holylovenia
commited on
Commit
•
11d88f0
1
Parent(s):
e45820c
Upload ara_close.py with huggingface_hub
Browse files- ara_close.py +194 -0
ara_close.py
ADDED
@@ -0,0 +1,194 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# coding=utf-8
|
2 |
+
# Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
|
3 |
+
#
|
4 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
+
# you may not use this file except in compliance with the License.
|
6 |
+
# You may obtain a copy of the License at
|
7 |
+
#
|
8 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
9 |
+
#
|
10 |
+
# Unless required by applicable law or agreed to in writing, software
|
11 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
12 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
13 |
+
# See the License for the specific language governing permissions and
|
14 |
+
# limitations under the License.
|
15 |
+
|
16 |
+
""" \
|
17 |
+
The dataset contribution of this study is a compilation of short fictional stories \
|
18 |
+
written in Bikol for readability assessment. The data was combined other collected \
|
19 |
+
Philippine language corpora, such as Tagalog and Cebuano. The data from these languages \
|
20 |
+
are all distributed across the Philippine elementary system's first three grade \
|
21 |
+
levels (L1, L2, L3). We sourced this dataset from Let's Read Asia (LRA), Bloom Library, \
|
22 |
+
Department of Education, and Adarna House.
|
23 |
+
"""
|
24 |
+
|
25 |
+
from pathlib import Path
|
26 |
+
from typing import Dict, List, Tuple
|
27 |
+
|
28 |
+
import datasets
|
29 |
+
|
30 |
+
from seacrowd.utils import schemas
|
31 |
+
from seacrowd.utils.configs import SEACrowdConfig
|
32 |
+
from seacrowd.utils.constants import Licenses, Tasks
|
33 |
+
|
34 |
+
_CITATION = """\
|
35 |
+
@inproceedings{imperial-kochmar-2023-automatic,
|
36 |
+
title = "Automatic Readability Assessment for Closely Related Languages",
|
37 |
+
author = "Imperial, Joseph Marvin and
|
38 |
+
Kochmar, Ekaterina",
|
39 |
+
editor = "Rogers, Anna and
|
40 |
+
Boyd-Graber, Jordan and
|
41 |
+
Okazaki, Naoaki",
|
42 |
+
booktitle = "Findings of the Association for Computational Linguistics: ACL 2023",
|
43 |
+
month = jul,
|
44 |
+
year = "2023",
|
45 |
+
address = "Toronto, Canada",
|
46 |
+
publisher = "Association for Computational Linguistics",
|
47 |
+
url = "https://aclanthology.org/2023.findings-acl.331",
|
48 |
+
doi = "10.18653/v1/2023.findings-acl.331",
|
49 |
+
pages = "5371--5386",
|
50 |
+
abstract = "In recent years, the main focus of research on automatic readability assessment (ARA) \
|
51 |
+
has shifted towards using expensive deep learning-based methods with the primary goal of increasing models{'} accuracy. \
|
52 |
+
This, however, is rarely applicable for low-resource languages where traditional handcrafted features are still \
|
53 |
+
widely used due to the lack of existing NLP tools to extract deeper linguistic representations. In this work, \
|
54 |
+
we take a step back from the technical component and focus on how linguistic aspects such as mutual intelligibility \
|
55 |
+
or degree of language relatedness can improve ARA in a low-resource setting. We collect short stories written in three \
|
56 |
+
languages in the Philippines{---}Tagalog, Bikol, and Cebuano{---}to train readability assessment models and explore the \
|
57 |
+
interaction of data and features in various cross-lingual setups. Our results show that the inclusion of CrossNGO, \
|
58 |
+
a novel specialized feature exploiting n-gram overlap applied to languages with high mutual intelligibility, \
|
59 |
+
significantly improves the performance of ARA models compared to the use of off-the-shelf large multilingual \
|
60 |
+
language models alone. Consequently, when both linguistic representations are combined, we achieve state-of-the-art \
|
61 |
+
results for Tagalog and Cebuano, and baseline scores for ARA in Bikol.",
|
62 |
+
}
|
63 |
+
"""
|
64 |
+
|
65 |
+
_DATASETNAME = "ara_close"
|
66 |
+
|
67 |
+
_DESCRIPTION = """\
|
68 |
+
The dataset contribution of this study is a compilation of short fictional stories \
|
69 |
+
written in Bikol for readability assessment. The data was combined other collected \
|
70 |
+
Philippine language corpora, such as Tagalog and Cebuano. The data from these languages \
|
71 |
+
are all distributed across the Philippine elementary system's first three grade \
|
72 |
+
levels (L1, L2, L3). We sourced this dataset from Let's Read Asia (LRA), Bloom Library, \
