Ebook: Legal Knowledge and Information Systems

We are delighted to announce the proceedings of the 32nd International Conference on Legal Knowledge and Information Systems (JURIX 2019). The JURIX annual conference, organized under the auspices of the Dutch Foundation for Legal Knowledge-Based Systems (http://www.jurix.nl), has been established as an internationally renowned forum for the exchange of ideas concerning theoretical models and practical applications developed in the broadly construed sphere of artificial intelligence (AI) and law research. Traditionally, this field has been concerned with legal knowledge representation and engineering, computational models of legal reasoning, and analyses of legal data. However, recent years have witnessed the application of machine learning tools to legally relevant tasks rising to prominence.

The constantly growing influence of AI on different spheres of social life has prompted the community’s emerging interest in the explainability, trustworthiness, and responsibility of intelligent systems—and not in vain, as a high-level expert group the European Commission convened this year published the Ethics Guidelines for Trustworthy AI. It declared that the very first attribute of trustworthy AI was “lawfulness.” The research presented at JURIX conferences is an excellent example of interdisciplinary research integrating the methods and approaches from different branches of jurisprudence and computer science.

The 2019 edition of JURIX, which runs from 11 to 13 December, is hosted by the Ontological Engineering Group at the Artificial Intelligence Department of the Technical University of Madrid (Universidad Politécnica de Madrid). For this edition, we have received 81 papers, from which 14 were selected as full papers (10 pages in the proceedings) and 17 as 6-page short papers. Moreover, three submissions have been accepted as demo presentations. These figures result in a total acceptance rate of 41.98% and a competitive 25.5% acceptance rate for full papers. The accepted papers cover a broad array of topics, from computational models of legal argumentation, case-based reasoning, legal ontologies, and evidential reasoning, through classification of different types of text in legal documents and comparing similarities and the relevance of judicial decisions, to issues of governmental transparency.

Two invited speakers have honored JURIX 2019 by kindly agreeing to deliver two keynote lectures: Danièle Bourcier and Francesca Toni. Daniéle Bourcier has been responsible for pioneering research in the field of law, computers, and linguistics—currently, she is a director of research emeritus at Centre Nationale de la Recherche Scientifique (CNRS) and leads the Law and Governance Technologies Department at the Centre for Administrative Science Research (CERSA) at the University of Paris II. She is actively involved in the AI and law community, currently serving as a member of the Executive Committee of the International Association of Artificial Intelligence and Law. Francesca Toni is one of the most significant representatives of the computational argumentation research community. She is Professor of Computational Logic in the Department of Computing at Imperial College London, a member of the AI research theme, and the leader of the Computational Logic and Argumentation research group (CLArg). Francesca Toni has contributed extensively to different topics in logic, agents-based systems, and argumentation, recently focusing her attention inter alia on the application of argumentation models to generate explanations.

Traditionally, the main JURIX conference is accompanied by co-located events comprising workshops and tutorials. This year’s edition welcomes seven workshops: the CEILI Workshop on Legal Data Analysis; GDPR Compliance—Theories, Techniques, Tools; IberLegal: NLP for Legal Domain in Languages of the Iberian Peninsula (Spanish, Catalan, Galician, Basque, and Portuguese); LegRegSW JURIX 2019 – A Legislation and Regulation Semantic Web; MIREL 2019 – Mining and Reasoning with Legal Texts; TeReCom – The 3rd Workshop on Technologies for Regulatory Compliance; XAILA 2019 – The EXplainable AI in Law Workshop; and Defeasible Logic for Normative Reasoning (a tutorial). The continuation of well-established events and the organization of entirely new ones provide a great added value to the JURIX conference, enhancing its thematic and methodological diversity and attracting members of the broader community. Since 2013, JURIX has also offered researchers entering the field as Ph.D. students the opportunity to present their work during the Doctoral Consortium session, and this edition is no exception. Finally, for the first time, this edition of JURIX offers the Industry SessionâĂŤa special event enabling business representatives to present their products to the academy to foster further discussions concerning state-of-the-art developments in legal tech.

