update

README.md

@@ -53,6 +53,8 @@ Tips:
 | para_pat | [ParaPat](https://aclanthology.org/2020.lrec-1.465.pdf); [Homepage](https://figshare.com/articles/dataset/ParaPat_The_Multi-Million_Sentences_Parallel_Corpus_of_Patents_Abstracts/12627632) | TRAIN: 10242500 | ParaPat：专利摘要的数百万个句子平行语料库 | [para_pat](https://huggingface.co/datasets/para_pat) |
 | pib | [CVIT-PIB](https://arxiv.org/abs/2008.04860) |  | 该数据集是 11 种印度语言的大规模句子对齐语料库，即： CVIT-PIB 语料库是印度语言可用的最大多语言语料库。 | [pib](https://huggingface.co/datasets/pib) |
 | poleval2019_mt |  |  | PolEval 是一项受 SemEval 启发的波兰语自然语言处理工具评估活动。 | [poleval2019_mt](https://huggingface.co/datasets/poleval2019_mt) |
+| wmt19 | [statmt.org](https://www.statmt.org/wmt19/translation-task.html) | 样本个数 | 我们的目标是尽可能使用公开的数据源。我们的训练数据主要来源是Europarl 语料库、 UN 语料库、新闻评论语料库和 ParaCrawl语料库。我们还发布了单语 新闻抓取语料库。将提供其他特定语言的语料库。 | [wmt/wmt19](https://huggingface.co/datasets/wmt/wmt19) |
+| ro_sts_parallel |  | 样本个数 | 我们提出 RO-STS-Parallel - 通过将 STS 英语数据集翻译成罗马尼亚语而获得的并行罗马尼亚语-英语数据集。 | [ro_sts_parallel](https://huggingface.co/datasets/ro_sts_parallel) |
 
 
 机器翻译
@@ -70,6 +72,38 @@ https://opus.nlpl.eu/
 | para_crawl | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) |  | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
 | php | [PHP](https://opus.nlpl.eu/PHP/corpus/version/PHP) | TRAIN: 44007 | 最初从 http://se.php.net/download-docs.php 中提取的并行语料库。该语料库相当嘈杂。 | [php](https://huggingface.co/datasets/php) |
 | tatoeba | [Tatoeba](https://opus.nlpl.eu/Tatoeba/corpus/version/Tatoeba); [tatoeba](https://tatoeba.org/); [Tatoeba Paper](https://arxiv.org/abs/1812.10464v2) | TRAIN: 702895 | Tatoeba 是句子和翻译的集合。 | [tatoeba](https://huggingface.co/datasets/tatoeba) |
+| qed_amara | [QED](https://opus.nlpl.eu/QED/corpus/version/QED) | 样本个数 |  | [qed_amara](https://huggingface.co/datasets/qed_amara) |
+| setimes | [SETIMES](https://opus.nlpl.eu/SETIMES/corpus/version/SETIMES) | 样本个数 | 英语和东南欧语言平行语料库 | [setimes](https://huggingface.co/datasets/setimes) |
+| spc | [SPC](https://opus.nlpl.eu/SPC/corpus/version/SPC) | 样本个数 |  | [spc](https://huggingface.co/datasets/spc) |
+| tanzil | [Tanzil](https://opus.nlpl.eu/Tanzil/corpus/version/Tanzil) | 样本个数 |  | [tanzil](https://huggingface.co/datasets/tanzil) |
+
+
+机器翻译
+
+https://opus.nlpl.eu/
+
+| 数据 | 原始数据/项目地址 | 样本个数 | 原始数据描述 | 替代数据下载地址 |
+| :--- |    :---:      |   :---: |    :---:    |     :---:     |
+| para_crawl_en_bg | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 1967082 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_cs | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 5601171 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_da | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 4617796 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_de | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 31041474 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_el | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 3799096 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_es (MemoryError) | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | MemoryError | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_et | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 1625870 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_fi | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 4071888 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_fr (MemoryError) | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | MemoryError | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_ga | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 686474 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_hr | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 1911081 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_hu | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 3292718 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_it | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 22718884 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_lt | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 1554000 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_lv | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 1059209 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_mt | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 379616 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_pl | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 6537110 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_pt | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 15186124 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+| para_crawl_en_ro | [ParaCrawl](https://opus.nlpl.eu/ParaCrawl/corpus/version/ParaCrawl); [ParaCrawl](https://aclanthology.org/2020.acl-main.417.pdf) | TRAIN: 3580912 | 欧洲官方语言的网络规模并行语料库。 | [para_crawl](https://huggingface.co/datasets/para_crawl) |
+
 
