Natural Language Processing Takes Off in State Government

How State and Local Agencies Are Using Natural Language Processing

“Across government, whether in defense, transportation, human services, public safety, or health care, agencies struggle with a similar problem — making sense out of huge volumes of unstructured text to inform decisions, improve services, and save lives,” notes a 2019 Deloitte report. “Text analytics, and specifically NLP, can be used to aid processes from investigating crime to providing intelligence for policy analysis.”

One way is through Deloitte’s AI platform RegExplorer, says William Eggers, executive director of Deloitte’s Center for Government Insights, who co-authored the report. The tool allows governments sift through large numbers of text documents, such as regulations, that would ordinarily take humans much longer to process.

“A lot of these regulations — there’s thousands of pages of regulations,” Eggers says. “A lot of them haven’t been updated in many years, a lot of them are duplicative and overlapping. And when governments are looking at regulatory reform or reducing burdens and so on and so forth, it’s really impossible for the human eye to see all those different connections between the corpus of regulations.”

The tool digests, analyzes and understands the links between documents to help governments make connections — including with regulations in other countries — and streamline changes.

EXPLORE: Virginia plans to expand its use of artificial intelligence in state agencies.

“It comes in quite handy when people are doing policy reform, regulation reform and also creating legislation,” he says. “Government has both a lot of data and a lot of text. And NLP becomes a key technology for helping you to turn that text into data that can be analyzed, which was previously not possible to do through the human eye.”

Another big area where state and local governments are using natural language processing is within the court and criminal justice systems.

Mike Morper, a vice president of product marketing at the AI tech company Veritone, says his company’s aiWARE platform is basically a search engine for unstructured data, such as audio and video, that helps agencies quickly find what they need. This has big implications for law enforcement, attorneys and others who are increasingly reliant on multimedia for evidence, he says.

“When you start contemplating the hundreds and thousands and tens of thousands of hours of media that are being recorded, it is literally impossible for individual humans to sit there and listen to all of it,” says Morper. “Technology like NLP can process that media so that it can build a transcript of what that conversation was or what was said in that video. Then you can apply keyword searches against that transcript, and you can start quickly culling down massive amounts of audio and video so that you can start building a compelling case to prosecute.”

In Texas, state officials are using NLP for chatbots and for reading emails to send them to the correct departments to aid workflow. For example, during the pandemic, one agency was receiving 2,000 service tickets a day, 80 percent of which were requests to help with password resets, says Krishna Edathil, director of the Texas Artificial Intelligence Center of Excellence (AI-CoE).

Through NLP, the password requests were isolated from the main help desk inbox so that the appropriate teams were able to address the most pressing issues.

Edathil says the biggest benefits of NLP have been the efficiency and productivity it has brought to his office.

“It is becoming a great digital assistant that helps with growing cases and demands for IT help desk support,” he says — especially when each state agency has full-time-employee caps they cannot exceed.

Other pandemic-era use cases have emerged as well.

IBM’s NLP technology is helping state and local governments with constituent care and messaging. The Rhode Island Department of Health’s has a virtual assistant called Rhoda that has helped residents locate COVID-19 testing sites and manage vaccine eligibility and travel restrictions, Dobrin says.

RELATED: How is the use of artificial intelligence accelerating in state government?

Top Natural Language Processing Techniques and Step

Some of the most common ways NLP is used is in speech-to-text and text-to-speech applications; in sentiment analysis to extract meaning from text such as emotions, tone or sarcasm; for word sense disambiguation to select the meaning of a word that may have multiple meanings; and for natural language generation to turn structured data into human language, Dobrin says.

Edathil’s center uses it mainly for sentiment analysis, where a bot could help to prioritize emails by understanding the content of customers’ complaints and concerns.

“This is not limited to email; it’s for any form of text. Social media response and chatbot conversations are great examples,” he says.

The steps to use NLP begin with lexical analysis, then parsing, semantic analysis, discourse integration and pragmatic analysis, she says. While some models require users to do these themselves, the Texas AI-CoE relies on major cloud providers, such as Microsoft, Google and Amazon, who have already implemented this technology through Platform as a Service.

MORE FROM STATETECH: Learn why states are deploying computer vision technology.

The Future of Natural Language Processing in Government Operations

For government entities that want to start using NLP, Edathil’s best advice is to start with simple use cases, such as prioritizing an email queue. Choose “a task that won’t raise concerns that this technology might take away humans’ jobs,” he says. “Instead, it can show that NLP can help them do their jobs much more efficiently and faster.”

Governments can also look to businesses that frequently use NLP to gain insights and drive analytics with techniques such as advanced sentiment analysis to get ideas for a range of “increasingly prevalent” uses that would translate well to the government sector, says Dobrin.

“With NLP, a city agency can ask an AI, ‘Which bus stop gets the most complaints?’ or ‘Where are noise complaints a problem?’ and the AI can retrieve that information,” he says. “This is all part of the greater trend of how we’re getting better at customizing AI so that it can eliminate repetitive work and research time and help unlock greater human potential.”