Monografico_3_EN_Doxa_29

texts of that same month written in previous years (that were introduced in the first phase), combined both elements, and created the news. The result is a text with a specific topic, natural language, and easy to understand, which has been represented in an XML or JSON file. This document was sent to customers through the API system, which allows the media to access all the information available in the company’s database via URL. In addition, all informative texts created by Gabriele are accompanied by an image, which is also created through algorithms. “Our program allows for greater content production, at more effective costs, which leads to greater commitment toward niche audiences and an increase in productivity of the editorial staff, who can focus on more complex stories”, as Martinez de Lecea emphasizes. At the same time, Moratilla argues that the system is pushing the limits of what was thought possible, both in terms of production quality and its wide variety of potential uses.

All the content generated is published on the website of the different media with which the startup collaborates, yet at no time does the media reveal to the reader the authorship of the news. In other words, they publish the text with the signature of the newspaper in question without alluding to the automatic generation techniques used by the startup company. For that reason, Narrativa Inteligencia Artificial recommends that the media indicate the fact that the content has been created through an algorithm and reveal the source of the data, as is already done in the United States.

3.3. Quality of the texts produced by Narrativa Inteligencia Artificial

Analysis of the journalistic quality of the texts generated by Gabriele as a result of the 145 journalists surveyed sends contradictory signals. On one hand, the sample places a positive value on the neutrality, veracity, syntax, brevity and coherence of the facts that are narrated, but considers that the originality, variety, style and rhythm are questionable, and therefore can be improved.

In the case of the news about Apple’s quarterly results, 78.87% of the sample “fully agree” with the syntax used in the text and 62.15% believe that the consistency in the presentation of ideas is correct. With regard to cohesion in the wording, 51.87% believe it is adequate, while the remaining 48.13% think that it could be improved, since there is no uniformity in the representation of symbols (21.15%). The journalists also agree on the point of grammatical correctness: 40.87% “partially agree” that there is clarity in the ideas and expositive transparency, while 15.12% “totally agree”. The remaining 44.01% are neither in agreement nor in disagreement. In terms of news coverage, 55.09% of those surveyed consider that the text is sufficiently current and important to be published in any media. In addition, 77.56% “totally agree” on the accuracy of the data that are narrated with regard to the news event since the software completely avoids words of broad meaning and uses precise terms. In the opposite way, the professionals gave a negative evaluation to other aspects related to the aesthetic features of the text. The figure of 92.18% criticise the fact that there is no diversity in points of view, and 83.21% think there should be a greater variety of narratives as well as a higher level of quality in information sources. In the same vein, 83.13% of the journalists who took part in this research believe the text lacks context and interpretation, which makes it superficial (29.51%), and they also think it is too restrained and concise (9.79%).

The news about the intention to vote in Spain in the general elections on 26 June 2016 show similar results. The figure of 63.56% of the sample surveyed “totally agree” with the order and relationship of the noun phrases in the information, and