Contributed Commentary By Rana Gujral, CEO Behavioral Signals
Our lives are interconnected by billions of chatbots and voice assistants who take in language and provide a response, often in just fractions of a second. And while this has become something of second-nature for most consumers, the technology behind those rapid responses is relatively new and has evolved quickly in recent years.
Natural language processing (NLP), the AI-powered means by which computers dissect the words you input – in either text or voice – and process your request, has benefited from years of training and advancements in the algorithms powering it.
NLP refers to any process by which a machine analyzes and extracts meaning from human speech. This means it is done at varying scales of complexity. An NLP algorithm can be used to evaluate a large data file to extract certain terms and phrases. A more complex NLP algorithm exists in Siri and Alexa, parsing user input, converting it to text, analyzing that text and providing an answer, and converting that answer back into audio output.
Even something as innocuous as Google search results now use NLP algorithms to try to interpret the meaning of your search rather than just mining for specific results that match the keywords you provided. AI allows us to move beyond user-input to user-intent, extracting the meaning behind shortened phrases and colloquialisms to fully understand the meaning of a query and provide a more specific answer. Let’s take a close look at what voice NLP can do and how it is impacting AI on a broader scale.
How Natural Language Processing Works
Computer languages follow a rigid set of rules. If you miss an operator or argument, it might break completely. Human language has no such restrictions. While someone’s speech might become significantly less precise, our languages are incredibly flexible in how they enable expression. Historically, this has made them difficult to capture and analyze with machines.
Deep learning has changed that, enabling AI-systems that can evaluate and parse human language with an incredible degree of accuracy. The result is NLP. Natural language processing is not new. For decades, scientists have been writing rules to parse grammar and spelling to try and extract meaning from text. But it wasn’t until scientists shifted to a machine-learning model in recent decades that progress became readily apparent. Still, it took many more years to reach a point of complete, accurate language analysis to a degree that it could become interchangeable. Google Translate, for example, was completely revamped in 2016 using Google’s TensorFlow framework, introducing neural machine translation in place of previous statistical models. The result is a highly accurate translation tool for most common language pairings.
NLP has since been used for a variety of additional tasks including language generation, entity recognition, speech tagging, sentiment analysis, and speech recognition (which we are all more familiar with).
Simple speech recognition had been around for some time prior to the advancements in NLP in recent years. Software was long able to capture speech in text form, albeit with less than perfect accuracy. NLP is different in that it goes beyond merely capturing speech. It synthesizes, analyzes, and assigns meaning to speech that enables developers to use that data in new forms – translating, consolidating, or searching for relevant information.
Applications of Natural Language Processing
NLP is being utilized in dozens of industries to streamline operations and address long-standing pain points. The healthcare industry is a prime example. With more than 95% of hospitals utilizing electronic health records (EHR), the amount of time being spent on documentation to meet new regulations has ballooned throughout the industry. The burnout this creates limits the time practitioners can spend with their patients and reduces their efficacy. NLP is being used to process those large volumes of data faster and with less practitioner input. It is being used to extract data and clinical information into a format that can be more readily accessed and consumed by doctors.
NLP is being used in a number of industries to streamline communication with customers. Chatbots have become ubiquitous, but have been especially impactful in high-touch industries like finance. They enable customers to access information that previously required a human agent, ask basic questions, and get routed to the correct agent when a problem is too complex.
Other applications for NLP include:
- Sentiment Analysis – Data is inherently objective, and the precursors of modern NLP relied heavily on quantitative input to extract meaning. Today’s NLP algorithms can evaluate things like social media posts, market research panels, questionnaires, and interviews to identify common elements of key emotions and get a general impression of someone’s sentiment, or in the case of a larger study, the public’s sentiment. Ohio based Cognovi Labs, for example, has used such an algorithm to build their Coronavirus Panic Index measuring how people are reacting to the pandemic in real-time.
- Data Summarization – We have the sum total of human knowledge in our hands every day, but the volume of information is overwhelming. NLP can reduce long texts to high-level summaries that are more actionable in business, academic research, and more.
- Market Intelligence – Increasingly, companies rely on NLP algorithms to identify market response to their products. Data comes in the form of website traffic, user reviews, social media posts, customer service calls, and even offline sales data – all of which can be analyzed for actionable patterns.
NLP applications are expanding rapidly as “new-age NLP” enables developers to skip the text-to-speech step in processing input. Behavioral analysis of speech data can now distinguish vocal cues like pitch and tonal variance, making the data much richer than raw text transcriptions of audio files, as NLP has traditionally operated. Instead of extracting meaning strictly from the text, these newer NLP algorithms are significantly more accurate in assigning intent and sentiment to speech.
Language Agnostic Machines
One of the biggest challenges in business and computing is the language barrier. There are more than 7,000 languages in the world, and hundreds used in international business. Some countries have dozens of dialects on their own, which can create major barriers to computing applications that rely on one of a handful of core languages.
Traditionally, to expand an NLP model to a new language, a new data set must be recorded and annotated. It’s a very expensive and time-consuming process that doesn’t scale well. Deep learning advancements, however, are allowing developers to create cross-lingual embeddings and systems like mBERT that can help expedite the development of new NLP models. We’re not there yet, but we now have a path to language-agnostic computing, which will more fully connect the world’s minds through AI.
Rana Gujral is an entrepreneur, speaker, investor, and CEO at Behavioral Signals, an emotion AI and behavioral recognition software company. His writing can be found in publications such as Inc., TechCrunch, and Forbes. He can be reached at firstname.lastname@example.org.