In this paper, we discuss previous research in automatic prediction of court decisions. We define the difference between outcome identification, outcome-based judgement categorisation and outcome forecasting, and review how various studies fall into these categories. We discuss how important it is to understand the legal data that one works with in order to determine which task can be performed. Finally, we reflect on the needs of the legal discipline regarding the analysis of court judgements.

Legal decision-support systems have the potential to improve access to justice, administrative efficiency, and judicial consistency, but broad adoption of such systems is contingent on development of technologies with low knowledge-engineering, validation, and maintenance costs. This paper describes two approaches to an important form of legal decision support—explainable outcome prediction—that obviate both annotation of an entire decision corpus and manual processing of new cases. The first approach, which uses an attention network for prediction and attention weights to highlight salient case (...) text, was shown to be capable of predicting decisions, but attention-weight-based text highlighting did not demonstrably improve human decision speed or accuracy in an evaluation with 61 human subjects. The second approach, termed semi-supervised case annotation for legal explanations, exploits structural and semantic regularities in case corpora to identify textual patterns that have both predictable relationships to case decisions and explanatory value. (shrink)

When courts started publishing judgements, big data analysis within the legal domain became possible. By taking data from the European Court of Human Rights as an example, we investigate how natural language processing tools can be used to analyse texts of the court proceedings in order to automatically predict judicial decisions. With an average accuracy of 75% in predicting the violation of 9 articles of the European Convention on Human Rights our approach highlights the potential of machine learning approaches in (...) the legal domain. We show, however, that predicting decisions for future cases based on the cases from the past negatively impacts performance. Furthermore, we demonstrate that we can achieve a relatively high classification performance when predicting outcomes based only on the surnames of the judges that try the case. (shrink)

Deep Learning has been widely used for tackling challenging natural language processing tasks over the recent years. Similarly, the application of Deep Neural Networks in legal analytics has increased significantly. In this survey, we study the early adaptation of Deep Learning in legal analytics focusing on three main fields; text classification, information extraction, and information retrieval. We focus on the semantic feature representations, a key instrument for the successful application of deep learning in natural language processing. Additionally, we share pre-trained (...) legal word embeddings using the word2vec model over large corpora, comprised legislations from UK, EU, Canada, Australia, USA, and Japan among others. (shrink)

Patent landscaping is the process of finding patents related to a particular topic. It is important for companies, investors, governments, and academics seeking to gauge innovation and assess risk. However, there is no broadly recognized best approach to landscaping. Frequently, patent landscaping is a bespoke human-driven process that relies heavily on complex queries over bibliographic patent databases. In this paper, we present Automated Patent Landscaping, an approach that jointly leverages human domain expertise, heuristics based on patent metadata, and machine learning (...) to generate high-quality patent landscapes with minimal effort. In particular, this paper describes a flexible automated methodology to construct a patent landscape for a topic based on an initial seed set of patents. This approach takes human-selected seed patents that are representative of a topic, such as operating systems, and uses structure inherent in patent data such as references and class codes to “expand” the seed set to a set of “probably-related” patents and anti-seed “probably-unrelated” patents. The expanded set of patents is then pruned with a semi-supervised machine learning model trained on seed and anti-seed patents. This removes patents from the expanded set that are unrelated to the topic and ensures a comprehensive and accurate landscape. (shrink)

This paper gives an account of dimensions in the reason model found in Horty : 1–33, 2011), Horty and Bench-Capon and Rigoni :133–160, 2015. doi: 10.1007/s10506-015-9166-x). The account is constructed with the purpose of rectifying problems with the approach to incorporating dimensions in Horty, namely, the problems arising from the collapse of the distinction between the reason model and the result model on that approach. Examination of the newly constructed theory revealed that the importance of dimensions in the reason model (...) lies in their ability to establish weighings between reasons/factors of the same polarity. This permits past cases to constrain future cases in ways they could not on the reason model with just factors. The paper then discusses how dimensions might be established from a case base and how dimensions that are incomplete in various aspects might be dealt with in the theory. It closes with comparisons to other work in the literature on AI and Law and suggestions for further improvement of the reason model. (shrink)

