I’ve just implemented the vision-impaired CAPTCHA workaround that I discussed before, wherein (what a great word) the image’s alternate text contains a second word, which, if you enter it, puts your comment in the moderation queue. It’d be nice if you got a message to that effect, but I couldn’t immediately see how to do that, so for now your comment will seem to disappear into the æther.

So I’m expecting one of three things to happen.

1. The thousands of vision-impaired readers I’ve picked up over the last two weeks since I introduced CAPTCHAs will break their silence and start leaving insightful (no pun intended) comments in droves.
2. Every spambot in existence will try submitting the alt text, and my inbox will be pummeled into oblivion with moderation emails offering various anatomical enlargements.
3. Deathly silence.

My money’s on 3. Stay tuned.

My custom CAPTCHA implementation is up and running. It’s currently a horrible collection of hacks into CapCC to call a Python script, so it’s in no state to distribute to anyone but myself at the moment. I might improve that situation at some point.

I have no idea how strong it is, other than a vague impression that I wouldn’t like to have to try to crack it. No vision-impaired option yet, but I’m probably going to do the moderation thing that I mentioned in my earlier post.

Coincidentally, Slashdot just reported (the news is actually a couple of weeks old) that Yahoo’s CAPTCHA has been broken with 35% accuracy.

If you have a very keen eye, you may have noticed that I added CAPTCHAs to the comment form. The problem isn’t exactly the volume of spam that I was getting (maybe two or three comments a day), or that the spam was getting through (WordPress has correctly marked almost all of the spam comments for moderation), but that there were very few nuggets of real comments in there to make the whole process worthwhile, and it was getting a bit depressing. I’m too paranoid about false positives to trust the filter to delete them altogether, so captchas seemed like the best option.

For now, I’m using a WordPress plugin called CapCC. However, on and off I’ve been working on my own text distortion doohickey which I’m planning to drop in at some point. It’s a really interesting problem – coming up with a system where you can generate a test in software, then determine in software that whatever completed the test was not software.

One annoying thing is that it there’s currently no vision-impaired option. “Well,” you say, “how many blind people actually read and decide to comment on your blog?”, and the answer is probably zero (only slightly lower than the number of able-sighted people who comment on my blog), but it annoys me on principle. It’s also an interesting addition to the general problem – now it has to be possible for software to read but not answer the test – so it’s sitting there mocking me as a technical challenge. The standard solution is to offer an audio version, but that has its own issues.

I did think of one solution, but it relies on spammers being rational actors, which is not an assumption I’m completely at ease with. Suppose there are two answers to the captcha – one encoded in the image, and one in the image’s alternate text. If you enter the one in the image, your comment is published, no questions asked. If you enter the alt text one, your comment is put in the moderation queue.

The alt text, of course, can be easily read by accessibility software and spambots alike. However, because the comment is going into the moderation queue, it’ll never see the light of day unless I recognise it as legitimate; so there’s no incentive in terms of published spam for a spammer to specifically try to defeat it. So as long as there aren’t any spambots that look at alt text as a matter of course, and as long as no one decides to spam my moderation queue out of spite, it should all work. (And if spambots do look at alt text, then I can put it in body text alongside or whatever.)

I’d really like to ask anyone using accessibility software what they think of this approach, but at the moment you won’t be able to leave a comment unless you find another way to parse the captcha. So if you have any thoughts, send me an email. (I do trust my email spam filter.)

Here’s an experiment: search Wikipedia for the title of a blog post, and, if the first search result gives a page with an unambiguous opening sentence, remove citations and anything in parentheses, prepend “But” and add as a comment.

Examples:

His Dark Materials

But His Dark Materials, a trilogy of fantasy novels by Philip Pullman, comprises Northern Lights, The Subtle Knife and The Amber Spyglass.

Falling off a building is a miracle

But Genki Bakuhatsu Ganbaruger is a 47 episode animated television series, and the second series produced for the Eldran franchise funded by Tomy and produced by Sunrise.

Scarlet ‘A’

But A Study in Scarlet is a detective mystery novel written by Sir Arthur Conan Doyle and published in 1887.

Jacoby-Shermer debate

But atheism, as a philosophical view, is the position that either affirms the nonexistence of gods or rejects theism.

God good, demons bad

But in religion, folklore, and mythology a demon is a supernatural being that has generally been described as a malevolent spirit, and in Christian terms it is generally understood as a Fallen angel, formerly of God.

The Mario Clone they Play in Hell

But the video game Super Mario RPG: Legend of the Seven Stars, created by the Nintendo and Square Enix companies, has a unique set of characters that derive both from previous installments of the Mario video game series and new characters made specifically for the game.

Head tracking with a Wii remote

But Metroid Prime 3: Corruption is a video game developed by Retro Studios and published by Nintendo for the Wii.

Northern Lights

But Northern Lights is a common name for the aurora borealis in the Northern Hemisphere.

Merry Christmas

But holiday greetings are a selection of greetings that are often spoken with good intentions to strangers, family, friends, or other people during the months of December and January.