First one is Maltese.

why many sentences include tom and mike? it's beyond strange.

yup and the "deluxe" version: only track translations that are in a set of languages. (because no offense for Latin, but I don't care when my sentences are translated in it)

For this I've nothing concrete, but I may have a start of idea, on using a specific software in a smart way to be able to do this. Because it's something a normal database is not suitable for.

A generalisation on this would be able to track event on the website sentences added/translaion added [ in X ] [ by Y ] and up to the user to define the set of events he tracks

pour la nouvelle version j'ai commencé a y reflechir, j'avais pensé à cela

http://tato.sysko.fr/eng/sentences/show/655679, avec vraiment une distinction claire
et peut-etre oui ajouté une infobulle près du "translations of translations [késako](affiche un texte quand on passe dessus)"

et je rajoute une 3e suggestion: sur les phrases traduites (pas la principale) il n'y a pas d'infobulle => il faudrait en mettre une « Faire de cette phrase la phrase principale à traduire »

Génial !

la procédure de déduplication était en cours, elle a été fusionné avec la phrase 683297, la bizarrerie est-elle toujours présente ?

Actually this "tool" is made possible by the totally new architecture of the version I'm developping. So we can't integrate only the part to view every indirect translations in the current version.

If you're used to understand a faster than normal recording, it may be so much easy to understand a normal conversation I think ^^

It's the big difficulty from the beginning, instead of a gradually difficulty :p

Actually we share these audio with the Shtooka project, which aim to produce "natural souding" (in speed /pitch etc.) audios, so I do agree they're not suitable for very beginner, but they're exactly said the way a native will say it to an other native. So if you're able to catch them and repeat them perfectly, you will be ready for "normal life" situation. The same as we do have a lot of sentences which include some hard pattern or some "shortcut" that you will never found in textbook aimed to beginner.

I think, but maybe people do not share my point of view, that Tatoeba itself is a "raw natural-content" provider, and is not aimed to be directly used by beginner. In a perfect world we would need some parallel project to turn this raw data into content suitable for learner (by filtering the sentence only with simple pattern/re-recording at "non-natural but suitable for learner speed")

Because we can't, in my mind restrict Tatoeba to a "learning purpose". For example in the case of Shanghainese, it is now widely recognized that this language (actually the Wu) is endangered and that there is very very few numerized data about it. So in that case of "language archiving/protecting" purpose, sound recorded to a too-slow to be natural speed will not be suitable.

I think by keeping the data as raw and near from the way they appear in the real world help to open the possibility of usage of the contents collected by Tatoeba.

I would be interested to know the other contributors point of view about this.