Robots are now associated with various aspects of our lives. These sophisticated machines have been increasingly used in different manufacturing industries and services sectors for decades. During this time, they have been a factor in causing significant harm to humans, prompting questions of liability. Industrial robots are presently regarded as products for liability purposes. In contrast, some commentators have proposed that robots be granted legal personality, with an overarching aim of exonerating the respective creators and users of these artefacts from (...) liability. This article is concerned mainly with industrial robots that exercise some degree of self-control as programmed, though the creation of fully autonomous robots is still a long way off. The proponents of the robot’s personality compare these machines generally with corporations, and sporadically with, inter alia, animals, and idols, in substantiating their arguments. This article discusses the attributes of legal personhood and the justifications for the separate personality of corporations and idols. It then demonstrates the reasons for refusal of an animal’s personality. It concludes that robots are ineligible to be persons, based on the requirements of personhood. (shrink)

Conferring legal personhood on purely synthetic entities is a very real legal possibility, one under consideration presently by the European Union. We show here that such legislative action would be morally unnecessary and legally troublesome. While AI legal personhood may have some emotional or economic appeal, so do many superficially desirable hazards against which the law protects us. We review the utility and history of legal fictions of personhood, discussing salient precedents where such fictions resulted in abuse or incoherence. We (...) conclude that difficulties in holding “electronic persons” accountable when they violate the rights of others outweigh the highly precarious moral interests that AI legal personhood might protect. (shrink)

Logic-based approaches to legal problem solving model the rule-governed nature of legal argumentation, justification, and other legal discourse but suffer from two key obstacles: the absence of efficient, scalable techniques for creating authoritative representations of legal texts as logical expressions; and the difficulty of evaluating legal terms and concepts in terms of the language of ordinary discourse. Data-centric techniques can be used to finesse the challenges of formalizing legal rules and matching legal predicates with the language of ordinary parlance by (...) exploiting knowledge latent in legal corpora. However, these techniques typically are opaque and unable to support the rule-governed discourse needed for persuasive argumentation and justification. This paper distinguishes representative legal tasks to which each approach appears to be particularly well suited and proposes a hybrid model that exploits the complementarity of each. (shrink)

This paper describes the Eunomos software, an advanced legal document and knowledge management system, based on legislative XML and ontologies. We describe the challenges of legal research in an increasingly complex, multi-level and multi-lingual world and how the Eunomos software helps users cut through the information overload to get the legal information they need in an organized and structured way and keep track of the state of the relevant law on any given topic. Using NLP tools to semi-automate the lower-skill (...) tasks makes this ambitious project a realistic commercial prospect as it helps keep costs down while at the same time allowing greater coverage. We describe the core system from workflow and technical perspectives, and discuss applications of the system for various user groups. (shrink)

In this article I argue for rule-based, non-monotonic theories of common law judicial reasoning and improve upon one such theory offered by Horty and Bench-Capon. The improvements reveal some of the interconnections between formal theories of judicial reasoning and traditional issues within jurisprudence regarding the notions of the ratio decidendi and obiter dicta. Though I do not purport to resolve the long-standing jurisprudential issues here, it is beneficial for theorists both of legal philosophy and formalizing legal reasoning to see where (...) the two projects interact. (shrink)

This paper describes one way in which a precise reason model of precedent could be developed, based on the general idea that courts are constrained to reach a decision that is consistent with the assessment of the balance of reasons made in relevant earlier decisions. The account provided here has the additional advantage of showing how this reason model can be reconciled with the traditional idea that precedential constraint involves rules, as long as these rules are taken to be defeasible. (...) The account presented is firmly based on a body of work that has emerged in AI and Law. This work is discussed, and there is a particular discussion of approaches based on theory construction, and how that work relates to the model described in this paper. (shrink)

We discuss several aspects of legal arguments, primarily arguments about the meaning of statutes. First, we discuss how the requirements of argument guide the specification and selection of supporting cases and how an existing case base influences argument formation. Second, we present,our evolving taxonomy of patterns of actual legal argument. This taxonomy builds upon our much earlier work on argument moves and also on our more recent analysis of how cases are used to support arguments for the interpretation of legal (...) statutes. Third, we show how the theory of argument used by CABARET, a hybrid case-based/rule-based reasoner, can support many of the argument patterns in our taxonomy. (shrink)

In this article the question is raised whether artificial intelligence has any psychological relevance, i.e. contributes to our knowledge of how the mind/brain works. It is argued that the psychological relevance of artificial intelligence of the symbolic kind is questionable as yet, since there is no indication that the brain structurally resembles or operates like a digital computer. However, artificial intelligence of the connectionist kind may have psychological relevance, not because the brain is a neural network, but because connectionist networks (...) exhibit operating characteristics which mimic operant behavior. Finally it is concluded that, since most of the work done so far in AI and Law is of the symbolic kind, it has as yet contributed little to our understanding of the legal mind. (shrink)

If we are eventually to understand the capability of higher organisms for perceptual recognition, generalization, recall, and thinking, we must first have answers to three fundamental questions: 1. How is information about the physical world sensed, or detected, by the biological system? 2. In what form is information stored, or remembered? 3. How does information contained in storage, or in memory, influence recognition and behavior? The first of these questions is in the.

