Regular expression for legal citation - regex

Question

How would you design a regular expression to capture a legal citation? Here is a paragraph that shows a two typical legal citations:
We have insisted on strict scrutiny in every context, even for
so-called “benign” racial classifications, such as race-conscious
university admissions policies, see Grutter v. Bollinger, 539 U.S.
306, 326 (2003), race-based preferences in government contracts, see
Adarand, supra, at 226, and race-based districting intended to improve
minority representation, see Shaw v. Reno, 509 U.S. 630, 650 (1993).
A citation will either be preceded by a comma and whitespace, a period and whitespace, or a "signal" such as "see" or "see, e.g.," and whitespace. I'm having trouble figuring out how to accurately specify the start of the citation.
I am most familiar with Perl regular expressions but can understand examples from other languages as well.

Answer 1

In your example, you've preceded the citations with what the BlueBook deems a 'signal' (Rule 1.2 on page 54 of the nineteenth edition). Other signals include but are not limited to : e.g., accord, also, cf., compare, and, with, contra, and but. These can be combined in surprising and unexpected ways . . . See also, e.g. Watts v. United States, 394 U.S. 705 (1969) (per curiam). Of course, there are also citations that are not preceded by signals
Then you'll also want to handle case citations with unexpected case names :
See v. Seattle, 387 U.S. 541 (1967)
Others have attacked this particular problem by first identifying the reporter reference (i.e. 387 U.S. 541) with a regular expression like (\d+)\s(.+?)\s(\d+) and then trying to expand the range from there. Case citations can be arbitrarily complex so this path is not without its own pitfalls. Reporter references can also take on some interesting forms as per BlueBook rules:
Jones v. Smith, _ F.3d _ (2011)
For decisions which are not yet published for instance. Of course, authors will use variations of the above including (but not limited to) --- F.3d ---

Answer 2

This certainly isn't perfect, but without more examples to test against it's the best I can think of. Thanks to #Paul H. for extra signal words to add.
#!/usr/bin/perl
$search_text = <<EOD;
"We have insisted on strict scrutiny in every context, even for so-called “benign” racial classifications, such as race-conscious university admissions policies, see Grutter v. Bollinger, 539 U.S. 306, 326 (2003), race-based preferences in government contracts, see Adarand, supra, at 226, and race-based districting intended to improve minority representation, see Shaw v. Reno, 509 U.S. 630, 650 (1993)."
In your example, you've preceded the citations with what the BlueBook deems a 'signal' (Rule 1.2 on page 54 of the nineteenth edition). Other signals include but are not limited to : e.g., accord, also, cf., compare, and, with, contra, and but. These can be combined in surprising and unexpected ways . . . See also, e.g. Watts v. United States, 394 U.S. 705 (1969) (per curiam). Of course, there are also citations that are not preceded by signals
Then you'll also want to handle case citations with unexpected case names :
See v. Seattle, 387 U.S. 541 (1967)
Others have attacked this particular problem by first identifying the reporter reference (i.e. 387 U.S. 541) with a regular expression like (\d+)\s(.+?)\s(\d+) and then trying to expand the range from there. Case citations can be arbitrarily complex so this path is not without its own pitfalls. Reporter references can also take on some interesting forms as per BlueBook rules:
EOD
while ($search_text =~ m/(\, |\. |\; )?(see(\,|\.|\;)? |e\.g\.(\,|\.|\;)? |accord(\,|\.|\;)? |also(\,|\.|\;)? |cf\.(\,|\.|\;)? |compare(\,|\.|\;)? |with(\,|\.|\;)? |contra(\,|\.|\;)? |but(\,|\.|\;)? )+(.{0,100}\d+ \(\d{4}\))/g) {
print "$12\n";
}
while ($search_text =~ m/[\n\t]+(.{0,100}\d+ \(\d{4}\))/ig) {
print "$1\n";
}
Output is:
Grutter v. Bollinger, 539 U.S. 306, 326 (2003)
Shaw v. Reno, 509 U.S. 630, 650 (1993)
Watts v. United States, 394 U.S. 705 (1969)
See v. Seattle, 387 U.S. 541 (1967)

Answer 3

Well you can use the following at the start. You will need more patterns for other starts.
/(, )|(see )/
The end will prove to be the bigger problem. For example in "see Adarand, supra, at 226, and race-based..." there is no clear end indicator. I suspect pure regular expressions won't be sufficient for this task, you need a higher form of language analysis. Or be content with matching only a subset of all citations, or matching too much sometimes.

Answer 4

I'm using http://gskinner.com/RegExr/ to test this syntax
(?<=see )\w+ v. \w+, \d{3} U\.S\. \d{3}, \d{3} \(\d{4}\)
As you can see, i'm using 'Positive lookbehind'

Answer 5

I've written a pattern (created for JavaScript since you didn't specify a language) which can be used to match the two citations you mention:
var regex = /(\w+\sv.\s\w+,\s\d*\s?[\w.]*[\d,\s]*\(\d{4}\))/ig;
You can see it in action here.
It will match others as long as they follow the same format of:
Name v. Name, 999 A.A.A... 999, 999 (1999)
Although presence of some parts is made optional. Please provide more information on citations that may not fit this pattern if it doesn't appear to meet your needs.

