Initial work of LAW-TRAIN aimed at reaching agreement on the nature and specifications of the system’s main modules, namely the
• Collaborative international Virtual Interrogation Game
• Virtual Suspect
• Training Tools.
The methodology of the interrogation process was crystallized from interviews and workshops, conducted with all end-user partners and criminologists.
At first, each trainee reviews and learns the existing material; subsequently, the members of the joint interrogation team review the evidence, share information from the different countries, discuss legal approach options, prepare content and objectives for the interrogation, and decide on the questioning strategy, including the setting of the interrogation room and its atmosphere. It has become clear that the Individual and Joint Preparation steps must be addressed in the training platform, since they are as important as the interrogation.
In the following interrogation phase, one or two interrogators are situated in the virtual interrogation room with the VS (and an optional passive lawyer). The interrogators question the VS and can see its facial expressions and body movement and hear its statements. They can influence the suspect by moving about or changing their tone of questioning. The team’s other interrogators can listen and observe the interrogation remotely. It was agreed that they should not interrupt the ongoing interrogation but can ask the trainee to step out into their shared space. This interrogation process repeats for several cycles until the team can reach conclusions. A joint debriefing phase follows
Use cases and specifications concerning an international drug trafficking crime have been established for the training sessions. In parallel, the judicial framework for transnational police and judicial cooperation was defined, and a concise review of international, European and national legislation and policies on transnational cooperation in criminal matters is being prepared to cover treaties, global and European conventions and bilateral agreements. Of the different methods for transnational collaboration, that of Joint Investigation Team was selected for LAW-TRAIN. The analysis also addressed specific legal requirements to allow accounting for difficulties and opportunities, and for considering additional conditions needed for successful transnational cooperation in the different legal frameworks of the consortium partners. This allowed defining the best practices for a transnational joint interrogation.
Besides the trainee and the VS, the system includes an automatic virtual trainer and a human trainer. The diagnostic tools that these trainers apply are designed to ensure that the steps need for training are provided. Possible ways of feedback have been summarized.
Additional work involved the design of the virtual interrogation room allowing the interrogators plan the interrogation room and to place its elements, as is done in real life. Work has also started on other elements of the game, such as communication and the Joint Team room.
The design of the VS involved generation of verbal and non-verbal behaviour that is affected by the questions and activities of the interrogators, and the specification of the VS’ graphics, expressions and abilities. Implementing a person’s cognition and behaviour involves creation of a set of possible statements and questions that the interrogator can use have been generated, with the patterns of possible responses for the VS. Choosing the appropriate statement by VS during the interrogation must take into account the ongoing conversation, the body language of the trainees and environmental information. In particular, selecting a response from among the true, false and uninformative response categories. Furthermore, when several answers from a category are available, choosing one should agree with the history of the interrogation until then. Based on preliminary successful attempts in using machine learning for the VS, we decided to combine Machine Learning and Dialogue Management approaches, and for this approach to work, we will annotate each question with static and dynamic features be applied when choosing the answer.
An additional important feature is whether a question or a statement is threatening. This feature is associated with a question or statement when it is added to the system, it can be changed during the interrogation, depending on its relevance to the crime.
Expert detectives from the consortium’s end-user partners will build a list of statements and questions that an interrogator can say during an interrogation.
