Using Natural Language Processing for government chatbots.

Chatbots are increasingly used across industry for customer service and support. The government sector is also beginning to adopt this technology to improve citizen engagement and provide 24/7 assistance.

One of the key components of a chatbot is natural language processing (NLP). NLP consists of both natural language understanding (NLU) and natural language generation (NLG).

The launch of generative text assistants like OpenAI’s Chat GPT and Google’s Bard, provide new opportunities but also some new problems for government agencies looking to improve citizen service.

In this blog post, I look at the advantages of using NLU to process citizen inquiries and the challenges that arise when using NLG to provide definitive answers.

Advantages of NLU in Government Chatbots

  1. Improved Citizen Engagement: NLU allows chatbots to understand and interpret citizen inquiries accurately. By doing so, it improves the overall citizen experience by providing relevant and accurate information in real-time.
  2. Cost Savings: Chatbots powered by NLU can automate the handling of frequently asked questions, freeing up government resources and reducing costs.
  3. Improved Service Delivery: With NLU, chatbots can provide personalized and contextualized responses to citizens, enhancing service delivery and responsiveness.
  4. Often, NLU is available across multiple languages so that more citizens can be served in their chosen language without the additional cost of translation or in the production of printed material.

Challenges of NLG in Government Chatbots

  1. Legal and Regulatory Compliance: Often, the response that the government assistant provides will need to be definitive. NLG poses a challenge in this area as it can generate responses that do not provide the correct information or even fail to comply with regulations.
  2. Ambiguity in Citizen Inquiries: Citizens often ask open-ended or ambiguous questions, making it difficult for chatbots to provide definitive answers.
  3. Limited Data Availability: NLG relies heavily on data availability. The government sector may have limited data, making it challenging to generate accurate responses to citizen inquiries.
  4. Timeliness: Amendments to legislation can be introduced that need to be reflected in the assistant from day zero. Most NLG systems are based on models that require a period of training and during this period the information may be out of date.

Most of the intelligent assistants that I have created for our public sector customers have contained legal advice or at least guidance for how people interact with public services.

This advice and guidance needs to be specific and accurate, something that has always been best handled through curated messaging created by working hand-in-hand with the departmental experts who have those discussions with citizens every day.

David Patterson


As ever, any government AI assistant needs to be designed from a citizen-first perspective. It needs to be designed to make things easier, automating back-end processes that would otherwise be time-consuming or confusing.

Using NLU to understand the citizen prompt helps us to achieve these aims. At the moment, however, it is probably fair to say that Natural Language Generation is not suitable for government purposes. Current systems are challenged when it comes to providing accurate, timely and definitive responses to citizen inquiries.

The best option in most situations will therefore be to use a service that understands the citizen but uses curated text responses to provide accurate information to the citizen.

You can read more on this topic in David’s previous articles