This paper discusses some engineering considerations that should be taken into account when building a knowledge based system, and recommends isomorphism, the well defined correspondence of the knowledge base to the source texts, as a basic principle of system construction in the legal domain. Isomorphism, as it has been used in the field of legal knowledge based systems, is characterised and the benefits which stem from its use are described. Some objections to and limitations of the approach are discussed. The (...) paper concludes with a case study giving a detailed example of the use of the isomorphic approach in a particular application. (shrink)

The paper focuses on a specific type of referring legal provisions, in which the referring phrase contains a component that indicates the position of a certain fragment of the same text of a normative act by determining the position of that fragment in relation to the fragment in which the given referring phrase is located. Despite the fact that these referrals, called deictic, may be perceived as uncomplicated in structure and as functioning correctly in legal texts, many theoretical as well (...) as practical problems arise during their drafting and legal interpretation. The article attempts to reveal, organize and name these problems basing on the conceptual grid formulated by Studnicki, Łachwa, Fall and Stabrawa as a universal, although not widely known outside Polish jurisprudence, tool for analysing and solving referring provisions as elements of various legal systems, especially continental ones. Texts of selected Polish Codes have been chosen as an illustrative material of deictic referrals which have been incorrectly formulated as leaving space for interpretative doubts in terms of place of indication and place of reference, which also causes a potential problem of the deferred ostension phenomenon noticed by Quine. The author argues that proper drafting of referring legal provisions with deictic indication requires that the legislator should have extensive pragmalinguistic knowledge and is an essential part of law-making oriented towards achieving comprehensibility of texts of normative acts, especially as important as Codes. (shrink)

A large number of cross-references to various bodies of text are used in legal texts, each serving a different purpose. It is often necessary for authorities and companies to look into certain types of these citations. Yet, there is a lack of automatic tools to aid in this process. Recently, citation graphs have been used to improve the intelligibility of complex rule frameworks. We propose an algorithm that builds the citation graph from a document and automatically labels each edge according (...) to its purpose. Our method uses the citing text only and thus works only on citations who’s purpose can be uniquely identified by their surrounding text. This framework is then applied to the US code. This paper includes defining and evaluating a standard gold set of labels that cover a vast majority of citation types which appear in the “US Code” but are still short enough for practical use. We also proposed a novel linear-chain conditional random field model that extracts the features required for labeling the citations from the surrounding text. We then analyzed the effectiveness of different clustering methods such as K-means and support vector machine to automatically label each citation with the corresponding label. Besides this, we talk about the practical difficulties of this task and give a comparison of human accuracy compared to our end-to-end algorithm. (shrink)

This paper presents a methodology to design and implement programs intended to decide cases, described as sets of factors, according to a theory of a particular domain based on a set of precedent cases relating to that domain. We useDialectical Frameworks, a recent development in AI knowledge representation, as the central feature of our design method. ADFs will play a role akin to that played by Entity–Relationship models in the design of database systems. First, we explain how the factor hierarchy (...) of the well-known legal reasoning system CATO can be used to instantiate an ADF for the domain of US Trade Secrets. This is intended to demonstrate the suitability of ADFs for expressing the design of legal cased based systems. The method is then applied to two other legal domains often used in the literature of AI and Law. In each domain, the design is provided by the domain analyst expressing the cases in terms of factors organised into an ADF from which an executable program can be implemented in a straightforward way by taking advantage of the closeness of the acceptance conditions of the ADF to components of an executable program. We evaluate the ease of implementation, the performance and efficacy of the resulting program, ease of refinement of the program and the transparency of the reasoning. This evaluation suggests ways in which factor based systems, which are limited by taking as their starting point the representation of cases as sets of factors and so abstracting away the particular facts, can be extended to address open issues in AI and Law by incorporating the case facts to improve the decision, and by considering justification and reasoning using portion of precedents. (shrink)

