yanaiela commited on
Commit
55ce402
1 Parent(s): 7ad28e5

Adding Text-based NP Enrichment (TNE) dataset (#4153)

Browse files

* Adding Text-based NP Enrichment (TNE) dataset

* Tests fix: explicit file encoding + yaml format in readme

* contribution section in the readme

* Update datasets/tne/README.md

Co-authored-by: Quentin Lhoest <[email protected]>

* Update datasets/tne/README.md

Co-authored-by: Quentin Lhoest <[email protected]>

* Update datasets/tne/README.md

Co-authored-by: Quentin Lhoest <[email protected]>

* Update datasets/tne/README.md

Co-authored-by: Quentin Lhoest <[email protected]>

* Update datasets/tne/README.md

Co-authored-by: Quentin Lhoest <[email protected]>

* converting to a single config + updated preposition labels. Moving to single config instead of one for train and one for test. Using an empty list in case of test files, that do not contain the np links labels publicly available. The preposition labels is now using the ClassLabel, with a predefined list of 24 possible values

* Updating the number of documents

Co-authored-by: Quentin Lhoest <[email protected]>

* Update datasets/tne/README.md

Co-authored-by: Quentin Lhoest <[email protected]>

* fixed remaining of previous separation between labaled/unlabaled

Co-authored-by: Quentin Lhoest <[email protected]>

Commit from https://github.com/huggingface/datasets/commit/329781e193dfc373c875d3dc9f8b91644d2b27b1

