Digitising Finnish history using crowdsourced volunteers

Performance indicators

Digitalkoot was launched in February 2011. As of August 2012:

• 107,528 people have visited the Digitalkoot website.
• Volunteers have contributed a total of 418,361 minutes of their time
• 7,634,174 microtasks have been completed by these volunteers

The accuracy of the tasks completed is estimated at over 99%! This figure was gauged by randomly selecting two long articles from the processed newspapers and manually calculating the number of mistakes found first in the OCR and then in the article fixed by the Digitalkoot effort. The result was staggering: in a sample article of 1,467 words, Digitalkoot had produced only 14 mistakes. Another article of 516 words was even higher in quality, as only one single word in the whole article was considered wrong.

By comparison, the OCR process had made a mistake in 228 words of the first article, and in 118 words of the second article. In other words, whereas OCR systems struggle to get pass 85% mark in accuracy, it seems possible to achieve well over 99% accuracy in digitizing words by simply getting people to play a computer game.

Because of its unique approach to improving newspaper archives, Digitalkoot has attracted a lot of media publicity. Along with coverage in all the major Finnish newspapers, television and radio stations, it has also featured in major international publications including the New York Times and Wired.