docs/evals/m1_data_eda_all.md

M1 data EDA — UniversalCEFR

Generated: 2026-06-11T14:43:50+00:00 Command: scripts/eda_universalcefr.py --langs all --report docs/evals/m1_data_eda_all.md

Subsets overview

dataset	lang	rows	categories	formats	licenses	words: median (p10, p90)
UniversalCEFR/caes_es	es	31149	{'learner': 31149}	{'document-level': 31149}	{'CC BY-NC 4.0': 31149}	150 (p10=66, p90=289)
UniversalCEFR/cambridge_exams_en	en	331	{'reference': 331}	{'document-level': 331}	{'CC BY-NC-SA 4.0': 331}	563 (p10=123, p90=923)
UniversalCEFR/cefr_asag_en	en	299	{'learner': 299}	{'document-level': 299}	{'CC BY-NC-SA 4.0': 299}	59 (p10=8, p90=150)
UniversalCEFR/cefr_sp_en	en	10004	{'reference': 10004}	{'sentence-level': 10004}	{'CC BY-NC-SA 4.0': 10004}	14 (p10=8, p90=24)
UniversalCEFR/cople2_pt	pt	942	{'learner': 942}	{'paragraph-level': 942}	{'CC BY-SA-NC 4.0': 942}	152 (p10=67, p90=264)
UniversalCEFR/elg_cefr_de	de	509	{'reference': 509}	{'document-level': 509}	{'CC BY-NC-SA 4.0': 509}	140 (p10=124, p90=148)
UniversalCEFR/elg_cefr_en	en	712	{'reference': 712}	{'document-level': 712}	{'CC BY-NC-SA 4.0': 712}	277 (p10=191, p90=799)
UniversalCEFR/elg_cefr_nl	nl	3596	{'reference': 3596}	{'document-level': 3596}	{'CC BY-NC-SA 4.0': 3596}	211 (p10=127, p90=230)
UniversalCEFR/elle_et	et	1697	{'learner': 1697}	{'paragraph-level': 420, 'document-level': 1277}	{'CC BY 4.0': 1697}	186 (p10=61, p90=463)
UniversalCEFR/hablacultura_es	es	713	{'reference': 713}	{'paragraph-level': 713}	{'CC BY NC 4.0': 713}	61 (p10=23, p90=137)
UniversalCEFR/icle500_en	en	495	{'learner': 495}	{'document-level': 495}	{'CC0 1.0 (Public Domain)': 495}	585 (p10=476, p90=762)
UniversalCEFR/kwiqiz_es	es	206	{'reference': 206}	{'document-level': 206}	{'CC BY NC 4.0': 206}	232 (p10=123, p90=447)
UniversalCEFR/kwiqiz_fr	fr	344	{'reference': 344}	{'document-level': 344}	{'CC BY NC 4.0': 344}	329 (p10=165, p90=631)
UniversalCEFR/learn_welsh_cy	cy	1372	{'reference': 1372}	{'dialogue-level': 115, 'document-level': 41, 'paragraph-level': 109, 'sentence-level': 1107}	{'public': 1372}	6 (p10=3, p90=61)
UniversalCEFR/merlin_cs	cs	441	{'learner': 441}	{'paragraph-level': 441}	{'CC BY-SA 4.0': 441}	135 (p10=61, p90=230)
UniversalCEFR/merlin_de	de	1033	{'learner': 1033}	{'paragraph-level': 1033}	{'CC BY-SA 4.0': 1033}	104 (p10=40, p90=222)
UniversalCEFR/merlin_it	it	813	{'learner': 813}	{'paragraph-level': 813}	{'CC BY-SA 4.0': 813}	92 (p10=44, p90=198)
UniversalCEFR/peapl2_pt	pt	481	{'learner': 481}	{'paragraph-level': 481}	{'CC BY SA NC 4.0': 481}	220 (p10=100, p90=319)
UniversalCEFR/readme_ar	ar	1945	{'reference': 1945}	{'sentence-level': 1945}	{'CC BY SA NC 4.0': 1945}	19 (p10=7, p90=47)
UniversalCEFR/readme_en	en	2822	{'reference': 2822}	{'sentence-level': 2822}	{'CC BY SA NC 4.0': 2822}	17 (p10=7, p90=36)
UniversalCEFR/readme_fr	fr	1669	{'reference': 1669}	{'sentence-level': 1669}	{'CC BY SA NC 4.0': 1669}	18 (p10=8, p90=41)
UniversalCEFR/readme_hi	hi	1491	{'reference': 1491}	{'sentence-level': 1491}	{'CC BY SA NC 4.0': 1491}	17 (p10=8, p90=35)
UniversalCEFR/readme_ru	ru	1758	{'reference': 1758}	{'sentence-level': 1758}	{'CC BY SA NC 4.0': 1758}	16 (p10=7, p90=33)
UniversalCEFR/zaebuc_ar	ar	215	{'learner': 215}	{'paragraph-level': 215}	{'CC BY-SA 4.0': 215}	155 (p10=72, p90=244)