|
73 |
+
Department of Education, and Adarna House. \
|
74 |
+
"""
|
75 |
+
|
76 |
+
_HOMEPAGE = "https://github.com/imperialite/ara-close-lang"
|
77 |
+
|
78 |
+
_LANGUAGES = ["bcl", "ceb"] # We follow ISO639-3 language code (https://iso639-3.sil.org/code_tables/639/data)
|
79 |
+
|
80 |
+
_LICENSE = Licenses.CC_BY_4_0.value # example: Licenses.MIT.value, Licenses.CC_BY_NC_SA_4_0.value, Licenses.UNLICENSE.value, Licenses.UNKNOWN.value
|
81 |
+
|
82 |
+
_LOCAL = False
|
83 |
+
|
84 |
+
_URLS = {
|
85 |
+
"bcl": "https://raw.githubusercontent.com/imperialite/ara-close-lang/main/data/bikol/bik_all_data.txt",
|
86 |
+
# 'tgl': '', # file for tgl language was deleted
|
87 |
+
"ceb": "https://raw.githubusercontent.com/imperialite/ara-close-lang/main/data/cebuano/ceb_all_data.txt",
|
88 |
+
}
|
89 |
+
|
90 |
+
_SUPPORTED_TASKS = [Tasks.READABILITY_ASSESSMENT]
|
91 |
+
|
92 |
+
_SOURCE_VERSION = "1.0.0"
|
93 |
+
|
94 |
+
_SEACROWD_VERSION = "2024.06.20"
|
95 |
+
|
96 |
+
|
97 |
+
class AraCloseDataset(datasets.GeneratorBasedBuilder):
|
98 |
+
f"""{_DESCRIPTION}"""
|
99 |
+
|
100 |
+
SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
|
101 |
+
SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION)
|
102 |
+
|
103 |
+
BUILDER_CONFIGS = [SEACrowdConfig(name=f"{_DATASETNAME}_{lang}_source", version=datasets.Version(_SOURCE_VERSION), description=f"{_DATASETNAME} source schema", schema="source", subset_id=f"{_DATASETNAME}",) for lang in _LANGUAGES] + [
|
104 |
+
SEACrowdConfig(
|
105 |
+
name=f"{_DATASETNAME}_{lang}_seacrowd_text",
|
106 |
+
version=datasets.Version(_SEACROWD_VERSION),
|
107 |
+
description=f"{_DATASETNAME} SEACrowd schema",
|
108 |
+
schema="seacrowd_text",
|
109 |
+
subset_id=f"{_DATASETNAME}",
|
110 |
+
)
|
111 |
+
for lang in _LANGUAGES
|
112 |
+
]
|
113 |
+
|
114 |
+
DEFAULT_CONFIG_NAME = f"{_DATASETNAME}_source"
|
115 |
+
|
116 |
+
def _info(self) -> datasets.DatasetInfo:
|
117 |
+
|
118 |
+
if self.config.schema == "source":
|
119 |
+
features = datasets.Features(
|
120 |
+
{
|
121 |
+
"title": datasets.Value("string"),
|
122 |
+
"text": datasets.Value("string"),
|
123 |
+
"label": datasets.Value("string"),
|
124 |
+
}
|
125 |
+
)
|
126 |
+
|
127 |
+
elif self.config.schema == "seacrowd_text":
|
128 |
+
features = schemas.text_features(["1", "2", "3"])
|
129 |
+
|
130 |
+
return datasets.DatasetInfo(
|
131 |
+
description=_DESCRIPTION,
|
132 |
+
features=features,
|
133 |
+
homepage=_HOMEPAGE,
|
134 |
+
license=_LICENSE,
|
135 |
+
citation=_CITATION,
|
136 |
+
)
|
137 |
+
|
138 |
+
def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
|
139 |
+
"""Returns SplitGenerators."""
|
140 |
+
|
141 |
+
lang = self.config.name.split("_")[2]
|
142 |
+
if lang in _LANGUAGES:
|
143 |
+
data_path = Path(dl_manager.download_and_extract(_URLS[lang]))
|
144 |
+
else:
|
145 |
+
data_path = [Path(dl_manager.download_and_extract(_URLS[lang])) for lang in _LANGUAGES]
|
146 |
+
|
147 |
+
return [
|
148 |
+
datasets.SplitGenerator(
|
149 |
+
name=datasets.Split.TRAIN,
|
150 |
+
gen_kwargs={
|
151 |
+
"filepath": data_path,
|
152 |
+
"split": "train",
|
153 |
+
},
|
154 |
+
)
|
155 |
+
]
|
156 |
+
|
157 |
+
def _generate_examples(self, filepath: Path, split: str) -> Tuple[int, Dict]:
|
158 |
+
"""Yields examples as (key, example) tuples."""
|
159 |
+
lang = self.config.name.split("_")[2]
|
160 |
+
if lang in _LANGUAGES:
|
161 |
+
file_content = open(filepath, "r").readlines()
|
162 |
+
else:
|
163 |
+
file_content = []
|
164 |
+
for path in filepath:
|
165 |
+
lines = open(path, "r").readlines()
|
166 |
+
file_content.extend(lines)
|
167 |
+
|
168 |
+
if self.config.schema == "source":
|
169 |
+
idx = 0
|
170 |
+
for line in file_content:
|
171 |
+
split_data = line.strip().split(",")
|
172 |
+
title = split_data[0]
|
173 |
+
label = split_data[1]
|
174 |
+
text = ",".join(split_data[2:])
|
175 |
+
ex = {"title": title, "text": text, "label": label}
|
176 |
+
yield idx, ex
|
177 |
+
idx += 1
|
178 |
+
|
179 |
+
elif self.config.schema == "seacrowd_text":
|
180 |
+
idx = 0
|
181 |
+
for line in file_content:
|
182 |
+
split_data = line.strip().split(",")
|
183 |
+
title = split_data[0]
|
184 |
+
label = split_data[1]
|
185 |
+
text = ",".join(split_data[2:])
|
186 |
+
ex = {
|
187 |
+
"id": idx,
|
188 |
+
"text": text,
|
189 |
+
"label": label,
|
190 |
+
}
|
191 |
+
yield idx, ex
|
192 |
+
idx += 1
|
193 |
+
else:
|
194 |
+
raise ValueError(f"Invalid config: {self.config.name}")
|