-
Notifications
You must be signed in to change notification settings - Fork 2
/
get_parallel_data_tatoeba.py
118 lines (95 loc) · 4.27 KB
/
get_parallel_data_tatoeba.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
"""
Tatoeba (https://tatoeba.org/) is a collection of sentences and translation, mainly aiming for language learning.
It is available for more than 300 languages.
This script downloads the Tatoeba corpus and extracts the sentences & translations in the languages you like
"""
import os
import sentence_transformers
import tarfile
import gzip
# Note: Tatoeba uses 3 letter languages codes (ISO-639-2),
# while other datasets like OPUS / TED2020 use 2 letter language codes (ISO-639-1)
# For training of sentence transformers, which type of language code is used doesn't matter.
# For language codes, see: https://en.wikipedia.org/wiki/List_of_ISO_639-2_codes
source_languages = set(["eng"])
target_languages = set(["swe"])
num_dev_sentences = 1000 # Number of sentences that are used to create a development set
tatoeba_folder = "tatoeba"
output_folder = "parallel-sentences/"
sentences_file_bz2 = os.path.join(tatoeba_folder, "sentences.tar.bz2")
sentences_file = os.path.join(tatoeba_folder, "sentences.csv")
links_file_bz2 = os.path.join(tatoeba_folder, "links.tar.bz2")
links_file = os.path.join(tatoeba_folder, "links.csv")
download_url = "https://downloads.tatoeba.org/exports/"
os.makedirs(tatoeba_folder, exist_ok=True)
os.makedirs(output_folder, exist_ok=True)
# Download files if needed
for filepath in [sentences_file_bz2, links_file_bz2]:
if not os.path.exists(filepath):
url = download_url + os.path.basename(filepath)
print("Download", url)
sentence_transformers.util.http_get(url, filepath)
# Extract files if needed
if not os.path.exists(sentences_file):
print("Extract", sentences_file_bz2)
tar = tarfile.open(sentences_file_bz2, "r:bz2")
tar.extract("sentences.csv", path=tatoeba_folder)
tar.close()
if not os.path.exists(links_file):
print("Extract", links_file_bz2)
tar = tarfile.open(links_file_bz2, "r:bz2")
tar.extract("links.csv", path=tatoeba_folder)
tar.close()
# Read sentences
sentences = {}
all_langs = target_languages.union(source_languages)
print("Read sentences.csv file")
with open(sentences_file, encoding="utf8") as fIn:
for line in fIn:
id, lang, sentence = line.strip().split("\t")
if lang in all_langs:
sentences[id] = (lang, sentence)
# Read links that map the translations between different languages
print("Read links.csv")
translations = {
src_lang: {trg_lang: {} for trg_lang in target_languages} for src_lang in source_languages
}
with open(links_file, encoding="utf8") as fIn:
for line in fIn:
src_id, target_id = line.strip().split()
if src_id in sentences and target_id in sentences:
src_lang, src_sent = sentences[src_id]
trg_lang, trg_sent = sentences[target_id]
if src_lang in source_languages and trg_lang in target_languages:
if src_sent not in translations[src_lang][trg_lang]:
translations[src_lang][trg_lang][src_sent] = []
translations[src_lang][trg_lang][src_sent].append(trg_sent)
# Write everything to the output folder
print("Write output files")
for src_lang in source_languages:
for trg_lang in target_languages:
source_sentences = list(translations[src_lang][trg_lang])
train_sentences = source_sentences[num_dev_sentences:]
dev_sentences = source_sentences[0:num_dev_sentences]
print("{}-{} has {} sentences".format(src_lang, trg_lang, len(source_sentences)))
if len(dev_sentences) > 0:
with gzip.open(
os.path.join(output_folder, "Tatoeba-{}-{}-dev.tsv.gz".format(src_lang, trg_lang)),
"wt",
encoding="utf8",
) as fOut:
for sent in dev_sentences:
fOut.write("\t".join([sent] + translations[src_lang][trg_lang][sent]))
fOut.write("\n")
if len(train_sentences) > 0:
with gzip.open(
os.path.join(
output_folder, "Tatoeba-{}-{}-train.tsv.gz".format(src_lang, trg_lang)
),
"wt",
encoding="utf8",
) as fOut:
for sent in train_sentences:
fOut.write("\t".join([sent] + translations[src_lang][trg_lang][sent]))
fOut.write("\n")
print("---DONE---")