Natural language processing (NLP) denotes the use of artificial intelligence (AI) to manipulate written or spoken languages. Like the air we breathe, NLP is so pervasive today that we hardly notice it. When you use Alexa, you are conversing with an NLP machine; when you type into your chatbot or search, NLP technology comes to the fore. When you use Machine Learning (ML) algorithms to extract data from documents, you use NLP once again. Similarly, when you use Zoom or Google Meet, it is NLP that transcribes your speech. The list is practically endless.
NLP itself is an umbrella term that refers to a bunch of related technologies. NLP is at the core of Sentiment analysis, text extraction, machine translation, conversational AI, document AI, text summarization … and the list goes one. As AI systems become more and more intelligent, these systems would need to interact with humans in a rich, context-aware manner. It is NLP that would make it possible for machines to understand the context in which they operate. For example, when a user says ‘bank’ in the context of a financial institution, NLP engines can differentiate it from a river ‘bank’ and so on. This higher level of intelligence is a primary requirement for humans to converse with machines smoothly.
The Technology Drivers Behind NLP’s Success
Traditionally, NLP has been a complex problem to solve. However, two significant advances—one in 2017 and another in 2019—brought substantial improvements to NLP. In 2017, a new form of deep learning model called Transformer made it possible to parallelize ML training more efficiently, resulting in vastly improved accuracies.
In 2019, Google introduced Bidirectional Encoder Representations from Transformers (BERT), which improves the above Transformer architecture. Straightaway BERT helped achieve state-of-the-art performance on several NLP tasks such as reading comprehension, text extraction, sentiment analysis, etc. These two advancements meant that NLP could easily outdo average humans in many tasks and in some cases, even exceed the performance of subject matter experts.
So, How Big is the NLP Market?
The NLP market is at a relatively nascent stage but is fast expanding. According to the research firm, MarketsandMarkets, the NLP market would grow at a CAGR of 20.3% (from 11.6 billion in 2020 to USD 35.1 billion by 2026). Research firm Statistica is even more optimistic. According to their October 2021 article, NLP would catapult 14-fold between the years 2017 and 2025. This is certainly a phenomenal growth for a technology that was pretty much confined to the labs even as late as a decade ago.
A Word of Caution
Even as the NLP market grows and becomes mainstream, practitioners should be careful while investing in NLP. First and foremost is the understanding that NLP is not a single technology, but a suite of technologies. Consequently, not all the underlying systems have the same maturity curve. In principle, practitioners should value NLP along two dimensions—one that measures business benefits and two, the propensity of the underlying NLP technology to become mainstream.
According to Gartner, technologies such as conversational AI, chatbots, and document AI are expected to bring high to very high (transformational) business benefits while promising to become mainstream in less than two years. Contrast this with technologies such as text summarization which, according to Gartner, will likely bring in moderate benefits and will take 5-10 years to mature! Thus it is clear that not all the underlying NLP technologies are born equal, and investments require careful scrutiny.
Another important consideration for practitioners is the choice of natural language. Most models are good at English language followed by Chinese while performing below par on several international languages. Similarly, these language models tend to show a cultural and regional bias as many of them are trained on public datasets that have a large exposure to the western world.
Lastly, the adoption of NLP varies widely between industries with Healthcare (Drug Discovery, Clinical Trial Analytics, EHRs) taking the lion’s share of the NLP usage, followed by paper-heavy industries such as insurance and mortgage.
The Future of NLP
The roadmap for NLP is dotted by two major trajectories—the first, powered by larger Transformer Models such as GPT-3 and its future cousins. The second significant advancement will be in dialogue models where Google, Facebook, and other companies pour millions of dollars into research and development. First, let us discuss transformer models.
GPT-3 was developed by Open AI, a research business co-founded by Elon Musk and has several big names such as Sam Altman to its repertoire. GPT-3 is a multitasking system that can do several things such as translate text, extract text, converse with a human, and if you are bored, it can humor you with its poems. However, where GPT-3 has become savvy (and practically useful) is in the field of generating software code. Given basic instructions, GPT-3 can develop complete programs in Python, Java, and several other languages paving the way for exciting future opportunities. The future beckons bigger and bigger transformer models such as GPT-4 or the Chinese version called Wu Dao 2.0 (which is 10 times that of GPT-3).
The second major trend in NLP involves research from Google and Facebook around dialog models and conversational AI. Google, for example, unveiled a demonstration of a conversational AI system called LAMDA. The power of LAMDA is that it can connect with humans on a seemingly endless number of topics, unlike the modern chatbots which are trained for narrow conversations. If successful, LAMDA would very likely disrupt help desk, customer support and as one Google blog puts it, it will usher, “entirely new categories of helpful applications.”
We can argue that recent developments in NLP make it alluring for investments by practitioners and tech aficionados. The NLP market itself is fast-growing with increased adoption in healthcare, finance, and insurance. NLP is a suite of technologies, and practitioners can do well to discern which of the underlying systems will bring the maximum business benefit and by when. The future of NLP is very promising as more advancements would bring better user experience, thus opening up newer markets.