Parties involved in disputes often lack the information they need to take rational decisions. As a consequence, they frequently enter into agreements that are not as advantageous as they could be. Having the right information in the right time would guide parties into taking more weighted and realistic decisions. Specifically, parties should consider their best, worst and most likely outcomes in litigation as well as all the possibilities in between. In this paper the importance of this information is highlighted and domain-dependent methods for compiling it are presented. Moreover, this work describes three case studies in which these methods are being applied with the objective of informing the parties, empowering their role in the dispute resolution process and helping them achieve more satisfactory outcomes.
Conflicts arise in the most different scenarios and are present in our day-to-day since our first years of life. Essentially, conflicts are due to the competitive nature of our society, in which each individual wants to maximize his personal gain. A conflict is generally defined as an opposition of interests between two persons. When two persons with opposing interests clash, a dispute arises, that will eventually need to be settled. Each of these persons has ideas and values of its own that will guide and support his actions throughout the dispute resolution process.
Until just a few years ago, these conflicts occurred mainly between two persons
that where on the psychical presence of each other. However, given the new
information society that we now live on, this is no longer necessarily true. In fact,
online activities, such as the use of e-commerce sites amazon.com and ebay.com, have
led to the development of on-line disputes. We argue that if a transaction occurs
online, then disputants are likely to accept online techniques to resolve their disputes.
Thus, the development of e-commerce requires new ways of resolving conflicts that
avoid courts. Different forms or methods of alternative dispute resolution for
electronic environments have been pointed out by legal doctrine. Thus being, we can
now speak of Online Dispute Resolution (ODR) as any method of dispute resolution
in which wholly or partially an open or closed network is used as a virtual location to
solve a dispute [
A relevant issue, in a first moment, will be to inquire in what way (and to what
point) traditional mechanisms such as negotiation [
The work presented here develops around these two main ideas. We will therefore
analyze to what extent technology can be used to help parties achieve more
satisfactory solutions. Specifically, we will look at several methods for efficient and
contextualized information retrieval as a way to provide meaningful information that
is not available in traditional procedures. Moreover, we will look at three novel ODR
prototypes developed with the objective of fastening and making more efficient
dispute resolution processes. These prototypes are supported by UMCourt, an
agentbased architecture that supports the development of ODR services [
Throughout this paper we will be guided by a main idea: there is not one single technological solution that can address all the problems. In that sense, techniques and procedures should be chosen and adapted according to each specific legal domain.
On-line dispute resolution methods can provide easy, efficient and fast ways for
resolving disputes, contrary to the judicial path which is usually expensive and time
consuming. First and second generation ODR [
This architecture is built on the notion of intelligent agent [
This is possible to do because there are procedures that are independent of domain. Let us take as example a negotiation process. This generally consists on several consecutive rounds in which each agent states an opinion about the proposal currently on the table. And this is independent of the domain of the negotiation. Thus, high level agents guide the process and determine when a new round should start, the turn of the several agents or when the process should finish. Each low level agent then has the autonomy to choose among the several actions in the knowledge base, the one that corresponds to the current domain of the negotiation.
Following this approach increases functionality reuse and allows to have a single architecture supporting services in a wide range of domains. It also simplifies the task of adding a new legal domain. In fact, in order to do so, it is generally only necessary to develop the ontology of the new domain, with all the actions, rules, constraints and specific concepts. This will tell the low level agents how to act when they receive a task from this new domain.
Let us now move to a close description of the architecture. At the moment, it
implements two high level functionalities: a case-based reasoning (CBR) [
All the agents and their roles are depicted in Table 1. These agents were defined after the specific requirements of this project, following an iterative cut-down process of increasing specification. Nevertheless, given the open nature of the architecture, it can be easily extended by adding new agents or ontologies, namely to address other legal domains or to implement new functionalities in a domain. Reviser
FSA
Indexer l Parser e v e l w lo Rules
Extension l o trn Coordinator o c actions to adapt a given case so that it can be used.
Looks at a group of cases in order to select an outcome/solution for a given case. Proposes the outcome to the coordinator as well as a justification and waits for the outcome. If the outcome does not comply with the one suggested provides a list of more probable reasons for the failure. Agents request case information to this agent. This agent provides the cases by interacting with the Parser agent. Each case that is requested is maintained in memory so that, should they be requested again, they are rapidly available in memory and do not need to be parsed again. This agent also checks constantly for changes in the files of the cases that are in memory in order to maintain them synchronized.
Contains a list of agents and the actions that they are authorized to perform.
Is responsible for implementing services to implement all the interactions with the database. It is also responsible for the security and integrity of the database, making sure that only the correct agents perform the authorized operations.
Contains a list of Jade FSM behaviors that describe the guidelines or steps necessary for an agent to perform defined actions.
