When it comes to implementation, government IT leaders and business line champions will need to tread thoughtfully as they embark on the road to NLP.

On the up side, all the major cloud vendors and many third-party providers offer NLP services, with APIs to make these offerings readily interoperable with existing systems (see for example here: Amazon Web Services, Google Cloud and Microsoft Azure.) This means that once a government agency is ready to make the leap, there should be ample infrastructure support available to enable an NLP application.

On the other hand, human language is extremely complex; verbal and written communication is incredibly nuanced. That means that while it is possible to buy the general tools that empower NLP, business leaders within government will likely need to refine those tools for their specific purposes. The more technical and specific the subject matter, the more fine-tuning you’ll need to do.

Suppose you’re a citizen-facing agency providing a government benefit. There are a finite number of forms in play and the ontology is fairly constrained. Now look at something more esoteric. Say you’re the FBI and you have a court order to read the cell phone texts of potential bad actors. There are emojis, slang, perhaps a foreign language thrown in. See how it gets tricky?

The sweet spot for NLP lies somewhere at the intersection of simplicity and repeatability.

For planning purposes, this suggests that the NLP effort will have to extend beyond the agency IT shop to embrace the business line owners and subject matter experts will have the specificity of experience needed to round out the knowledge base. The technologists still will have a critical role to play: They know where the data lives and how it moves; and they are attuned to the regulatory and compliance nuances to which government is bound.

But it’s the subject matter experts who will have to help write the playbook. They will do that first by defining the problem: Do we spend too much time on these forms? Do citizens wait on hold for too long? Then they will break down the business processes, looking for places where simplicity and repetition open a potential window for AI.

When it’s time to take that initial plunge, it may pay to train the AI to search for winners rather than losers. Say you’re using NLP to review claims forms. Instead of seeking the typos, the mismatches and the incomplete fields, you can ask NLP to flag those forms that look perfect, where everything matches and all you need is a human to sign off. Take the easy wins first, to pave the way for further, more complex iterations.

The road ahead

Looking ahead, it’s clear that private-sector market momentum is building behind NLP.

Markets and Markets analysts put the natural language processing market at $16.07 billion by 2021, while analysts at Mordor Intelligence project the market will reach $12.88 billion by 2023, a 22.5 percent annual growth rate.

The recent introduction of Apple’s Siri-enabled watch sounded either a starting gun or a warning shot for government agencies. By ratcheting up citizen expectation, such implementations put new pressure on government agencies: If consumers can speak natural-language questions into their wrists and get immediate intelligible answers, that creates a whole new service expectation.

It’s quickly getting to be second nature for people to interact via voice when in search of basic information. According to the 2018 Accenture Digital Consumer Survey, 37 percent of Americans were expected to have an Alexa-style smart speaker by the end of the year. The survey also found that usage and satisfaction with standalone intelligent virtual assistants surpassed mobile devices, with two-thirds of virtual assistant owners using their smartphone less as a result. As these tools continue to gain traction in the commercial sector, government will find itself pressed to offer its services on at least as sophisticated a footing.

Given that the technology is still evolving, the more forward-thinking agencies will begin an internal self-scrutiny now. The vendors will come up with the mechanical supports, but the business process details that form the core of the NLP value proposition—those have to come from within. Internal subject matter experts can be getting the ducks in a row now, highlighting the places where automation can yield the greatest benefit. Business line owners can start to think about small experiments, looking for the easy wins that will help to break down cultural barriers and position agencies for future gains.

Bryan Rich

Managing Director – Accenture Federal Services, Applied Intelligence Lead

Ian McCulloh, Ph.D

Chief Data Scientist – Accenture Federal Services


The coming federal AI productivity boom
Federal workers ready to thrive in the age of AI

Subscription Center
Stay in the know with our newsletters Stay in the know with our newsletters