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
    // =========================================================================
    // Coreference Datasets
    // =========================================================================
    GAP {
        name: "GAP",
        description: "Gender Ambiguous Pronoun resolution. Google's benchmark for exposing gender bias in coreference systems.",
        url: "https://raw.githubusercontent.com/google-research-datasets/gap-coreference/master/gap-test.tsv",
        entity_types: ["PER"],
        language: "en",
        domain: "wikipedia",
        license: "Apache-2.0",
        citation: "Webster et al. (2018)",
        paper_url: "https://aclanthology.org/Q18-1042/",
        year: 2018,
        format: "TSV",
        size_hint: "8,908 pronoun-name pairs",
        example: "ID\tText\tPronoun\tA\tB\tA-coref\ntest-1\tZoe met Alice and she waved.\tshe\tZoe\tAlice\tFALSE",
        notes: "Designed to expose gender bias; Kaggle shared task; balanced male/female",
        tasks: ["coref"],
        hf_id: "google-gap-coreference/gap",
        categories: [coref, bias_evaluation],
    },
    PreCo {
        name: "PreCo",
        description: "Large-scale coreference from PreCo reading comprehension corpus. 10x larger than OntoNotes.",
        url: "https://huggingface.co/datasets/coref-data/preco_raw",
        entity_types: ["PER", "LOC", "ORG"],
        language: "en",
        domain: "reading_comprehension",
        license: "CC-BY-4.0",
        citation: "Chen et al. (2018)",
        paper_url: "https://aclanthology.org/D18-1016/",
        year: 2018,
        format: "JSONL",
        annotation_scheme: "CoNLLCoref",
        size_hint: "38k documents, includes singletons",
        notes: "Preschool vocabulary for cleaner evaluation; largest public coref corpus",
        splits: ["train", "dev", "test"],
        tasks: ["coref"],
        hf_id: "coref-data/preco_raw",
        categories: [coref],
    },
    LitBank {
        name: "LitBank",
        description: "Literary coreference. 100 public-domain English fiction works (1719-1922) with ACE-style entities.",
        url: "https://raw.githubusercontent.com/dbamman/litbank/master/coref/brat/1023_bleak_house_brat.ann",
        entity_types: ["PER", "LOC", "ORG", "GPE", "FAC", "VEH"],
        language: "en",
        domain: "literature",
        license: "CC-BY-4.0",
        citation: "Bamman et al. (2019)",
        paper_url: "https://aclanthology.org/P19-1353/",
        year: 2019,
        format: "BRAT",
        annotation_scheme: "Standoff",
        size_hint: "100 novels, ~2k tokens each",
        notes: "Focus on character coreference; includes event coref; public domain texts",
        splits: ["all"],
        tasks: ["ner", "coref", "event_coref"],
        expected_docs: 100,
        categories: [coref, literary],
    },
    ECBPlus {
        name: "ECB+",
        description: "Event Coreference Bank Plus. Standard benchmark for cross-document event coreference resolution.",
        url: "https://raw.githubusercontent.com/cltl/ecbPlus/master/ECB%2B_LREC2014/ECB%2B.zip",
        entity_types: ["EVENT", "TIME", "LOC", "PARTICIPANT"],
        language: "en",
        domain: "news",
        license: "CC-BY-3.0",
        citation: "Cybulska & Vossen (2014)",
        paper_url: "https://aclanthology.org/L14-1646/",
        year: 2014,
        format: "XML-ZIP",
        size_hint: "43 topics, 982 docs, ~7k events",
        example: "Doc1: 'The earthquake [struck] at 3am.' Doc2: 'The [tremor] caused damage.'\nEvents: struck_1, tremor_2 -> coreferent (same event)",
        notes: "De facto CDCR standard; topic-clustered structure may cause overfitting",
        splits: ["train", "dev", "test"],
        tasks: ["coref", "event_coref", "cdcr"],
        expected_docs: 982,
        categories: [coref, event_coref],
    },

    OntoNotesCoref {
        name: "OntoNotes Coreference",
        description: "OntoNotes 5.0 coreference annotations. Gold-standard multi-genre coref including WSJ, broadcast, web.",
        url: "https://catalog.ldc.upenn.edu/LDC2013T19",
        entity_types: ["PER", "ORG", "GPE", "NORP"],
        language: "en",
        domain: "mixed",
        license: "LDC",
        citation: "Pradhan et al. (2012)",
        paper_url: "https://aclanthology.org/W12-4501/",
        year: 2012,
        format: "CoNLL",
        annotation_scheme: "CoNLLCoref",
        size_hint: "3,493 documents, ~1.6M tokens",
        notes: "De facto standard for within-document coreference evaluation",
        categories: [coref],
    },

