@Article{Vincent2003,
author = {Aleven, Vincent},
title = {Using background knowledge in case-based legal reasoning: A computational model and an intelligent learning environment.},
journal = {Artificial Intelligence},
year = {2003},
volume = {150},
Tpages = {183--237},
}
In the law profession, legal experts have to organize a written argument based on multiple cases and make arguments about the similarity of cases. In order to help with the task, the authors have devised an expert system called CATO that uses case based reasoning (CBR) to represent and apply middle-level normative background knowledge in formulating the legal arguments for lawyers to use.
The first claim that the authors make in this article is that their program CATO
“…includes a representation of middle-level normative background knowledge for one
area of the law, trade secrets law, in the form of a Factor Hierarchy. Further,
it contains computational methods for using the background knowledge to
(1) generate multi-case arguments organized by issues,
(2) make context-sensitive
arguments about the significance of differences between cases, and
(3) select the best cases to cite in an argument.”[p.185]
The second claim of the paper is that the argumentation model CATO can be used effectively in computer-based instruction. Students and teachers can use the argumentation skills provided by the model in argumentation examples of legal case. Therefore, this research paper is relevant to the fields of AI and Law as well as AI and Education.
Important skills for beginning law students to have:
A representation of middle-level normative background knowledge
Background: Use of factors to represent cases
“Factors are stereotypical collections of facts that, experts agree,
influence the outcome of a case.” [p.190]. For example, in a case where the
defendant has reproduced the plaintiff’s unpatented product, the plaintiff’s
position is strengthened by the fact that its product was unique on the market.
Also, one of the defendants took product development information and tools from the
plaintiff’s business and the defendant manufactured a product that was highly similar
to the plaintiff’s. On the other hand, the factor that is in favor of the defendant is
that the plaintiff did not take any measures to protect the secrecy of its information
and disclosed information to outsiders. “The construction of such arguments involves
the use of background knowledge about the meaning of the factors used to represent cases.” [p.191]
The Factor Hierarchy
The factors in the cases that are in CATO are organized hierarchically and they are
linked together by positive (strong) and negative (weak) links. A positive link indicates
that a factor supports the conclusion of a particular side of the case. A negative link however,
means that the arguments are in favor of the adversary in a given case.
Using background knowledge to organize multi-case arguments by issues
Identifying issues
The CATO model focuses on issues of how the law should be applied to the facts of a case,
related to some of the important concepts in trade secrets law. The system searches for
the legal implication in a factor hierarchy database related to the factors specific to the case.
If a solution is not found then previous cases with similar arguments are searched. The CATO system
“… uses the same representation of background knowledge to organize arguments by issues and to reason
about the significance of differences of cases.” [p.194]
Organizing arguments by issues
After having found the issues that are relevant to the case, CATO organizes them by identifying the
strengths and weaknesses of the arguments that concern plaintiff and defendant.
It also gives explanations of why a particular factual strength matters to an issue.
“In order to generate an Issue-Based Argument, CATO must be given a problem situation
and a set of precedent cases, all represented in terms factors, and a side on whose behalf to argue (plaintiff or defendant).”[P.197]
Example arguments about the significance of distinctions
In the cases being compared by CATO, it is important to assign weights to similarities and differences
that are sensitive to the context of the arguments.
“The significance of differences must depend on the specifics of the problem and past
case being compared and the purpose for which the comparison is made.”[p.200]
CATO can identify the significance of a distinction between two cases’ arguments by characterizing that the difference exists at a higher abstract level in the factor hierarchy.
Downplaying a distinction by drawing an abstract parallel between cases
In order for CATO to make an abstract conclusion about the similarities of two cases,
it uses a search method that propagates between factors taking into account the weak and strong ones.
If a factor comes up as a distinct from the case that it is being compared to, the system searches a higher
level of abstraction to determine if this distinction is overruled by a stronger similarity or not.
If the search comes up with a stronger similarity factor at a more abstract level,
then the CATO downplays the distinction and draws an abstract parallel between the cases
based on the stronger similar factors. “Weak evidence in favor of a conclusion can be blocked by strong opposing evidence.
Strong evidence in favor of a conclusion however cannot be blocked.
When the evidence for the two opposing conclusions associated with a given abstract factor is of equal strength,
we consider both conclusions to be supported.” [p.202]
Testing CATO’s Performance
The authors tested the CATO performance against other CBR methods available to solve legal cases.
They used a case database of 184 trade secrets cases, and 26 argument factors,
half of which are in favor of the plaintiff and half in favor of the defendant.
Seven CBR methods were evaluated on predicting case outcomes based on the relevance criterion R, as follows [p.213]:
find all cases that satisfy relevance criterion R
if there are cases that satisfy R and all of them were won by the same side
then predict that this side wins
otherwise, abstain
The results show that there were significant differences in the performance of CATO compared to the other CBR methods that don’t use background knowledge. CATO classified 89% of the cases, abstained on 11% of them and incorrect predictions accounted for only 1% of the cases. This was significantly higher correct predictions than the next best CBR method which classified 69% of the cases abstaining from the rest and had 5% incorrect predictions. This performance evaluation lends support to the accuracy of the CATO system over the others, an improvement credited in part to background knowledge.
Evaluating CATO instructional effectiveness
An experiment was conducted with 30 law students in order
to evaluate CATO’s effectiveness as an instructional tool.
The students were divided into two groups. In the control group,
14 students attended their regular argument classes. For the experimental group,
16 students worked in pairs on the CATO program. At the end of the 9 periods, the
students were tested on their argument skills and on their Legal Memo-Writing.
. The results of the experiment show that the there is no significant difference in
performance on the argument test for the two groups. The CATO group scored 70% compared
to the 68% scored by the control group. On the legal memo-writing however, the control group
scored better than the experimental group with scores of 79% and 70% respectively.
The researchers concluded that CATO can be used as a substitute for the purposes of
training law students on the argumentation skills. The weaknesses of the CATO program
lies in the fact that it can not reason holistically about cases, which is the type of
reasoning required for the legal memo-writing part of the test.
Conclusion
In this article, the authors “…have described and illustrated novel methods for using background knowledge
(1) to organize multi-case arguments by issues,
(2) to generate context-sensitive arguments about the similarity of cases, which focus
on alternative characterizations of a distinction’s significance,and
(3) to assess the similarity of cases in order to select the best cases to cite in an argument.” [p.231]
The authors have evaluated the performance of their CATO program compared to other CBR methods,
and also evaluated its effectiveness for law instruction both of which showed significant results in favor of CATO.