Statement_types_in_legal_argument | Al-Abdulkarim, Atkinson, Bench-Capon, JURIX 2016
Abstract:
In this paper we present an overview of the process of argumentation with legal cases, from evidence to verdict. We identify the various different types of statement involved in the various stages, and describe how the various types relate to one another. In particular we show how we can obtain the legally accepted facts which form the basis for consideration of the law governing the cases from facts about the world. We also explain how we can determine which particular facts are relevant. In so doing we bring together several important pieces of AI and Law research and clarify their relationships.
Introduction
Can we identify the various stages of reasoning from evidence to verdict? Typically models and approaches in this field only focus on one particular type of stage. This paper instead focuses on transitions between stages and the different types of statements used in each stage.
Previous attempt to distinguish stages of reasoning: K. Atkinson and T. J. M. Bench-Capon. Legal case-based reasoning as practical reasoning. Artif. Intell. Law, 13(1):93–131, 2005
Attention points:
- Differentiate different types of statement in legal reasoning
- Can we distinguish and model the transition between reasoning about facts and reasoning about law?
- How do we deal with uncertainty?
- "Although many legal concepts are characterised by open texture [32] and so have a penumbra of uncertainty [22], the verdict itself can permit no qualification. How and where the doubts are resolved remains an open question." (3)
- "Most approaches (e.g. [27]) represent legal concepts only as Booleans. For example factors are considered present or absent and values are promoted or demoted, whereas we believe that differing degrees of presence and absence (e.g. [10], [19], [1]) need to be recognised."
- "By "Verdict" we mean a binary decision such as guilty or innocent and recognise that, when appropriate, sentence and damages remain to be determined. In the US Supreme Court, much discussed in AI and Law research, often the decision of the lower court is affirmed or quashed and the case returned to he lower court for resolution" (footnote p. 3)
- TODO an innovation of my own project is that sentences are modeled as vectors, including the height of the sentence and the possibility of co-occurring sentences.
- So in these terms I model the "sentences" rather than the "verdict".
Layers of legal reasoning
Focus is on US civil cases based on common law.
According to the authors:
- "In criminal law there is more focus on what the evidence establishes" (4)
- "in European civil law precedents are given less emphasis."
My own project focuses on Dutch (so European) criminal law. To which degree do the findings of this paper generalize to this situation?
The stages identified in this paper are:
- Evidence
- Intermediate Fact
- Legal Fact
- Base Level Factor
- Abstract Factor
- Issue
- Intermediate Issue
- Verdict
Discussed in top-down manner, so starting with verdict and ending with evidence.
Verdict
"The verdict is binary: the verdict must be “yea or nay”: the judge cannot refuse to answer the question, nor express uncertainty or doubt as to the verdict. Leave to appeal may be given, but the verdict remains in force until it is overturned or quashed on appeal." (4)
The verdict thus has the character of a decision. In this sense it is not helpful to consider it as "true" or "false"; instead it is a performance utterance in the sense that if a judge decides "guilty" then it is so. Verdict statements thus have "assertability" or "felicity" (4).
The justification of the verdict is supposed to be in terms of "issues"; moreover, there is apparently wide consensus that the relation between issues and verdict is deductive:
Issues
"The court is required to justify the verdict. Justification is in terms of issues. The relation between issues and verdict, is, as is agreed by all of [30], [18] and [27], deductive: the verdict is expected to follow logically once the issues have been resolved." (4)
Such reference to "issues" may be a reference to a:
- statute
- commentary
- summaries of the common law
- precedent cases
It seems to me this reasoning should be deductive. But has it been shown that this reasoning is in fact deductive? The phrase "justifying" a verdict could also be interpreted as saying that judge's reach a verdict and then need to justify it. I.e. the deductive inferences could be a post-hoc reconstruction of the judge's reasoning which may itself be largely based on common sense reasoning or perhaps be influenced by psychological biases. And even so, this idea of a neat deduction leading up to a verdict seems to be at odds with the non-controversial statement that the law leaves room for a judge's (subjective) judgment, although this could be more evident for the determination of the height of the sentence rather than the verdict itself.
Even though I have my doubts about whether a judge's reasoning is indeed logically deductive, it is clear that they are modelled in AI systems. The authors discuss "logical models of legislation systems" (4): "Thus the logical model tells us precisely what arguments can be presented, what issues are relevant, how a case should be presented in terms of these issues, and the consequences of resolving these issues in particular ways. (4-5)"
Factors
Another approach is the reasoning based on "factors", e.g. see Falakmasir 2017 and Westermann 2019. The detection of factors is often done using machine learning techniques, the reasoning over the detected factors themselves is typically based on logical rules.
