2011/2/5 David Gerard <dgerard(a)gmail.com> Well, we it.source fellow are writing our communication about (it will be published into wikisource-l), but a brief mention to good news is mandatory here. We have a simple script that extracts word images, corresponding to doubtful OCR interpretation, from any djvu file with a text layer; scripts to upload into djvu layer again fixed words are simple too. We posted first communication into John Vandenberg en.source user page, and a wikicaptcha is now something possible. See John's lalk here: http://en.wikisource.org/wiki/User_talk:John_Vandenberg#reCAPTCHA_for_source Alex brollo

On Saturday 05 February 2011 10:51 PM, Diederik van Liere wrote: There is atleast one successful captcha php script for Malayalam Language (http://sourceforge.net/projects/mlcaptcha/ , http://mlcaptcha.blogspot.com/2010/02/blog-post_24.html ). I don't know whether it can work with mediawiki.

On 05/02/11 22:36, Platonides wrote: Alternatively, you could add a Malayalam wordlist to the current Wikipedia captcha, which would have the same effect, and has (I hope) a better visual obfuscation method than this, designed specifically to resist some of the most recent bot decoding methods. Perhaps we could have a place to add these wordlists on the meta-wiki or on translatewiki, to allow people without transmit rights to create them? All that is needed is about 2,000 short words for each language, which can be used to create around 4,000,000 possible challenge words, which will in turn will be used to create an endless stream of captcha images, no two of which should ever be alike. The wordlists themselves need not be secret: they are only needed to create easily-typed strings that are sufficiently large in number to provide a moderate challenge to brute force guessing. -- Neil

David Gerard wrote: My intention wasn't to come across as dismissive. On the other hand, if people begin new conversations without having read the old conversations, it sets back progress dramatically. The opening post didn't make any mention of the old bugs or their progress, so I was trying to point out that these issues were already known and there were already forums in which they could and should be discussed. David Gerard (also) wrote: I think it's a real pity that CAPTCHAs are needed at all. They're a pain-in-the-ass and their effectiveness against coordinated or sophisticated attacks is dubious at best. Diederik van Liere wrote: Ideas for Google Summer of Code 2011 should go here: http://www.mediawiki.org/wiki/Summer_of_Code_2011/Project_ideas Platonides wrote: Occasionally the CAPTCHA will create offensive combinations, which are a bit worse than double Es and the like. ;-) This is the subject of bugs 10408, 16166, and 21025. Alex wrote (referring to a GSOC project involving better CAPTCHA support): There's certainly a right balance to be struck between projects that are far too large and complex (and thus never get finished) and projects that are too small and get finished within a day or two. Personally, I'd much rather have a bunch of small projects get finished (and re-worked as necessary) than have one large project get started, but never finished (LiquidThreads, interwiki transclusion, etc.). If you looked at all of the CAPTCHA-related bugs as a group (including possibly removing the Python dependency), there's more than enough to be at least considered for Summer of Code 2011. MZMcBride

2011/2/5 MZMcBride <z(a)mzmcbride.com>om>: I was thinking that this first step into entering a wiki (as a registered user) might be the most difficult thing people will ever be required to perform in their life as a Wikimedia user. It is like we require our users to have an IQ above 130, while very simple on-wiki tasks such as correcting typing mistakes don't require more than an average IQ.

2011/2/5 Marcus Buck <wiki(a)marcusbuck.org> Just to let you know that Aubrey just prestented it.source idea for wikicaptcha into wikisource-l :-) Obviously, if a wikicaptcha tool will be built and will run, we can do anything .... and while interpreting words (in any language) any user will contribute to source transcriptions in a very valuable way. Alex brollo

2011/2/5 David Gerard <dgerard(a)gmail.com> Please consider that only a test script run - just to show that it's possible that a python script loads djvu text layer, selects doubtful words, selects the image of such words and saves them into a file. Now it's matter for excellent developers: how to select djvu files, where to upload them, how to build the database os words/images, how to build a user interface to show images and to get user input.... our it.source test documents that the first step can be done. It's so rewarding to give to that script the numbero of a djvu page, nothing re, then to see tiff images popping out into the folder... :-) Alex

2011/2/5 River Tarnell <r.tarnell(a)ieee.org> IMHO, it could be an opportunity to think again to the role of Commons as a central library. I imagine something like this: 1. as soon as a djvu file with a text layer is uploaded, a complete set of pages text layers is extracted, saving words coordinates too; 2. such text layers could be browsed by a script, extracting all words marked as doubtful (usually with a ^ characters), but extracting too words which don't match with a good dictionary; 3. a dynamic recaptcha database is updated and word images are submitted to wiki contributors, both as a formal captcha for unlogged user edits, and as a volunteer job to help wikisource projects; updates will fix text files; 4. a tool should be build, to upload "pure text" from such text files into any wikisource project; 5. finally refined text could be re-uploaded into djvu file, so converting it into a "djvu file with a wiki text layer". Alex 4.