The evolution of the fuzziness of legal concepts, explored by our research, is an area that has so far resisted any attempts at formal modelling. Making conceptual ambiguities measurable and understanding their evolutionary dynamics help expand the toolbox for achieving coherent and predictable decision-making in the functioning of the legal system. The semantic variants of legal concepts are modelled by density functions over the semantic space, whereas the dynamics of legal ambiguity is modelled by variations in the shapes and spatial distances of such functions. Similar formal methodological approaches are being developed in several affiliated research centres abroad, for example by analysing the ambiguity in conceptual classification patterns in various fields of art. These foreign research efforts are also in the testing phase of such field-specific theoretical frameworks. Therefore, our research aims to set up the domestic leg of an emerging new consortium, to which we would contribute by analysing the dynamics of legal concepts. Formal modelling and the manual collection of empirical material are to be followed by contextual Deep Learning tests.

In the initial phase of the research, we investigated the semantic variants of basic legal concepts and identified their characteristic cognitive dimensions. To that end, we analysed detailed assessments of legal concepts by the Constitutional Court, looking for specific descriptive dimensions. The experiences drawn during the first phase on mapping into cognitive space were applied to second-instance and Curia-level rulings on domestic criminal judgments (pilot study). The aim was to operationalize concepts and clarify their cognitive spatial dimensions. A presentation on the results of the exploratory phase of the research and the consequent adaptation of our initial formal model was held at an international conference in Berlin in June 2022.

In addition to the above, we manually collected data on the frequency of occurrences of the semantic variants of concepts in the relevant cognitive framework in the criminal law database. Understanding the probabilities of occurrence allows estimates of the cognitive distances of legal concepts (a relative entropy calculation based on the Kullback-Leibler divergence). The empirical study will be published by submitting two articles in English and an article in Hungarian to a law journal. Further research materials are produced while carrying on with manual coding. As we begin to prepare for machine learning-based coding, we draw on parallel research by participants at the Berlin conference, which have adapted the same formal cognitive model to contexts not related to legal concepts.

 

Project participants
Viktória Közantal
Viktor Lőrincz
Gábor Péli
László Pólos

 

Conference paper
Pólos László. Perception, Communication, and Inferencing. 2022 Nagymaros Conference, ESMT Berlin, 19–22 June 2022

 

Keywords: Formal modeling, Legal discretion, Deep learning, Social Cognition, AI in Law