nickprock/bert-italian-finetuned-ner
Token Classification
•
Updated
•
263
•
7
The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider
removing the
loading script
and relying on
automated data support
(you can use
convert_to_parquet
from the datasets
library). If this is not possible, please
open a discussion
for direct help.
WikiAnn NER dataset formatted in a part of TNER project.
LOC
, ORG
, PER
An example of train
of ja
looks as follows.
{
'tokens': ['#', '#', 'ユ', 'リ', 'ウ', 'ス', '・', 'ベ', 'ー', 'リ', 'ッ', 'ク', '#', '1', '9','9','9'],
'tags': [6, 6, 2, 5, 5, 5, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6]
}
The label2id dictionary can be found at here.
{
"B-LOC": 0,
"B-ORG": 1,
"B-PER": 2,
"I-LOC": 3,
"I-ORG": 4,
"I-PER": 5,
"O": 6
}
language | train | validation | test |
---|---|---|---|
ace | 100 | 100 | 100 |
bg | 20000 | 10000 | 10000 |
da | 20000 | 10000 | 10000 |
fur | 100 | 100 | 100 |
ilo | 100 | 100 | 100 |
lij | 100 | 100 | 100 |
mzn | 100 | 100 | 100 |
qu | 100 | 100 | 100 |
su | 100 | 100 | 100 |
vi | 20000 | 10000 | 10000 |
af | 5000 | 1000 | 1000 |
bh | 100 | 100 | 100 |
de | 20000 | 10000 | 10000 |
fy | 1000 | 1000 | 1000 |
io | 100 | 100 | 100 |
lmo | 100 | 100 | 100 |
nap | 100 | 100 | 100 |
rm | 100 | 100 | 100 |
sv | 20000 | 10000 | 10000 |
vls | 100 | 100 | 100 |
als | 100 | 100 | 100 |
bn | 10000 | 1000 | 1000 |
diq | 100 | 100 | 100 |
ga | 1000 | 1000 | 1000 |
is | 1000 | 1000 | 1000 |
ln | 100 | 100 | 100 |
nds | 100 | 100 | 100 |
ro | 20000 | 10000 | 10000 |
sw | 1000 | 1000 | 1000 |
vo | 100 | 100 | 100 |
am | 100 | 100 | 100 |
bo | 100 | 100 | 100 |
dv | 100 | 100 | 100 |
gan | 100 | 100 | 100 |
it | 20000 | 10000 | 10000 |
lt | 10000 | 10000 | 10000 |
ne | 100 | 100 | 100 |
ru | 20000 | 10000 | 10000 |
szl | 100 | 100 | 100 |
wa | 100 | 100 | 100 |
an | 1000 | 1000 | 1000 |
br | 1000 | 1000 | 1000 |
el | 20000 | 10000 | 10000 |
gd | 100 | 100 | 100 |
ja | 20000 | 10000 | 10000 |
lv | 10000 | 10000 | 10000 |
nl | 20000 | 10000 | 10000 |
rw | 100 | 100 | 100 |
ta | 15000 | 1000 | 1000 |
war | 100 | 100 | 100 |
ang | 100 | 100 | 100 |
bs | 15000 | 1000 | 1000 |
eml | 100 | 100 | 100 |
gl | 15000 | 10000 | 10000 |
jbo | 100 | 100 | 100 |
map-bms | 100 | 100 | 100 |
nn | 20000 | 1000 | 1000 |
sa | 100 | 100 | 100 |
te | 1000 | 1000 | 1000 |
wuu | 100 | 100 | 100 |
ar | 20000 | 10000 | 10000 |
ca | 20000 | 10000 | 10000 |
en | 20000 | 10000 | 10000 |
gn | 100 | 100 | 100 |
jv | 100 | 100 | 100 |
mg | 100 | 100 | 100 |
no | 20000 | 10000 | 10000 |
sah | 100 | 100 | 100 |
tg | 100 | 100 | 100 |
xmf | 100 | 100 | 100 |
arc | 100 | 100 | 100 |
cbk-zam | 100 | 100 | 100 |
eo | 15000 | 10000 | 10000 |
gu | 100 | 100 | 100 |
ka | 10000 | 10000 | 10000 |
mhr | 100 | 100 | 100 |
nov | 100 | 100 | 100 |
scn | 100 | 100 | 100 |
th | 20000 | 10000 | 10000 |
yi | 100 | 100 | 100 |
arz | 100 | 100 | 100 |
cdo | 100 | 100 | 100 |
es | 20000 | 10000 | 10000 |
hak | 100 | 100 | 100 |
kk | 1000 | 1000 | 1000 |
mi | 100 | 100 | 100 |
oc | 100 | 100 | 100 |
sco | 100 | 100 | 100 |
tk | 100 | 100 | 100 |
yo | 100 | 100 | 100 |
as | 100 | 100 | 100 |
ce | 100 | 100 | 100 |
et | 15000 | 10000 | 10000 |
he | 20000 | 10000 | 10000 |
km | 100 | 100 | 100 |
min | 100 | 100 | 100 |
or | 100 | 100 | 100 |
sd | 100 | 100 | 100 |
tl | 10000 | 1000 | 1000 |