Level distribution per subset

dataset	A1	A2	B1	B2	C1	C2	odd labels
UniversalCEFR/caes_es	8261	8582	6510	4729	3067	0	0
UniversalCEFR/cambridge_exams_en	0	64	60	71	67	69	0
UniversalCEFR/cefr_asag_en	18	59	113	74	30	5	0
UniversalCEFR/cefr_sp_en	124	1271	3305	3330	1744	230	0
UniversalCEFR/cople2_pt	236	236	163	163	72	72	0
UniversalCEFR/elg_cefr_de	31	90	90	115	88	95	0
UniversalCEFR/elg_cefr_en	24	110	158	140	97	69	114
UniversalCEFR/elg_cefr_nl	51	216	782	738	219	85	1505
UniversalCEFR/elle_et	0	395	588	407	307	0	0
UniversalCEFR/hablacultura_es	0	33	271	273	118	0	18
UniversalCEFR/icle500_en	0	1	29	99	91	64	211
UniversalCEFR/kwiqiz_es	21	33	54	59	39	0	0
UniversalCEFR/kwiqiz_fr	11	24	131	126	52	0	0
UniversalCEFR/learn_welsh_cy	764	608	0	0	0	0	0
UniversalCEFR/merlin_cs	1	188	165	81	4	0	2
UniversalCEFR/merlin_de	57	306	331	293	42	4	0
UniversalCEFR/merlin_it	29	381	394	2	0	0	7
UniversalCEFR/peapl2_pt	78	89	204	70	40	0	0
UniversalCEFR/readme_ar	81	252	514	565	350	183	0
UniversalCEFR/readme_en	182	673	623	896	377	71	0
UniversalCEFR/readme_fr	140	366	444	352	241	126	0
UniversalCEFR/readme_hi	263	283	286	263	222	174	0
UniversalCEFR/readme_ru	402	293	409	326	237	91	0
UniversalCEFR/zaebuc_ar	0	7	111	80	11	0	6

Odd labels detail: {'UniversalCEFR/elg_cefr_en': {'A1+': 1, 'A2+': 31, 'B1+': 48, 'B2+': 34}, 'UniversalCEFR/elg_cefr_nl': {'B1+': 708, 'A2+': 316, 'B2+': 305, 'A1+': 60, 'C2+': 14, 'C1+': 102}, 'UniversalCEFR/hablacultura_es': {'B': 18}, 'UniversalCEFR/icle500_en': {'B2+': 119, 'B1+': 85, 'NA': 7}, 'UniversalCEFR/merlin_cs': {'EMPTY': 2}, 'UniversalCEFR/merlin_it': {'unrated': 6, 'EMPTY': 1}, 'UniversalCEFR/zaebuc_ar': {'Unassessable': 6}}

Reference pool (the M1 reading-classifier candidates)

Subsets whose category includes reference, i.e. texts written for learners rather than by them. Learner-production subsets are the M3 candidates (grading learner writing), not M1 training data.

lang	reference rows	from subsets
ar	1945	readme_ar
cy	1372	learn_welsh_cy
de	509	elg_cefr_de
en	13869	cambridge_exams_en, cefr_sp_en, elg_cefr_en, readme_en
es	919	hablacultura_es, kwiqiz_es
fr	2013	kwiqiz_fr, readme_fr
hi	1491	readme_hi
nl	3596	elg_cefr_nl
ru	1758	readme_ru

---

Notes: word counts use whitespace tokenisation (approximate for ar/hi). Licenses are aggregated from the per-row license field; decisions and exclusions are recorded in docs/adr/0003-datasets-and-licensing.md.