27 September 2016 (Tel Aviv) - For the past three years I have tried to keep to a reduced conference circuit ... but I have failed. I added DLD Tel Aviv a few years ago, more as an opportunity to meet the key folks behind Google's Search organization, and it has not been without yearly surprises. If I learned nothing else from this conference it has been that a bunch of very, very smart people are continually evolving their thinking by making the most of today's digital tools, and that "out thinking" will continue to evolve as new tools enter our lives. That is the new, marvelous cognitive landscape.

Previously Google has said that it uses neural networks in Google Translate, but specifically for its  real-time visual translation feature. And  earlier this year in my Mobile World Congress coverage I noted that Google was working on incorporating deep learning into more of Google Translate. 

Sure enough: today's contributions are the result of that work.

Google has been incorporating deep neural networks into more and more of its applications, including  Google Allo  and  Inbox by Gmail . It's also  helping Google more efficiently run its data centers . It is also creating a stand-alone application for mass translation in situations like legislative texts, and e-discovery.

For GNMT, the company is relying on eight-layer long short-term memory recurrent neural networks (LSTM-RNNs), "with residual connections between layers to encourage gradient flow," quoting from the full technical paper (link below). Once the neural networks have been sufficiently trained with the help of graphics processing units (GPUs), Google relies on its recently unveiled  tensor processing units (TPUs) to make inferences about new data.

Easier, faster and more reliable translations, a world of seamless and immediate translation within our grasp. Yes, we still the usual issues: context, syntax, intonation and ambiguity. Because a computer system is not context aware, it could grab the wrong word. Additionally, it doesn't understand the language at all. It just tries to decode words, instead of decoding the meaning. Many languages are not similar at all, and do not have corresponding common words and/or their usage is not the same at all.

But with Google's announcement today, we are getting much, much better.

And before my summer holiday break I was in Paris and attended a Microsoft Research event for its Natural Language Processing group and saw the further developments (over the last year) on their Machine Translation (MT) project which is focused on creating MT systems and technologies that cater to the multitude of translation scenarios today, including legal. The key is Statistical Machine Translation (SMT) and that breaks down into areas such as syntax-based SMT and phrase-based SMT. Plus there is Word Alignment and Language Modeling technologies.

These advanced language modeling toolkits mean that problems with morphology, syntax, semantics and word sense disambiguation are being solved. Not completely solved yet, but coming. For the vendors and the multinational companies who need it, the business model is a no brainer. The value of an automated, instant, seamless translation platform to a corporation means the vendor that solves it could charge a substantial amount of money for such a tool.