고등학교 영어Ⅰ (미래엔 김성연) 연계 글

Around 2000, free email services faced an unusual problem. Spammers had written programs that could fill out sign-up forms and open large numbers of accounts for sending junk mail. That year, the computer scientist Manuel Blum, his graduate student Luis von Ahn, and their colleagues at Carnegie Mellon University began looking for a quick way of telling real users from programs.

They answered with short puzzles called CAPTCHAs, which people could solve easily but the programs of the time could not. In one version, the screen showed a single word with distorted letters on a cluttered background. Oddly, the computer that generated each puzzle could not read the word either; it merely checked if the typed answer matched the word it had chosen. After Yahoo added this version to its sign-up page, visitors could open an account only by reading the twisted letters and typing them correctly.

As CAPTCHAs spread, von Ahn began to ask if the time people spent on them had to be wasted. Each puzzle took only seconds, but people solved millions of them every day, so the seconds accumulated into more than 150,000 hours of human effort daily. He and his colleagues began searching for a way of collecting those seconds and putting them to good use.

In 2007, von Ahn's team launched reCAPTCHA, a new form of the test that also helped turn old printed pages into digital text. Libraries and archives were preserving old books and newspapers by scanning their pages, and software then transformed each image into text that could be searched. On pages with faded ink and yellowed paper, however, the software failed to recognize as many as one word in five.

The new test showed users two words cut from such scans instead of random letters that a computer had made up. The system already knew one of the words, so it could check if the user was human. The other was a word that the software had been unable to read. By showing each unknown word to several users and comparing their answers, the system accepted a spelling once enough users had submitted identical guesses.

Its transcriptions proved more than 99 percent accurate, matching the standard that professional transcription services promise. Within its first year, users solved more than 1.2 billion reCAPTCHAs and deciphered over 440 million words, the equivalent of more than 17,600 books. Instead of simply proving that they were human and moving on, millions of people were now rescuing words from fading pages, most of them without realizing it.