MLP-1 — Controlled Legal Writing
MLP-1 is an open standard for writing legal text under rules. Michigan Legal Publishing Ltd. publishes it, and anyone may use it. The full text, the checking tools, and the change log are on GitHub at github.com/michlp/mlp-1.
The standard is at Issue 2, a draft for review. We welcome comments and change requests.
The problem it solves
A controlled language is a restricted form of English. You write inside a fixed list of words and a fixed set of sentence patterns, so that readers cannot take the text two ways. Aircraft manuals have used them for forty years.
Every one of those languages works the same way. When a word is hard, you swap it for an easier one.
That approach destroys legal text. A term of art is not a hard word. It is the name of a rule. "Summary judgment" is not a fancy way of saying "a quick decision by the judge." It is a specific procedure, with its own standard of proof and its own body of case law. Swap the term for the plain phrase and you have not made the rule clearer. You have deleted it.
So a controlled language for law has to start somewhere else.
Three things that make it different
Terms of art are kept, not simplified. Rewriting a legal standard in your own words breaks the standard. When a reader needs help with a term, you keep the term and add a short explanation next to it.
Ambiguity is kept and marked. Some rules can honestly be read two ways. The standard does not let you pick one. You write the restatement so it can still be read both ways, and you record that the ambiguity is there. Choosing a reading silently replaces the law with your opinion of it.
Faithfulness is proven, not claimed. A restatement does not count as accurate because its author believes it is. It counts as accurate because someone followed a written procedure, and the record of that work can be inspected afterward.
How the standard is arranged
MLP-1 contains 79 rules and 11 general recommendations, in three parts.
Part 1 covers writing. Eleven sections on words, terms of art, verbs and modality, sentence structure, conditions and exceptions, cross-references, operative text, explanatory text, consequences, punctuation and numbers, and faithful restatement.
Part 2 is a dictionary, and a deliberately small one. It covers only general vocabulary: the ordinary verbs, prepositions and conjunctions that hold a sentence together. It does not try to list approved legal terms. Such a list would be out of date the day it was published, because courts invent new terms constantly. Worse, it would tempt writers to reach for an approved near-synonym instead of the real term, which is the exact mistake the standard exists to prevent.
Part 3 covers checking, and it is described below.
Three appendices record where the standard departs from technical-writing languages, list plain replacements for legalese, and give the form for proposing a change.
Who the text is for
The same rule can be written for a judge or for someone representing themselves. The standard calls this the register, and it sets four of them. The register decides how you handle terms of art and how long your sentences may run.
- R1, operative. For courts, and for the people bound by the text. Terms of art are used with no explanation, because this text is the law itself. Sentences run to 30 words.
- R2, professional. For lawyers, judges, clerks and law students. Terms of art are kept, with no explanation. Sentences run to 30 words.
- R3, public. For self-represented litigants, reporters and general readers. Terms of art are kept and explained the first time they appear. Sentences run to 25 words.
- R4, introductory. For readers with no legal background. Terms of art are kept, and each is followed by a description of what it means. Sentences run to 20 words.
Text written at the introductory level must carry a notice saying it is not equivalent to the source. The other three levels must not carry that notice, because at those levels the text is required to be equivalent.
How much of the standard you are claiming
There are three levels of conformance, and a document states which one it meets.
- Level 1, style. You followed the writing rules. Suitable for drafting from scratch, internal memos, and reference copy.
- Level 2, controlled. Level 1, plus the dictionary. Suitable for drafting text that will operate as law, and for commentary written in house.
- Level 3, verified. Level 2, plus the faithfulness rules and a completed check. Any published restatement of a real rule must reach Level 3. Nothing below it may be called a restatement or printed next to a source text.
How a restatement is checked
This is the part of the standard that does the real work, so it is worth setting out in full.
Start with a rule that says:
A party must file a notice of appeal within 30 days after entry of judgment.
Step one: list what the source actually says
Before writing anything, you break the rule into its separate claims. The standard calls each one a proposition. A claim is small enough when removing any part of it would drop something legally important.
The rule above makes several claims at once. Someone has to act, and that someone is a party. The action is filing a notice of appeal. It is required rather than allowed. There is a deadline of 30 days. The deadline runs from a specific event, the entry of judgment, not from the decision or from when anyone was told about it.
You write each claim on its own line, in a table called the ledger. For each one you record who acts, what word the source used to require it, what the action is, what conditions apply, and which part of the source it came from. If the source is genuinely unclear about something, or never says who has to act, you mark that too, rather than filling the gap yourself.