Indexes new cases in the Database and creates the specific xml files in the correct system folders.
Verifies the validity of XML files against the defined schemas. Valid cases are parsed and returned as a Java Object that can be handled by other agent.
Embodies rules of type if condition then action that provide basic reactive actions for guiding agents in the decision making processes. Can be used to model a rule-based legal domain. Multiple instances of this agent exist that implement different pre-selection algorithms. Some have already been implemented, such as the Template Retrieval while others (e.g. Clustering algorithms) are now being developed.
Defines several search and similarity settings according to which retrieve parameters can be changed.
Multiple instances of this agent embody different similarity algorithms.
This agent represents a group of agents that have as task to gather information from html forms. Then, they compile that information and forward them according to the defined format to the agents in the backend. As each html page in the frontend generally has, at least, one agent and they perform relatively simple tasks, we will not detail them further here.
This agent receives requests from external agents and forwards them to the correct agents inside the platform, provided the external agents are trying to perform an action that they are authorized for.
Receives task requests from other agents (e.g. external agents, interface agents) and take the necessary steps (requesting tasks to other agents) in order to perform them. This agent maintains a list of active tasks and has access to a list of finite state Fault Manager
automata that define the next action for each task, provided by the FSA agent.
Starts all the necessary agents according to an editable XML configuration file. It then constantly checks the agent registry and, if any agent has crashed, restarts another instance of that agent.
This agent controls the pending requests to specific secondary agents and starts new instances of agents that have a significant amount of work load.
The organization of this agents is depicted in Figure 1. In this figure, rounded rectangles represent agents and the lines represent the main communication paths. Note the existence of the DF agent (Directory Facilitator), an agent that makes part of the Jade platform and provides support for service registry and lookup.
The ideal dispute resolution process is one in which the two parties are better at the end than they were at the beginning. Unfortunately, not all disputes have this conclusion. In order to improve this, we believe that it is of ultimate importance to: (1) provide the parties with more knowledge about the dispute and (2) potentiate the role of the parties throughout all the process. In fact, parties that have poor access to important information generally end making bad choices or, at least, they hardly make the best choice. Moreover, parties usually have a reduced role on the resolution process, resulting in suspicion about the outcome, mostly because they do not understand how it was achieved. Thus, in this section we describe the approach followed in the development of UMCourt in order to attack these problems. 3.1
The first step is thus to identify the knowledge that is meaningful for the parties,
according to the legal domain of the dispute, and then determine the methodologies
suited for compiling that knowledge. On a first instance, it would be interesting for a
party to determine to which extent is it reasonable to engage in a dispute resolution
process. That is, are there significant advantages against a traditional litigation? This
question can be analyzed from several points of view. On the one hand, alternative
dispute resolution processes are generally faster, cheaper, more private and
personalized [
It would thus be really important for each party to know its BATNA - Best
Alternative to a Negotiated Agreement, or the possible best outcome “along a
particular path if I try to get my interests satisfied in a way that does not require
negotiation with the other party” [
However, the use of the BATNA alone is not enough to take informed decisions as
parties often tend “to develop an overly optimistic view on their chances in disputes”
[
However, it could also be interesting to consider the whole space between the
BATNA and WATNA as a useful element to be taken into account for making (or
accepting) a proposal. If we consider for instance, in the labor law domain, the
scenario of a worker being fired, litigation will most likely occur. Under many legal
systems, a huge deal of legal parameters have to be considered, including antiquity,
supplementary work, just cause for dismissal, among others. For the worker, the
amounts involved are not irrelevant: being fired without good indemnities may be
seen as a double sacrifice. But he might, on the other side, receive significant
financial compensation. In order to clearly see the advantages of a proposed
agreement, parties should thus also consider the spectrum between their BATNA and
their WATNA. Of course, the less space there is between BATNA and WATNA, the
less dangerous it becomes for the party not to accept the agreement (unless, of course,
their BATNA is really disadvantageous). A wider space between BATNA and
WATNA would usually mean that it can become rather dangerous for the party not to
accept the ODR agreement (except in situations when the WATNA is really not
inconvenient at all for the party). We can thus argue that knowledge about the space
between the BATNA and the WATNA is also very important. This space is evidently
related to the Zone of Possible Agreement proposed by Raiffa (1982) [
Moreover, it would be interesting for a party to understand the region of this space
in which an outcome is more likely. That is, if the parties are to solve the dispute
through litigation, what is the most likely outcome? In fact, sticking only with the
BATNA and WATNA may not be realistic as these are usually not the most likely
outcomes but merely informative values that establish boundaries. Thus, an informed
party should also consider the MLATNA – Most Likely Alternative to a Negotiated
Agreement [
Concluding, in order for a party that is engaging in an alternative dispute resolution to take informed and rational decisions, he should consider knowledge about: (1) the best possible outcome in litigation; (2) the worst possible outcome in litigation; (3) the space between the two previous values; (4) the most likely outcome in litigation and (5) the probability of each outcome within the zone of possible agreement.