 
 ### 参考来源

data/para_crawl_en_bg.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c109bf7b905f430233f8038fe1ca006f66033527a8b7c1001be96e671a0c638e
+size 488593740

data/para_crawl_en_cs.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8e8849dc243b163b01a96ceafc9c7d271f07ea9be9f7eed6c4f083a5dbb5fca
+size 1028423849

data/para_crawl_en_da.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b291196df34491ade3212d5899cfc1d5b90e5b26ea548e11170c8156340b6350
+size 921672376

data/para_crawl_en_de.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:443fcd45e0fcd216f19372c08a7965aec08c3ba3061edaca0e6aeed78b7facde
+size 6174854971

data/para_crawl_en_el.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88d8db935b1088940e81d079c6b447a581afa3c4877a6756a9d86e2b0d236785
+size 949899855

data/para_crawl_en_es.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:50c05dda5ae094fb7ffe829c3b035ba42ced5ebe43a8d150e25647baedb4979c
+size 8125092148

data/para_crawl_en_et.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1e0fc5a1074456d55fae1fa8942bad1d347ad44314997938ad3ea51759803ff
+size 314654700

data/para_crawl_en_fi.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b3797f6e391ad19ba7faef910ab8f2ffc9fe03d2a3820277e8a53a55e7ec976
+size 802164519

data/para_crawl_en_ga.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:253967043892e55ff4afb1fb10665b64dd3defa2149ee357e05358107184470a
+size 152261650

data/para_crawl_en_hr.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c36e7cbf09b429cedbef16c68a3e47c50c8abdf22be7e6ea58e6ebb17a71cf0b
+size 382384467

data/para_crawl_en_hu.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:52025e26e42a1430b362b6df8143905d34740623307f11d3c5adfdcf23285152
+size 614589993

data/para_crawl_en_it.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae72cdd6a68623c189552f46fb71b676daed7cc2edb5157e979573185682b13b
+size 4887378831

data/para_crawl_en_lt.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc2c6919af380d8c61b72d2f957420b3b719883604dcd19e25a295020de880a6
+size 303016979

data/para_crawl_en_lv.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d53fcd39cb7135c6ab7c0f452a576c4917aab9c4bf1d88cd71ab1a05e0c01734
+size 216578225

data/para_crawl_en_mt.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad2a8e2de3a94a3bc8c63e14af6edf53c467711dbe39924c4930c06662e08d3f
+size 79926847

data/para_crawl_en_nl.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee3510dc2de4375c1f95c4a650958813cefc1e899ac1631cc4250aa287d2a6a3
+size 2121286142

data/para_crawl_en_pl.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:edbf5552e0ba4ea8c22002a340c570c530bd94e0b2f07e8cb6f8b423f33f6fa7
+size 1304234343

data/para_crawl_en_pt.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63d2f4ff183a05c7c8bdcaaa60195bf63ee7f9329a4e63e8338cbb6eba6f6961
+size 3071571433

data/para_crawl_en_ro.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:972f5b540ae63617e635055215f6a6beb06df8485f98954d76a54773c5ea603e
+size 756706879

examples/load_data/plan_1.py

@@ -1,5 +1,10 @@
 #!/usr/bin/python3
 # -*- coding: utf-8 -*-
+"""
+训练模型后发现它对短句子的识别能力很差。
+当句子长度足够的时候，它能够识别准确，并给出接近1.0 的概率值。
+但对于短句子，识别结果让人难以接受。例如把“你好”识别成 de 德语。
+"""
 import argparse
 import json
 import os

examples/preprocess/preprocess_para_crawl.py

@@ -4,6 +4,7 @@ import argparse
 from collections import defaultdict
 import json
 import os
+from pathlib import Path
 import sys
 
 pwd = os.path.abspath(os.path.dirname(__file__))
@@ -26,7 +27,7 @@ def get_args():
     )
     parser.add_argument(
         "--output_file",
-        default=(project_path / "data/para_crawl.jsonl"),
+        default=(project_path / "data/para_crawl_en_ro.jsonl"),
         type=str
     )
 
@@ -38,8 +39,28 @@ def main():
     args = get_args()
 
     name_list = [
-        "enbg", "encs", "enda", "ende", "enel", "enes", "enet", "enfi", "enfr", "enga", "enhr", "enhu", "enit",
-        "enlt", "enlv", "enmt", "ennl", "enpl", "enpt", "enro", "ensk", "ensl", "ensv"
+        # "enbg",
+        # "encs",
+        # "enda",
+        # "ende",
+        # "enel",
+        # "enes",
+        # "enet",
+        # "enfi",
+        # "enfr",
+        # "enga",
+        # "enhr",
+        # "enhu",
+        # "enit",
+        # "enlt",
+        # "enlv",
+        # "enmt",
+        # "ennl",
+        # "enpl",
+        # "enpt",
+        "enro",
+        # "ensk",
+        # "ensl", "ensv"
     ]
 