Files changed (4) hide show
  1. README.md +188 -0
  2. dataset_infos.json +1 -0
  3. dummy/1.1.0/dummy_data.zip +3 -0
  4. tne.py +210 -0
README.md ADDED
@@ -0,0 +1,188 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ annotations_creators:
3
+ - crowdsourced
4
+ language_creators:
5
+ - found
6
+ languages:
7
+ - en
8
+ licenses:
9
+ - mit
10
+ multilinguality:
11
+ - monolingual
12
+ pretty_name: Text-based NP Enrichment
13
+ size_categories:
14
+ - 1K<n<10K
15
+ source_datasets:
16
+ - original
17
+ task_categories:
18
+ - text-retrieval
19
+ task_ids:
20
+ - document-retrieval
21
+ ---
22
+
23
+ # Dataset Card for Text-based NP Enrichment
24
+
25
+ ## Table of Contents
26
+ - [Dataset Description](#dataset-description)
27
+ - [Dataset Summary](#dataset-summary)
28
+ - [Supported Tasks](#supported-tasks-and-leaderboards)
29
+ - [Languages](#languages)
30
+ - [Dataset Structure](#dataset-structure)
31
+ - [Data Instances](#data-instances)
32
+ - [Data Fields](#data-instances)
33
+ - [Data Splits](#data-instances)
34
+ - [Dataset Creation](#dataset-creation)
35
+ - [Curation Rationale](#curation-rationale)
36
+ - [Source Data](#source-data)
37
+ - [Annotations](#annotations)
38
+ - [Personal and Sensitive Information](#personal-and-sensitive-information)
39
+ - [Considerations for Using the Data](#considerations-for-using-the-data)
40
+ - [Social Impact of Dataset](#social-impact-of-dataset)
41
+ - [Discussion of Biases](#discussion-of-biases)
42
+ - [Other Known Limitations](#other-known-limitations)
43
+ - [Additional Information](#additional-information)
44
+ - [Dataset Curators](#dataset-curators)
45
+ - [Licensing Information](#licensing-information)
46
+ - [Citation Information](#citation-information)
47
+
48
+ ## Dataset Description
49
+
50
+ - **Homepage:** https://yanaiela.github.io/TNE/
51
+ - **Repository:** https://github.com/yanaiela/TNE
52
+ - **Paper:** https://arxiv.org/abs/2109.12085
53
+ - **Leaderboard:** [TNE](https://leaderboard.allenai.org/tne/submissions/public)
54
+ [TNE OOD](https://leaderboard.allenai.org/tne-ood/submissions/public)
55
+ - **Point of Contact:** [Yanai Elazar](mailto:[email protected])
56
+
57
+ ### Dataset Summary
58
+
59
+ Text-based NP Enrichment (TNE) is a natural language understanding (NLU) task, which focus on relations between noun phrases (NPs) that can be mediated via prepositions. The dataset contains 5,497 documents, annotated exhaustively with all possible links between the NPs in each document.
60
+
61
+ The main data comes from WikiNews, which is used for train/dev/test. We also collected an additional set of 509 documents to serve as out of distribution (OOD) data points, from the Book Corpus, IMDB reviews and Reddit.
62
+
63
+ ### Supported Tasks and Leaderboards
64
+
65
+ [Needs More Information]
66
+
67
+ ### Languages
68
+
69
+ The text in the dataset is in English, as spoken in the different domains we include. The associated BCP-47 code is `en`.
70
+
71
+ ## Dataset Structure
72
+
73
+ ### Data Instances
74
+
75
+ The original files are in a jsonl format, containing a dictionary of a single document, in each line.
76
+ Each document contain a different amount of labels, due to the different amount of NPs.
77
+ The test and ood splits come without the annotated labels.
78
+
79
+ ### Data Fields
80
+
81
+ A document consists of:
82
+
83
+ * `id`: a unique identifier of a document, beginning with `r` and followed by a number
84
+ * `text`: the text of the document. The title and subtitles (if exists) are separated with two new lines. The paragraphs
85
+ are separated by a single new line.
86
+ * `tokens`: a list of string, containing the tokenized tokens
87
+ * `nps`: a list of dictionaries, containing the following entries:
88
+ * `text`: the text of the np
89
+ * `start_index`: an integer indicating the starting index in the text
90
+ * `end_index`: an integer indicating the ending index in the text
91
+ * `start_token`: an integer indicating the first token of the np out of the tokenized tokens
92
+ * `end_token`: an integer indicating the last token of the np out of the tokenized tokens
93
+ * `id`: the id of the np
94
+ * `np_relations`: these are the relation labels of the document. It is a list of dictionaries, where each
95
+ dictionary contains:
96
+ * `anchor`: the id of the anchor np
97
+ * `complement`: the id of the complement np
98
+ * `preposition`: the preposition that links between the anchor and the complement. This can take one out of 24 pre-defined preposition (23 + member(s)-of)
