Argument Mining for Intelligence Analysis

The Hunt Lab has an ongoing interest in the application of methods from the field of computational argumentation and argument mining to challenges in intelligence analysis.

What is argument mining?

Argument mining is the automated extraction of reasoning structure from text. An argument mining system takes sets of documents (e.g., intelligence products, articles, transcripts or web pages) as input, and outputs a set of the claims made therein, along with the argumentative relations between them. Argument mining is a rapidly-developing area of artificial intelligence, in the sub-field of natural language processing.

How is argument mining relevant to intelligence analysis?

Intelligence analysts often need to find out what arguments or evidence exist in relation to a specific issue. For example an analyst might be trying to determine whether a hostile actor has a particular capability. This assessment would be based in part on arguments previously made on this topic, whether by other analysts, outside experts, or by the hostile actor itself. Identifying these arguments is currently a slow and laborious activity requiring a high level of analytic acumen, particularly if the arguments are thinly distributed in large sets of documents or transcripts. A fully-developed argument mining tool would be able to perform this aspect of the analyst’s task automatically.


The Hunt Lab has recently developed Navigator, a prototype application designed to enhance an analyst’s ability to extract insight from large collections of documents related to a given topic. It uses argument mining to extract arguments from a set of user-uploaded texts, and presents this argument map to the analyst via a user interface that builds on recent innovations in the design of ‘tools for thought’. The application helps answer questions such as:

  • What claims are being made, and with what frequency?
  • Who is making them?
  • What arguments are presented for or against each claim?
  • Which claims are most/least contested?

The prototype is publicly accessible and can be explored at

Explore the prototype

Read the report