Answer 6

For this kind of potentially complex regexes, I tend to break it down into simple pieces that can be individually unit-tested and evolved.
I use REL, a DSL (in Scala) that allows you to reassemble and reuse your regex pieces. This way, you can define your regex like this:
val NAME = α+
val VS = """v\.?""" ~ """s\.?""".?
val CASE = NAME("name1") ~ " " ~ VS ~ " " ~ NAME("name2")
val NUM = ß ~ (δ+) ~ ß
val US = """U\.? ?S\.? """
val YEAR = ( ("1[7-9]" | "20") ~ δ{2} )("year")
val REF = CASE ~ ", " ~ // "Doe v. Smith, "
(NUM ~ " ").? ~ // "123 " (optional)
US ~ NUM ~ // "U.S. 456"
(", " ~ NUM).* ~ // ", 678" 0 to N times
" \\(" ~ YEAR ~ "\\)" // "(1999)"
Then test every bit like this:
"NUM" should {
"Match 1+ digits" in {
"1" must be matching(NUM)
"12" must be matching(NUM)
"123" must be matching(NUM)
"1234" must be matching(NUM)
"12345" must be matching(NUM)
"123456" must be matching(NUM)
}
"Match only standalone digits" in {
NUM.findFirstIn(" 123 ") must beSome("123")
NUM.findFirstIn(" n123 ") must beNone
}
}
Also, your unit/spec tests can double as your doc for this bit of regex, indicating what is matched and what is not (which tends to be important with regexes).
I made a gist for this example with a first naive implementation.
In the upcoming version of REL (0.3), you will be directly able to export the Regex in, say, PCRE flavor to use it independently… For now only JavaScript and .NET translations are implemented, so you can just run the sample using SBT and it will output a Java-flavored regex (although pretty simple, I think it can be copy/pasted to Perl).

Answer 7

I'm not overly familiar with perl, but if I wanted to do this, I would use some web lookups. First, I would find a good set of patterns.
I went with this regex:
(\d{3})\sU\.S\.\s(\d{3})
Breakdown of regex:
(\d{3}) --> looks for 3 numbers, places them into $1
\sU.S.\s --> looks for a whitespace followed by U.S. followed by another whitespace
(\d{3}) --> looks for 3 numbers again, placing them into $2
What this does, is looks for the pattern 539 U.S. 306 and places them into captures groups. This puts the following values into variables:
$1 = 539
$2 = 306
I would loop through, and find each instance of the pattern, then I would use something to grab this site from the web:
http://supreme.justia.com/cases/federal/us/$1/$2/case.html
Which in this case would become:
http://supreme.justia.com/cases/federal/us/539/306/case.html
Once I had this, I could go through the site tree for the following (I put the whole tree here, since depending on language the way you do this might be changed):
<body>
<div id="main">
<div id="container">
<div id="maincontent">
<h1> HERE IS THE TITLE OF THE CASE </h1>
The xpath of this is //*[#id="maincontent"]/h1.
From here you now have the full reference:
Grutter v. Bollinger - 539 U.S. 306 (2003)
I'm not a legal expert, so I don't know if there are other ways they can be declared (one of the other answers mentioned something like F.3d) then another approach would be needed to capture that. If I get some time later, I might write this out in PowerShell just to see how I would do it.

Answer 8

I manage the CourtListener.com project, which extracts millions of citations from opinions. The task at hand is a hard one. I wouldn't rely on signals like "see" and "also".
What we've done instead is create a JSON object containing all of the possible reporter abbreviations (it's about 10k lines long), which we store in our reporter-db repo:
https://pypi.org/project/reporters-db/1.0.12/
That's in Python, but the JSON object is in the source repo as well (there's a CSV if you like too).
Using the JSON object, we build up a regex that knows what all the reporter abbreviations are.
For your example, that'd be U.S., but we have hundreds more including misspellings and typos, and the regex is essentially:
\d{1,3} (REPORTER|REPORTER2|REPORTER3|...) \d{1,3}
It gets complicated because some page numbers use roman numerals just to mess with you and there's a half dozen other corner cases to think about.
So...you've kind of asked a really big question, if you want to do it well. The good news is that in addition to the reporters-db Python project and JSON object, we also open sourced our citation finder. It's not a standalone project unfortunately, but you can see how it works and use it as a starting place:
https://github.com/freelawproject/courtlistener
Look in cl/citations for the related code.
I hope this helps! We're a non-profit dedicated to these kinds of problems, so definitely reach out if you have questions or ideas.