     # TODO: 数据集太大，加载不完。

language_identification.py

@@ -30,7 +30,26 @@ _URLS = {
     "nbnn": "data/nbnn.jsonl",
     "nordic_langid": "data/nordic_langid.jsonl",
     "open_subtitles": "data/open_subtitles.jsonl",
-    # "para_crawl": "data/para_crawl.jsonl",
+    "para_crawl_en_bg": "data/para_crawl_en_bg.jsonl",
+    "para_crawl_en_cs": "data/para_crawl_en_cs.jsonl",
+    "para_crawl_en_da": "data/para_crawl_en_da.jsonl",
+    "para_crawl_en_de": "data/para_crawl_en_de.jsonl",
+    "para_crawl_en_el": "data/para_crawl_en_el.jsonl",
+    "para_crawl_en_es": "data/para_crawl_en_es.jsonl",
+    "para_crawl_en_et": "data/para_crawl_en_et.jsonl",
+    "para_crawl_en_fi": "data/para_crawl_en_fi.jsonl",
+    "para_crawl_en_fr": "data/para_crawl_en_fr.jsonl",
+    "para_crawl_en_ga": "data/para_crawl_en_ga.jsonl",
+    "para_crawl_en_hr": "data/para_crawl_en_hr.jsonl",
+    "para_crawl_en_hu": "data/para_crawl_en_hu.jsonl",
+    "para_crawl_en_it": "data/para_crawl_en_it.jsonl",
+    "para_crawl_en_lt": "data/para_crawl_en_lt.jsonl",
+    "para_crawl_en_lv": "data/para_crawl_en_lv.jsonl",
+    "para_crawl_en_mt": "data/para_crawl_en_mt.jsonl",
+    "para_crawl_en_nl": "data/para_crawl_en_nl.jsonl",
+    "para_crawl_en_pl": "data/para_crawl_en_pl.jsonl",
+    "para_crawl_en_pt": "data/para_crawl_en_pt.jsonl",
+    "para_crawl_en_ro": "data/para_crawl_en_ro.jsonl",
     "para_pat": "data/para_pat.jsonl",
     "php": "data/php.jsonl",
     "scandi_langid": "data/scandi_langid.jsonl",
@@ -138,7 +157,26 @@ class LanguageIdentification(datasets.GeneratorBasedBuilder):
         datasets.BuilderConfig(name="nbnn", version=VERSION, description="nbnn"),
         datasets.BuilderConfig(name="nordic_langid", version=VERSION, description="nordic_langid"),
         datasets.BuilderConfig(name="open_subtitles", version=VERSION, description="open_subtitles"),
-        # datasets.BuilderConfig(name="para_crawl", version=VERSION, description="para_crawl"),
+        datasets.BuilderConfig(name="para_crawl_en_bg", version=VERSION, description="para_crawl_en_bg"),
+        datasets.BuilderConfig(name="para_crawl_en_cs", version=VERSION, description="para_crawl_en_cs"),
+        datasets.BuilderConfig(name="para_crawl_en_da", version=VERSION, description="para_crawl_en_da"),
+        datasets.BuilderConfig(name="para_crawl_en_de", version=VERSION, description="para_crawl_en_de"),
+        datasets.BuilderConfig(name="para_crawl_en_el", version=VERSION, description="para_crawl_en_el"),
+        datasets.BuilderConfig(name="para_crawl_en_es", version=VERSION, description="para_crawl_en_es"),
+        datasets.BuilderConfig(name="para_crawl_en_et", version=VERSION, description="para_crawl_en_et"),
+        datasets.BuilderConfig(name="para_crawl_en_fi", version=VERSION, description="para_crawl_en_fi"),
+        datasets.BuilderConfig(name="para_crawl_en_fr", version=VERSION, description="para_crawl_en_fr"),
+        datasets.BuilderConfig(name="para_crawl_en_ga", version=VERSION, description="para_crawl_en_ga"),
+        datasets.BuilderConfig(name="para_crawl_en_hr", version=VERSION, description="para_crawl_en_hr"),
+        datasets.BuilderConfig(name="para_crawl_en_hu", version=VERSION, description="para_crawl_en_hu"),
+        datasets.BuilderConfig(name="para_crawl_en_it", version=VERSION, description="para_crawl_en_it"),
+        datasets.BuilderConfig(name="para_crawl_en_lt", version=VERSION, description="para_crawl_en_lt"),
+        datasets.BuilderConfig(name="para_crawl_en_lv", version=VERSION, description="para_crawl_en_lv"),
+        datasets.BuilderConfig(name="para_crawl_en_mt", version=VERSION, description="para_crawl_en_mt"),
+        datasets.BuilderConfig(name="para_crawl_en_nl", version=VERSION, description="para_crawl_en_nl"),
+        datasets.BuilderConfig(name="para_crawl_en_pl", version=VERSION, description="para_crawl_en_pl"),
+        datasets.BuilderConfig(name="para_crawl_en_pt", version=VERSION, description="para_crawl_en_pt"),
+        datasets.BuilderConfig(name="para_crawl_en_ro", version=VERSION, description="para_crawl_en_ro"),
         datasets.BuilderConfig(name="para_pat", version=VERSION, description="para_pat"),
         datasets.BuilderConfig(name="php", version=VERSION, description="php"),
         datasets.BuilderConfig(name="scandi_langid", version=VERSION, description="scandi_langid"),

load_data.md

@@ -59,10 +59,74 @@
 |   zu   |  zulu, south africa  |  10000   |    autshumato    |
 