    WikiCoref {
        name: "WikiCoref",
        description: "Wikipedia coreference corpus. 30 documents with full coreference annotation.",
        url: "",
        entity_types: ["PER", "ORG", "LOC"],
        language: "en",
        domain: "wikipedia",
        license: "CC-BY-SA-4.0",
        citation: "Ghaddar & Langlais (2016)",
        paper_url: "https://aclanthology.org/C16-1252/",
        year: 2016,
        format: "CoNLL",
        size_hint: "30 documents, ~60k tokens",
        notes: "Long documents averaging 2k tokens; challenging for span-based models. Prior download URL has an expired TLS cert; needs a fresh mirror.",
        access_status: Deprecated,
        categories: [coref],
    },

    ARRAU3 {
        name: "ARRAU 3.0",
        description: "Anaphora Resolution and Underspecification corpus version 3. Multi-genre with rich annotation.",
        url: "https://aclanthology.org/2024.codi-1.12/",
        entity_types: ["PER", "ORG", "LOC", "Event"],
        language: "en",
        domain: "mixed",
        license: "Research",
        citation: "Uryupina et al. (2024)",
        paper_url: "https://aclanthology.org/2024.codi-1.12/",
        year: 2024,
        format: "MMAX2",
        annotation_scheme: "ARRAU",
        size_hint: "~350k tokens across multiple genres",
        notes: "Rich annotation including bridging, discourse deixis, and ambiguity",
        categories: [coref],
    },

    AMIMeeting {
        name: "AMI Meeting",
        description: "Meeting transcripts with coreference and dialogue act annotation.",
        url: "https://groups.inf.ed.ac.uk/ami/download/",
        entity_types: ["PER", "ORG", "LOC"],
        language: "en",
        domain: "dialogue",
        license: "CC-BY-4.0",
        citation: "Carletta et al. (2005)",
        paper_url: "https://groups.inf.ed.ac.uk/ami/icsi/",
        year: 2005,
        format: "XML",
        size_hint: "100 hours of meetings",
        notes: "Multi-party dialogue; includes prosody and head gestures",
        categories: [coref, dialogue],
    },

    CLEFClinicalCoref {
        name: "CLEF Clinical Coreference",
        description: "Clinical coreference from ShARe/CLEF eHealth. Patient records with coref.",
        url: "",
        entity_types: ["Disorder", "Drug", "Procedure"],
        language: "en",
        domain: "clinical",
        license: "PhysioNet",
        citation: "Suominen et al. (2013)",
        paper_url: "https://clef2013.clef-initiative.eu/index.php?page=pages/proceedings.php",
        year: 2013,
        format: "Standoff",
        size_hint: "298 discharge summaries",
        notes: "Clinical concept coreference; disorder mentions across sentences. Was hosted on PhysioNet; URL appears gone/renamed; requires controlled access.",
        access_status: Registration,
        categories: [coref, biomedical],
    },

    RSTDT {
        name: "RST Discourse Treebank",
        description: "Penn Discourse Treebank with RST annotations. Discourse relations and structure.",
        url: "https://catalog.ldc.upenn.edu/LDC2002T07",
        entity_types: [],
        language: "en",
        domain: "news",
        license: "LDC",
        citation: "Carlson et al. (2001)",
        paper_url: "https://aclanthology.org/A00-1036/",
        year: 2001,
        format: "Custom",
        size_hint: "385 WSJ articles",
        notes: "RST discourse structure; useful for discourse-aware coreference",
        categories: [coref],
    },

    WinoBias {
        name: "WinoBias",
        description: "Coreference bias benchmark. Winograd-schema sentences testing occupational gender stereotypes.",
        url: "https://raw.githubusercontent.com/uclanlp/corefBias/master/WinoBias/wino/data/anti_stereotyped_type1.txt.dev",
        entity_types: ["PER"],
        language: "en",
        domain: "evaluation",
        license: "MIT",
        citation: "Zhao et al. (2018)",
        paper_url: "https://aclanthology.org/N18-2003/",
        year: 2018,
        format: "Custom",
        size_hint: "3,160 sentences",
        notes: "Type 1 (syntactic) and Type 2 (semantic) splits; tests BLS occupational stats",
        categories: [coref, bias_evaluation],
    },

