Work package 1WP1 aims at identifying novel formal techniques for modelling legal knowledge and interpretation, drawing from analyses and tutorials on past approaches.
Task 1.1 Transdisciplinary Research and Conceptual Models for Legal Knowledge Representation and Reasoning. This task will offer suitable theoretical foundations for the subsequent tasks of this WP. Studies in legal theory on issues such as context sensitivity in normative language (concept holism, ordinary vs. legal concepts) and the structure of legal interpretation of legal provisions have reached a sophisticated level of analysis but often fail to offer operative and rigorous models of legal argumentation. On the contrary, efforts in Artificial Intelligence and Law are complementary to the ones in legal theory: they have computational and formal concerns that make them effective; however, they often move from simplifying theoretical assumptions. This tasks aims at identifying new directions for further cross-fertilization of these disciplines.
Task 1.2 Formal Languages for Representing Norms, Policies, and Values in the Law. Given work of Task 1.1, we expect to provide new ideas for developing formal languages and a standard extensible language for representing norms, policies, and values. Relevant aspects to be considered are: (i) temporal properties of policies (the times when they are is in force, when they produce effects, and when such effects hold) and other temporal dimensions; (ii) different effects of policies (deontic, rights, powers, interdictions, etc.); and (iii) epistemic dimension of policies for modelling norms, policies, and values and information flow control.
Task 1.3 Logics for Modelling the Interpretation of Legal Provisions. We expect to deliver ideas and models for handling the formal language in Task 1.2; we expect to devise defeasible deontic, epistemic and temporal logics which allow to determine what and when certain normative requirements apply in a given context (in terms obligations, prohibitions, and other deontic statuses) as well as to establish which subjects know and who is allowed to know which information. The task will present a suitable logic for reasoning about interpretation of norms, policies and values and a computational analysis of it.
Work package 2WP2 aims at designing ontologies for normative knowledge and developing and extending existing NLP systems for mining both concept to be linked to the T-BOX and named entities in order to populate the A-BOX.
Task 2.1 Designing an ontology for normative reasoning. This task will design the ontology for normative reasoning. The T-Box of the ontology will specify defeasible neo-Davidsonian axioms for modelling deontic concepts such as "obligation", "prohibition", etc. Each concept will be associated with templates of linguistic patterns that express the concept in natural language such as "Every XXX ought to do YYY", "It is forbidden to do ZZZ", etc. In legal text, the arguments (e.g. XXX, YYY, and ZZZ) are named entities or other sentences represented by neo-Davidsonian eventuality variables.
Task 2.2 Develop NLP systems for mining named entities and concepts, in order to populate the ontology. This task will develop concrete NLP procedures for mining named entities and processing recurring patterns in legal documents. Named entities will populate the A-BOX of the ontology while recurring patterns are associated with concepts in the T-BOX. Previous systems developed by the partners, either rule-based or statistical, will be extended in Task2.2. Strategies will be applied to optimize human effort, applying automated ontology inference and population techniques, and combining automated methods and human assessment with Active Learning techniques.
Task 2.3 Building GOLD standard corpora for syntactic and semantic analysis. As long as new legal text s processed in Task2.2, a GOLD standard corpora of processed legal text will be built. The GOLD standard is used as a testbed to guide the incremental updates of the procedures, in order to check that these updates do not introduce errors in the data that were previously processed correctly. Corpus annotation will also be guided by approaches like Active Learning or Semi-supervised learning. With these techniques we will be exploring the universe of possible examples for annotation, and choosing those that are most representative of undescribed areas of the instance space, thus obtaining a more representative annotated corpus.
Work package 3WP3 aims at providing ontology-based access to normative knowledge and developing novel, reasoning based solutions and tools for decision making based on normative knowledge, and for checking and enforcing compliance.
Task 3.1 Ontology-based access to normative knowledge. This task will produce a machine-processable version of the norms ontology produced in WP2, and provide access to normative knowledge using an appropriate query language.
Task 3.2 Computational solutions for decision making and compliance. This task will develop concrete computational solutions for decision making based on normative knowledge, and for c0pliance checking and enforcement. Reasoning approaches to be studied are: (i) defeasible logic; (ii) defeasible logic programming; and (iii) answer set programming.
Task 3.3 Massive parallelization. This task will develop massively parallel algorithms for a selection of solutions developed in Task 3.2, capable of dealing with huge amounts of data.
Work package 4WP4 aims at studying the user needs, defining the case studies to exploit the technical contributions of WP2-WP3, and evaluating the proposed techniques on concrete case studies.
Task 4.1 User needs study. This task will listen, understand and collect users' needs for the applications to be developed. What they would like to do with them, and what sort of services would they like to see. This task will look for interaction and feedback with both common users (i.e., not legal experts) and users working in the legal domain: What do they need? How will they build applications? How would they visualize the legal data?
Task 4.2 Case study: licenses and contracts. This task will develop concrete computational solutions for mining licenses and contracts expressed in natural language, and then reason over the extracted information. A specific usage of such a kind of texts is that of manually extracting the set of obliged/prohibited/permitted actions to check whether certain business processes are compliant with the legal text they should refer to.
Task 4.3 Case study: Technical Documents. This task will develop concrete computational solutions for mining technical documents expressed in natural language, and then reason over the extracted information. Given the huge dimension of this particular kind of legal texts and their diffusion in the companies, the advantages of mining norms of such texts is to obtain a summary of the main constraints expressed in the documents, with invaluable time saving gain.
Task 4.4 Case study: Multilingual corpora of norms. This task will develop concrete computational solutions for mining multilingual corpora of norms and legal cases expressed in natural language, and then reason over the extracted information. Having one shared language-agnostic representation for such a kind of norms will help a lot in avoiding costly and time consuming translations, and misunderstandings.
Task 4.5 Evaluation. This task has the goal to evaluate what are the concrete results of the proposed techniques and see whether they fit to the original needs and whether they contribute to create new usages of the legal information. This task will give feedback about the performances (e.g., accuracy, time) of the proposed applications for the case studies.