99
+ * `complement_coref_cluster_id`: the coreference id, which the complement is part of.
100
+ * `coref`: the coreference labels. It contains a list of dictionaries, where each dictionary contains:
101
+ * `id`: the id of the coreference cluster
102
+ * `members`: the ids of the nps members of such cluster
103
+ * `np_type`: the type of cluster. It can be either
104
+ * `standard`: regular coreference cluster
105
+ * `time/date/measurement`: a time / date / measurement np. These will be singletons.
106
+ * `idiomatic`: an idiomatic expression
107
+ * `metadata`: metadata of the document. It contains the following:
108
+ * `annotators`: a dictionary with anonymized annotators id
109
+ * `coref_worker`: the coreference worker id
110
+ * `consolidator_worker`: the consolidator worker id
111
+ * `np-relations_worker`: the np relations worker id
112
+ * `url`: the url where the document was taken from (not always existing)
113
+ * `source`: the original file name where the document was taken from
114
+
115
+
116
+ ### Data Splits
117
+
118
+ The dataset is spread across four files, for the four different splits: train, dev, test and test_ood.
119
+ Additional details on the data statistics can be found in the [paper](https://arxiv.org/abs/2109.12085)
120
+
121
+ ## Dataset Creation
122
+
123
+ ### Curation Rationale
124
+
125
+ TNE was build as a new task for language understanding, focusing on extracting relations between nouns, moderated by prepositions.
126
+
127
+ ### Source Data
128
+
129
+ #### Initial Data Collection and Normalization
130
+
131
+ [Needs More Information]
132
+
133
+ #### Who are the source language producers?
134
+
135
+ [Needs More Information]
136
+
137
+ ### Annotations
138
+
139
+ #### Annotation process
140
+
141
+ [Needs More Information]
142
+
143
+ #### Who are the annotators?
144
+
145
+ [Needs More Information]
146
+
147
+ ### Personal and Sensitive Information
148
+
149
+ [Needs More Information]
150
+
151
+ ## Considerations for Using the Data
152
+
153
+ ### Social Impact of Dataset
154
+
155
+ [Needs More Information]
156
+
157
+ ### Discussion of Biases
158
+
159
+ [Needs More Information]
160
+
161
+ ### Other Known Limitations
162
+
163
+ [Needs More Information]
164
+
165
+ ## Additional Information
166
+
167
+ ### Dataset Curators
168
+
169
+ The dataset was created by Yanai Elazar, Victoria Basmov, Yoav Goldberg, Reut Tsarfaty, during work done at Bar-Ilan University, and AI2.
170
+
171
+ ### Licensing Information
172
+
173
+ The data is released under the MIT license.
174
+
175
+ ### Citation Information
176
+
177
+ ```bibtex
178
+ @article{tne,
179
+ author = {Elazar, Yanai and Basmov, Victoria and Goldberg, Yoav and Tsarfaty, Reut},
180
+ title = "{Text-based NP Enrichment}",
181
+ journal = {Transactions of the Association for Computational Linguistics},
182
+ year = {2022},
183
+ }
184
+ ```
185
+
186
+ ### Contributions
187
+
188
+ Thanks to [@yanaiela](https://github.com/yanaiela), who is also the first author of the paper, for adding this dataset.
dataset_infos.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"default": {"description": "TNE is an NLU task, which focus on relations between noun phrases (NPs) that can be mediated via prepositions.\nThe dataset contains 5,497 documents, annotated exhaustively with all possible links between the NPs in each document.\n", "citation": "@article{tne,\n author = {Elazar, Yanai and Basmov, Victoria and Goldberg, Yoav and Tsarfaty, Reut},\n title = \"{Text-based NP Enrichment}\",\n journal = {Transactions of the Association for Computational Linguistics},\n year = {2022},\n}\n", "homepage": "https://yanaiela.github.io/TNE/", "license": "MIT", "features": {"id": {"dtype": "string", "id": null, "_type": "Value"}, "text": {"dtype": "string", "id": null, "_type": "Value"}, "tokens": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "nps": [{"text": {"dtype": "string", "id": null, "_type": "Value"}, "first_char": {"dtype": "int32", "id": null, "_type": "Value"}, "last_char": {"dtype": "int32", "id": null, "_type": "Value"}, "first_token": {"dtype": "int32", "id": null, "_type": "Value"}, "last_token": {"dtype": "int32", "id": null, "_type": "Value"}, "id": {"dtype": "string", "id": null, "_type": "Value"}}], "np_relations": [{"anchor": {"dtype": "string", "id": null, "_type": "Value"}, "complement": {"dtype": "string", "id": null, "_type": "Value"}, "preposition": {"num_classes": 