Solving CAPTCHAs to Transcribe Books

We took detailed looks at reCAPTCHA and how it works last September and in early 2007. In short, reCAPTCHA has found a nifty way to crowdsource book transcriptions. When users solve a CAPTCHA through reCAPTCHA, the software will give users two words: one with a known answer (the control word) and one where the OCR software wasn't quite sure what the word was. Once a certain number of users have solved the suspicious word with the same result, it becomes a control word itself and the OCR software can learn this word.

Now, Google will be able to use this same technology to improve its own OCR efforts. Google currently makes over 1 million out-of-copyright books available for download through Google Books and one of the main arguments against these books has been the fact that these texts are not edited and include a lot of OCR errors. With reCAPTCHA, Google could potentially bring the error rate down dramatically and make Google Books even more useful.

In fact, the ability for mobile web developers to get their apps on a number of platforms without having to start from scratch each time has been made even easier as of late thanks to a new project called Phonegap. This open source development tool is actually a web-based platform that lets developers build apps in HTML and JavaScript while also allowing those apps to take advantage of the core features in the iPhone, Android, and Blackberry.

When you look at the type of applications built using this technology, you may be surprised. If you didn't already know they were made using PhoneGap, you may have thought that they were native applications. Take the Blok-Buster game (iTunes URL) for example, a Tetris-inspired game that involves clearing similarly-colored blocks from the screen. You can tilt the phone horizontally, vertically, or even flip it upside down and it behaves just as a native application would. Yet it is just a web app using the PhoneGap's functionality.

Technology like this can speed up the process of getting applications deployed across a number of platforms. That's something that is more critical than ever as we move forward with mobile web development. Today, there are just too many languages and platforms to code for and in the future there could be even more. So, perhaps we shouldn't be focused solely on native applications as the future of the mobile web after all. Maybe the future of the mobile web will just be the future of the web.

There are a number of fascinating figures and observations from Grigsby's presentation (embedded below) about the state of the mobile web and the challenges ahead. It's definitely worth a look.

The way MapReduce works is a fairly straightforward concept: You take a problem that requires working with a giant (and we're talking massive - sometimes petabytes) dataset, distribute working with the dataset over thousands of separate processes (called mapping) and then taking the thousands of results you get back and reducing those results into a single master result. For certain tasks, MapReduce can vastly improve the efficiency of these types of tasks, and adding more computing power gives you a linear improvement in speed.

Yahoo! has been using its own version of Hadoop for a while now. And even before this offering, larger Amazon Cloud Computing customers have already begun to use Hadoop in EC2. This is from Wikipedia's article on Hadoop:

As an example The New York Times used 100 Amazon EC2 instances and a Hadoop application to process 4TB of raw image TIFF data (stored in S3) into 1.1 million finished PDFs in the space of 24 hours at a computation cost of about $240 (not including bandwidth).

As Amazon says on its blog, "After a while [developers] tend to report that they begin to think in terms of the new style, and then see more and more applications for it." Which we believe means that MapReduce is the new, big hammer, and as developers start looking around, every dataset starts looking like a nail. This is good news for Amazon as it only stands to profit.

On the search results page, OneRiot emphasizes links, though you can also click on a small link at the bottom of each result to see the tweets that included this link. From there, you can also reply to a tweet. If OneRiot finds new tweets that fit your search, it will automatically prompt you to reload the page.

To get data about current trends on Twitter, OneRiot uses both Twitter's own API, as well as data it gathers from users who have installed OneRiot's own toolbar in their browsers.

Looking for Feedback

As OneRiot's GM Tobias Peggs told us yesterday, the company decided to test its new Twitter search feature on a separate page for now (twitter.oneriot.com), in order to get feedback from customers.

We, for one, would like to see the ability to organize results by how many users have linked to a certain page, for example. It would also be nice if OneRiot gave us the option to search for conversations around a specific link by copying and pasting a URL into OneRiot's search box.

Overall, we like OneRiot's approach to searching Twitter. In some ways, it is similar to MicroPlaza's search functionality, though OneRiot's results focus more on the real-time experience.

Name: Y Combinator
Location: Cambridge, MA and Bay Area, CA
Investment: $5,000 + $5,000 per founder (i.e., $15,000 for two founders, $20,000 for three)
Stake Taken: 2-10% (usually about 6%)
Companies funded: Too many to list (about 58), but many you've heard of, like Reddit, Scribd, and Xobni (Wikipedia has a full list)
Next application deadline: April 2

Name: TechStars
Location: Boulder, CO
Investment: $5,000 per founder, up to $15,000 (3 founders)
Stake Taken: 5%
Companies funded: 9 so far, including Villij, and Intense Debate
Next application deadline: March 31

Name: SeedCamp
Location: London, UK
Investment: 50,000€ (about US$74,000)
Stake Taken: 10%
Companies funded: 6 so far, including Tablefinder
Next application deadline: August 12

Name: YEurope
Location: Vienna, Austria
Investment: 5,000€ per founder, up to 15,000€ (3 founders)
Stake Taken: 2-10%
Companies funded: Soup.io
Next application deadline: None

Name: Summer@Highland
Location: Lexington, MA
Investment: $7,500 for individuals or $15,000 for teams (split evenly)
Stake Taken: ?
Companies funded: 8
Next application deadline: Not yet announced

Name: LaunchBox
Location: Washington, DC
Investment: Between $15,000 and $30,000
Stake Taken: 4-8%
Companies funded: None yet
Next application deadline: March 14

Name: DreamIt Ventures
Location: Philadelphia, PA
Investment: Between $10,000 and $30,000
Stake Taken: 4-8%
Companies funded: None yet
Next application deadline: March 12

Name: Bootup Labs
Location: Vancouver, BC, Canada
Investment: ?
Stake Taken: ?
Companies funded: None yet
Next application deadline: None / not yet launched

Name: Bootphase
Location: Atlanta, GA
Investment: ?
Stake Taken: ?
Companies funded: None
Next application deadline: None / Not yet launched

See also: Charles River Ventures' QuickStart loan program, in which seed round startups receive a loan of up to $250,000 against a future Series A venture round (which CRV has the option to participate in). And see the annual Google Summer of Code program, in which stipends are awarded to students working on open source projects.