 
-代码：
-```python
-
+备注：
+```text
+问题1：
+训练模型后发现它对短句子的识别能力很差。
+当句子长度足够的时候，它能够识别准确，并给出接近1.0 的概率值。
+但对于短句子，识别结果让人难以接受。例如把“你好”识别成 de 德语。
 
+解答1：
+（1）数据增强，将句子截短。
+（2）我认为可考虑加大 dropout，狠狠的 dropout。
 
 ```
 
+### 方案2
+
+语种：
+
+|  语种   |     语种全称     |  样本个数  |     数据来源      |
+|  :---  |     :---:      |   :---:   |      :---:       |
+|   ar   |     arabic     |  100000   |     iwslt2017    |
+|   bg   |   bulgarian    |  100000   |       xnli       |
+|   bn   |    bengali     |  36064    |  open_subtitles  |
+|   bs   |    bosnian     |  10212    |  open_subtitles  |
+|   cs   |     czech      |  100000   |       emea       |
+|   da   |     danish     |  100000   |  open_subtitles  |
+|   de   |     german     |  100000   |     iwslt2017    |
+|   el   |  modern greek  |  100000   |       emea       |
+|   en   |    english     |  100000   |     iwslt2017    |
+|   eo   |   esperanto    |  94101    | tatoeba; open_subtitles |
+|   es   |    spanish     |  100000   |       xnli       |
+|   et   |    estonian    |  100000   |       emea       |
+|   fi   |    finnish     |  100000   |    ecb; kde4     |
+|   fo   |    faroese     |  23807    |  nordic_langid   |
+|   fr   |    french      |  100000   |     iwslt2017    |
+|   ga   |    irish       |  100000   | multi_para_crawl |
+|   gl   |    galician    |  3096     |     tatoeba      |
+|   hi   |     hindi      |  100000   |       xnli       |
+|  hi_en |     hindi      |  7180     | cmu_hinglish_dog |
+|   hr   |    croatian    |  95844    |   hrenwac_para   |
+|   hu   |    hungarian   |  3801     |   europa_ecdc_tm; europa_eac_tm   |
+|   hy   |    armenian    |   660     |  open_subtitles  |
+|   id   |   indonesian   |  23940    |   id_panl_bppt   |
+|   is   |   icelandic    |  100000   | multi_para_crawl |
+|   it   |    italian     |  100000   |     iwslt2017    |
+|   ja   |    japanese    |  100000   |     iwslt2017    |
+|   ko   |    korean      |  100000   |     iwslt2017    |
+|   lt   |   lithuanian   |  100000   |       emea       |
+|   lv   |    latvian     |  100000   | multi_para_crawl |
+|   mr   |    marathi     |  51807    |     tatoeba      |
+|   mt   |    maltese     |  100000   | multi_para_crawl |
+|   nl   |    dutch       |  100000   |       kde4       |
+|   no   |   norwegian    |  100000   | multi_para_crawl |
+|   pl   |    polish      |  -   |       ecb        |
+|   pt   |   portuguese   |  10000   |     tatoeba      |
+|   ro   |    romanian    |  10000   |       kde4       |
+|   ru   |    russian     |  10000   |       xnli       |
+|   sk   |    slovak      |  10000   | multi_para_crawl |
+|   sl   |   slovenian    |  4589    |   europa_ecdc_tm; europa_eac_tm   |
+|   sw   |    swahili     |  10000   |       xnli       |
+|   sv   |    swedish     |  10000   |       kde4       |
+|   th   |     thai       |  10000   |       xnli       |
+|   tl   |    tagalog     |  10000   | multi_para_crawl |
+|   tn   |    serpeti     |  10000   |   autshumato     |
+|   tr   |    turkish     |  10000   |       xnli       |
+|   ts   |    dzonga      |  10000   |    autshumato    |
+|   ur   |     urdu       |  10000   |       xnli       |
+|   vi   |   vietnamese   |  10000   |       xnli       |
+|   yo   |     yoruba     |  9970    |    menyo20k_mt   |
+|   zh   |    chinese     |  10000   |       xnli       |
+|   zu   |  zulu, south africa  |  10000   |    autshumato    |
+