I. Legal Reasoning - Generally

All legal reasoning follows one path. No legal argument can be accepted or rejected without all of the following pieces

1) Issue - What specifically is being debated?

2) Rule - What legal rule governs this issue?

3) Facts - What are the facts relevant to this Rule?

4) Analysis - Apply the rule to the facts.

5) Conclusion - Having applied the rule to the facts, what's the outcome?

II. Legal Reasoning - Explained with an Example

So, what does this mean?

1) Issue

The "issue" is the legal issue. It doesn't ask just any interesting question. It only asks whether THE LAW has anything to say about a particular topic. A classic example of this is a potential legal client who comes in and says that her boss is mean and rude -- he yells and screams and makes work wholly unpleasant. The client wants to know if she has a claim. I already know that there is no law (no rule) that generally prohibits a boss from being a jerk. However, experience tells me that the question should be:

(For those of you with quick logical minds, yes, this means there are forms of lawful discrimination.)

2) Rule

The "Rule" has two important parts. A lawyer, a judge, or whomever has to say what a rule is and where it comes from.

a) State the Rule

That rule says (paraphrasing): "It is unlawful to treat someone in a manner that negatively affects the terms and conditions of employment, if the affected person is in a 'protected class' and is treated differently from a 'similarly situated person' not in her protected class."

Each of the logical pieces you can break it into are called the "elements" of the rule. So, you could say the "elements" of discrimination are

Each of these pieces contain legal terms of art, terms that have their own legal rules. So, you'd actually end up with some nesting here.

b) Cite the Rule

The law is based on existing rules. Even when a decision is based upon what is "fair" (which isn't that often), it's because there's a rule that says that the decision of this type of issue will be based on fairness. And, there are so many rules that no one can know them all. So, an argument has no weight unless it says exactly which rule is being relied upon. As you've seen already this presents a variety of challenges:

If the lawyer provides the wrong law, she can lose the case (or the legal analysis), even if in the cosmos she should have won (ie., there's a law out there that gives her the result she wants). The law has mistakes in it, so the lawyer has to cite the law that exists and then provide some sort of annotation that explains why it's a mistake and/or where the mistake is. Like the Web, there are lots of versions of the same or similar things. You'll see this in case law. There can be lots of decisions that say the same thing on a particular issue. There are often decisions that cite other decisions for support. And, there's more than one publisher, so there's more than one citation to the exact same court decision. Within certain boundaries, any of these citations might be used.

In this example, the rules are:

3) Fact

There are lots and lots of facts that make up the client's story. For the purpose of legal analysis, we look for "material" facts. These are the facts that fit the elements of the rule.

So, in the example, we need to know: if the boss' behavior "affected" a "term or condition of employment"; if the potential client is in a "protected class"; if there are "similarly situated" employees; and if they've been treated in the same manner or differently. The facts that turn out to be relevant are:

4) Analysis

At this stage, we see if our material facts fit the law. So, in the example, we'll say

5) Conclusion

We see that all "elements" of the rule are met and conclude that her boss engaged in unlawful discrimination.

III. Our "Deadbeat Dad" Example

Applying all of this to our previously discussed "deadbeat dad" example, we see:

1) Issue:

Is Joe a Deadbeat Dad?

2) Rule:

a) State the rule:

Anyone is a Deadbeat Dad, if:

b) Cite the rule:

I know this is a rule, because it is a law passed by Congress. It's published at:

3) Fact:

We put together a hypothetical in which

4) Analysis:

Our analysis should say:

5) Conclusion:

When we following the chain of legal reasoning, we find out that I didn't provide enough information. We really don't know if Joe is a "deadbeat dad" because we missed one of the factual elements. We don't know if the debt is more than one year old.

IV. Tying it all to TAMI

For the TAMI project, we want to create a system that can look at a government action and tell us whether the action was permissible and/or it was based on logically insufficient inferences. Legal reasoning will provide a significant part of the framework.

A. Is a government action legally permissible?

When we try to determine if a government action was legally permissible, we follow the same steps of legal reasoning. We frame issues such as "Were the officers allowed to conduct a search of the bedroom dresser under these circumstances?" "Was the TSA allowed to base its 'no fly' decision on the data it used?" These questions increase the difficulty and complexity of the process because, to reach a conclusion, you must address multiple rules and the rules are more readily (regularly?) expressed conceptually than logically.

1. How do I express that as code?

We started with the "deadbeat dad" example above because it represents the simplest form of legal reasoning.

As a group, we wrote it in N3. Sample code appears at:


We wanted the answer more or less in English, so Lalana wrote a filter:


Now, when you run it in cwm, you should get some run information and:

:Joe a :DeadbeatDad .

2. But, it wasn't complete.

The reality in the practice of law is that it is not common to follow every branch of every rule that's potentially relevant. Whoever is doing the analysis picks the breadth of rule with which they're going to work. That's only an issue if there's another party who challenges the scope. Because law, like the Web, is massive, it seems that incremental building is the only possible methodology.

3. How do I convert this into something meaningful for TAMI?

TAMI is focused on data mining. This generally means looking at data from more than one source and means that a lot of the TAMI questions will be "Was X person entitled to get Y data under Z circumstances?" Although there will be other rules too, I've explained that the Privacy Act is a nearly universal rule that will apply and I've provided a detailed description of the analysis:


I think the big question is what process will TAMI employ? In the long term, I think people hope for a system that looks at a system log of transactions, knows what rules to apply, and applies them. If the issue is always "Was X person entitled to get Y data under Z circumstances?" you could begin to incrementally build the system. This would likely have a productive intersection with a series of information sharing environments being planned for the federal governmen, because they must produce access control systems that can attempt to process the same rules proactively.

I think in the short term, we could build a prototype that expected the government to name the Rule(s) and provide the Facts; our system would perform the Analysis and reach a Conclusion.

B. Did the government draw incorrect inferences?

We'd like for our system to be able to tell if an activity involved drawing bad conclusions from available data. I think that legal reasoning can only assist in that effort to the extent that it can identify the specific facts that are relevant to the government act. In general, this is one of the things that most distinguishes lawyers from lay people. Lay people tell a story; lawyers harvest a story for the facts relevant to the legal issue. Once we know which facts were legally relevant to the outcome, we probably will need to apply other forms of logic to determine if (a) the fact was an inference drawn from other facts and (b) the inference was warranted.


This document was written by K. Krasnow Waterman in her hat as Research Assistant to TAMI/DIG/CSAIL/MIT.