24, "names": ["about", "for", "with", "from", "among", "by", "on", "at", "during", "of", "member(s) of", "in", "after", "under", "to", "into", "before", "near", "outside", "around", "between", "against", "over", "inside"], "id": null, "_type": "ClassLabel"}, "complement_coref_cluster_id": {"dtype": "string", "id": null, "_type": "Value"}}], "coref": [{"id": {"dtype": "string", "id": null, "_type": "Value"}, "members": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "np_type": {"num_classes": 3, "names": ["standard", "time/date/measurement", "idiomatic"], "id": null, "_type": "ClassLabel"}}], "metadata": {"annotators": {"coref_worker": {"dtype": "int32", "id": null, "_type": "Value"}, "consolidator_worker": {"dtype": "int32", "id": null, "_type": "Value"}, "np-relations_worker": {"feature": {"dtype": "int32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}}, "url": {"dtype": "string", "id": null, "_type": "Value"}, "source": {"dtype": "string", "id": null, "_type": "Value"}}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "tne_dataset", "config_name": "default", "version": {"version_str": "1.1.0", "description": null, "major": 1, "minor": 1, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 41308170, "num_examples": 3988, "dataset_name": "tne_dataset"}, "validation": {"name": "validation", "num_bytes": 5495419, "num_examples": 500, "dataset_name": "tne_dataset"}, "test": {"name": "test", "num_bytes": 2203716, "num_examples": 500, "dataset_name": "tne_dataset"}, "test_ood": {"name": "test_ood", "num_bytes": 2249352, "num_examples": 509, "dataset_name": "tne_dataset"}}, "download_checksums": {"https://github.com/yanaiela/TNE/raw/main/data/train-v1.1.jsonl.gz": {"num_bytes": 11135145, "checksum": "d616372e92e6334dfddaaca3c7898f7aeed7b8827c9821aa555ad1d1a8e8eb1c"}, "https://github.com/yanaiela/TNE/raw/main/data/dev-v1.1.jsonl.gz": {"num_bytes": 1466591, "checksum": "3f3e466330908270ee742725c6b68f92b3243b988aa8b8c9bfef69ad8e68fe14"}, "https://github.com/yanaiela/TNE/raw/main/data/test_unlabeled-v1.1.jsonl.gz": {"num_bytes": 781824, "checksum": "a461ef30beba25feefbb1777ecb42e63a1ce20f3675e2c551a442c393f4118c7"}, "https://github.com/yanaiela/TNE/raw/main/data/ood_unlabeled-v1.1.jsonl.gz": {"num_bytes": 811018, "checksum": "9278c13c381f1f3b6623e599086984695c00e31840c121c2053d3f466f2d2170"}}, "download_size": 14194578, "post_processing_size": null, "dataset_size": 51256657, "size_in_bytes": 65451235}}
dummy/1.1.0/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d4185db56411091e0247d7598cec73fe636ca36d6a0db710d1d9aade6e871687
3
+ size 51537
tne.py ADDED
@@ -0,0 +1,210 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
2
+ #
3
+ # Licensed under the Apache License, Version 2.0 (the "License");
4
+ # you may not use this file except in compliance with the License.
5
+ # You may obtain a copy of the License at
6
+ #
7
+ # http://www.apache.org/licenses/LICENSE-2.0
8
+ #
9
+ # Unless required by applicable law or agreed to in writing, software
10
+ # distributed under the License is distributed on an "AS IS" BASIS,
11
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12
+ # See the License for the specific language governing permissions and
13
+ # limitations under the License.
14
+ """TNE: Text-based NP Enrichment"""
15
+
16
+ import json
17
+
18
+ import datasets
19
+
20
+
21
+ # Find for instance the citation on arxiv or on the dataset repo/website
22
+ _CITATION = """\
23
+ @article{tne,
24
+ author = {Elazar, Yanai and Basmov, Victoria and Goldberg, Yoav and Tsarfaty, Reut},
25
+ title = "{Text-based NP Enrichment}",
26
+ journal = {Transactions of the Association for Computational Linguistics},
27
+ year = {2022},
28
+ }
29
+ """
30
+
31
+ # You can copy an official description
32
+ _DESCRIPTION = """\
33
+ TNE is an NLU task, which focus on relations between noun phrases (NPs) that can be mediated via prepositions.
34
+ The dataset contains 5,497 documents, annotated exhaustively with all possible links between the NPs in each document.
35
+ """
36
+
37
+ _HOMEPAGE = "https://yanaiela.github.io/TNE/"
38
+
39
+ _LICENSE = "MIT"
40
+
41
+ _VERSION = "v1.1"
42
+
43
+ # The HuggingFace Datasets library doesn't host the datasets but only points to the original files.
44
+ # This can be an arbitrary nested dict/list of URLs (see below in `_split_generators` method)
45
+ _URL = "https://github.com/yanaiela/TNE/raw/main/data/"
46
+ _URLS = {
47
+ "train": _URL + f"train-{_VERSION}.jsonl.gz",
48
+ "dev": _URL + f"dev-{_VERSION}.jsonl.gz",
49
+ "test_unlabeled": _URL + f"test_unlabeled-{_VERSION}.jsonl.gz",
50