Organizing this edition of the conference would not have been possible without the support of many people and institutions. Special thanks are due to the local organizing team chaired by Víctor Rodríguez-Doncel and Elena Montiel Ponsoda (https://jurix2019.oeg-upm.net), and the enthusiasm of UPM’s Vice Chancellor for Research and the outstanding AI researcher, Asunción Gómez-Pérez. We would like to thank the workshops’ and tutorials’ organizers for their excellent proposals and for the effort involved in organizing the events. We owe our gratitude to Monica Palmirani, who kindly assumed the function of the Doctoral Consortium Chair. We are particularly grateful to the 91 members of the Program Committee for their excellent work in the rigorous review process and for their participation in the discussions concerning borderline papers. Finally, we would like to thank the former and current JURIX executive committee and steering committee members not only for their support and advice but also generally for taking care of all the JURIX initiatives.

Michał Araszkiewicz, JURIX 2019 Program Chair

Víctor Rodríguez-Doncel, JURIX 2019 Organization Chair

Automatically understanding the rhetorical roles of sentences in a legal case judgement is an important problem to solve, since it can help in several downstream tasks like summarization of legal judgments, legal search, and so on. The task is challenging since legal case documents are usually not well-structured, and these rhetorical roles may be subjective (as evident from variation of opinions between legal experts). In this paper, we address this task for judgments from the Supreme Court of India. We label sentences in 50 documents using multiple human annotators, and perform an extensive analysis of the human-assigned labels. We also attempt automatic identification of the rhetorical roles of sentences. While prior approaches towards this task used Conditional Random Fields over manually handcrafted features, we explore the use of deep neural models which do not require hand-crafting of features. Experiments show that neural models perform much better in this task than baseline methods which use handcrafted features.

In the legal domain, documents of various types are created in connection with a case. Some are transcripts prepared by court reporters, based on notes taken during the proceedings of a trial or deposition. For example, deposition transcripts capture the conversations between attorneys and deponents. These documents are mostly in the form of question-answer (QA) pairs. Summarizing the information contained in these documents is a challenge for attorneys and paralegals because of their length and form. Having automated methods to convert a QA pair into a canonical form could aid with the extraction of insights from depositions. These insights could be in the form of a short summary, a list of key facts, a set of answers to specific questions, or a similar result from text processing of these documents. In this paper, we describe methods using NLP and Deep Learning techniques to transform such QA pairs into a canonical form. The resulting transformed documents can be used for summarization and other downstream tasks.

Determining if a claim is accepted given judge arguments is an important non-trivial task in court decisions analyses. Application of recent efficient machine learning techniques may however be inappropriate for tackling this problem since, in the Legal domain, labelled datasets are most often small, scarce and expensive. This paper presents a deep learning model and a methodology for solving such complex classification tasks with only few labelled examples. We show in particular that mixing one-shot learning with recurrent neural networks and an attention mechanism enables obtaining efficient models while preserving some form of interpretability and limiting potential overfit. Results obtained on several types of claims in French court decisions, using different vectorization processes, are presented.

In the last years governments started to adapt new types of Artificial Intelligence (AI), particularly sub-symbolic data-driven AI, after having used more traditional types of AI since the mid-eighties of past century. The models generated by such sub-symbolic AI technologies, such as machine learning and deep learning are generally hard to understand, even by AI-experts. In many use contexts it is essential though that organisations that apply AI in their decision-making processes produce decisions that are explainable, transparent and comply with the rules set by law. This study is focused on the current developments of AI within governments and it aims to provide citizens with a good motivation of (partly) automated decisions. For this study a framework to assess the quality of explanations of legal decisions by public administrations was developed. It was found that communication with the citizen can be improved by providing a more interactive way to explain those decisions. Citizens could be offered more insights into the specific components of the decision made, the calculations applied and sources of law that contain the rules underlying the decision-making process.

Consumer contracts often contain unfair clauses, in apparent violation of the relevant legislation. In this paper we present a new methodology for evaluating such clauses in online Terms of Services. We expand a set of tagged documents (terms of service), with a structured corpus where unfair clauses are liked to a knowledge base of rationales for unfairness, and experiment with machine learning methods on this expanded training set. Our experimental study is based on deep neural networks that aim to combine learning and reasoning tasks, one major example being Memory Networks. Preliminary results show that this approach may not only provide reasons and explanations to the user, but also enhance the automated detection of unfair clauses.