Deontic logic as a way of formally reasoning about norms, an important area in AI and law, has traditionally concerned itself about formalising provisions of general statutes. Despite the long history of deontic logic, given the wide scope of the logic, it is difficult, if not impossible, to formalise all these notions in a single formalism, and there are still ongoing debates on appropriate semantics for deontic modalities in different contexts. In this paper, we restrict our attention to contracts between (...) interactive parties, which are both general enough to be an interesting object of study but specific enough so as to narrow down the debates regarding the meaning of modalities, and present a formalism for reasoning about them. (shrink)

There is an increasing need for norms to be embedded in technology as the widespread deployment of applications such as autonomous driving, warfare and big data analysis for crime fighting and counter-terrorism becomes ever closer. Current approaches to norms in multi-agent systems tend either to simply make prohibited actions unavailable, or to provide a set of rules which the agent is obliged to follow, either as part of its design or to avoid sanctions and punishments. In this paper we argue (...) for the position that agents should be equipped with the ability to reason about a system’s norms, by reasoning about the social and moral values that norms are designed to serve; that is, perform the sort of moral reasoning we expect of humans. In particular we highlight the need for such reasoning when circumstances are such that the rules should arguably be broken, so that the reasoning can guide agents in deciding whether to comply with the norms and, if violation is desirable, how best to violate them. One approach to enabling this is to make use of an argumentation scheme based on values and designed for practical reasoning: arguments for and against actions are generated using this scheme and agents choose between actions based on their preferences over these values. Moral reasoning then requires that agents have an acceptable set of values and an acceptable ordering on their values. We first discuss how this approach can be used to think about and justify norms in general, and then discuss how this reasoning can be used to think about when norms should be violated, and the form this violation should take. We illustrate how value based reasoning can be used to decide when and how to violate a norm using a road traffic example. We also briefly consider what makes an ordering on values acceptable, and how such an ordering might be determined. (shrink)

Recent work has demonstrated how data-driven AI methods can leverage consumer protection by supporting the automated analysis of legal documents. However, a shortcoming of data-driven approaches is poor explainability. We posit that in this domain useful explanations of classifier outcomes can be provided by resorting to legal rationales. We thus consider several configurations of memory-augmented neural networks where rationales are given a special role in the modeling of context knowledge. Our results show that rationales not only contribute to improve the (...) classification accuracy, but are also able to offer meaningful, natural language explanations of otherwise opaque classifier outcomes. (shrink)

In previous work, I showed how the “reason model” of precedential constraint could naturally be generalized from the standard setting in which it was first developed to a richer setting in which dimensional information is represented as well. Surprisingly, it then turned out that, in this new dimensional setting, the reason model of constraint collapsed into the “result model,” which supports only a fortiori reasoning. The purpose of this note is to suggest a modification of the reason model of constraint (...) that distinguishes it from the result model even in the dimensional setting. (shrink)

This paper shows how two models of precedential constraint can be broadened to include legal information represented through dimensions. I begin by describing a standard representation of legal cases based on boolean factors alone, and then reviewing two models of constraint developed within this standard setting. The first is the “result model”, supporting only a fortiori reasoning. The second is the “reason model”, supporting a richer notion of constraint, since it allows the reasons behind a court’s decisions to be taken (...) into account. I then show how the initial representation can be modified to incorporate dimensional information and how the result and reason models can be adapted to this new dimensional setting. As it turns out, these two models of constraint, which are distinct in the standard setting, coincide once they are transposed to the new dimensional setting, yielding exactly the same patterns of constraint. I therefore explore two ways of refining the reason model of constraint so that, even in the dimensional setting, it can still be separated from the result model. (shrink)

Legal reasoning requires identification through search of authoritative legal texts (such as statutes, constitutions, or prior judicial opinions) that apply to a given legal question. In this paper, using a network representation of US Supreme Court opinions that integrates citation connectivity and topical similarity, we model the activity of law search as an organizing principle in the evolution of the corpus of legal texts. The network model and (parametrized) probabilistic search behavior generates a Pagerank-style ranking of the texts that in (...) turn gives rise to a natural geometry of the opinion corpus. This enables us to then measure the ways in which new judicial opinions affect the topography of the network and its future evolution. While we deploy it here on the US Supreme Court opinion corpus, there are obvious extensions to large evolving bodies of legal text (or text corpora in general). The model is a proxy for the way in which new opinions influence the search behavior of litigants and judges and thus affect the law. This type of “legal search effect” is a new legal consequence of research practice that has not been previously identified in jurisprudential thought and has never before been subject to empirical analysis. We quantitatively estimate the extent of this effect and find significant relationships between search-related network structures and propensity of future citation. This finding indicates that “search influence” is a pathway through which judicial opinions can affect future legal development. (shrink)

