In articles about machine translation or auto-writing algorithms, the question is always the same: is a machine capable of doing a human’s job? I believe the correct approach would be to consider how machines and humans could take the communication process as a whole one step further, together. I can think of at least four very specific circumstances in which software helps us to do things that were impossible a short time ago.
High Volume of Data
Crowdsourcing software allows the services or the content that is needed to be obtained by soliciting contributions from people around the world. Crowd-assisted translation, for instance, requires the participation of human translators via social media to deal with a large body of text. The journalistic versions are tools that create stories from social media content, or through the participation of citizen journalists. They process enormous quantities of data, and may serve also to discover patterns in coverage, such as when someone has been quoted previously.
Machines can, in many instances, detect human error or deception. In translation, we use quality assurance software that compares segments of bilingual texts to automatically detect formal errors and enable their correction. In journalism, the latest is Truth Teller, a fact-checking service created by The Washington Post that analyzes political speeches in real time “because we think that politicians lie,” said the Post‘s executive producer for digital news, Cory Haik. And there is nothing that software can do about that.
A story-writing algorithm can relieve journalists of some routine tasks, the same way translation memories do for translators. Here is how the algorithm works: first, the developer feeds it with a bunch of data and helps the machine find the interesting aspects of an event. Then the robot journalist has to be provided with narrative structures assigned to different points of the story before it is ready for the last step, natural language generation. Some media are already employing this technology for financial news, sports and, like Los Angeles Times and their Quakebot, to offer last-minute information about earthquakes.
Translating journalistic content meets all the previous requirements: a large quantity of text to be translated in almost no time, the need to check for omissions, inconsistencies, formatting problems, and terminology errors, and a routine task that may have use for some kind of machine translation. One of the leading newspapers in Barcelona, El Periódico de Catalunya, overcomes the technical challenge of publishing in Catalan and Spanish on a daily basis thanks to its translation software. The software, surprisingly enough, is not based on computational linguistics like the robot journalist. It cannot analyze a sentence, distinguish between genders, or conjugate verbs. The method used by El Periódico consists simply of two databases; one contains words and the other contains sequences. It’s perfectly suited for written news, and it is preferable for one powerful reason: speed.
According to a study published by the Karlstad University in Sweden, stories written by robot journalists tend to be more boring than stories written by humans, but the interesting part of the study is that people can’t tell them apart. Machines definitely give us the opportunity to do the unthinkable; for instance, reporting about earthquakes around the clock immediately after they happen. Besides, if search engines are all over the new Thomas Pynchon novel, they can’t be that boring in the end.