If interested, this text is written by man

Original author: Steve Lohr
  • Transfer
“Wisconsin is clearly on a winning path, winning 51-10 after the third quarter of the match. The team increased its advantage when Russell Wilson found Jacob Petersen pass, and he made a touchdown after a seven-meter spurt, making the score 44-3. ”

With these words, the fragment of the news article begins , published 60 seconds after the end of the third quarter of the football match between the university teams of Wisconsin and Nevada. Although at first glance it is difficult to understand, but this note is completely written by a computer program.

Generator of journalistic texts created in the company Narrative Science , which is developing in the field of artificial intelligence.

The program takes data, such as sports statistics, financial statements of companies or data on real estate sales, and turns them into newspaper articles. For years, programmers have tried to create such journalistic robots for sports reporting, but the result is not brilliant: the texts were dry and it was clear that the article was computer generated. Professor of Journalism Chris Hammond and Computer Science Professor Larry Birnbaum from the Laboratory of Intelligent Systems at Northwestern University in Illinois have been working in this field for ten years and still managed to achieve a qualitatively better result. So last year, the Narrative Science startup appeared.

“I thought it was some kind of magic,” says Roger Lee, an investment company partner.Battery Ventures , which raised a $ 6 million investment for a startup. “It looks like a person wrote it.”

Artificial intelligence experts are also impressed with the success of Narrative Science. In their opinion, Narrative Science demonstrates "an increase in the complexity of automatic recognition and, now, the generation of syntactic structures."

The innovative work of Narrative Science raises a wide range of questions as to whether such AI programs can help professional journalists in their work or replace them. High technology has already affected the economic reality of the media, when online print revenue has fallen due to online advertising. Will robotic journalists become another nail in the coffin of traditional journalism?

The leaders of Narrative Science look at things soberly and position their program solely as a tool for low-budget work in a time-constrained environment. At the moment, the company has only 20 customers, and some of them are only experimenting with an outlandish novelty. Among the clients are traditional media who want to increase the generation of content due to news notes on the financial results of local companies and more complete coverage of sports matches in youth leagues.

“In this way, we help write articles that otherwise would not have been written at all,” says the director of Narrative Science, answering the question of crowding out human journalists with computer rivals.

For example, the aforementioned company The Big Ten Network(BTN) began using text generators in the spring of 2010 for short notes on baseball and softball games. They appeared on the site one to two minutes after the end of the game. To generate the texts, the final statistics of the match and the per-minute online transcript of the game's progress, which is conducted in real time in all American sports leagues, were used. The program has improved over time thanks to the help of professional BTN editors.

The development of Narrative Science can draw logical conclusions based on historical data, as well as the sequence and results of previous games. To generate meaningful designs, the program uses the concepts of “individual contribution”, “team effort”, “strong-willed victory” (gain after a lag in the account), “lost advantage”, “record result of the season”, “players on the rise” (which differ several matches in a row), “team positions” in different sections of sports statistics. Then the program decides which element is most important in a particular match - and it is selected for the first paragraph of the article. The data collected determines the choice of words. For example, in the case of a big advantage in the account, the program may choose the word “defeat” instead of “victory”.

“The main thing is composition, the construction of the text,” says Chris Hammond. “You can't just take numbers and convert them into words.”

Last fall, BTN expanded its use of the program also to football and basketball games. This content allows you to attract huge traffic with Google by keywords, because on this site reports on the game appear earlier than competitors, and Google very much appreciates operational content on hot topics. Last year, traffic to the BTN football section grew by 40% compared to 2009.

Another Narrative Science client is the publishing and real estate company Hanley Wood , which, with the help of a robot, publishes monthly on its website builderonline.comReal estate market reports in 350 nearby towns and areas. You can immediately see where prices are rising or falling, where there is a change in the volume of transactions. Hiring people to track these trends is too expensive, and information is in demand.

This customer worked with Narrative Science for several months to customize the program for a specific data structure, but the result exceeded all expectations. One of Hanley Wood’s executives, a former professional journalist from Thomson Reuters, admitted to be amazed at the quality of the articles. “They crossed the great linguistic barrier,” he notes. “The articles are in no way alike.”

Although they have only 20 customers, Narrative Science is already making good profits. The company takes $ 10 for each article with a volume of 500 words from its customer Hanley Wood (for comparison, 881 words in this text), that is, about $ 3,500 per month. At the same time, this is a very good deal for media sites - even cheaper than the cost of articles from content factories like Demand Media .

No doubt, in the future the quality of Narrative Science texts will only improve, and the cost will gradually go down. In the future, anyone can use such programs. According to experts, this opens up new horizons for journalism, as professional reporters get their hands on a powerful tool for data mining. You can study the topic from all sides and find such relationships that it was impossible to even imagine.

Well, Chris Hammond himself quotes the press with a grin, which writes that in 20 years the computer will be able to receive the Pulitzer Prize in Journalism. The professor believes that this is not so. The computer will receive the prize not in twenty, but in five years, and the prize should be awarded to the authors of the program.

Also popular now: