Newsletter 
Researchers from Google have recently introduced two new applications that were made possible by the search giant’s work on natural language understanding. In an effort to develop algorithms that better understand the human language, the search giant focused on developing hierarchical vector models. These models use word vectors, a tool that enables algorithms to learn and understand the relationships between words in terms of similarity of ideas and language. Furthermore, the search giant noted that it has also started using vectors in determining relationships between larger clusters of words like sentences and short paragraphs. The hierarchical vector model is the same machine learning model that powers the Smart Reply functionality of Gmail.
The two services introduced by the search giant’s researchers can be accessed through the “Semantic Experiences” website. One of the two applications is dubbed as the “Talk to Books” service, which allows users to search for books by asking questions or formulating statements. The algorithm developed by Google then parses the contents of the books and retrieves specific statements that respond to the questions or statements made by the user. However, the tech firm mentioned that the service can still be improved significantly. For example, the search giant stated that there are occasions wherein a statement is taken out of context by the algorithm. In addition, it could also be difficult for the algorithm to understand more complex questions and statements.
Users may also access the Semantris game through the Semantic Experiences website. Semantris is a word association game that uses the machine learning algorithm in judging the relationship between the words on the screen and the answers of the users. There are two versions of the game, dubbed as the arcade and the block versions. The key difference between the two versions is the lack of time pressure in the block version, which may allow users to answer not only words but also phrases and sentences. In the near future, the tech firm hopes that its machine learning algorithm can also be utilized in use cases like classification, semantic clustering, and whitelisting applications. Interested developers may also experiment and develop their own applications for the search giant’s algorithm by using a pre-trained semantic model, which can be obtained from the TensorFlow platform.
