Microsoft Technology of Machine Reading

The research Microsoft Research Asia command in January, 2018 announced creation of technology of machine reading on the basis of artificial intelligence technologies. We will compare the result of accuracy of answers to questions received by researchers to results of the answer of the person, developers claim. Achievements in this sphere will allow technology services to interact with people more naturally.

For an experiment the data set for machine reading and understanding of texts of SQuAD (Stanford Question Answering Dataset) developed by specialists from Stanford University was used. It contains the questions based on articles from Wikipedia. On January 3, 2018 Microsoft offered the model which received in the test for search of exact compliance result 82.650. Average result of the people answering the same set of questions — 82.304.

As envisioned by developers, technologies of machine reading and understanding of texts will allow computers to process quickly contents of books and documents, and then to provide to people necessary information in a simple and clear type. For example, drivers will be able quickly to find the answer in the extensive reference book on the car, having saved time and efforts in difficult situations. Doctors, lawyers and other specialists will quicker cope with the tasks requiring studying of many documents for search of rare precedents and other specific data. Technologies in general optimize their work and will allow to find more time for treatment of patients and settlement of legal issues.

For example, Microsoft develops the technologies allowing the computer to answer not only an initial question, but also on the subsequent and also on the questions consisting of several offers.

The previous versions of the models of Microsoft used for passing of the SQuAD test are already used in the search system Bing. The company already works on their application for the solution of more difficult tasks.