Hi and welcome to Tatoeba. This is what should be done to have a new language added: https://en.wiki.tatoeba.org/art...guage-request#

BONNE ANNÉE à tous.

You can remove the old tag from a sentence and add a new tag with a different name instead. If you want to change all occurrences of a tag, it's going to be a lot of work, of course. (Unless you know how to automate it. That's how I added "quote" tags to a bunch of sentences tagged "by <Somebody>")

And you cannot change tags that someone else added, but you can add the new tag next to the old tag.

It's not completely redundant, as there are a bunch of Uzbek sentences in the database using Cyrillic script. And the transliteration feature makes it possible to find them even when you're using Latin script to search: https://tatoeba.org/en/sentence...ry=dushmanning

Mozilla's Common Voice is similar in collecting sentences and recordings thereof. It does not have the translation aspect of Tatoeba.

See https://commonvoice.mozilla.org/

If you just need English sentences, there are a few. However, I have looked myself, and found Tatoeba to be of the best quality, especially for English.

English-only:
• English Penn Treebank (Pennsylvania State University)
... is not something I know much about.
• English Web Treebank (Universal Dependencies)
... is mostly composed of biased sentence picks, but each has a grammatical breakdown. Stanford's NLP project Stanza uses it.
• Common Voice (Mozilla Foundation)
... as Augustus said!

With translation:
• OpenSubtitles2018 Corpus (OpenSubtitles)
... isn't very good for high-fidelity translation, but is rather natural, apart from its dramatizations.

Honorable mentions:
• Google Books Ngram Dataset (Google)
... only has a few languages. For example, their Japanese dataset is old and can only be accessed via purchase in yen.
• Wikipedia and Wiktionary (Wikimedia Foundation)

• Any other English (meta)corpora out there

https://www.google.com/search?q...s"%7C"dataset"

It really depends on your intentions and usage, as all corpora have their biases, unfortunately.

Interesting link, CK. Thanks!