The headline of a May 24, 2016 press release reads, “Nortek Acquires Next-Generation Language Processing Platform Nuiku.” The release calls Nuiku “a next-generation natural language processing platform that facilitates sophisticated voice control across home automation and other applications.”
I’m into natural language processing — I follow NLP and applications as an industry analyst and consultant — so when my friends at Slator asked my take on the transaction, I thought I’d take a closer look. How could I not, reading that “Nuiku is a leader in natural language processing,” even though the claimant is biased: Nortek CEO Michael J. Clarke? So was this a successful exit pure and simple — a favorable sign for other NLP start-ups — or is there more one can learn and infer from the acquisition?
I’ll telegraph my conclusion: Nuiku is far less than a “next-generation NLP platform.” The acquisition circumstances tell a cautionary tale about how difficult production NLP is to bring to market and how quickly the NLP state-of-the-applied-art is advancing.
Further, Nortek is betting that it’s advantageous to own its own voice interface for the company’s HVAC, security, and control systems. Will maintaining a single-target voice interface prove tenable strategy in the face of super-heavyweight competition from the likes of Amazon Alexa? We will discuss this point as well.
According to Nuiku co-founder Sean Thompson’s LinkedIn profile (retrieved May 30, 2016), “Nuiku, Inc is building the next-generation user experience platform that allows people to talk to things using natural language (voice or text).” His profile states Nuiku “is using natural language processing and data intelligence to power a revolutionary new user experience.”
I wrote Thompson and co-founder Bill Baker, whom I know of from his days heading Microsoft SQL Server development, for comment. I didn’t receive a response, so my analysis will be based on publicly posted material.
The co-founders raised $1.6 million in seed funding in September 2013, according to Crunchbase. What they were shipping, per a YouTube video posted July 20, 2015, was “a platform that language-enables other applications,” a device-form interface that accepts, interprets, and processes inquiries and directives aimed at sales and customer-relationship management (CRM) systems including Salesforce and SAP.
But check out a series for three stills from Nuiku-published videos, the first two from the July 2015 video and the third from a July 16, 2014 video stating the aspiration, “we hope this application is going to help you spend less time, being frustrated with interacting with Salesforce and more time selling.”
What I see in Nuiku is what I’d call a natural-language command interface. It seems designed for a defined set of directives, engineered for specific target business systems. Actually, revise the label: A defined set of directives and constraints on the syntax and vocabulary you can use add up to what might be better termed a structured-language interface.
I don’t see anything resembling a “next-generation NLP platform,” what Nortek thinks it bought.
A Next-Generation NLP Platform?
Natural language processing (NLP) involves application of a language model, whether hand-crafted or built via machine learning, to common text and/or voice processing functions such as information extraction, summarization, and machine translation. What makes a system a platform, in my view, is that you can build on it, that you can add functions, adapt it to diverse data sources and outputs, whether via application programming interfaces (APIs) or code plug-ins or some other method. For example, Salesforce is the pioneer in delivering an enterprise services platform.
Nuiku handles a very narrow NLP function — it parses structured language rather than broader natural language — and extensibility is very limitation, which I’ll get into in a bit. Further, it is not “next generation,” not by a long shot.
A next-generation NLP platform is flexibly adaptable to new data sources (social, online, enterprise, and personal), languages, and analysis goals such as sentiment, event, and intent extraction. Nortek’s claims about Nuiku are grandiose, which isn’t surprising, because public-relations puffery is all too common, and because you can get only so far with $1.6 million funding.
Despite the overstated PR claims, Nuiku does seem like a strong fit for Nortek. But is today’s strong fit a good business bet, given the pace of technical innovation?
Natural Language Control
Nortek manufactures heating, cooling, and ventilation, security, and audio, video, and control solutions. A voice interface seems a natural progression from buttons, knobs, and sliders, whether physical or on-screen, and typed commands. That’s the premise behind consumer-device voice assistants. Macworld’s comparative review is interesting reading, “Apple Siri vs Microsoft Cortana vs Google Now vs Amazon Echo Alexa: Which is the best voice control technology?” It looks at accents, question-handling versatility, breadth of knowledge, and app and smart-home control capabilities. There’s no hint of those capabilities in what I can see of Nuiku, nor of the complexity-handling promised by providers such VocalIQ, acquired last year by Apple.
True next-generation NLP often turns to machine learning to handle diverse applications, languages, accents, and regionalisms. Looking at Nuiku’s approach, I wonder whether the acquired technology will scale for field deployment?
Nuiku’s Web site is history, but I found a cached “how it works” document. Pulling bullets from a slide on Logical Models:
• We have a rich logical model for the sales domain
• We understand the core entities, their relationships, etc. and have a lot of semantic and linguistic tags
• If your data model looks like our sales model, you’re in luck
• If not, we work together to build a logical model of your domain
• We can re‐use some of our model; to some extent, people are people regardless of the domain
• But your application has some unique elements
• We estimate a new domain model requires 3 to 12 person‐months of effort
That last point is a huge Ouch. It says the Nuiku offering was far from fully productized. (It’s also imperfect. Note, in the snaps above, mistranscription of “controller” as “control and.”)
The set of Logical Model points suggests an architecture antithetical to agility. Combine with the market’s preference for consolidated access to multiple systems, expressed in the current trend toward conversational interfaces replacing apps, and you start to question Nortek’s investment… unless what we have here was really a talent buy. It may have been.
NLP Market Lessons
Finally, NLP is hot. Why did Nuiku sell itself? My guess is that the company ran out of money and couldn’t attract a Series A, or perhaps they didn’t try. (Again, I did not receive a response to a note to Nuiku’s co-founders.)
It’s hard for a start-up to bring competitive, commercial-grade NLP to market nowadays.
I see the most promising route as exploiting an open-source machine-learning or NLP framework, trained with/for industry-specific data and a well-defined business function. Apply an open-source or start-up friendly engine such as Gate, Stanford NLP, Carrot2/Lingo3G, and Basis Technology’s Rosette engine for basic functions such as named entity recognition.
Nonetheless, there are some really interesting NLP start-ups out there, Aylien for one. I bet they’ll be bought by end-2017. That’s because it’s also not easy to make money and scale as an NLP provider. Lexalytics has done it. MeaningCloud, as part of a larger group, and Revealed Context, a spin-off of social-agency Converseon, are trying to make it work. Otherwise, options are acquisition by a larger tech provider (example: AlchemyAPI by IBM) or pivot to an application focus (like Luminoso, Kanjoya, and People Pattern, and like Clarabridge dating back 6-7 years) or acquisition by a solution provider… per Nortek’s acquisition of Nuiku.
Disclosure: Lexalytics, MeaningCloud, and Revealed Context are sponsors of the 2016 Sentiment Analysis Symposium, which I own and organize. (The symposium takes place July 12 in New York.)