Some time ago, I was typing using Google Docs, when a red underline appeared under a word indicating a error. Nothing new: it happens dozens of times a day due to typos or some blunder. However, when I checked the word that the correction program had reported, I noticed something strange: Docs indicated as an error the form “what is” (without apostrophe), suggesting that I replace it with “what is” (with the apostrophe).
In a bizarre reversal of the parties, the Google software asked me to use the incorrect form, reporting as incorrect the one that is the correct form. Because you know (or you should know) that “what is” is written without apostrophe: the reason, as Treccani explains, is that the adjective and pronoun “which” also exists in the autonomous form “qual”, consequently it does not need the apostrophe when it is used. The same the same reasoning applies to the form “another”, which Docs suggested on several occasions to replace “another” with the wrong form: the indefinite masculine article “a” exists in an autonomous form, therefore it does not want the apostrophe.
If Google’s mistake is (also) our fault
Why is Docs reporting them as errors? The cause lies in the way the software has been trained to recognize our potential grammatical and spelling mistakes. Unlike classic correction programs, Docs did not receive the rules of Italian from programmers, but uses artificial intelligence: an algorithm statistically analyzes the use of language made by people and learn to distinguish the shapes that are most likely to be correct.
An example is useful to understand: if 90% of Italians writes “if I were to receive a gift I would be happy” e only 10% writes “if I should receive …”, the algorithm can learn that the first form is correct and the second is a blunder. The problem with “what is” and “another” is that the wrong use of the apostrophe (and therefore of the forms “what is” and “another”) is so widespread that the algorithm, reasoning only on a statistical basis, he convinces himself that it is the correct form.
To explain it is precisely Google on a dedicated page: “Suggestions are powered by machine learning. Because language understanding models use billions of common phrases to automatically learn about the world, they can also reflect mistakes e i prejudices humans. Being aware of this is a good start and the discussion on how to manage them is continuing. “
Artificial intelligences between racism and the ethical question
by Emanuele Capone
The limits of AI
It is one of the limits of artificial intelligence: accepting the form used by the majority as correct (and working only on a statistical basis, without even a smattering of rules), the algorithm excludes the possibility that the majority can make mistakes. It is a form of majority dictatorship applied to the linguistic sphere, which as such can also have practical consequences. It cannot be excluded that, by dint of reporting incorrect but widely used forms, Google Docs will further contribute to their dissemination, to the point of making them the canonical forms of use and therefore making them accepted.
Nothing particularly serious: language is by definition something that changes and evolves over time. If that weren’t enough, the use of “what is” has already been spotted in some literary classics, including Pinocchio by Carlo Collodi. In short, it would not be a tragedy if a form now considered wrong became the norm by dint of being used anyway. It remains to understand if we want an algorithm to contribute to the changes of the Italian language.
.