Having seen the knowledge that a party should have in order to take rational
decisions, let us now depict the methods for compiling that knowledge. Considering
the BATNA and the WATNA values, we are using a rule-based approach. Rule-based
Systems (RbS) are generally the simplest way of implementing intelligent behaviors
[
Let us take, as an example, the Portuguese labor law domain, as depicted in Decree of Law (DL) 7/2009 (Portuguese laws), considering a scenario in which a worker wants to end the labor contract claiming that the employer did not pay the last three salaries. According to Article 394th, nr. 2 a), the lack of regular payment of the salary constitutes a just cause for a worker to end the contract. Moreover, Article 394th, nr. 1 when there is a just cause, the worker can immediately end the labor contract. The first question is thus to determine the existence or not of the lack of payment, and thus, of a just cause for ending the contract. Assuming that this has been proved, let us try to determine the best and worst scenarios, from the point of view of the worker. The most important norms are found in Article 396th, numbers 1, 3 and 4. Number 1 states that, if Article 394th is true (there is just cause for ending contract), the worker is entitled to 15 to 45 days of salary plus indemnity for each year of contract. It also states that this value varies according to the degree of wrongfulness of the employer and that the total indemnity paid to the worker should not be inferior to three salaries plus indemnity. However, number 3 states that the indemnity paid can be higher whenever the worker suffered property damage or other damage, of higher value. Finally, number 4 states that, in the cases of a temporary employment contract, the value of the indemnity cannot be smaller than the value of the salaries that would be received until the end of the contract. We can thus formalize the computation of the BATNA and WATNA in the form of IF-THEN rules.
A simplification of the rules that allow the computation of the BATNA and WATNA values according to the Portuguese labour law. This example code considers only the case in which a worker ends the contract with a just cause. M_SALARY denotes the monthly salary; D_SALARY denotes the daily salary; M_REMAINING denotes the months remaining until the end of the temporary contract; +VARIABLE denotes an unknown value, higher than VARIABLE.
Def_Rule 396 if RULE_394 then
WATNA := 3 * (M_SALARY + SENIORITY) if TEMPORARY_CONTRACT then if WATNA < M_REMAINING *(M_SALARY + SENIORITY) then
WATNA := M_REMAINING *(M_SALARY + SENIORITY) if WATNA < 15 * (D_SALARY + SENIORITY) then
WATNA := 15 * (D_SALARY + SENIORITY) BATNA := 45 * (D_SALARY + SENIORITY) if BATNA < DAMAGE then
BATNA := +DAMAGE
There are some interesting advantages in following a rule-based approach. The main is that this is a relatively simple way of implementing legal norms, being also easy to maintain and update. Having defined the values of the BATNA and the WATNA, it is immediately possible to compute the range of the ZOPA: it is given by the distance between these two values. More challenging is the determination of the possible outcomes and its corresponding likeliness.
In this case a purely rule-based approach would not be appropriate as it is
necessary to evaluate a group of cases and categorize them according to their
likeliness. Moreover, a mechanism to select cases according to its similarity is also
obligatory. Therefore, following a case-based approach would be an appropriate
choice [
In order to determine the possible cases, their likeliness and the MLATNA,
UMCourt relies on the previously mentioned CBR algorithm. However, as our focus
is merely on compiling information for informing parties rather than suggesting an
outcome for the dispute, only the first stage of the algorithm is relevant [
To do it, a hybrid approach is being used that combines a template algorithm with a
nearest neighbor one [
In the next step, the nearest neighbor algorithm must only be applied to the set of pre-selected cases instead of applying it to all the cases in the case memory, a task that could be very time consuming as the nearest neighbor algorithm has linear complexity (equation 1).
In this algorithm, the weights are, at this moment, determined by a law expert, based on the importance that, according to his experience, each of the components of the similarity measure has. However, it is our objective that, in the future, the system changes these values dynamically, looking at past iterations, in an attempt to select the most appropriate weights for each case.
The output of this phase of the algorithm is a list of similar cases, ordered according to their value of similarity with the current case. We only need to add the main assumption of CBR: if a case is similar to another one, then its conclusion is also expected to be similar. Based on this, we can look at the solution of the retrieved case with the higher degree of similarity and assume that it is the most likely solution to the current case. Likewise, we can look at the region in which the similarity of the cases is higher and assume that the probable solution lies in that region, i.e., the MLATNA. Following the same approach, for each case retrieved a value of similarity is also provided to the user, which will indicate its likeliness to occur.
