AI and machine learning have allowed the generation of software capable of generating informative texts in several languages, including Spanish
Journalists are no longer alone when exercising their trade: Artificial Intelligence (AI) and machine learning have allowed us to create software capable of generating informative texts in Spanish. Leo, developed by LeoRobotIA and Dail Software – a company specializing in AI and Natural Language Processing (PLN) – is a good example. Another example is Narrativa, a Spanish startup that has aroused the interest of United Arab Emirates investors, the Xataka site reported.
Despite existing less than 5 years ago, both tools have achieved a remarkable implementation in the media in Spanish. In mid-July Narrativa and EFE signed an agreement whereby the agency will distribute and market Narrativa products. The agreement allows EFE to incorporate into its offer “content generated by artificial intelligence” related to sports, financial, meteorological information, statistical data or surveys and electoral results.
Narrative too He works with several specialized media in Spanish, both on paper and digital: Sport, 20 Minutes, El Periódico, El Confidencial, El Español, Heraldo, El Independiente or La Información. It also provides content for specialized websites and companies, such as Noon, Intigral, Property Finder, The Social Audience or the Matchapp sports app.
The CEO of Narrativa, David Llorente, affirms that the startup collaborates today “with more than 25 media”. “For each one we generate a different version. We produce real-time and unique news per customer,” he says. Both Narrativa and Leo obtain content using tables of measurable data: statistics with sports, financial results, lotteries, meteorology, economic balances, etc. that convert into informative texts. At the moment, this leaves out genres such as literary reporting, political, judicial and travel chronicles, event coverage or interviews. “The idea is to transform structured data, for example an Excel sheet, with identified columns and data, into news,” says Llorente.
The mechanics are similar in Narrative, although those responsible influence the use of the PLN. “Based on structured data provided by a provider, we teach the machine to write football, smartphones, cars, financial results … and we also teach you how to write as the client wants, “says Juan Carlos F. Galindo, CEO and co-founder of LeoRobotIA:” I put the style book of that client or media and the machine He is able to write as he wishes and also with infinite variability; that is, every time, with the same data, I can elaborate a different text infinitely many times “.
“Once I have trained the machine, that I have defined how the data is, that I have trained it as the client wants, it is automatic. The machine drinks from the data source directly, automatically the media decides when and how publishes and in thousandths of a second the robot is able to publish information of one thousand, two thousand, three thousand … characters, whatever the client wants, “Galindo points out before emphasizing that, unlike other similar systems, Leo does not resort to standard templates that the program is responsible for filling out.
News generating softwares are not exclusive to Spanish. Neither new. International agencies such as the Press Association, Associated Press (AP), Reuters or Bloomberg have long used AI to produce content based on structured data and plan to strengthen their commitment over the next two years.
Reuters and Bloomberg, in fact, already generate more than 30% of their informative pieces with this type of systems. As EFE recalls in the note announcing its agreement with Narrativa, in Britain it has even launched Reporters and Data and Robots (RADAR), a pioneering project promoted by the Press Association (PA) in collaboration with Urbs Media that allows media premises have parts generated automatically. Since mid-2018, it has more than 180,000 articles.
In recent years, media such as the New York Times, Forbes or the Los Angeles Times have been on the car. The LA newspaper in fact starred in a controversial sound in 2017, when it notified by mistake of an earthquake that had actually been recorded in 1925. The author of the news was Quakebot, a system connected to the US Geological Survey database ( USGS).
Slips like Quakebot or Tay, the chatbot that Microsoft had to withdraw in 2016 because it posted xenophobic and sexist comments on Twitter, has not prevented the use of robots is increasing in the newsrooms. The Washington Post, for example, uses Heliograf to write news about sports or certain political pieces. It was also at the beginning of the year that Forbes was testing a tool called Bertie. Google itself has invested around 700,000 euros through its Digital News Initiative division in the promotion of RADAR.
Galindo remembers how the robots of four or five years ago took a minute to write a page. Today your system is capable of generating information of thousands of characters at an unimaginable speed makes a little five years. “What will be the coexistence with journalists? It will help them to do a job of higher quality, more human and more interest,” Llorente agrees. Thanks to the robots, the editors – the Narrative CEO reflects – will be relieved of routine work, such as writing pieces with the results of the lottery or the weather. “We cover news that the media cannot cover or for which you need a small version that your journalist can edit,” he notes.
Its other great advantage is that it allows journalists to make more personal pieces: “All these texts are editable. They can be published as is or add value for the text to be differential. “The Associated Press assures in fact that thanks to the use of robots, approximately 20% of its workforce has stopped developing simple articles to take on larger investigations.
In spite of the virtues and the multiple advantages that robots will have, the debate continues on the table as to whether they will end up eliminating employment. In February Magnet collected that at stake could be 7% of political analysts and 89% of technical writers. The University of Oxford and the American Society of News Publishers reduce the percentage of workers at risk for automation to 10%.
Day by day The main ally of human reporters is the limitations of robots. When working with statistics and numerical data, the new systems do not yet access the most elaborate genres and with the greatest presence of subjectivity. Galindo affects another key point: the credibility of the sources, especially when handling tools that, used with a doubtful intention, can become great propagators of fake news.