Reasoning with evidence is error prone, especially when qualitative and quantitative evidence is combined, as shown by infamous miscarriages of justice, such as the Lucia de Berk case in the Netherlands. Methods for the rational analysis of evidential reasoning come in different kinds, often with arguments, scenarios and probabilities as primitives. Recently various combinations of argumentative, narrative and probabilistic methods have been investigated. By the complexity and subtlety of the subject matter, it has proven hard to assess the specific strengths and points of attention of different methods. Comparative case studies have only recently started, and never by one team. In this paper, we provide an analysis of a single case in order to compare the relative merits of two methods recently proposed in AI and Law: a method using Bayesian networks with embedded scenarios, and a method using case models that provide a formal analysis of argument validity. To optimise the transparency of the two analyses, we have selected a case about which the final decision is undisputed. The two analyses allow us to provide a comparative evaluation showing strengths and weaknesses of the two methods. We find a core of evidential reasoning that is shared between the methods.

Identification of relevant or similar court decisions is a core activity in legal decision making for case law researchers and practitioners. With an ever increasing body of case law, a manual analysis of court decisions can become practically impossible. As a result, some decisions are inevitably overlooked. Alternatively, network analysis may be applied to detect relevant precedents and landmark cases. Previous research suggests that citation networks of court decisions frequently provide relevant precedents and landmark cases. The advent of text similarity measures (both syntactic and semantic) has meant that potentially relevant cases can be identified without the need to manually read them. However, how close do these measures come to approximating the notion of relevance captured in the citation network? In this contribution, we explore this question by measuring the level of agreement of state-of-the-art text similarity algorithms with the citation behavior in the case citation network. For this paper, we focus on judgements by the Court of Justice of the European Union (CJEU) as published in the EUR-Lex database. Our results show that similarity of the full texts of CJEU court decisions does not closely mirror citation behaviour, there is a substantial overlap. In particular, we found syntactic measures surprisingly outperform semantic ones in approximating the citation network.

In this paper several existing dimension-based models of precedential constraint are compared and an alternative is proposed, which unlike existing models does not require that for each value assignment to a dimension it is specified whether it is for or against the case’s outcome. This arguably makes the model easier to apply in practice. In addition, it is shown how several factor- and dimension-based models of precedential constraint can be embedded in a Dung-style argumentation-based form, so that general tools from the formal study of argumentation become applicable.

In this work we describe a method to identify document pairwise relevance in the context of a typical legal document collection: limited resources, long queries and long documents. We review the usage of generalized language models, including supervised and unsupervised learning. We observe how our method, while using text summaries, overperforms existing baselines based on full text, and motivate potential improvement directions for future work.

This paper extends previous work by presenting a framework for modelling legislative deliberation in the form of dialogues with incomplete information. Roughly, in such legislative dialogues coalitions are initially equipped with different theories which constitute their private knowledge. Under this assumption they can dynamically change and propose new legislation associated with different utility functions.

This paper examines to what extent distributional approaches to induce bilingual lexica can capture correspondences between bilingual terms in international treaties. Recent developments in bilingual distributional representation learning methods have improved bilingual textual processing performances, and the application of these methods to processing specialised texts and technical terms has increased, including in the legal domain. Here we face at least two issues. Firstly, whether technical terms follow the distributional hypothesis or not is both theoretically and practically a critical concern. Theoretically, corresponding technical terms in different languages are the labels of the same concept and thus their equivalence is independent of the textual context. From this point of view, the distributional hypothesis holds only when the terms totally bind the context. This leads to the second issue, i.e. to verify the extent to which word embedding models trained on texts with different levels of specialisation are useful in capturing cross-lingual equivalences of terms. This paper examines these issues by conducting experiments in which different models trained on the texts with different degree of specialisations are evaluated against three different sets of equivalent bilingual pairs in the legal domain, i.e. of legal terms, of sub-technical terms and of general words. The results show that models learned on large-scale general texts fall far behind models learned on specialised texts in representing equivalent bilingual terms, while the former models have better performances for sub-technical terms and general words than the latter.

As legal regulations evolve, companies and organizations are tasked with quickly understanding and adapting to regulation changes. Tools like legal knowledge bases can facilitate this process, by either helping users navigate legal information or become aware of potentially relevant updates. At their core, these tools require legal references from many sources to be unified, e.g., by legal entity linking. This is challenging since legal references are often implicitly expressed, or combined via a context. In this paper, we prototype a machine learning approach to link legal references and retrieve combinations for a given context, based on standard features and classifiers, as used in entity resolution. As an extension, we evaluate an enhancement of those features with topic vectors, aiming to capture the relevant context of the passage containing a reference.We experiment with a repository of authoritative sources on German law for building topic models and extracting legal references and report that topic models do indeed contribute in improving supervised entity linking and reference retrieval.

