Natural language processing has been developing through the years with the introduction of word vectors which give the ability for the AI to learn and understand sentences, words, phrases from any language.
Vector Representation Of Language
With the word to vector representation of the language in space, huge sentences and small paragraphs are represented in the form of a vector. Relation to other words, synonymy, antonymy, meronymy, holonymy, and many other types of relationships are all represented in vector space language models. Google’s new Talk to Books is a way to explore books with a help of a sentence and find the best match based on over 1,00,000 books in their collection. This is where Semantris comes into picture. It is a word association arcade powered by machine learning, where you submit words which are relevant to the highlighted words.
Talk To Books
With Talk to Books one can find different ways to explore literature. You type in a sentence or ask a question, and the model finds sentences in books that are the found in the books. It takes into the account the keyword matching metric to find relevant books. This way you can also get to know if the books interest you or not.
The model is trained on millions of phrases and sentences, where it learns to figure out what is the best response based on the user input. Once the user inputs a question or sentence, the models find all the phrases among the 1,00,000 books to find the best matches and respond based on the query. The principle of the technique is finding the semantic meaning of the sentence.
At this level, the models only work on a sentence level (a paragraph input is not available, like in Smart Reply for Gmail) but a good response from the model can be expected. One can also find books and passages that are not so relevant and the highlighted paragraph might not be an obvious one. You may also notice that being well-known does not make a book sort to the top. This experiment looks only at how well the individual sentences match up. However, one benefit of this is that the tool may help people discover unexpected authors and titles, and give a glimpse into books in a way that is fresh and innovative.
Google’s arcade game Semantris is based on word association application and powered by ML. A word or a phrase input is given based on the highlighted word and it then scores them on how well the model responds to what you have typed. Similarity, opposites and neighboring concepts are all fair-game using this semantic model. The time pressure in the Arcade version (shown below) will tempt you to enter single words as prompts. The Blocks version has no time pressure, which makes it a great place to try out entering in phrases and sentences. You may enjoy exploring how obscure you can be with your hints.
These are only a few examples of how Google is working on experience and application designs using machine learning tools. Other applications relating to the classification, semantic similarity, semantic clustering, whitelist applications (selecting the right response from many alternatives), and semantic search (of which Talk To Books is an example) are currently being worked on.
Smart Reply in Gmail
Responding to emails is tedious job, but with Google’s Smart Reply one can reply to these emails effortlessly by opting one of the quick responses. They are predicted on the basis of the content in email itself. The feature already drives 12 percent of replies in the Inbox on mobile. And starting today, Smart Reply is coming to Gmail on web too.
Smart Reply suggests three responses based on the email you received. Once you’ve selected one, you can send it immediately or edit your response starting with the Smart Reply text. Either way, you’re saving time.
Credit: The Driod Lawyer
Smart Reply makes use of the machine learning techniques to give the best responses based on the user emails. If you are more inclined to use “Sure, thanks” than “Sure”, it suggests the response based on the information gathered by the user previous habits. A new version of Smart Reply will roll out globally on Android and iOS in English first, followed by Spanish in the coming weeks.
As the technology is rising, natural language processing is reaching high limits of accuracy and deceiving humans. It is sometimes hard to predict the right phrase or a word, but with more and more users using the tech, the models will learn fast and shall be more accurate.