Members of Tatoeba have been contributing original sentences in dead languages for a long time, including Latin, ancient Greek, old French, middle French, ancient Hebrew, old Spanish, old Turkish, just to name a few. Changing the policy now would not be very practical, and some sentences may have been sourced from historical texts without attribution.

Of course there are no "native speakers" of these languages who may claim they are correct and natural, but I think we can approach a similar quality level based on non-native checks. I have a feeling that people who go all the way learning dead languages must be so passionate about it that they can be trusted.

Everything seems OK except one thing. I cannot change flags.

Kieltä tuntematta: tuleeko vastaan jotain ongelmia, jos molempia merkkilajeja voi käyttää yksinään tai sekaisin, jos kieli kerran näin tekee?

Good question. It could be interesting to know what led to the decision of using POJ only on Minnan Wikipedia https://zh-min-nan.wikipedia.org/wiki/Th%C3%A2u-ia%CC%8Dh

I guess we can have a transcription system where users manually enter alternative scripts (without autogeneration), as long as there is a some way to validate consistency between the sentence and the alternative script, and of the alternative script alone.

As for mixing both scripts within a single sentence, I am not sure if we want to explicitly allow it. On the one hand, if it reflects real world usage of the language, it should be on Tatoeba. On the other hand, it makes it difficult to classify and search. Can you clarify if a typical sentence would include a majority of Chinese characters and a just few POJ, or if it's 50/50, or the opposite?

For your information, Tatoeba already has quite a few languages that use two scripts without automatic conversion. Just to name a few:

- Arabic/Latin:
https://tatoeba.org/sentences/s...ta&sort=random
https://tatoeba.org/sentences/s...hg&sort=random

- Cyrillic/Latin
https://tatoeba.org/sentences/s...fn&sort=random
https://tatoeba.org/sentences/s...rp&sort=random

Some contributors add both scripts and link both sentences to one another. Not ideal, and not officially recommended, but it helps with searching.

Hi rul, thanks for reporting this issue, I just had a look into it. I confirm there is a problem, I see a difference between what the search returns and what’s actually in the corpus.

It was a actually a rare and temporary problem. We may fix the root cause in the future, but for now I have simply reindexed the affected sentences on the "Translated by Tatoebans" list. Your search should now return the latest sentences.

@gillux is really the only person who can answer that question directly. However, I just want to insert my opinion that even if someday we are able to capture audio on the site, it shouldn't be made audible until it has been reviewed. Audio differs from text in two ways:

(1) A native speaker can spot errors in text virtually immediately, but listening to audio takes time.

(2) Audio can suffer from background noise and lack of clarity, which is not a factor with text.

I can easily imagine a large number of submissions of poor quality damaging the usefulness of the site if they "went live" immediately.

> "When if ever Tatoeba will be able to capture audio directly on site?"

This year hopefully!

I think I was waiting to get something working before making any announcement here, but since you brought up the topic, let me explain what has been going on behind the scenes.

I have been in contact with Hugo from Lingua Libre for some time on a
different channel. Hugo was part of the Shtooka project back in the days, along with its creator Nicolas. Later, they worked on a cloud version, which was rebranded as Lingua Libre. So essentially Lingua Libre and Shtooka are just the same piece of software at two different points in time.

A quick introduction of Lingua Libre. Lingua Libre [1] is the name of a recording tool, but also a Wikimedia France project used to gather recordings. It uses Wikimedia Commons for audio storage, and members of the Wikimedia community help connecting with native speakers to have their voice recorded. The recordings are used by other Wikimedia projects such as Wiktionary or Wikipedia, so they mainly focus on recording words.

I connected with Hugo in early 2025. Hugo was actually astonished to learn that Tatoeba audio contributors still rely on the good old Shtooka. We quickly figured out there was room for collaboration. Lingua Libre has a strong recorder and is starting to support sentences in addition to words, but it needs open text content, so it could benefit from Tatoeba’s linguistically diverse corpus. Tatoeba lacks an easy-to-use recorder and audio support is rather basic, so it could benefit from Lingua Libre’s tooling and Wikimedia Foundation’s infrastructure and "aura".

In 2025, Hugo has been working on a new version of the recorder that makes it easier for other projects to reuse. This new version has been in "beta test" for some time now and I think it will become their new official recorder soon.

At some point in late 2025, Hugo and I tried to apply to a Microsoft grant [2] to kickstart collaboration between Tatoeba and Lingua Libre, but our application got rejected. This means it will take more time to get things done but we will get there eventually.

After discussing with Trang and CK, I drafted an initial technical plan to allow using Lingua Libre to record Tatoeba sentences, and you can follow the progress on GitHub [3]. Basically, it is much harder to make two pieces of software collaborate than to develop everything in-house, but I believe it will pay in the long run. Generally speaking, Tatoeba and Lingua Libre share common goals of creating open and diverse linguistic ressources and preserving endangered languages, so I believe our project and communities should be more connected and aware of what the other party is doing. I wish the recorder could be the first step in that direction.

[1] https://lingualibre.org/
[2] https://www.microsoft.com/en-us...-voices-in-ai/
[3] https://github.com/Tatoeba/tatoeba2/issues/3183

If you want your own sentence be deleted, you can edit and replace the text with DELETE.

I unlinked it from the Turkish. Now just copy and paste it to the right place.