At this point, the worker has all the main information that he/she may need in order to make rational decisions throughout the dispute resolution process. It is possible to use a visual tool to represent all this information in a single and intuitive graphic that the party can consult. In Figure 2 a graphic for a fictitious case is presented. Each dot represents a case, with an associated value of similarity and an utility value, which represents the indemnity that the worker will receive. The case with the highest value of similarity is the MLATNA and tells the worker the most likely outcome if he decides to go into a court. The positions of the BATNA and WATNA are also represented, depicting the best and worst possible scenarios. The dashed line is given by a 3rd degree polynomial function and represents an overall view of similarity versus utility. Looking at this line, the worker can conclude that, if he goes into a litigation process, according to the known cases, the indemnity will most likely be between a value of 500 and 650.
MLATNA
BATNA
Combining cases with rules, it was possible to develop a methodology that compiles all the important information at the outset of the dispute resolution process. This will, in a first moment, help the disputant party decide if he/she should advance into litigation or if it would be better to continue with this process. At this point, the disputant party is able to weight the consequences of his possible decisions, assess its chances of success, determine its consequences and thus achieve more satisfactory outcomes. Let us now depict three case studies in which the techniques depicted in this paper are being applied, all focusing on the Portuguese legal domain.
UMCourt Divorce and Heritage’s Share [
However, the parties may not be completely honest when assigning the points. As an example, if one of the parties makes the allocation of points considering the monetary value of the items, and the other party doesn’t have a notion about the prices or simply assigns the points according to the preferences (unaware of the bad intention of the other party), the second party is in disadvantage. In order to address this fair division problem, considering the monetary perspective, this work introduces some changes in the AW algorithm: the Adjusted Winner by Value. Therefore, a component depicting the monetary value of each item, in which the arithmetic manipulation is performed, was added.
Although the division presented by AW by Value may be considered fair according to the preferences and the values of the items, the parties may choose not to accept it as it was proposed. In order to address this issue, a negotiation mechanism that can mediate the process and present other alternatives was developed. This mechanism is supported by the techniques presented in this paper. In that sense, parties can access other similar cases, know their best and worst scenarios, know the most likely scenario and thus cooperate in a more rational and realistic way in the negotiation process.
The legal domain of this case study is the Portuguese consumer law. Because this domain is a quite wide one, we restricted it to the problematic of buy and sell of consumer goods and respective warranty contracts. Thus, concrete solutions for the conflicts arising from the supply of defective goods (embodied mobiles or real estate) were modeled. Financial services are also considered, as well as the cases in which there are damages arising out of defective products. Regarding the legal boundaries established, solutions for conflicts are being modeled as they are depicted in Decree of Law (DL) 67/2003, as published by DL 84/2008 (Portuguese laws).
Using this system, an unsatisfied buyer can use a web site1 or a mobile application
(Figure 3) in two different ways: for simulating a defect of a product that he intends to
buy of for asking for a solution for a dispute arising out of a defective product already
bought [
In that sense, the information mentioned above can help workers throughout such processes. Specifically, a worker can consult past similar cases and know the likeliness of their solutions in his dispute, know his BATNA and WATNA and know his MLATNA. Moreover, a multi-party negotiation tool supported by the CBR mechanism is also available (Figures 4 and 5). The main purpose of this tool is to support effective negotiation between two or more parties involved in a labor law dispute. The tool starts by proposing the solution of the MLATNA and the parties engage in a sequence of turns in which, in each turn, all parties can accept, change, ignore or refuse the current proposal for solution. In each round, if there is no consensus, the system will build a solution from the suggestions of the parties (if possible) or will suggest a solution from a similar case. The process goes on until a consensus is achieved or the system runs out of suggestions for solutions. In this case study, not only the party has access to the information described above but can also use a negotiation tool that will improve the efficacy of the alternative dispute resolution process.
The main idea that is present throughout this paper is that informed parties are in a better position to take weighted and rational decisions. Concretely, a party should consider, in a first moment, the usefulness of entering into an alternative dispute resolution process. In order to take this decision, concepts like the BATNA, WATNA and MLATNA are of utter importance. Moreover, the party should also be aware of all the possible outcomes and their likeliness, in order to have a clear picture of all the possibilities. In this paper we identified the relevant knowledge for taking rational decisions and pointed out to domain-dependent methods for creating that knowledge. In a parallel work, we are developing an abstract architecture that implements these concepts in several legal fields by using ontologies. In this approach, agents are abstract entities that provide services useful for all the domains addressed. However, a service is implemented differently in each domain, according to the specificities of the legislation. In order to implement the services for each specific domain the agents use domain ontologies, which define how each action should be implemented. This approach results in simpler architectures in which functionality reuse is maximized. Acknowledgments. The work described in this paper is included in TIARAC Telematics and Artificial Intelligence in Alternative Conflict Resolution Project (PTDC/JUR/71354/2006), which is a research project supported by FCT (Science & Technology Foundation), Portugal.