+ "ood_unlabeled": _URL + f"ood_unlabeled-{_VERSION}.jsonl.gz",
51
+ }
52
+
53
+
54
+ class TNEDataset(datasets.GeneratorBasedBuilder):
55
+ """TNE: Text-based NP Enrichment"""
56
+
57
+ VERSION = datasets.Version("1.1.0")
58
+
59
+ def _info(self):
60
+ features = datasets.Features(
61
+ {
62
+ "id": datasets.Value("string"),
63
+ "text": datasets.Value("string"),
64
+ "tokens": datasets.Sequence(datasets.Value("string")),
65
+ "nps": [
66
+ {
67
+ "text": datasets.Value("string"),
68
+ "first_char": datasets.Value("int32"),
69
+ "last_char": datasets.Value("int32"),
70
+ "first_token": datasets.Value("int32"),
71
+ "last_token": datasets.Value("int32"),
72
+ "id": datasets.Value("string"),
73
+ }
74
+ ],
75
+ "np_relations": [
76
+ {
77
+ "anchor": datasets.Value("string"),
78
+ "complement": datasets.Value("string"),
79
+ "preposition": datasets.features.ClassLabel(
80
+ names=[
81
+ "about",
82
+ "for",
83
+ "with",
84
+ "from",
85
+ "among",
86
+ "by",
87
+ "on",
88
+ "at",
89
+ "during",
90
+ "of",
91
+ "member(s) of",
92
+ "in",
93
+ "after",
94
+ "under",
95
+ "to",
96
+ "into",
97
+ "before",
98
+ "near",
99
+ "outside",
100
+ "around",
101
+ "between",
102
+ "against",
103
+ "over",
104
+ "inside",
105
+ ]
106
+ ),
107
+ "complement_coref_cluster_id": datasets.Value("string"),
108
+ }
109
+ ],
110
+ "coref": [
111
+ {
112
+ "id": datasets.Value("string"),
113
+ "members": datasets.Sequence(datasets.Value("string")),
114
+ "np_type": datasets.features.ClassLabel(
115
+ names=[
116
+ "standard",
117
+ "time/date/measurement",
118
+ "idiomatic",
119
+ ]
120
+ ),
121
+ }
122
+ ],
123
+ "metadata": {
124
+ "annotators": {
125
+ "coref_worker": datasets.Value("int32"),
126
+ "consolidator_worker": datasets.Value("int32"),
127
+ "np-relations_worker": datasets.Sequence(datasets.Value("int32")),
128
+ },
129
+ "url": datasets.Value("string"),
130
+ "source": datasets.Value("string"),
131
+ },
132
+ }
133
+ )
134
+
135
+ return datasets.DatasetInfo(
136
+ # This is the description that will appear on the datasets page.
137
+ description=_DESCRIPTION,
138
+ # This defines the different columns of the dataset and their types
139
+ features=features, # Here we define them above because they are different between the two configurations
140
+ # If there's a common (input, target) tuple from the features, uncomment supervised_keys line below and
141
+ # specify them. They'll be used if as_supervised=True in builder.as_dataset.
142
+ # supervised_keys=("sentence", "label"),
143
+ # Homepage of the dataset for documentation
144
+ homepage=_HOMEPAGE,
145
+ # License for the dataset if available
146
+ license=_LICENSE,
147
+ # Citation for the dataset
148
+ citation=_CITATION,
149
+ )
150
+
151
+ def _split_generators(self, dl_manager):
152
+ urls = _URLS
153
+ data_dir = dl_manager.download_and_extract(urls)
154
+
155
+ return [
156
+ datasets.SplitGenerator(
157
+ name=datasets.Split.TRAIN,
158
+ # These kwargs will be passed to _generate_examples
159
+ gen_kwargs={
160
+ "filepath": data_dir["train"],
161
+ "split": "train",
162
+ },
163
+ ),
164
+ datasets.SplitGenerator(
165
+ name=datasets.Split.VALIDATION,
166
+ # These kwargs will be passed to _generate_examples
167
+ gen_kwargs={
168
+ "filepath": data_dir["dev"],
169
+ "split": "dev",
170
+ },
171
+ ),
172
+ datasets.SplitGenerator(
173
+ name=datasets.Split.TEST,
174
+ # These kwargs will be passed to _generate_examples
175
+ gen_kwargs={"filepath": data_dir["test_unlabeled"], "split": "test_unlabeled"},
176
+ ),
177
+ datasets.SplitGenerator(
178
+ name=datasets.Split("test_ood"),
179
+ # These kwargs will be passed to _generate_examples
180
+ gen_kwargs={"filepath": data_dir["ood_unlabeled"], "split": "ood_unlabeled"},
181
+ ),
182
+ ]
183
+
184
+ # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
185
+ def _generate_examples(self, filepath, split):
186
+ # The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.
187
+ with open(filepath, "r", encoding="utf-8") as f:
188
+ for key, row in enumerate(f):
189
+ data = json.loads(row)
190
+
191
+ ex_id = data["id"]
192
+ text = data["text"]
193
+ tokens = data["tokens"]
194
+ nps = data["nps"]
195
+ if split in ["test_unlabeled", "ood_unlabeled"]:
196
+ np_relations = []
197
+ else:
198
+ np_relations = data["np_relations"]
199
+ coref = data["coref"]
200
+ metadata = data["metadata"]
201
+
202
+ yield key, {
203
+ "id": ex_id,
204
+ "text": text,
205
+ "tokens": tokens,
206
+ "nps": nps,
207
+ "np_relations": np_relations,
208
+ "coref": coref,
209
+ "metadata": metadata,
210
+ }