Page 25 - The EDIT | Q3 2017

new bot is in the market and that it is our fault that they don’t understand us. But this would just raise the question: is it user error or is it really the current state of AI?
Instead of (just) processing information based on possible statistical matches, NLU looks for the meaning of information in context. It looks beyond statistics to actual meaning so it can understand what has been said. NLU would understand your “Call Beth... no, John”, it would know that you meant “Call John.”
The NLP solutions that we see today rely partly on statistical analysis of word order and frequency, which leads to a probability of accuracy — but we need NLU to create meaning and understanding with human-like accuracy and enable future chatbots.
What does the future hold?
The next generation of chatbots will depend on engineers, linguists and AI specialists coming together and realizing that NLP alone is not going to be enough. Acknowledging that, adopting context and meaning within the underlying framework that enables chatbots is critical.
NLU is really the only answer. It will truly allow the market of chatbots to deliver on the promise of providing optimized customer experiences, and more natural engagement with consumers to cross the uncanny valley of AI.
(Source: http://bit.ly/2vvTAj6)
It is a marathon – not a sprint
There is little doubt that tech companies are walking in the right direction when it comes to NLU. A recent
Discovery
article published on BusinessInsider reveals that Microsoft’s speech recognition efforts have hit a significant milestone as it can now transcribe human speech with a 5.1% error rate. Why is this so impressive? Because it is the same error rate as humans.
One of the reasons why Microsoft managed to lower its error rate was because they considered the context of the speech to make better guesses as to what unclear words are — exactly like humans do.
For example: It might not be clear from the audio whether someone is saying “that’s not fair” or “that’s not fur.” Traditionally, this ambiguity might lead to transcription errors. But speech recognition tech can now look at context for clues. If it is a speech about the risks of gambling, then it’s probably “that’s not fair”, but if it’s a conversation about fabrics, “that’s not fur” probably fits better.
Read the full article here: http://read.bi/2wg8RFP
Chatbots will include NLUI (Natural Language User Interface) to enable voice conversation. And they will appear as a layer across websites, display ads, apps and out-of-home experiences. The technology will enable people to intercept video ads and seamlessly engage with the characters in the ads.
[Merge – The closing gap between technology and us, 2017]
25
“Understanding language is the holy grail of machine learning.” [John Giannandrea, British computer scientist]
THE EDIT ISSUE 7 | Q3 2017