Work on a computer program called SMILE + IBP (SMart Index Learner Plus Issue-Based Prediction) bridges case-based reasoning and extracting information from texts. The program addresses a technologically challenging task that is also very relevant from a legal viewpoint: to extract information from textual descriptions of the facts of decided cases and apply that information to predict the outcomes of new cases. The program attempts to automatically classify textual descriptions of the facts of legal problems in terms of Factors, a (...) set of classification concepts that capture stereotypical fact patterns that effect the strength of a legal claim, here trade secret misappropriation. Using these classifications, the program can evaluate and explain predictions about a problem’s outcome given a database of previously classified cases. This paper provides an extended example illustrating both functions, prediction by IBP and text classification by SMILE, and reports empirical evaluations of each. While IBP’s results are quite strong, and SMILE’s much weaker, SMILE + IBP still has some success predicting and explaining the outcomes of case scenarios input as texts. It marks the first time to our knowledge that a program can reason automatically about legal case texts. (shrink)

We shall introduce a set of fundamental legal concepts, providing a definition of each of them. This set will include, besides the usual deontic modalities (obligation, prohibition and permission), the following notions: obligative rights (rights related to other’s obligations), permissive rights, erga-omnes rights, normative conditionals, liability rights, different kinds of legal powers, potestative rights (rights to produce legal results), result-declarations (acts intended to produce legal determinations), and sources of the law.

In this paper we describe AGATHA, a program designed to automate the process of theory construction in case based domains. Given a seed case and a number of precedent cases, the program uses a set of argument moves to generate a search space for a dialogue between the parties to the dispute. Each move is associated with a set of theory constructors, and thus each point in the space can be associated with a theory intended to explain the seed case (...) and the other cases in the domain. The space is large and so an heuristic search method is needed. This paper describes two methods based on A* and alpha/beta pruning and also a series of experiments designed to explore the appropriateness of different evaluation functions, the most useful precedents to use as seed cases and the quality of the resulting theories. (shrink)

This paper proposes several recurrent neural network-based models for recognizing requisite and effectuation parts in Legal Texts. Firstly, we propose a modification of BiLSTM-CRF model that allows the use of external features to improve the performance of deep learning models in case large annotated corpora are not available. However, this model can only recognize RE parts which are not overlapped. Secondly, we propose two approaches for recognizing overlapping RE parts including the cascading approach which uses the sequence of BiLSTM-CRF models (...) and the unified model approach with the multilayer BiLSTM-CRF model and the multilayer BiLSTM-MLP-CRF model. Experimental results on two Japan law RRE datasets demonstrated advantages of our proposed models. For the Japanese National Pension Law dataset, our approaches obtained an \ score of 93.27% and exhibited a significant improvement compared to previous approaches. For the Japan Civil Code RRE dataset which is written in English, our approaches produced an \ score of 78.24% in recognizing RE parts that exhibited a significant improvement over strong baselines. In addition, using external features and in-domain pre-trained word embeddings also improved the performance of RRE systems. (shrink)

There seems to be no clear consensus in the existing literature about the role of deontic logic in legal knowledge representation — in large part, we argue, because of an apparent misunderstanding of what deontic logic is, and a misplaced preoccupation with the surface formulation of legislative texts. Our aim in this paper is to indicate, first, which aspects of legal reasoning are addressed by deontic logic, and then to sketch out the beginnings of a methodology for its use in (...) the analysis and representation of law.The essential point for which we argue is that deontic logic — in some form or other —needs to be taken seriously whenever it is necessary to make explicit, and then reason about, the distinction between what ought to be the case and what is the case, or as we also say, between the ideal and the actual. We take the library regulations at Imperial College as the main illustration, and small examples from genuinely legal domains to introduce specific points. In conclusion, we touch on the role of deontic logic in the development of the theory of normative positions. (shrink)

