entity_types: ["PER", "LOC", "ORG", "GROUP", "TITLE", "PROD", "EVENT", "TIME", "OTHER"],
        language: "mul",
        domain: "social_media",
        license: "Research",
        citation: "Aguilar et al. (2018)",
        paper_url: "https://aclanthology.org/W18-3219/",
        year: 2018,
        format: "CoNLL",
        annotation_scheme: "BIO",
        notes: "Spanglish; first major code-switching NER shared task",
        categories: [ner, social_media, multilingual, low_resource],
    },
    HinglishNER {
        name: "Hinglish NER",
        description: "Hindi-English code-mixed social media NER. Roman script Hindi mixed with English.",
        url: "https://github.com/murali1996/CodemixedNLP",
        entity_types: ["PER", "LOC", "ORG"],
        language: "mul",
        domain: "social_media",
        license: "CC-BY-4.0",
        citation: "Priyadharshini et al. (2020)",
        paper_url: "https://aclanthology.org/2020.calcs-1.6/",
        year: 2020,
        format: "JSONL",
        annotation_scheme: "BIO",
        notes: "GLUECoS/LinCE benchmark; download via CodemixedNLP toolkit; Romanized Hindi; ~400M speakers use code-switching daily",
        splits: ["train", "dev", "test"],
        tasks: ["ner"],
        categories: [ner, social_media, multilingual, low_resource],
    },

    // =========================================================================
    // Medieval / Historical NER
    // =========================================================================
    MedievalCharterNER {
        name: "Medieval Charter NER",
        description: "Multilingual medieval charter NER. Latin, French, Spanish from major charter collections.",
        url: "https://zenodo.org/records/6463699",
        entity_types: ["PER", "LOC", "ORG", "DATE"],
        language: "mul",
        domain: "historical",
        license: "CC-BY-4.0",
        citation: "Camps et al. (2022)",
        paper_url: "https://aclanthology.org/2022.lrec-1.530/",
        year: 2022,
        format: "CoNLL",
        size_hint: "~100k tokens across 4 charter collections",
        notes: "HOME-ALCAR, CBMA, Diplomata Belgica, CODEA; medieval Latin/vernacular",
        categories: [ner, historical, multilingual, low_resource],
    },
    CBMACharters {
        name: "CBMA Charters",
        description: "Burgundian medieval Latin charters NER. 9th-14th century diplomatic documents.",
        url: "",  // Access via Zenodo/project
        entity_types: ["PER", "LOC", "ORG", "DATE", "TITLE"],
        language: "la",
        domain: "historical",
        license: "Research",
        citation: "Perreaux (2021)",
        paper_url: "https://dhq-static.digitalhumanities.org/pdf/000574.pdf",
        year: 2021,
        format: "CoNLL",
        notes: "Chartae Burgundiae Medii Aevi; medieval Latin; notarial hands",
        access_status: ContactAuthors,
        categories: [ner, ancient, historical, low_resource],
    },

    // =========================================================================
    // Speech NER
    // =========================================================================
    MSNER {
        name: "MSNER",
        description: "Multilingual Spoken NER. Speech-to-NER on VoxPopuli parliamentary speeches.",
        url: "https://rdr.kuleuven.be/dataset.xhtml?persistentId=doi:10.48804/ZTVMIX",
        entity_types: ["PER", "LOC", "ORG", "MISC"],
        language: "mul",
        domain: "speech",
        license: "CC-BY-4.0",
        citation: "Evain et al. (2024)",
        paper_url: "https://aclanthology.org/2024.isa-1.2/",
        year: 2024,
        format: "CoNLL",
        annotation_scheme: "BIO",
        size_hint: "~590h train, 17h gold test",
        notes: "Dutch, French, German, Spanish; ASR transcripts; first multilingual speech NER corpus",
        categories: [ner, speech, multilingual],
    },

    // =========================================================================
    // Robustness / Noise Benchmarks
    // =========================================================================
    NoiseBench {
        name: "NoiseBench",
        description: "Robustness benchmark for NER. 6 real noise types: expert, crowd, LLM, distant/weak supervision.",
        url: "https://raw.githubusercontent.com/elenamer/NoiseBench/main/data/annotations/clean.traindev",
        entity_types: ["PER", "LOC", "ORG", "MISC"],
        language: "en",
        domain: "evaluation",
        license: "MIT",
        citation: "Merhej et al. (2024)",
        paper_url: "https://aclanthology.org/2024.emnlp-main.1011/",
        year: 2024,
        format: "CoNLL",
        size_hint: "CoNLL-03 subset with 7 label variants",
        notes: "Compares simulated vs real label noise; includes German variant; using clean subset",
        tasks: ["ner"],
        access_status: Public,
        categories: [ner, adversarial],
    },
    RockNER {
        name: "RockNER",
        description: "Robustness benchmark for NER. Real-world adversarial examples with boundary ambiguity.",
        url: "https://github.com/INK-USC/RockNER",
        entity_types: ["PER", "LOC", "ORG", "MISC"],
        language: "en",
        domain: "evaluation",
        license: "Apache-2.0",
        citation: "Lin et al. (2021)",
        paper_url: "https://aclanthology.org/2021.acl-long.340/",
        year: 2021,
        format: "CoNLL",
        size_hint: "~1.5k challenging examples",
        notes: "ACL 2021; entity boundary attacks, rare entities, syntactic perturbations; robustness stress test",
        splits: ["test"],
        tasks: ["ner"],
        categories: [ner, adversarial],
    },
    CrossWeigh {
        name: "CrossWeigh",
        description: "Cross-lingual adversarial NER evaluation. Tests multilingual model robustness.",
        url: "https://raw.githubusercontent.com/ZihanWangKi/CrossWeigh/master/data/conllpp_test.txt",
        entity_types: ["PER", "LOC", "ORG", "MISC"],
        language: "en",
        domain: "evaluation",
        license: "MIT",
        citation: "Wang et al. (2019)",
        paper_url: "https://aclanthology.org/D19-1519/",
        year: 2019,
        format: "CoNLL",
        size_hint: "Adversarial cross-lingual test sets; includes CoNLL++ cleaned version",
        notes: "Tests cross-lingual transfer robustness; character/word perturbations; zero-shot evaluation; CoNLL++ fix",
        splits: ["test"],
        tasks: ["ner"],
        access_status: Public,
        categories: [ner, adversarial, multilingual],
    },