Encoding Schemes for a Discourse Support System for Legal Argument Henry Prakken and Gerard Vreeswijk 1 Abstract. This paper reports on the ongoing development of a dis- course support system for legal argument named PROSUPPORT.A description is given of the system’s encoding schemes with which the user can enter his or her analysis of the discourse. These schemes, which are implemented as web browser forms linked to a database, serve to capture support relations of propositions within arguments, and dialectical relations between arguments. In addition, they sup- port the recording of relevant argumentative and procedural speech acts made with respect to these arguments, such as disputing or con- ceding a claim, and allocating the burden of proof. The main issue in developing these encoding schemes is how expressiveness of the schemes can be reconciled with ease of use, on a suitable theoretical basis. 1 Introduction In several related areas of computer science there is a growing inter- est in software support for such discourse processes as discussion, negotiation, dispute resolution and collective decision making. Un- like with ‘conventional’ decision-support tools (such as knowledge- based systems), the task of such systems is not to produce or suggest solutions to a problem with the help of domain knowledge, but to help the participants in discursive interactions to structure their rea- soning and discourse, so that they can make sense of the discourse and interact effectively. One professional area where such systems are of great poten- tial use is the law. Participants in legal procedures (including alter- native procedures such as online dispute resolution) often face the complex task of managing the information they are confronted with and the communication and reasoning they are expected to engage in. Discourse support systems can provide important assistance for these tasks: they could facilitate the structured inputting of a vari- ety of discursive data, such as which claims have been made, con- ceded or challenged, how the burden of proof was assigned, which grounds and evidence have been adduced and counterattacked, how these grounds and evidence can be assessed, and whether the parties have respected the rules of procedure. The system could then usefully display, combine and restructure this input, and compute the conse- quences of the user’s evaluative decisions (e.g. who wins given a cer- tain allocation of the burden of proof and assessment of evidence?). Such systems could also support the (semi-) automatic generation of case summaries or even verdicts. These functionalities can be put to use in a variety of contexts. Individual users can be supported in making their own analysis of the discourse, invisible for other partici- 1 Institute of Information and Computing Sciences, Utrecht Univer- sity, PO Box 80089, 3508 TB Utrecht, The Netherlands, email: {henry,gv}@cs.uu.nl pants. The joint participants can be supported in their communicative and disputational interactions. Or the supporting staff of a judge or other official can be supported in their task to preprocess an analysis of a case, and to pass on the results to the official. Finally, in online versions of dispute resolution discourse systems could be a principal means of interaction between the participants. In the field of AI & Law there is a growing body of theoretical research on discourse support for legal argument and legal procedure (e.g. [3, 1, 4, 12]). However, substantial research on architectures for implementation and on user experiences is still sparse. We know of only two systems that have been implemented with practical use in mind, viz. Loui’s Room 5 system [8] and Verheij’s ArguMed tool [18], and one further system that is currently being implemented, viz. Lodder & Huygen’s support tool for online dispute resolution [7]. In other application areas, such as meeting support and intelligent tutoring, more practical experience with discourse support systems has been gained (see e.g. [9, 16, 15, 2]). These experiences raise im- portant issues for legal discourse support systems. One of the main lessons learned is that it is very easy to overestimate the users’ abil- ity and willingness to learn a new codification scheme [15, 2]. The PROSUPPORT project, on which this paper reports, intends to take this lesson at heart. Its aim is to develop a discourse support system for legal procedure that provides useful computational power to the user but that is also easy to use. Naturally, these two goals tend to conflict. The desire to offer use- ful computational power to the user requires that the user’s input is structured as much as possible, in a way that reflects the essential ele- ments of legal discourse. The more these elements are made explicit by the user, the more the system can do with it. However, the de- sire to make these elements explicit requires complex representation schemes for the user’s input, which leads to a tension with the lessons on usability learned in other areas. Put simply, the more expressive a language, the harder it is to learn and use. Resolving this tension in an optimal way is one of the main research themes of the PROSUP- PORT project. In other words, the project aims to discover conditions under which “formality” in interactive systems of the studied kind is helpful instead of harmful (cf. [15]). To elaborate on the desired expressiveness, the following features of legal reasoning are especially relevant. Firstly, legal reasoning is adversarial, which means that arguments pro and con a claim are exchanged and conflicts between arguments must be resolved. Sec- ondly, legal reasoning contains several specialised reasoning forms, such as combining rules and precedents, attacking the application of a rule, using and attack witness or expert evidence, reasoning about causation, and so on. Finally, legal reasoning takes place in a proce- dural context, where the notions of presumptions and burden of proof are essential, and where not only arguments but also other speech acts