The first issue of Artificial Intelligence and Law journal was published in 1992. This paper provides commentaries on nine significant papers drawn from the Journal’s second decade. Four of the papers relate to reasoning with legal cases, introducing contextual considerations, predicting outcomes on the basis of natural language descriptions of the cases, comparing different ways of representing cases, and formalising precedential reasoning. One introduces a method of analysing arguments that was to become very widely used in AI and Law, namely (...) argumentation schemes. Two relate to ontologies for the representation of legal concepts and two take advantage of the increasing availability of legal corpora in this decade, to automate document summarisation and for the mining of arguments. (shrink)

In this paper, we present an informal introduction to Logical English and illustrate its use to standardise the legal wording of the Automatic Early Termination clauses of International Swaps and Derivatives Association Agreements. LE can be viewed both as an alternative to conventional legal English for expressing legal documents, and as an alternative to conventional computer languages for automating legal documents. LE is a controlled natural language, which is designed both to be computer-executable and to be readable by English speakers (...) without special training. The basic form of LE is syntactic sugar for logic programs, in which all sentences have the same standard form, either as rules of the form conclusion if conditions or as unconditional sentences of the form conclusion. However, LE extends normal logic programming by introducing features that are present in other computer languages and other logics. These features include typed variables signalled by common nouns, and existentially quantified variables in the conclusions of sentences signalled by indefinite articles. Although LE translates naturally into a logic programming language such as Prolog or ASP, it can also serve as a neutral standard, which can be compiled into other lower-level computer languages. (shrink)

In normative multi-agent systems, the question of “how an agent identifies norms in an open agent society” has not received much attention. This paper aims at addressing this question. To this end, this paper proposes an architecture for norm identification for an agent. The architecture is based on observation of interactions between agents. This architecture enables an autonomous agent to identify prohibition norms in a society using the prohibition norm identification (PNI) algorithm. The PNI algorithm uses association rule mining, a (...) data mining approach to identify sequences of events as candidate norms. When a norm changes, an agent using our architecture will be able to modify the norm and also remove a norm if it does not hold in the society. Using simulations of a park scenario we demonstrate how an agent makes use of the norm identification framework to identify prohibition norms. (shrink)

Terms of service of on-line platforms too often contain clauses that are potentially unfair to the consumer. We present an experimental study where machine learning is employed to automatically detect such potentially unfair clauses. Results show that the proposed system could provide a valuable tool for lawyers and consumers alike.

A model of normative provisions and related axioms represented by using RDF/owl are presented as a contribution to implement the semantic web in the legal domain. In particular, a pattern able to implement the Hohfeldian legal fundamental relations between provisions using OWL-DL expressivity is proposed. Moreover, a query-based approach able to deal with relations between provision instances is described. An example of advanced access and reasoning over provisions using the proposed approach, as well as a prototype architecture of a provision (...) retrieval system, are shown. The main benefit is represented by the ability of the approach to keep the complexity of the problem within a description logic computational tractability. (shrink)

Modeling law search and retrieval as prediction problems has recently emerged as a predominant approach in law intelligence. Focusing on the law article retrieval task, we present a deep learning framework named LamBERTa, which is designed for civil-law codes, and specifically trained on the Italian civil code. To our knowledge, this is the first study proposing an advanced approach to law article prediction for the Italian legal system based on a BERT (Bidirectional Encoder Representations from Transformers) learning framework, which has (...) recently attracted increased attention among deep learning approaches, showing outstanding effectiveness in several natural language processing and learning tasks. We define LamBERTa models by fine-tuning an Italian pre-trained BERT on the Italian civil code or its portions, for law article retrieval as a classification task. One key aspect of our LamBERTa framework is that we conceived it to address an extreme classification scenario, which is characterized by a high number of classes, the few-shot learning problem, and the lack of test query benchmarks for Italian legal prediction tasks. To solve such issues, we define different methods for the unsupervised labeling of the law articles, which can in principle be applied to any law article code system. We provide insights into the explainability and interpretability of our LamBERTa models, and we present an extensive experimental analysis over query sets of different type, for single-label as well as multi-label evaluation tasks. Empirical evidence has shown the effectiveness of LamBERTa, and also its superiority against widely used deep-learning text classifiers and a few-shot learner conceived for an attribute-aware prediction task. (shrink)

The first issue of _Artificial Intelligence and Law_ journal was published in 1992. This paper provides commentaries on landmark papers from the first decade of that journal. The topics discussed include reasoning with cases, argumentation, normative reasoning, dialogue, representing legal knowledge and neural networks.

