Alexa and Google improve their Spanish: more correct and inclusive

Meeting on the Spanish language and artificial intelligence in the RAE, chaired by its director, Santiago Muñoz Machado, in the center.

Marie Mulot from Amazon Alexa, left, Chema Alonso from Telefonica, David Carmona from Microsoft (on screen) and Ester Marinas from Google.Jaime Villanueva

A Spaniard, regardless of his accent, can speak with Alexa, the virtual assistant developed by Amazon, without communication problems. Then you can send a message or write an email through Google platforms with a certain guarantee that both systems will correct spelling errors, now also grammatical errors, and if you need to translate a profession the results will include both gender options. And when you have to write a document in Word, the Microsoft word processor will offer you solutions to semantic questions. These three companies are teaching machines that work with artificial intelligence to speak correct Spanish thanks to the Spanish Language and Artificial Intelligence (LEIA) project of the RAE, the 22 sister academies around the world,Telefónica and the technology companies that, two years ago, when this idea was presented, have joined the initiative.

More information

• Androids dream in correct Spanish

• The RAE wants the machines to speak academic Spanish

"I was born programmed to learn, what I need is for you to be my teachers," says a female voice, the gender that has been chosen for machines that use artificial intelligence, in the LEIA advertising campaign. These systems, as Chema Alonso, director of Telefónica's global Digital Consumption unit and technical director of the project, recalled this Monday morning, are programmed with ones and zeros; From there, the challenge they have proposed is that the algorithms used for artificial intelligence do not have English as their first language, but rather start from "a universal and natural language". "We have made the leap with models so that machines can learn with words," said Alonso, while acknowledging that until now algorithms were trained "with massive data without prior review,without analyzing that correct Spanish was used ”.

This universal and natural language model trains machines in any task with the benefit of breaking down language barriers. "The same day these features are available in English, they are also available in Spanish," said David Carmona, CEO of AI at Microsoft Corporation, by videoconference. "These are massive models that reason about language in much the same way as humans do," continued Carmona, who has given an example of how their children learn mathematics in the US. "They only need to do it in a language to be able to develop the task in English or Spanish ”.

The technological ones not only advance so that their machines use correct Spanish, but they adapt gradually to the different versions of the 22 Spanish-speaking countries, more than 600 million people with their accents and language peculiarities. Alexa, explained Marie Mulot, general manager of Amazon Alexa in Spain, proposes synonyms, knows sayings, recites poems, has learned that sentences in Castilian have a different structure from those of English and has been adapted to the Spanish of Spain, Mexico and that of Spanish-speakers in the United States.

Alexa is also making progress in trying to overcome the burdens of machismo and does not wish its users to be happy or content, but rather to "be in a good mood", as a solution to gender bias. The Google translator, as exemplified by Ester Marinas, responsible for brand and reputation of Google Spain and Portugal, circumvents these structural problems that not only affect the language, by offering, for example, the translation of professions in masculine and feminine and thus avoiding perpetuate the imposition of a single gender in trades traditionally linked to men and in those related to care, which are usually identified with women. "This solution is complicated in the context of a sentence," he stressed, "trying to distinguish the translations of independent words from longer texts."

The road, the companies involved have affirmed, is still long. On the horizon is the goal of getting machines with artificial intelligence that are already beginning to create texts to do so not only correctly, but with some kind of sense. A future in which these systems compete with writers and, as Santiago Muñoz Machado, director of the RAE, has said, “perhaps they feel next to academics”. Until that day arrives, the technology companies have another stop on the route: get the certificate that the Academy will issue to certify the proper use of Spanish. "It is more difficult for us to impose that quality on the machines, but the owners of those machines who want to teach them to speak well, with approved criteria, do understand us", summarized Muñoz Machado.