You do this from the source alone. Not from a summary, not from a treatise, and not from a draft you have already written. Reading someone else's version first anchors you to their reading of the rule.
Step two: write the restatement
Now you write, following the rules in Part 1, for whichever reader you named.
Step three: check forward, from the list to your text
Take each claim on your list and find the sentence in your restatement that carries it.
A claim with no matching sentence has been dropped. Suppose your restatement says "You have 30 days to appeal." The deadline survived. But the trigger, entry of judgment, is gone, and a reader now has no idea when the clock starts. So has the fact that a party must do it, and that a notice of appeal is the specific document required.
A claim that matches a sentence saying something different has been altered. "You have 30 days" is permission. "A party must file" is an obligation. Those are not the same, and the change is easy to miss because the sentence reads well.
Step four: check backward, from your text to the list
Now reverse it. Take each sentence you wrote and find the claim on the list that it came from.
A sentence with no matching claim is something you added. This is where the writer's own knowledge leaks in. You know that appeal deadlines are usually strict, so you write "This deadline cannot be extended." It may well be true. If the source did not say it, it does not belong in a restatement of the source.
This backward check is harder than the forward one, and it catches mistakes the forward check misses. The standard suggests doing it on a different day.
Step five: keep the record
You keep the ledger. It is the evidence that the checking actually happened, and it is where you start when the rule is amended. An amendment usually changes only a few claims, and the ledger tells you exactly which of your sentences those were.
Step six: have someone rebuild the source from your text alone
Steps three and four share a weakness. The person doing them is holding the source, so they know what the rule says. When they read your restatement, their memory quietly fills in anything you left out, and the gap becomes invisible.
So the last step takes the source away. You hand your restatement to a reader who has never seen the original, and ask them to write down what the rule must have said. Every duty, every deadline, every condition, every consequence.
Then you compare their version against the real rule. Anything they stated that the rule does not say, or anything the rule says that they could not recover, is a defect in your writing. It is not a failure on their part.
Note what this step proves and what it does not. If a reader rebuilds the rule correctly, your text carries enough. That is not the same as your text carrying only what the source says. Step four is what shows that.
Signing off
A second person, who did not write the text, reviews it against the source and the ledger, and reads the restatement last. Their job is not to improve the writing. It is to answer two questions. Does the restatement say anything the source does not? Does the source say anything the restatement does not?
That reviewer signs the record. A document also stores a fingerprint of the exact source text it was built from, so that when the source is amended, the mismatch is detected automatically and the document is marked unverified rather than quietly republished.
If you use an AI model to help
The standard allows it, with strict limits, and the reason is worth understanding. A checker who has already seen your draft will read the source looking to confirm the draft. Models do this even more readily than people.
So the model must not see your restatement while it is breaking down the source. It must not see the ledger while drafting. The forward and backward checks must be separate conversations that share no history. The blind rebuild must be a conversation that has never received the source at all, because a model that read the source earlier in a conversation is not blind to it, whatever you tell it. And a model's approval is never verification. A model that wrote a sentence will judge that sentence faithful. A person signs.
What the standard yields to
MLP-1 is the last authority, not the first. Where something else governs the text, the standard steps aside, in this order: a statute, court rule or code of construction; the drafting manual of the body publishing the text, some of which are required by law; the drafting manual of the jurisdiction; the institution's own style guide; and only then this standard. Where MLP-1 conflicts with any of those, the other one wins, and the writer records the conflict instead of quietly resolving it.
Tools
The repository includes a checking program, the ledger format in machine-readable form, and worked examples built from real court rules. The examples cover Georgia appellate practice, Federal Rule of Criminal Procedure 29, and a Tennessee problem that draws on three instruments that disagree with each other.
The program runs nineteen checks. Some are matters of style, such as sentence length. Others compare your ledger against your text, such as whether every number, citation and defined term in the source survived into the restatement.
It has one important limit. The program can only check the ledger you gave it. It never reads the source itself, so it cannot tell you about a claim you failed to write down in the first place. Breaking down a rule is human work, and so is the signature at the end.
Jurisdiction profiles
A few rules depend on where you are. The repository includes a template listing what a profile for a jurisdiction has to answer. No profile has been written yet. Each one is a research project, and an answer without a citation would be worse than no answer at all.
Getting the standard
The standard is available as a PDF, an ODT file, and Markdown, all from the repository. To propose a change, open an issue on the issue tracker or use the change form in Appendix C. It is released under a Creative Commons license that permits non-commercial use with attribution.
MLP-1 and Controlled Legal Writing are trademarks of Michigan Legal Publishing Ltd.