Law search is fundamental to legal reasoning and its articulation is an important challenge and open problem in the ongoing efforts to investigate legal reasoning as a formal process. This Article formulates a mathematical model that frames the behavioral and cognitive framework of law search as a sequential decision process. The model has two components: first, a model of the legal corpus as a search space and second, a model of the search process that is compatible with that environment. The (...) search space has the structure of a “multi-network”—an interleaved structure of distinct networks—developed in earlier work. In this Article, we develop and formally describe three related models of the search process. We then implement these models on a subset of the corpus of U.S. Supreme Court opinions and assess their performance against two benchmark prediction tasks. The first is to predict the citations in a document from its semantic content. The second is to predict the search results generated by human users. For both benchmarks, all search models outperform a null model with the learning-based model outperforming the other approaches. Our results indicate that through additional work and refinement, there may be the potential for machine law search to achieve human or near-human levels of performance. (shrink)

This paper provides an analysis of how concepts pertinent to legal contracts can influence certain aspects of their digital implementation through smart contracts, as inspired by recent developments in distributed ledger technology. We discuss how properties of imperative and declarative languages including the underlying architectures to support contract management and lifecycle apply to various aspects of legal contracts. We then address these properties in the context of several blockchain architectures. While imperative languages are commonly used to implement smart contracts, we (...) find that declarative languages provide more natural ways to deal with certain aspects of legal contracts and their automated management. (shrink)

This paper describes one way in which a precise reason model of precedent could be developed, based on Grant Lamond’s general idea that a later court is constrained to reach a decision that is consistent an earlier court’s assessment of the balance of reasons. The account provided here has the additional advantage of showing how this reason model can be reconciled with the traditional idea that precedential constraint involves rules, as long as these rules are taken to be defeasible.

In this paper several recent factor- and dimension-based models of precedential constraint are formally investigated and an alternative dimension-based model is proposed. Simple factor- and dimension-based syntactic criteria are identified for checking whether a decision in a new case is forced, in terms of the relevant differences between a precedent and a new case, and the difference between absence of factors and negated factors in factor-based models is investigated. Then Horty’s and Rigoni’s recent dimension-based models of precedential constraint are critically (...) examined. An alternative to their reason models is proposed which is less expressive but arguably easier to apply in practice. (shrink)

Norms provide a valuable mechanism for establishing coherent cooperative behaviour in decentralised systems in which there is no central authority. One of the most influential formulations of norm emergence was proposed by Axelrod :1095–1111, 1986). This paper provides an empirical analysis of aspects of Axelrod’s approach, by exploring some of the key assumptions made in previous evaluations of the model. We explore the dynamics of norm emergence and the occurrence of norm collapse when applying the model over extended durations. It (...) is this phenomenon of norm collapse that can motivate the emergence of a central authority to enforce laws and so preserve the norms, rather than relying on individuals to punish defection. Our findings identify characteristics that significantly influence norm establishment using Axelrod’s formulation, but are likely to be of importance for norm establishment more generally. Moreover, Axelrod’s model suffers from significant limitations in assuming that private strategies of individuals are available to others, and that agents are omniscient in being aware of all norm violations and punishments. Because this is an unreasonable expectation, the approach does not lend itself to modelling real-world systems such as online networks or electronic markets. In response, the paper proposes alternatives to Axelrod’s model, by replacing the evolutionary approach, enabling agents to learn, and by restricting the metapunishment of agents to cases where the original defection is observed, in order to be able to apply the model to real-world domains. This work can also help explain the formation of a “social contract” to legitimate enforcement by a central authority. (shrink)

Few automated legal reasoning systems have been developed in domains of law in which a judicial decision maker has extensive discretion in the exercise of his or her powers. Discretionary domains challenge existing artificial intelligence paradigms because models of judicial reasoning are difficult, if not impossible to specify. We argue that judicial discretion adds to the characterisation of law as open textured in a way which has not been addressed by artificial intelligence and law researchers in depth. We demonstrate that (...) systems for reasoning with this form of open texture can be built by integrating rule sets with neural networks trained with data collected from standard past cases. The obstacles to this approach include difficulties in generating explanations once conclusions have been inferred, difficulties associated with the collection of sufficient data from past cases and difficulties associated with integrating two vastly different paradigms. A knowledge representation scheme based on the structure of arguments proposed by Toulmin has been used to overcome these obstacles. The system, known as Split Up, predicts judicial decisions in property proceedings within family law in Australia. Predictions from the system have been compared to those from a group of lawyers with favourable results. (shrink)