Datasets:

Modalities:
Text
Formats:
json
ArXiv:
Libraries:
Datasets
pandas
License:
cointegrated commited on
Commit
642a0ea
1 Parent(s): a7bb7e6

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +35 -22
README.md CHANGED
@@ -29,16 +29,16 @@ task_ids: []
29
 
30
  ## Dataset Description
31
 
32
- - **Homepage:**
33
- - **Repository:**
34
- - **Paper:**
35
- - **Leaderboard:**
36
- - **Point of Contact:**
37
 
38
  ### Dataset Summary
39
 
40
  This is a corpus of parallel Erzya-Russian words, phrases and sentences, collected in the paper "The first neural machine translation system for the Erzya language".
41
 
 
 
42
  The corpus consists of the following parts:
43
 
44
  | name | size | composition |
@@ -81,7 +81,7 @@ data_extended = load_dataset(
81
 
82
  ### Languages
83
 
84
- The main part of the dataset (`train`, `dev` and `test`) consists of "natural" Erzya and Russian sentences, translated to the other language by humans. There is also a larger Erzya-only part of the corpus (`mono`), translated to Russian automatically.
85
 
86
  ## Dataset Structure
87
 
@@ -102,33 +102,46 @@ All data instances have three string fields: `myv`, `ru` and `src` (the last one
102
 
103
  ### Data Splits
104
 
105
- [More Information Needed]
 
 
 
106
 
107
  ## Dataset Creation
108
 
109
  ### Curation Rationale
110
 
111
- [More Information Needed]
 
112
 
113
  ### Source Data
114
 
115
  #### Initial Data Collection and Normalization
116
 
117
- [More Information Needed]
 
 
 
 
 
118
 
119
  #### Who are the source language producers?
120
 
121
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
122
 
123
  ### Annotations
124
 
125
- #### Annotation process
126
-
127
- [More Information Needed]
128
-
129
- #### Who are the annotators?
130
-
131
- No human annotation was involved in data collection.
132
 
133
  ### Personal and Sensitive Information
134
 
@@ -139,7 +152,7 @@ However, some sentences collected, for example, from news articles or LiveJourna
139
 
140
  ### Social Impact of Dataset
141
 
142
- [More Information Needed]
143
 
144
  ### Discussion of Biases
145
 
@@ -149,18 +162,18 @@ because they appear in the corresponding Wikipedias on similar positions.
149
 
150
  ### Other Known Limitations
151
 
152
- [More Information Needed]
153
 
154
  ## Additional Information
155
 
156
  ### Dataset Curators
157
 
158
- [More Information Needed]
159
 
160
  ### Licensing Information
161
 
162
- [More Information Needed]
163
 
164
  ### Citation Information
165
 
166
- [More Information Needed]
 
29
 
30
  ## Dataset Description
31
 
32
+ - **Repository:** https://github.com/slone-nlp/myv-nmt
33
+ - **Paper:**: [TBD]
34
+ - **Point of Contact:** @cointegrated
 
 
35
 
36
  ### Dataset Summary
37
 
38
  This is a corpus of parallel Erzya-Russian words, phrases and sentences, collected in the paper "The first neural machine translation system for the Erzya language".
39
 
40
+ Erzya (`myv`) is a language from the Uralic family. It is spoken primarily in the Republic of Mordovia and some other regions of Russia and other post-Soviet countries. We use the Cyrillic version of its script.
41
+
42
  The corpus consists of the following parts:
43
 
44
  | name | size | composition |
 
81
 
82
  ### Languages
83
 
84
+ The main part of the dataset (`train`, `dev` and `test`) consists of "natural" Erzya (Cyrillic) and Russian sentences, translated to the other language by humans. There is also a larger Erzya-only part of the corpus (`mono`), translated to Russian automatically.
85
 
86
  ## Dataset Structure
87
 
 
102
 
103
  ### Data Splits
104
 
105
+ - train: parallel sentences, words and phrases, collected from various sources. Most of them are aligned automatically. Noisy.
106
+ - dev: 1500 parallel sentences, selected from the 6 most reliable and diverse sources.
107
+ - test: same as dev.
108
+ - mono: Erzya sentences collected from various sources, with the Russian counterpart generated by a neural machine translation model.
109
 
110
  ## Dataset Creation
111
 
112
  ### Curation Rationale
113
 
114
+ This is, as far as we know, the first publicly available parallel Russian-Erzya corpus, and the first medium-sized translation corpus for Erzya.
115
+ We hope that it sets a meaningful baseline for Erzya machine translation.
116
 
117
  ### Source Data
118
 
119
  #### Initial Data Collection and Normalization
120
 
121
+ The dataset was collected from various sources (see below).
122
+
123
+ The texts were spit in sentences using the [razdel]() package.
124
+ For some sources, sentences were filtered by language using the [slone/fastText-LID-323](https://huggingface.co/slone/fastText-LID-323) model.
125
+ For most of the sources, `myv` and `ru` sentences were aligned automatically using the [slone/LaBSE-en-ru-myv-v1](https://huggingface.co/slone/LaBSE-en-ru-myv-v1) sentence encoder
126
+ and the code from [the paper repository](https://github.com/slone-nlp/myv-nmt).
127
 
128
  #### Who are the source language producers?
129
 
130
+ The dataset comprises parallel `myv-ru` and monolingual `myv` texts from diverse sources:
131
+ - 12K parallel sentences from the Bible (http://finugorbib.com);
132
+ - 3K parallel Wikimedia sentences from OPUS;
133
+ - 42K parallel words or short phrases collected from various online dictionaries ();
134
+ - the Erzya Wikipedia and the corresponding articles from the Russian Wikipedia;
135
+ - 18 books, including 3 books with Erzya-Russian bitexts (http://lib.e-mordovia.ru);
136
+ - Soviet-time books and periodicals (https://fennougrica.kansalliskirjasto.fi);
137
+ - The Erzya part of Wikisource (https://wikisource.org/wiki/Main_Page/?oldid=895127);
138
+ - Short texts by modern Erzya authors (https://rus4all.ru/myv/);
139
+ - News articles from the Erzya Pravda website (http://erziapr.ru);
140
+ - Texts found in LiveJournal (https://www.livejournal.com) by searching with the 100 most frequent Erzya words.
141
 
142
  ### Annotations
143
 
144
+ No human annotation was involved in the data collection.
 
 
 
 
 
 
145
 
146
  ### Personal and Sensitive Information
147
 
 
152
 
153
  ### Social Impact of Dataset
154
 
155
+ Publication of this dataset may attract some attention to the endangered Erzya language.
156
 
157
  ### Discussion of Biases
158
 
 
162
 
163
  ### Other Known Limitations
164
 
165
+ The dataset is noisy: some texts in it may be ungrammatical, in a wrong language, or poorly aligned.
166
 
167
  ## Additional Information
168
 
169
  ### Dataset Curators
170
 
171
+ The data was collected by David Dale (https://huggingface.co/cointegrated).
172
 
173
  ### Licensing Information
174
 
175
+ The status of the dataset is not final, but after we check everything, we hope to be able to distribute it under the [CC-BY-SA license](http://creativecommons.org/licenses/by-sa/4.0/).
176
 
177
  ### Citation Information
178
 
179
+ [TBD]