Which authors of this paper are endorsers? Disable MathJax (What is MathJax?)

APPLE released the iPhone 7 last week along with iOS 10, a major upgrade for its popular mobile operating system. What better time to download some apps to take the new hardware and software for a spin? In one update, Apple's messaging system, iMessage, got a turbo boost. It now lets people embellish conversations with stickers, interactive drawings and animations. The messaging system even gets its own App Store for downloading third-party stickers and games.

This tutorial will explain some practical tips about how to train a neural machine translation system. It is partly based around examples using the lamtram toolkit. Note that this will not cover the theory behind NMT in detail, nor is it a survey meant to cover all the work on neural MT, but it will show you how to use lamtram, and also demonstrate some things that you have to do in order to make a system that actually works well (focusing on ones that are implemented in my toolkit). This tutorial will assume that you have already installed lamtram (and the cnn backend library that it depends on) on Linux or Mac. Then, use git to pull this tutorial and the corresponding data. The data in the data/ directory is Japanese-English data that I have prepared doing some language-specific preprocessing (tokenization, lowercasing, etc.). Machine translation is a method for translating from a source sequence F with words f_1, ..., f_J to a target sequence E with words e_1, ..., e_I. This usually means that we translate between a sentence in a source language (e.g.

Watson APIs are now available for public use, albeit only through IBM's Bluemix cloud services platform. IBM's Watson Developer Cloud now offers eight services for building what IBM describes as cognitive apps, with more services promised later on. The Relationship Extraction system seems less limited by available data than Machine Translation, but it is limited in different ways. When the Relationship Extraction system is fed the sentence "Nick Cave's new film '20,000 Days on Earth' debuted yesterday," it understood that "Nick Cave" was a person and that "yesterday" was a date, but didn't understand that "20,000 Days" referred to the title of a work.

Those who have been chomping at the bit to use IBM's Watson machine-intelligence service with their apps need gnaw no longer. Watson APIs are now available for public use, albeit only through IBM's Bluemix cloud services platform. IBM's Watson Developer Cloud now offers eight services for building what IBM describes as cognitive apps, with more services promised later on. Of the services offered so far, Visualization Rendering seems the most immediately useful and powerful, since it isn't limited by data training many of Watson's other services rely on. Most of the services rely on a "corpus," or cultivated body of data that Watson can use as raw material, so the breadth of several Watson offerings is limited by the size of their existing corpora.

Samsung has filed for a patent covering a mobile device that could run a second operating system via virtualization. The Samsung patent application, reported on SamMobile and MSPowerUser, covers all sorts of digital devices, according to Samsung: smartphones, tablet PCs, notebook computers, and more. "In this specification, the case where as an example, the first operating system is the android, the second operating system is the window (Windows) is shown," the patent's translation reads. But the patent also makes clear than any OS could be used, including Tizen, Linux, or MacOS. In the world conceived by the patent, the host OS would run the device, and the secondary OS would be run essentially as an app.

Google Crowdsource lets you help the company with language translation, handwriting recognition and map translation accuracy. Google has a formidable artificial intelligence team working on everything from photo recognition to email spam filtering. The online giant on Monday released a new Android app called Crowdsource that lets you contribute your own suggestions to language translation, handwriting recognition and street sign transcription. "Each microtask takes no more than five to 10 seconds, so knock away a few the next time you find yourself with a few moments to kill," Google suggests in the app description. "Every time you use it, you know that you've made the internet a better place for your community."

Online retailer eBay is attempting to extend its machine language capabilities beyond automatic language translation to e-commerce uses designed to make product searches more relevant. As automation improves, the company said one goal eliminating the search box. Meanwhile, development cycles have been reduced as more machine learning libraries are released to the open source community. "As machines get better at decoding natural language, commerce should become increasingly conversational -- eventually rendering the search box redundant," eBay CEO Devin Wenig noted recently. Wenig added that the pace of machine intelligence development has quickened over the last year.

So how do we program a computer to translate human language? The simplest approach is to replace every word in a sentence with the translated word in the target language. This is easy to implement because all you need is a dictionary to look up each word's translation. But the results are bad because it ignores grammar and context. So the next thing you might do is start adding language-specific rules to improve the results.

Neural machine translation aims at building a single large neural network that can be trained to maximize translation performance. The encoder-decoder architecture with an attention mechanism achieves a translation performance comparable to the existing state-of-the-art phrase-based systems on the task of English-to-French translation. However, the use of large vocabulary becomes the bottleneck in both training and improving the performance. In this paper, we propose an efficient architecture to train a deep character-level neural machine translation by introducing a decimator and an interpolator. The decimator is used to sample the source sequence before encoding while the interpolator is used to resample after decoding. Such a deep model has two major advantages. It avoids the large vocabulary issue radically; at the same time, it is much faster and more memory-efficient in training than conventional character-based models. More interestingly, our model is able to translate the misspelled word like human beings.