Microsoft bought SwiftKey – why? – Tabletowo.pl

>

Once Microsoft tried to sell you a product, which has not yet had. Currently, Microsoft is buying products that you use … a long time. That happened with the post Acomplia, calendar Sunrise, Minecraft or Wunderlist.

Some editors, including Tom Warren (The Verge) and Paul Thurrott, proclaimed the death of Windows Mobile and suggest letting go of this platform. Regardless of whether you agree with this diagnosis, plan B Microsoft is very clear – a clear presence on all platforms with its software and services. Mainly for iOS and Android. Or maybe this is not a plan B just plan A, and Windows 10 Mobile is simply artificially kept alive?

The question of earning a free service and applications in the case of Microsoft is on-river, but perhaps more interesting are technical and technological issues. Why Microsoft bought the company responsible for SwiftKey, which is the most popular keyboard for Android and iOS? After WP8 keyboard for writing gestures (word flow) was abundant and recognized as one of the best keyboards for mobile devices? It is true. However, do not know why, W10 Keyboard for Mobile is a disaster when it comes to action AutoCorrect and autopodpowiedzi. This is partly due to the integrated mobile and desktop keyboards, but it does not explain all that fateful its activities.

All the known commercially keyboard with auto correction and hints are based on the model of the so-called n-grams. N-grams is based on statistics and are used for predicting the next item in the sequence. The use of n-grams requires gathering a sufficiently large resource statistics – the body (in this case, massive sets of texts and text typed by the user). Creating a model of n-gram begins counting occurrences of a sequence of fixed length n in the existing language resources. Each user, by introducing their own texts on the keyboard creates its own model, which allows for additional “personalization” of the keyboard. Due to performance limitations, mobile keyboard limited to 2-gram model, which robs them of some contexts and peeled only quantitative statistical method. However good it would not be a model, such a solution will never be enough “smart” to “know” what we mean and in what context we create a given statement.

That was until October 2015, when it SwiftKey presented keyboard neural Alpha based on a deep neural networks. In contrast to the model based on n-grams, the neural network captures the relationships between words, grouping them into clusters, which allows you to display more “contextual” hints taking into account more than a few words earlier in the sequence. And this is only the first stage. In the future, the keyboard may prompt not only single words but whole sentences, depending on the context in which the user is located.

Just what is all this to Microsoft? The basic version of SwiftKey is and will be free, and purchases inside the app will never be a notable source of income for Microsoft. So what we pay for the gratuity? As usual – data. Anonymous telemetry data that reach a much wider vocabulary expressions SwiftKew (soon Microsoft), which is a dictionary on a regular basis is used for learning neural networks used in a new type of keyboards. A potential amount of data coming from the keyboards billion mobile devices will be unprecedented. And now SwiftKey user base is huge – and led by Microsoft can only grow even more. This database is a priceless treasure for all future projects of machine learning related to the auto correction, predictive text, translating phrases into other languages ​​or pattern recognition.

SwiftKey will remain available and maintained for android and iOS, and in the coming months, Microsoft has say more about the integration of this keyboard with Windows. It is not clear whether it will be an alternative to the default keyboard as windows or however its complete replacement. I keep my fingers crossed for the second option – szczeólnie that the team SwiftKey joined the department of Microsoft Research (MSR), which clearly shows the great ambition of this project related to machine learning and predictive algorithms. And all of this, that within 5 years, thanks to the tactile keyboard autopodpowiedziom become faster than conventional physical keyboard. And it would be a breakthrough. Entering the longer pieces of text is one of the last bastions of devices with a physical keyboard for the average user. If this advantage disappears, smartphones and other devices touch even more dominate the market, with wider and wider application of (semi) professional.

source: Microsoft