Commit
·
327b1cf
1
Parent(s):
954a6bb
Upload indocoref.py with huggingface_hub
Browse files- indocoref.py +248 -0
indocoref.py
ADDED
@@ -0,0 +1,248 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
from pathlib import Path
|
3 |
+
from typing import Dict, List, Tuple
|
4 |
+
|
5 |
+
try:
|
6 |
+
from typing import TypedDict
|
7 |
+
except:
|
8 |
+
from typing_extensions import TypedDict
|
9 |
+
|
10 |
+
import datasets
|
11 |
+
|
12 |
+
from nusacrowd.nusa_datasets.indocoref.utils.text_preprocess import \
|
13 |
+
TextPreprocess
|
14 |
+
from nusacrowd.utils import schemas
|
15 |
+
from nusacrowd.utils.configs import NusantaraConfig
|
16 |
+
from nusacrowd.utils.constants import Tasks
|
17 |
+
|
18 |
+
_CITATION = """\
|
19 |
+
@inproceedings{artari-etal-2021-multi,
|
20 |
+
title = {A Multi-Pass Sieve Coreference Resolution for {I}ndonesian},
|
21 |
+
author = {Artari, Valentina Kania Prameswara and Mahendra, Rahmad and Jiwanggi, Meganingrum Arista and Anggraito, Adityo and Budi, Indra},
|
22 |
+
year = 2021,
|
23 |
+
month = sep,
|
24 |
+
booktitle = {Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2021)},
|
25 |
+
publisher = {INCOMA Ltd.},
|
26 |
+
address = {Held Online},
|
27 |
+
pages = {79--85},
|
28 |
+
url = {https://aclanthology.org/2021.ranlp-1.10},
|
29 |
+
abstract = {Coreference resolution is an NLP task to find out whether the set of referring expressions belong to the same concept in discourse. A multi-pass sieve is a deterministic coreference model that implements several layers of sieves, where each sieve takes a pair of correlated mentions from a collection of non-coherent mentions. The multi-pass sieve is based on the principle of high precision, followed by increased recall in each sieve. In this work, we examine the portability of the multi-pass sieve coreference resolution model to the Indonesian language. We conduct the experiment on 201 Wikipedia documents and the multi-pass sieve system yields 72.74{\%} of MUC F-measure and 52.18{\%} of BCUBED F-measure.}
|
30 |
+
}
|
31 |
+
"""
|
32 |
+
|
33 |
+
_LOCAL = False
|
34 |
+
_LANGUAGES = ["ind"] # We follow ISO639-3 language code (https://iso639-3.sil.org/code_tables/639/data)
|
35 |
+
_DATASETNAME = "indocoref"
|
36 |
+
_DESCRIPTION = """\
|
37 |
+
Dataset contains articles from Wikipedia Bahasa Indonesia which fulfill these conditions:
|
38 |
+
- The pages contain many noun phrases, which the authors subjectively pick: (i) fictional plots, e.g., subtitles for films,
|
39 |
+
TV show episodes, and novel stories; (ii) biographies (incl. fictional characters); and (iii) historical events or important events.
|
40 |
+
- The pages contain significant variation of pronoun and named-entity. We count the number of first, second, third person pronouns,
|
41 |
+
and clitic pronouns in the document by applying string matching.We examine the number
|
42 |
+
of named-entity using the Stanford CoreNLP
|
43 |
+
NER Tagger (Manning et al., 2014) with a
|
44 |
+
model trained from the Indonesian corpus
|
45 |
+
taken from Alfina et al. (2016).
|
46 |
+
The Wikipedia texts have length of 500 to
|
47 |
+
2000 words.
|
48 |
+
We sample 201 of pages from subset of filtered
|
49 |
+
Wikipedia pages. We hire five annotators who are
|
50 |
+
undergraduate student in Linguistics department.
|
51 |
+
They are native in Indonesian. Annotation is carried out using the Script d’Annotation des Chanes
|
52 |
+
de Rfrence (SACR), a web-based Coreference resolution annotation tool developed by Oberle (2018).
