** Sentences & Translations Stats **

These stats have been updated.

http://tatoeba.j-langtools.com/transtop/

Various graphs have been updated, too:

http://tatoeba.j-langtools.com/graphs.html

It seems that RCVD_REMOVED check was added to SpamAssassin rules on December, 16th, and December, 21st is maybe the date your provider updated his SpamAssassin. So, if your provider's SMTP server removes 'Received:' headers, this fact was just unnoticed until the update.

Judging by the mail headers, Tatoeba doesn't have its own MTA. It sends mail via Google SMTP server.

Tatoeba e-mail notification system has nothing to do with this issue, as far as I can judge.
'Received:' headers are being inserted by SMTP servers at every hop (i.e., each SMTP server that relays or delivers a message adds its own 'Received:' header for tracking purposes).
In the SpamAssassin v.3.4.1, RCVD_REMOVED check fails in three cases only:
- the message was received via a mailing list running ezmlm software configured to remove 'Received:' headers except the top one.
- all 'Received:' headers were removed from the message (normal SMTP servers and mail programs never do this)
- the message was received via MSN Groups (this service removes 'Received:' headers added by SMTP-relays the message has passed through)

To nail the problem down, please check the headers of the messages you receive from Tatoeba.

It would be nice to have a mass deletion function in personal messaging system (in all folders).

С Новым Годом!
Happy New Year!
明けましておめでとうございます！
:-)

** Sentences & Translations Stats **

These stats have been updated. It was the last update this year. :-)

http://tatoeba.j-langtools.com/transtop/

Various graphs are now available on one page:

http://tatoeba.j-langtools.com/graphs.html

It works like a charm now. Thanks, gillux!

It seems that the database that the search engine uses should be re-indexed to reflect recent changes. Indeed, judging from my example the lag is two days at least. I don't know how often re-indexing is being carried out but even one day interval is too long since it makes advanced search feature almost unusable in certain scenarios.

Судя по всему, индексирование не проводится почему-то. Если перевод сделан недавно, то поисковик не исключает его при выводе результата.

Если результат у вас другой, то, значит, поиск меня лично не любит. :-)
Другой браузер, другой компьютер, другая ОС, другой провайдер - результат тот же:
https://dl.dropboxusercontent.c...advsearch2.png

Условия поиска на картинке пробовали повторить? :-)

** Advanced search feature is slightly broken? **

Recently I've noticed that advanced search sometimes fails to filter unwanted translations from the result. Please take a look:
https://dl.dropboxusercontent.c.../advsearch.png
According to the conditions, direct Russian translations should be excluded from result but still we can observe them there.

** Sentences & Translations Stats **

These stats have been updated.

http://tatoeba.j-langtools.com/transtop/

Various graphs are now available on one page:

http://tatoeba.j-langtools.com/graphs.html

> whether adding romanization for “any other language that's not "romanized" (that don't use the Roman script)” is a good thing...

I think, for different languages we could add different data that is actual to a particular language. For Russian, for example, we could provide a version of a sentence with stress marks. Romanization is just one of the options.

** Sentences & Translations Stats **

These stats and the underlying software have been updated.

http://tatoeba.j-langtools.com/transtop/

The following graphs have been updated, too:

Top 13 Tatoeba languages monthly dynamics:
http://tatoeba.j-langtools.com/langs.png

English sentences distribution by word count:
http://tatoeba.j-langtools.com/dist-eng-wc.png

English sentences distribution by translation count:
http://tatoeba.j-langtools.com/dist-eng-tc.png

Looks like a localization bug. I hope @sadhen, the Chinese UI translator, or some other authorized person will eventually fix it.

Всё зависит от того, как и чем вы открываете скачиваемый файл и с какой целью.

** Sentences & Translations Stats **

These stats have been updated.

http://tatoeba.j-langtools.com/transtop/

The following graphs are now available:

Top 13 Tatoeba languages monthly dynamics (updated):
http://tatoeba.j-langtools.com/langs.png

English sentences distribution by word count (updated):
http://tatoeba.j-langtools.com/dist-eng-wc.png

English sentences distribution by translation count (new):
http://tatoeba.j-langtools.com/dist-eng-tc.png

> It might be useful to have the selection default to a member's registered native language when the script has trouble detecting a language

I think this really can narrow the problem down a bit.

The short answer is: the Tatoeba language autodetection engine is not perfect.

The algorithm it built upon (character n-gram language models, as far as I know) is pretty probabilistic and, thus, not accurate by design, especially for languages that use similar alphabets and (in the worst case) belong to the same language group. The accuracy can be increased (but not radically at the moment, I'm afraid) by regular n-gram database training and proper n-gram weighting. The problem is that it requires rather big corpus for each language and Tatoeba corpus is still not big enough for the majority of its languages.

Another approach is to amplify existing algorithm with various auxiliary heuristic algorithms for each language or at least for the most problematic ones. This can increase the accuracy dramatically but it requires collaboration of the members who know those languages well and have enough time and knowledge to assist Allan Simon, the author of the Tatodetect engine, provided that he himself has enough time and motivation.