    // =========================================================================
    // Indigenous Language Datasets
    // =========================================================================
    QxoRef {
        name: "qxoRef",
        description: "First coreference corpus for Conchucos Quechua. Historically significant as first indigenous coref resource.",
        url: "https://raw.githubusercontent.com/elizabethpankratz/qxoRef/f0eb5716573b3f428bfcfdda923b195d0e7967b8/qxoRef_AZ23.conll",
        entity_types: ["PER", "LOC", "ORG"],
        language: "qxo",
        domain: "narrative",
        license: "CC-BY-NC-SA-4.0",
        citation: "Rios (2021)",
        paper_url: "https://aclanthology.org/2021.americasnlp-1.1/",
        year: 2021,
        format: "CoNLL",
        annotation_scheme: "CoNLLCoref",
        size_hint: "12 docs, 332 mentions",
        notes: "First indigenous coreference corpus; pro-drop language; agglutinative morphology",
        categories: [coref, indigenous, low_resource],
    },
    AmericasNLI {
        name: "AmericasNLI",
        description: "NLI for 10 Indigenous American languages (Quechua, Guaraní, Nahuatl, etc.).",
        url: "https://raw.githubusercontent.com/nala-cub/AmericasNLI/be3c351b7e1ae69936c61bfde3e24f30757db9ac/test.tsv",
        entity_types: [],
        language: "mul",
        domain: "general",
        license: "CC-BY-4.0",
        citation: "Ebrahimi et al. (2022)",
        paper_url: "https://aclanthology.org/2022.acl-long.435/",
        year: 2022,
        format: "TSV",
        notes: "Tests zero-shot transfer from multilingual models; 10 indigenous languages",
        categories: [indigenous, multilingual, low_resource],
    },
    CherokeeNER {
        name: "Cherokee NER",
        description: "Cherokee-English parallel corpus for NER transfer. Uses Syllabary script.",
        url: "",
        entity_types: ["PER", "LOC", "ORG"],
        language: "chr",
        domain: "indigenous",
        license: "Research",
        citation: "Zhang et al. (2020)",
        paper_url: "https://aclanthology.org/2020.findings-emnlp.464/",
        year: 2020,
        format: "Custom",
        notes: "Syllabary script (85 characters); polysynthetic language; ~7k speakers. Prior URL is dead; needs a fresh mirror.",
        access_status: Deprecated,
        categories: [ner, indigenous, low_resource],
    },
    NahuatlNER {
        name: "Nahuatl NER",
        description: "Named entity recognition for Nahuatl (Aztec language). Colonial-era texts and modern usage.",
        url: "",
        entity_types: ["PER", "LOC", "ORG"],
        language: "nah",
        domain: "historical",
        license: "CC-BY-4.0",
        citation: "Gutierrez-Vasquez et al. (2023)",
        year: 2023,
        format: "CoNLL",
        notes: "Polysynthetic Uto-Aztecan language; ~1.7M speakers; includes colonial manuscripts. Prior URL is dead; needs a fresh mirror.",
        access_status: Deprecated,
        categories: [ner, indigenous, low_resource, historical],
    },
    MaoriNER {
        name: "Māori NER",
        description: "Named entity recognition for Te Reo Māori. New Zealand indigenous language corpus.",
        url: "",
        entity_types: ["PER", "LOC", "ORG"],
        language: "mi",
        domain: "indigenous",
        license: "Research",
        citation: "Te Hiku Media (2022)",
        year: 2022,
        format: "JSONL",
        notes: "Polynesian language; ~50k fluent speakers; limited training data available",
        access_status: ContactAuthors,
        categories: [ner, indigenous, low_resource],
    },
    WelshNER {
        name: "Welsh NER",
        description: "Named entity recognition for Welsh (Cymraeg). Celtic language NER corpus.",
        url: "",
        entity_types: ["PER", "LOC", "ORG", "MISC"],
        language: "cy",
        domain: "news",
        license: "CC-BY-4.0",
        citation: "Roberts et al. (2021)",
        year: 2021,
        format: "CoNLL",
        notes: "Celtic language; ~900k speakers; supports Welsh-specific entity types. Prior URL is dead; needs a fresh mirror.",
        access_status: Deprecated,
        categories: [ner, indigenous, low_resource],
    },
    BasqueNER {
        name: "Basque NER",
        description: "Named entity recognition for Basque (Euskara). Language isolate NER corpus.",
        url: "",
        entity_types: ["PER", "LOC", "ORG", "MISC"],
        language: "eu",
        domain: "news",
        license: "CC-BY-4.0",
        citation: "Alegria et al. (2019)",
        year: 2019,
        format: "CoNLL",
        size_hint: "~80k tokens",
        notes: "Language isolate; agglutinative morphology; ~750k speakers; ergative-absolutive alignment. Prior URL is dead; needs a fresh mirror.",
        access_status: Deprecated,
        categories: [ner, indigenous, low_resource],
    },