The authors also point this out in the sense that they recognize that the concepts represented as factors are not "sharp concepts" (5). Specifically, if the logical models are defined on "issues", then an additional step from factor to issue is needed. But this step requires a form of judgment, namely whether the information of the detected factor is valuable and then whether this information is in favor or against. Determining the value of this information in turns requires looking at precedent cases...
"What is needed is a judgement as to whether the value of the information was sufficient for it to be considered a trade secret. It is the declaration that it is sufficient that enables the move from the extent to which plaintiff is (or is not) favoured, which is a matter of degree, to the sharp concepts needed to drive the logical model in the next stage of reasoning." (5)
The authors point out that even though the theoretical legal formulation of factors, e.g. whether information is a trade secret, is phrased in terms as "the extent to which the information is known outside of his business". However, reasoning applications of factors (see refs above) treat factors as present or not present!
Dimensions
But what is the relation between facts and factors?
Example of a dimension: "Thus HYPO has a dimension Security-Measures which has none as the extreme pro-defendant position and then steps through a series of more rigorous measures until the extreme pro-plaintiff position is reached." (7)
So if I'm not mistaken this could be suggestion on how to record factors in a non-binary way. Various measures or "dimension points" (8) can be detected that contribute to a dimension either in the direction of the defendant or the plaintiff. This is the first role of dimensions.
Thus dimensions provide a way of determining what are the relevant features of the case. The set of dimensions circumscribes what can be considered relevant aspects of the case, and the dimensions which are applicable on the facts are the aspects which are relevant to the case under consideration [12]. Moreover since dimensions range from an extreme pro-plaintiff point to an extreme pro-defendant point they can be used to indicate the degree to which the plaintiff is favoured on that aspect. However, they do not determine the legal significance of the fact pattern: unless the facts put the case at one of the extremes the dimension cannot be said to favour either party. In order to determine the significance, some point or points on the dimension must be determined as cross over points, where the dimension ceases altogether to favour the plaintiff and where it begins to favour (increasingly) the defendant. The crossover may be a single point, or there may be a neutral buffer zone. (7)
Legal facts
Legal facts are, within the scope of the case, considered to be certain; but their support in a system is a matter of degree. In this framework "it is the allocation to this range that is the legal fact." (8) Moreover, this framework inherently has a criterion for legal relevance: if a fact does not relate to a dimension, it is considered legally irrelevant. This agreement on the "legal" facts is the second role of dimensions.
A third role is that assignment of a fact to a dimension can be understood as conclusions of reasoning about the presented evidence.
Thus the factual reasoning establishes, on the basis of the evidence, a degree of belief or confidence in a statement that the case should be assigned to particular range on a dimension, and then this is accepted or rejected as a legal fact according as to whether or not it meets the standard required by the particular procedural context. (9)
World facts
A proof standard describes how much confidence is required to consider a certain proposition to be true. An example in criminal law is that a fact should be "beyond reasonable doubt". What this establishes is a transition from a "world fact" into a "legal fact". There is always some degree of uncertainty in case of the former, but in a legal case the established facts are a fixed point of departure and treated as certain. That is "the passage from the world to law renders all facts equal: the degree to which they emerged from the factual reasoning." (9)
Evidence
Example of a chain starting from evidence (testimony in this example):
Testimony is presented, intermediate facts are derived with varying degrees of certainty, to establish a degree of belief in the particular location of the case on various dimensions, whereupon application of the appropriate standard of proof admits (or excludes) the claim as a legally accepted fact. (10)
Representation
Given the above a statement will be a 4-tuple:
- Statement type
- Type of parents
- Type of children
- Value
Remarks on the need for ontologies and feasbility:
Additionally we will need a small ontology recording dimensions, the recognised ranges on the dimensions, the cross overs and the degree to which these points are said to favour the plaintiff. The size is tightly constrained by the use of dimensions to indicate relevance, and so does not fall foul of the problems identified in [6], where producing the ontology seems a dauntingly unlimited task. The problems noted there do, however, come to the fore when trying to design the factual reasoning from evidence to the legal facts. This is because the facts - and the appropriate styles of reasoning - are so disparate. But there is no necessity for a system to include this level of detail. (11)