|
53 |
+
From the 201 texts, there are 16,460 mentions
|
54 |
+
tagged by the annotators
|
55 |
+
"""
|
56 |
+
|
57 |
+
_HOMEPAGE = "https://github.com/valentinakania/indocoref/"
|
58 |
+
_LICENSE = "MIT"
|
59 |
+
_URLS = {
|
60 |
+
_DATASETNAME: "https://github.com/valentinakania/indocoref/archive/refs/heads/main.zip",
|
61 |
+
}
|
62 |
+
_SUPPORTED_TASKS = [Tasks.COREFERENCE_RESOLUTION]
|
63 |
+
# Does not seem to have versioning
|
64 |
+
_SOURCE_VERSION = "1.0.0"
|
65 |
+
_NUSANTARA_VERSION = "1.0.0"
|
66 |
+
|
67 |
+
|
68 |
+
class Indocoref(datasets.GeneratorBasedBuilder):
|
69 |
+
"""A collection of 210 curated articles from Wikipedia Bahasa Indonesia with Coreference Annotations"""
|
70 |
+
|
71 |
+
SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
|
72 |
+
NUSANTARA_VERSION = datasets.Version(_NUSANTARA_VERSION)
|
73 |
+
|
74 |
+
BUILDER_CONFIGS = [
|
75 |
+
NusantaraConfig(
|
76 |
+
name="indocoref_source",
|
77 |
+
version=SOURCE_VERSION,
|
78 |
+
description="Indocoref source schema",
|
79 |
+
schema="source",
|
80 |
+
subset_id="indocoref",
|
81 |
+
),
|
82 |
+
NusantaraConfig(
|
83 |
+
name="indocoref_nusantara_kb",
|
84 |
+
version=NUSANTARA_VERSION,
|
85 |
+
description="Indocoref Nusantara schema",
|
86 |
+
schema="nusantara_kb",
|
87 |
+
subset_id="indocoref",
|
88 |
+
),
|
89 |
+
]
|
90 |
+
|
91 |
+
DEFAULT_CONFIG_NAME = "indocoref_source"
|
92 |
+
|
93 |
+
def _info(self) -> datasets.DatasetInfo:
|
94 |
+
# The dataset does not really come with a schema, the features here come from the returned value
|
95 |
+
# of the accompanying utils files.
|
96 |
+
if self.config.schema == "source":
|
97 |
+
features = datasets.Features(
|
98 |
+
{
|
99 |
+
"id": datasets.Value("int64"),
|
100 |
+
"passage": datasets.Value("string"),
|
101 |
+
"mentions": [
|
102 |
+
{
|
103 |
+
"id": datasets.Value("int64"),
|
104 |
+
# Two entities which share a label are coreferences
|
105 |
+
"labels": datasets.Sequence(datasets.Value("string")),
|
106 |
+
"class": datasets.Value("string"),
|
107 |
+
"text": datasets.Value("string"),
|
108 |
+
"pronoun": datasets.Value("bool"),
|
109 |
+
"proper": datasets.Value("bool"),
|
110 |
+
"sent": datasets.Value("int32"),
|
111 |
+
"cluster": datasets.Value("int32"),
|
112 |
+
"per": datasets.Value("bool"),
|
113 |
+
"org": datasets.Value("bool"),
|
114 |
+
"loc": datasets.Value("bool"),
|
115 |
+
"ner": datasets.Value("bool"),
|
116 |
+
# "offset" is only available after modifying the original util class
|
117 |
+
# "offset": datasets.Sequence(datasets.Value("int32"))
|
118 |
+
# POS tags were originally available but removed due to polyglot icu dependency
|
119 |
+
# polyglot.Text(passage, hint_language_code='id')
|
120 |
+
}
|
121 |
+
],
|
122 |
+
}
|
123 |
+
)
|
124 |
+
elif self.config.schema == "nusantara_kb":
|
125 |
+
features = schemas.kb_features
|
126 |
+
|
127 |
+
return datasets.DatasetInfo(
|
128 |
+
description=_DESCRIPTION,
|
129 |
+
features=features,
|
130 |
+
homepage=_HOMEPAGE,
|
131 |
+
license=_LICENSE,
|
132 |
+
citation=_CITATION,
|
133 |
+
)
|
134 |
+
|
135 |
+
class ReadPassage(TypedDict):
|
136 |
+
passage: str
|
137 |
+
annotated: str
|
138 |
+
mentions: List[any]
|
139 |
+
|
140 |
+
def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
|
141 |
+
"""Returns SplitGenerators."""
|
142 |
+
urls = _URLS[_DATASETNAME]
|
143 |
+
base_path = Path(dl_manager.download_and_extract(urls)) / "indocoref-main" / "data"
|
144 |
+
passage_path = base_path / "passage"
|
145 |
+
annotated_path = base_path / "annotated"
|
146 |
+
mentions_per_file = TextPreprocess(annotated_path).run(0)
|
147 |
+
|
148 |
+
data: List[self.ReadPassage] = []
|
149 |
+
for passage_file_name, annotated_file_name in zip(sorted(os.listdir(passage_path)), sorted(os.listdir(annotated_path))):
|
150 |
+
passage_file_path, annotated_file_path = passage_path / passage_file_name, annotated_path / annotated_file_name
|
151 |
+
|
152 |
+
if os.path.isfile(passage_file_path) and os.path.isfile(annotated_file_path):
|
153 |
+
with open(passage_file_path, "r") as fpassage, open(annotated_file_path, "r") as fannotated:
|
154 |
+
data.append(self.ReadPassage(passage=fpassage.read(), annotated=fannotated.read(), mentions=mentions_per_file[annotated_file_name]))