    // =========================================================================
    // Historical NER Datasets
    // =========================================================================
    HIPE2022 {
        name: "HIPE-2022",
        description: "Multilingual Historical NER. 6 datasets across 11 languages including Latin.",
        url: "https://raw.githubusercontent.com/hipe-eval/HIPE-2022-data/147f5bc3c7fb7e5c6b024a9ffd6503cd019fb9ea/data/v2.1/hipe2020/de/HIPE-2022-v2.1-hipe2020-test-de.tsv",
        entity_types: ["PER", "LOC", "ORG", "PROD"],
        language: "mul",
        domain: "historical",
        license: "CC-BY-NC-4.0",
        citation: "Ehrmann et al. (2022)",
        paper_url: "https://ceur-ws.org/Vol-3180/paper-83.pdf",
        year: 2022,
        format: "TSV",
        annotation_scheme: "IOB2",
        notes: "CLEF-HIPE shared task; includes Latin and Classical commentary; OCR noise",
        splits: ["test"],
        tasks: ["ner"],
        categories: [ner, historical, multilingual],
    },
    HistNERo {
        name: "HistNERo",
        description: "Romanian historical newspaper NER. First Romanian historical NER corpus from four regions.",
        url: "https://github.com/avramandrei/histnero",
        entity_types: ["PER", "LOC", "ORG", "DATE", "MISC"],
        language: "ro",
        domain: "historical",
        license: "CC-BY-4.0",
        citation: "HistNERo Team (2024)",
        paper_url: "https://arxiv.org/abs/2405.00155",
        year: 2024,
        format: "CoNLL",
        size_hint: "~323k tokens, 19th-20th century newspapers",
        notes: "Four historical Romanian regions (Bessarabia, Moldavia, Transylvania, Wallachia); diachronic benchmark",
        splits: ["train", "dev", "test"],
        tasks: ["ner"],
        categories: [ner, historical, low_resource],
    },
    QuaeroOldPress {
        name: "Quaero Old Press",
        description: "French historical newspaper NER from 1890. OCR-corrected with manual NE annotations.",
        url: "",
        entity_types: ["PER", "LOC", "ORG", "TIME", "PROD"],
        language: "fr",
        domain: "historical",
        license: "Research",
        citation: "Galibert et al. (2012)",
        year: 2012,
        format: "XML",
        size_hint: "295 pages, 1890 newspapers",
        notes: "French historical NER benchmark; manual OCR corrections; reasonably clean historical text",
        splits: ["test"],
        tasks: ["ner"],
        access_status: ContactAuthors,
        categories: [ner, historical],
    },
    HistoricalChineseNER {
        name: "Historical Chinese NER",
        description: "Multi-task historical Chinese corpus. NER + entity linking + coreference + relations.",
        url: "",
        entity_types: ["PER", "LOC", "ORG", "TIME", "OFFICIAL"],
        language: "zh",
        domain: "historical",
        license: "Research",
        citation: "LREC-COLING (2024)",
        paper_url: "https://aclanthology.org/2024.lrec-main.35.pdf",
        year: 2024,
        format: "JSONL",
        size_hint: "Historical Chinese newspapers + documents",
        notes: "LREC-COLING 2024; multi-task historical IE benchmark; cross-genre historical Chinese",
        splits: ["train", "dev", "test"],
        tasks: ["ner", "el", "coref", "re"],
        access_status: ContactAuthors,
        categories: [ner, entity_linking, coref, historical, multilingual],
    },
    CHisIEC {
        name: "CHisIEC",
        description: "Chinese Historical Information Extraction Corpus. Ancient Chinese NER + RE with 12 relation types.",
        url: "https://raw.githubusercontent.com/tangxuemei1995/CHisIEC/main/data/re/coling_test.json",
        entity_types: ["PER", "LOC", "OFI", "BOOK"],
        language: "lzh",
        domain: "historical",
        license: "Research",
        citation: "Tang et al. (2024)",
        paper_url: "https://aclanthology.org/2024.lrec-main.283/",
        year: 2024,
        format: "JSON",
        size_hint: "3,891 paragraphs, 13,520 entities, 8,228 relations",
        notes: "Ancient Chinese dynastic histories (24史); 12 domain-specific relations for historical socio-political structures; pre-modern Chinese (文言文)",
        splits: ["train", "dev", "test"],
        tasks: ["ner", "re"],
        categories: [ner, relation_extraction, historical, ancient],
    },