Google has not finished having fun with books, and more in its work around artificial intelligence. We know that Mountain View’s natural language processing is at the heart of the strategy. And the experiments are not lacking. This one, Talk to Books, is most captivating.
Google Research, the digital experiment lab, has just launched a site, Talk to Books, that really serves what it refers to. The idea is still rather limited – and first of all because research is limited to the English language. But one can ask a question, or grab a quote, and the site will find sentences from books, related to the subject of the research.
The big difference is that the system does not rely on paired key words: the artificial intelligence that makes the tool rotate has been goinfrée of ” billion pairs of sentences, like conversations “. This allows him to identify what could be a good answer to the question asked. Talk to Books is therefore a model of bot chat based on digitized books, and whose machine extracts appropriate formulations and responses.
On the other hand, the answers are not really human: the AI mainly uses keywords within a sentence. But the ultimate idea is to discover books, asking simple questions. ” Talking to him with phrases will provide better results than keywords. And that, because AI is based on human conversations “Says Google Research.
More than 100,000 books have been used for Talk to Books, with some limitations in the results we obtain. Mainly because the result sentences are extracted from any context, and the machine does not know how to contextualize. The book could therefore be completely off topic as a whole, and have only a simple keyword connection with the search.
On the other hand, the mass of documents used makes it possible to bring out a maximum of diversity in the presented books.
When Google, Amazon and Facebook will have shaped this world without thought or soul …
All this engineering, says Ray Kurzweil, director of the department, and Rachel Bernstein, product manager at Google Research, continues the exploration of natural language and its evolution. The whole program plays on the multiple strata of language, integrating linguistic tools such as synonyms, antonyms, metonymy, or homonymy. The established relationships are then translated into a spatial vector language, to provide a relevant answer.
In any case, as much as possible.
Try so Talk To Books .
Artificial intelligence at the service of books and reading