|
155 |
+
|
156 |
+
# Dataset has no predefined splits, using datasets.Split.TRAIN for all of the data.
|
157 |
+
return [
|
158 |
+
datasets.SplitGenerator(
|
159 |
+
name=datasets.Split.TRAIN,
|
160 |
+
gen_kwargs={
|
161 |
+
"data": data,
|
162 |
+
"split": "train",
|
163 |
+
},
|
164 |
+
),
|
165 |
+
]
|
166 |
+
|
167 |
+
class DisjointSet:
|
168 |
+
parent = {}
|
169 |
+
|
170 |
+
def __init__(self, items):
|
171 |
+
for item in items:
|
172 |
+
self.parent[item] = item
|
173 |
+
|
174 |
+
def find(self, k):
|
175 |
+
if self.parent[k] == k:
|
176 |
+
return k
|
177 |
+
return self.find(self.parent[k])
|
178 |
+
|
179 |
+
def union(self, a, b):
|
180 |
+
x = self.find(a)
|
181 |
+
y = self.find(b)
|
182 |
+
self.parent[x] = y
|
183 |
+
|
184 |
+
def _generate_examples(self, data: List[ReadPassage], split: str) -> Tuple[int, Dict]:
|
185 |
+
"""Yields examples as (key, example) tuples."""
|
186 |
+
if self.config.schema == "source":
|
187 |
+
for index, example in enumerate(data):
|
188 |
+
passage, mentions = example["passage"], example["mentions"]
|
189 |
+
row = {
|
190 |
+
"id": index,
|
191 |
+
"passage": passage,
|
192 |
+
"mentions": [
|
193 |
+
{
|
194 |
+
"id": mention["id"],
|
195 |
+
"labels": mention["labels"],
|
196 |
+
"class": mention["class"],
|
197 |
+
"text": mention["text"],
|
198 |
+
"pronoun": mention["pronoun"],
|
199 |
+
"proper": mention["proper"],
|
200 |
+
"sent": mention["sent"],
|
201 |
+
"cluster": mention["cluster"],
|
202 |
+
"per": mention["per"],
|
203 |
+
"org": mention["org"],
|
204 |
+
"loc": mention["loc"],
|
205 |
+
"ner": mention["ner"],
|
206 |
+
}
|
207 |
+
for mention in mentions
|
208 |
+
],
|
209 |
+
}
|
210 |
+
yield index, row
|
211 |
+
|
212 |
+
elif self.config.schema == "nusantara_kb":
|
213 |
+
for index, example in enumerate(data):
|
214 |
+
passage, mentions = example["passage"], example["mentions"]
|
215 |
+
# Annotated text does not have any line breaks but the original passage does
|
216 |
+
passage = passage.replace(" \n", " ")
|
217 |
+
passage = passage.replace("\n", " ")
|
218 |
+
all_labels = {label for mention in mentions for label in mention["labels"]}
|
219 |
+
labels_disjoint_set = self.DisjointSet(all_labels)
|
220 |
+
for mention in mentions:
|
221 |
+
for i in range(1, len(mention["labels"])):
|
222 |
+
labels_disjoint_set.union(mention["labels"][i], mention["labels"][i - 1])
|
223 |
+
coreferences = {}
|
224 |
+
for mention in mentions:
|
225 |
+
coreference_id = labels_disjoint_set.find(mention["labels"][0])
|
226 |
+
if coreference_id not in coreferences:
|
227 |
+
coreferences[coreference_id] = []
|
228 |
+
coreferences[coreference_id].append(str(mention["id"]))
|
229 |
+
|
230 |
+
row_id = str(index)
|
231 |
+
row = {
|
232 |
+
"id": row_id,
|
233 |
+
"passages": [{"id": "passage-" + row_id, "type": "text", "text": [passage], "offsets": [[0, len(passage)]]}],
|
234 |
+
"entities": [
|
235 |
+
{
|
236 |
+
"id": row_id + "-entity-" + str(mention["id"]),
|
237 |
+
"type": mention["class"],
|
238 |
+
"text": [mention["text"]],
|
239 |
+
"offsets": [list(mention["offset"])],
|
240 |
+
"normalized": [],
|
241 |
+
}
|
242 |
+
for mention in mentions
|
243 |
+
],
|
244 |
+
"coreferences": [{"id": row_id + "-coreference-" + str(coref_id), "entity_ids": [row_id + "-entity-" + entity_id for entity_id in entity_ids]} for coref_id, entity_ids in enumerate(coreferences.values())],
|
245 |
+
"events": [],
|
246 |
+
"relations": [],
|
247 |
+
}
|
248 |
+
yield index, row
|