zea | 100 | 100 | 100 |
ast | 1000 | 1000 | 1000 |
ceb | 100 | 100 | 100 |
eu | 10000 | 10000 | 10000 |
hi | 5000 | 1000 | 1000 |
kn | 100 | 100 | 100 |
mk | 10000 | 1000 | 1000 |
os | 100 | 100 | 100 |
sh | 20000 | 10000 | 10000 |
tr | 20000 | 10000 | 10000 |
zh-classical | 100 | 100 | 100 |
ay | 100 | 100 | 100 |
ckb | 1000 | 1000 | 1000 |
ext | 100 | 100 | 100 |
hr | 20000 | 10000 | 10000 |
ko | 20000 | 10000 | 10000 |
ml | 10000 | 1000 | 1000 |
pa | 100 | 100 | 100 |
si | 100 | 100 | 100 |
tt | 1000 | 1000 | 1000 |
zh-min-nan | 100 | 100 | 100 |
az | 10000 | 1000 | 1000 |
co | 100 | 100 | 100 |
fa | 20000 | 10000 | 10000 |
hsb | 100 | 100 | 100 |
ksh | 100 | 100 | 100 |
mn | 100 | 100 | 100 |
pdc | 100 | 100 | 100 |
simple | 20000 | 1000 | 1000 |
ug | 100 | 100 | 100 |
zh-yue | 20000 | 10000 | 10000 |
ba | 100 | 100 | 100 |
crh | 100 | 100 | 100 |
fi | 20000 | 10000 | 10000 |
hu | 20000 | 10000 | 10000 |
ku | 100 | 100 | 100 |
mr | 5000 | 1000 | 1000 |
pl | 20000 | 10000 | 10000 |
sk | 20000 | 10000 | 10000 |
uk | 20000 | 10000 | 10000 |
zh | 20000 | 10000 | 10000 |
bar | 100 | 100 | 100 |
cs | 20000 | 10000 | 10000 |
fiu-vro | 100 | 100 | 100 |
hy | 15000 | 1000 | 1000 |
ky | 100 | 100 | 100 |
ms | 20000 | 1000 | 1000 |
pms | 100 | 100 | 100 |
sl | 15000 | 10000 | 10000 |
ur | 20000 | 1000 | 1000 |
bat-smg | 100 | 100 | 100 |
csb | 100 | 100 | 100 |
fo | 100 | 100 | 100 |
ia | 100 | 100 | 100 |
la | 5000 | 1000 | 1000 |
mt | 100 | 100 | 100 |
pnb | 100 | 100 | 100 |
so | 100 | 100 | 100 |
uz | 1000 | 1000 | 1000 |
be-x-old | 5000 | 1000 | 1000 |
cv | 100 | 100 | 100 |
fr | 20000 | 10000 | 10000 |
id | 20000 | 10000 | 10000 |
lb | 5000 | 1000 | 1000 |
mwl | 100 | 100 | 100 |
ps | 100 | 100 | 100 |
sq | 5000 | 1000 | 1000 |
vec | 100 | 100 | 100 |
be | 15000 | 1000 | 1000 |
cy | 10000 | 1000 | 1000 |
frr | 100 | 100 | 100 |
ig | 100 | 100 | 100 |
li | 100 | 100 | 100 |
my | 100 | 100 | 100 |
pt | 20000 | 10000 | 10000 |
sr | 20000 | 10000 | 10000 |
vep | 100 | 100 | 100 |
@inproceedings{pan-etal-2017-cross,
title = "Cross-lingual Name Tagging and Linking for 282 Languages",
author = "Pan, Xiaoman and
Zhang, Boliang and
May, Jonathan and
Nothman, Joel and
Knight, Kevin and
Ji, Heng",
booktitle = "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2017",
address = "Vancouver, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/P17-1178",
doi = "10.18653/v1/P17-1178",
pages = "1946--1958",
abstract = "The ambitious goal of this work is to develop a cross-lingual name tagging and linking framework for 282 languages that exist in Wikipedia. Given a document in any of these languages, our framework is able to identify name mentions, assign a coarse-grained or fine-grained type to each mention, and link it to an English Knowledge Base (KB) if it is linkable. We achieve this goal by performing a series of new KB mining methods: generating {``}silver-standard{''} annotations by transferring annotations from English to other languages through cross-lingual links and KB properties, refining annotations through self-training and topic selection, deriving language-specific morphology features from anchor links, and mining word translation pairs from cross-lingual links. Both name tagging and linking results for 282 languages are promising on Wikipedia data and on-Wikipedia data.",
}