In this paper, we present a method of building strong, explainable classifiers in the form of Boolean search rules. We developed an interactive environment called CASE (Computer Assisted Semantic Exploration) which exploits word co-occurrence to guide human annotators in selection of relevant search terms. The system seamlessly facilitates iterative evaluation and improvement of the classification rules. The process enables the human annotators to leverage the benefits of statistical information while incorporating their expert intuition into the creation of such rules.We evaluate classifiers created with our CASE system on 4 datasets, and compare the results to machine learning methods, including SKOPE rules, Random forest, Support Vector Machine, and fastText classifiers. The results drive the discussion on trade-offs between superior compactness, simplicity, and intuitiveness of the Boolean search rules versus the better performance of state-of-the-art machine learning models for text classification.

We address the legal text understanding task, and in particular we treat Japanese judgment documents in civil law. Rhetorical status classification (RSC) is the task of classifying sentences according to the rhetorical functions they fulfil; it is an important preprocessing step for our overall goal of legal summarisation. We present several improvements over our previous RSC classifier, which was based on CRF. The first is a BiLSTM-CRF based model which improves performance significantly over previous baselines. The BiLSTM-CRF architecture is able to additionally take the context in terms of neighbouring sentences into account. The second improvement is the inclusion of section heading information, which resulted in the overall best classifier. Explicit structure in the text, such as headings, is an information source which is likely to be important to legal professionals during the reading phase; this makes the automatic exploitation of such information attractive.We also considerably extended the size of our annotated corpus of judgment documents.

The increase of data-driven M&A transactions have raised apprehension over potential violations of data privacy rights. The economic significance attributed to Big Data has also called into question whether data privacy could be a parameter in merger control proceedings. Our purpose is to address the privacy and monopoly concerns arising from data-driven transactions within the scope of both the EC Regulation and the purpose limitation principle under the GDPR.

ANGELIC is a methodology for encapsulating knowledge of a body of case law. Logiak is a system intended to support the development of logic programs by domain experts, and provides an excellent environment for the rapid realisation of ANGELIC designs. We report our use of Logiak to realise ANGELIC designs, using both Boolean factors and factors with magnitude.

The paper deals with the problem of formalizing the renvoi in private international law. A rule based (first-order) fragment of a multimodal logic including context modalities as well as a (simplified) notion of common knowledge is introduced. It allows context variables to occur within modalities and context names to be used as predicate arguments, providing a simple combination of meta-predicates and modal constructs. The nesting of contexts in queries is exploited in the formalization of the renvoi problem.

We describe a set of dialogue moves which give a procedure to model the development of case law over a sequence of cases.

Different formalisms for defeasible reasoning have been used to represent legal knowledge and to reason with it. In this work, we provide an overview of the following logic-based approaches to defeasible reasoning: Defeasible Logic, Answer Set Programming, ABA+, ASPIC+, and DeLP. We compare features of these approaches from three perspectives: the logical model (knowledge representation), the method (computational mechanisms), and the technology (available software). On this basis, we identify and apply criteria for assessing their suitability for legal applications. We discuss the different approaches through a legal running example.

An approach for legal compliance representation and checking within a Linked Open Data framework is presented. It is based on modeling deontic norms in terms of ontology and ontology property restrictions. It is also shown how the approach can handle norm defeasibility. Such methodology is implemented by decidable fragments of OWL 2, while legal reasoning is implemented by available decidable reasoners.

We identify some legal reasoning patterns concerning deontic closure and conflicts in defeasible deontic logics. First, whether the logic allows the derivation of permissions from conflicting norms. Second, whether the logic treats norms as closed under logical implication. We suggest appropriate approaches for legal settings.

We introduce a computational model based on Deontic Defeasible Logic to handle the issue of Pragmatic Oddity. The key idea is that a conjunctive obligation is allowed only when each individual obligation is independent of the violation of the other obligations. The solution makes essential use of the constructive proof theory of the logic.