The Current State of NLP, Voice Search, and Voice Assistants
Part one of our series " The current state of AI - Where we are right now, and what developments can be expected soon?"
I learned a lot about voice search capabilities when I was in Scotland for a few months earlier this year.
My husband and his family have heavy Scottish accents. And Alexa, well, she struggles with heavy accents. I’ll never forget the image of his entire family crowding around the Alexa, trying to get her to play a song, (Pleey oos a wee song, Alexa). Exasperated, they finally all turned to me, looks on their faces indicating I should just ask her myself. But even with my easy-to-understand Canadian accent, I too struggled at times, often having to repeat myself two or three times before Alexa understood what I was saying.
That said, I was fascinated to read about the leaps and bounds being made in Natural Language Processing, which is the key to unlocking voice assistant capabilities, powered by voice search technology, which is where we will start part I of this series.
“Dr. Pérez believes that the unicorns may have originated in Argentina, where the animals were believed to be descendants of a lost race of people who lived there before the arrival of humans in those parts of South America."
“While their origins are still unclear, some believe that perhaps the creatures were created when a human and a unicorn met each other in a time before human civilization. According to Pérez, ‘In South America, such incidents seem to be quite common.”
Would you believe that the above text was entirely written by a machine? Its only prompt was the following: “In a shocking finding, scientist discovered a herd of unicorns living in a remote, previously unexplored valley, in the Andes Mountains. Even more surprising to the researchers was the fact that the unicorns spoke perfect English.”
Natural Language Processing has made huge strides this year. So huge, in fact, that on 20 June, 2019, it surpassed the human baseline for understanding language.
That means NLP now has a better understanding of language than the human baseline average for tasks like logic, common sense understanding, and lexical semantics. That’s pretty awesome.
However, one of the major limitations of NLP at the moment remains its inability to deduce intent from a query, according to Gartner. In order to understand intent, it’s necessary for voice search to identify the specific domain/product set to be searched.
“This helps generate appropriate answers or next questions for the customer as the conversation leads to a product selection,” the report explains. “Some vendors have NLP and natural language understanding (NLU), but few have as yet extended understanding to the product domain, providing steps toward conversational search.” We’ll get there with NLP, but we’re just not fully there yet.
With the incredible NLP advancements in mind, it’s not surprising that voice assistants are the hot AI topics of the moment. But, believe it or not, we’ve not seen the same impressive progress with voice assistants that we have with so many other aspects of AI.
In fact, a digital personal assistant study that posed 4,999 questions to the most popular voice assistants on the market showed that every personal assistant -- from Google Assistant to Siri to Cortana -- had dropped in accuracy since last year.
According to the study,
“This indicates that current technologies may be reaching their peak capabilities. The next big uptick will likely require a new generation of algorithms. This is something all the major players are surely working on.”
By the way, if you’re curious which assistant ranked the most accurate, it was the Google Assistant on a smartphone.
That said, the report did show that all digital assistants attempted to answer more questions than they did in 2018. With this in mind, we can reasonably assume that once some more complex algorithms are developed, digital assistants will really take off on a whole new level in the coming year(s).
That’s important, because as of June 2019, one in five UK households had a smart speaker, and one in 10 Irish and German homes featured a device as well. In Q1 of 2019, 3.35 million smart speakers shipped to homes across Europe, up 58 percent since Q1 2018. Forecasts predict smart speaker sales in Europe to hit 23 million for the entirety of 2019.
Alongside voice assistants, voice search is being touted as the next big thing in 2020. A popularly quoted statistic is that half of all searches in 2020 will be voice. But upon further digging, it turns out that this statistic was actually a prediction by the Chief Scientist at Baidu back in 2014, who was predicting something in five years’ time.
Now, as we ease our way into 2020, that forecast looks unlikely to come true, according to an eConsultancy report by voice search expert Rebecca Sentance, who explains,
“This means that the amount of voice queries on Google would need to increase by another 22 percentage points over the next year and a half for Ng’s prediction to come true. To reach 50%...we would need to find an additional 1.3 billion voice searches per day from somewhere.”
The report goes further to expand on the problems being encountered by voice search at this time, including problems understanding user questions (which can mainly be attributed to poor NLP performance).
Voice search is next to impossible for people with heavy accents and speech impediments, as evidenced by my anecdote about my Scottish family and Alexa. And while that paints an amusing picture, the struggle is real.
The major reason, though, that Sentance doesn’t think voice search will take off in its current form? “It doesn’t do what it’s supposed to,” she explains. And she’s right. The primary use of voice search is to respond to a user query with a single, definitive result. But right now, it doesn’t really do that most of the time. While sometimes, voice search will respond to a question with information pulled from the most reliable source on the topic, other times, a voice search query will simply turn up the same results that a typed search would have, with a list of websites that rank well for your particular search terms.
The next steps
Between NLP, voice assistants, and voice search, it may feel like we’re taking two steps forward and one step back. Well, that’s because we are. That’s how technological advancements work -- through trial and error. And as increasing numbers of people populate their homes with digital voice devices, the number of trials is set to increase exponentially. This fact alone will speed up the development of NLP, voice assistants, and voice search, and I expect to be writing a much different article on this topic at this time next year -- maybe even entirely dictated by voice.
- https://openai.com/blog/better-language-models/. Accessed 13 Nov. 2019.
- https://www.slideshare.net/StateofAIReport/state-of-ai-report-2019-151804430. Accessed 14 Nov. 2019.
- https://www.gartner.com/en/documents/3902963/market-guide-for-digital-commerce-search. Accessed 14 Nov. 2019.
- https://www.perficientdigital.com/insights/our-research/digital-personal-assistants-study. Accessed 14 Nov. 2019.
- https://voicebot.ai/2019/10/11/over-20-of-uk-households-have-smart-speakers-while-germany-passes-10-and-ireland-approaches-that-milestone/. Accessed 14 Nov. 2019.
- https://www.fastcompany.com/3035721/baidu-is-taking-search-out-of-text-era-and-taking-on-google-with-deep-learning. Accessed 14 Nov. 2019.
- https://econsultancy.com/the-future-of-voice-search-2020-and-beyond/. Accessed 14 Nov. 2019.
Stay in the loop
Subscribe to our email newsletter. No spam, just occasional insight from our experts.