APR 18, 2018 8:45 AM PDT

Talk to Books and Play Word Tetris With Google

WRITTEN BY: Julia Travers

Would you like to choose a new book to read by sending interview questions to a library of 100,000? If you had asked your favorite story a question before reading, do you think you would have enjoyed the response? Google Research is inviting the public to try out a new feature that allows people to conversate with books, in a fashion. Users can enter a question or statement and the artificial intelligence (AI) will respond with a sentence from a book that it finds related. Talk to Books is one of two Semantic Experiences Google is inviting people to try – the second is a Tetris-like word-association game called Semantris.

images from Google's Semantic Experiences, credit: Google

Talk to Books

Talk to Books is a demo and game built on their research into training AI to use natural language processing – to use language like people. To do this, they relied on machine learning, which Google defines as “a program or system that builds (trains) a predictive model from input data." The input this AI was trained with was a billion pairs of statements, wherein the second statement answers or responds to the first. The team also utilized word vectors, in which words are represented in a space “where semantically similar words are mapped to nearby points,” to empower algorithms to learn about relationships between words and statements.

“Since language is composed of hierarchies of concepts, we create the vectors using a hierarchy of modules, each of which considers features that correspond to sequences at different temporal scales,” Google Research Director of Engineering Ray Kurzweil and Product Manager Rachel Bernstein explain in an announcement.

“Relatedness, synonymy, antonymy, meronymy, holonymy and many other types of relationships may all be represented in vector space language models if we train them in the right way and then pose the right “questions,” they add.

The developers hope this tool could help people to find interesting books by starting with their internal text, rather than at the title or author level; to “surface books in a way that is fresh and innovative.”

Samples are below.

Question: Where is the best place to travel in the summer?

“...heat load which the exposed skin is receiving from the sky. The British Isles, despite the advantage of long summer days, experience a cloudy climate compared to Southern Spain, where the sun shines brightly for as much as 80 percent of the daylight hours. Ultraviolet (UV) radiation is even more intense in tropical … ” from Worldwide Destinations: The Geography of Travel and Tourism

Statement: Tell me a love story.

“He said, 'Listen, and I will tell you a story. Once upon a time a boy and girl met each other and they fell in love. They loved each other so much they got married … ” from Miguel Street.

Semantris

Google’s second Semantic Experience is a word game that uses the same AI, with two versions.image of Semantris Blocks and classic Tetris, credit: Google and public domain In the Blocks game, players enter a clue they think matches one of the words on the screen, which are written on colored blocks. The AI then tries to select the associated block. A match gives the player points and the word block disappears. If the stack of blocks reaches the top limit, the game is won.

In the Arcade version, which is faster, players type a response to a given word. Points are awarded and all the words the AI thinks are related then move to the bottom and “break,” somewhat like pieces in Tetris. Also, as in Tetris, the goal is to keep the word blocks from reaching the top of the screen.

Both AI and word-lovers may enjoy playing with and learning about the language skills, habits and imperfections of Google’s natural language AI in these experiences. The Google researchers remind the public that this is a working demo, and not a precise search tool. They invite people to try their sample queries and experiment with new questions and statements; “Play with it,” they suggest. They also welcome feedback and recognize the possibility that responses could be nonsensical or even offensive, due to the AI’s potential to reflect human cognitive biases.

 

Source:

Google Research

About the Author
Bachelor's (BA/BS/Other)
Julia Travers is a writer, artist and teacher. She frequently covers science, tech, conservation and the arts. She enjoys solutions journalism. Find more of her work at jtravers.journoportfolio.com.
You May Also Like
Loading Comments...