tatoebatools 0.2.3

A library for downloading and reading data from Tatoeba

tatoebatools

tatoebatools helps you to integrate Tatoeba into your app more quickly by allowing you to easily download and parse Tatoeba weekly exports.

Installation

This library supports Python 3.8+.

pip install tatoebatools

Basic Usage

Use the high-level ParallelCorpus class to automatically download and iterate through all sentence/translation pairs from a source language to a target language.

>>> from tatoebatools import ParallelCorpus
>>> for sentence, translation in ParallelCorpus("cmn", "eng"):
        print((sentence.text, translation.text))
...
('那里有八块小圆石。', 'There were eight pebbles there.')
('这个椅子坐着不舒服。', 'This chair is uncomfortable.')
('我会在这里等着到他回来的。', 'Until he comes back, I will wait here.')

Advanced Usage

The Tatoeba data files are handled by the tatoeba object.

from tatoebatools import tatoeba

By default, the fetched Tatoeba data files are stored inside the tatoebatools package. But you can download them to another location.

tatoeba.dir = "/path/to/my/tatoeba/dir"

Use the all_tables attribute to list the Tatoeba data tables you can have access to.

>>> tatoeba.all_tables
['jpn_indices', 'links', ... , 'user_languages', 'user_lists']

Each table has its own set of attributes:

Table	Attributes
sentences_detailed	sentence_id, lang, text, username, date_added, date_last_modified
sentences_base	sentence_id, base_of_the_sentence
sentences_CC0	sentence_id, lang, text, date_last_modified
links	sentence_id, translation_id
tags	sentence_id, tag_name
sentences_in_lists	list_id, sentence_id
jpn_indices	sentence_id, meaning_id, text
sentences_with_audio	sentence_id, audio_id, username, license, attribution_url
user_languages	lang, skill_level, username, details
transcriptions	sentence_id, lang, script_name, username, transcription
user_lists	list_id, username, date_created, date_last_modified, list_name, editable_by

Find out more about the Tatoeba data files and their fields here.

Most Tatoeba languages are identified by their IS0 639-3 codes. The asterisk character '*' designates all languages supported by Tatoeba. Call the all_languages attribute to list the languages supported by Tatoeba.

>>> tatoeba.all_languages
['abk', 'acm', ... , 'zul', 'zza']

Reading Tatoeba data with iterators

To read a table, just call its iterator. The downloading of the data file will be automatically handled in the background.

Set the scope argument to 'added' to only read rows that did not exist in the previous local version of an updated file. Set it to 'removed' to iterate over the rows that don't exist anymore.

# list all sentences in English
english_texts = [s.text for s in tatoeba.sentences_detailed("eng")]

# list all German sentences that were added by the latest local update
new_german_texts = [s.text for s in tatoeba.sentences_detailed("deu", scope="added")]

# list all links between French and Italian sentences
french_italian_links = [(lk.sentence_id, lk.translation_id) for lk in tatoeba.links("fra", "ita")]

# list all French native speakers
native_french = [x.username for x in tatoeba.user_languages("fra") if x.skill_level == 5]

# map German sentences to their audios
german_audios = {}
for audio in tatoeba.sentences_with_audio("deu"):
    german_audios.setdefault(audio.sentence_id, []).append(audio.audio_id)

Extracting Tatoeba data as dataframe

Since tatoebatools relies heavily on the pandas library, it is possible to load any supported table into memory as a dataframe.

# get the dataframe of the English sentences table
english_sentences_dataframe = tatoeba.get("sentences_detailed", ["eng"])

# get the dataframe of all links for which French is the source language
french_links_dataframe = tatoeba.get("links", ["fra", "*"])

Ingesting Tatoeba data into a database

The tatoebatools library includes SQLAlchemy models that help you to ingest Tatoeba data in the database of you choice.

In the example below, all sentence and user data is loaded into a local SQLite database, and then an index is added to speed up database queries.

from sqlalchemy import create_engine, Index

from tatoebatools import tatoeba
from tatoebatools.models import Base, SentenceDetailed

engine = create_engine( "sqlite:///./tatoeba.sqlite")

table_names = ["sentences_detailed", "user_languages"]

# create the tables in the database
tables = [
    Base.metadata.tables[table_name] 
    for table_name in table_names
]
Base.metadata.create_all(bind=engine, tables=tables)

# insert data into the tables
for table_name in table_names:
    with tatoeba.get(table_name, ["*"], chunksize=100000) as reader:
        for chunk in reader:
            chunk.to_sql(table_name, con=engine, index=False, if_exists="append")

# add an index on the 'username' column of the 'sentences_detailed' table
ix = Index('ix_sentences_detailed_username', SentenceDetailed.username)
ix.create(engine)