A new, dubious linguistic tool joins the debate.
In Heller and McDonald, two cases decided more than a decade ago, the Supreme Court held that the Second Amendment protects an individual right to own a gun. Neither case presented the issue of carrying a gun — but in the course of interpreting the amendment for his Heller opinion, the late Antonin Scalia laid out his thoughts. As the phrase is used in the Second Amendment, he wrote, to “bear arms” is to carry weapons in case of confrontation.
A new case, New York State Pistol & Rifle Association v. Bruen, seeks to cash that check. It’s a challenge to New York’s requirement that applicants for concealed-carry licenses show “good cause,” or a non-speculative need to carry a gun for self-defense, with local officials in charge of making the determination. Especially in the state’s most left-leaning areas, it’s incredibly difficult for the average person to get a permit.
But some academics are claiming, based on a new type of research called “corpus analysis,” to have found fresh evidence that Scalia was wrong. “Bearing arms,” they say, overwhelmingly referred to military service in the Founding era, rather than simply referring to the carrying of weapons. These scholars further suggest that Heller might have been wrong more deeply — even when it comes to keeping arms — though that is not the focus of the current case (and their “bear arms” argument is generally recognized as their strongest).
In corpus analysis, researchers start with a massive collection of writings or speech from a given time period (a “corpus”) and search it for uses of a contested term. They can look at how the term was used in context and count up how many times it was used in various ways. The potential appeal to constitutional originalists, who want laws to be interpreted by their “original public meaning” — the way an ordinary person would have read them at the time they were enacted — is obvious. This is, after all, a look at how people actually used words across a wide range of contexts.
So should the originalists on the Court reverse course, holding that there’s no right to carry a gun outside of the militia, and open the door to a broader reconsideration of Heller? Not unless they were already on the fence.
This type of analysis can be useful, perhaps even convincing in some circumstances. But despite its superficial appeal, it suffers from severe limitations that render it no more compelling than any other approach to constitutional interpretation. It must be treated skeptically, interpreted with care, and viewed as one small piece of the puzzle alongside other forms of evidence — the evidence that the Court already considered in the previous cases.
General Issues with Corpus Analysis
When a court interprets a tricky term, the question is often whether the term’s “ordinary” meaning is broad enough to cover some disputed conduct. But corpus analysis doesn’t tell us that. What it tells us, instead, is how frequently people actually use the term to refer to the conduct in question, relative to how often they use the term in other ways. This number is not necessarily all that informative.
As Josh Blackman and James C. Phillips once noted, for example, if you checked a corpus for uses of the verb “to read,” you’d probably find far more examples of people reading newspapers than of people reading street signs — but the low frequency of street-sign references would tell you nothing of importance. It would merely reveal that people read newspapers more than they read street signs, or at least more frequently discuss reading newspapers. A law forbidding people “to read,” or a constitutional provision giving them a right to, would presumably apply to street signs (and nutritional labels and ancient scrolls) just as much as it applied to newspapers.
Even if you fail to find any examples of a disputed use, it’s not dispositive. Some modern corpora don’t contain a single example of the word “vehicle” being used to refer to an airplane, for instance.
So when a corpus analysis reveals that a term was used in a disputed sense X percent of the time, that means . . . what, exactly? At what cutoff can we confirm or deny that the disputed use is part of the ordinary meaning? The appropriate thresholds will vary from case to case, often falling all the way to zero, and reasonable people will disagree as to where they should be. How often a term is used to describe a given situation depends not only on the term’s meaning, but also on how common or newsworthy the situation is, as well as the alternative words that speakers have at their disposal.
This problem is compounded by the fact that corpus analysis systematically produces narrow senses of what a term means, as a paper in the Harvard Law Review last year by Kevin P. Tobia showed.
In a series of experiments run on ordinary people, law students, and judges, Tobia divided his subjects into three groups. One group was simply given a term (say, “vehicle”) and asked whether various things counted as part of the category (cars, airplanes, shoulder baby carriers, etc.), to gauge the scope of the term’s ordinary meaning. The other two groups, by contrast, were not given the actual word, so they wouldn’t rely on their preexisting sense of what the word meant. Instead, they were given a fake word (such as “ailac”), accompanied by either (A) a dictionary definition of the real word Tobia was testing or (B) some results from a corpus analysis of that word, including examples of the word used in context and other terms the word was frequently used alongside.
The upshot of these experiments is that corpus analysis tends to highlight a narrow subset of a term’s potential uses. Those relying on corpus results, for example, tended to say that an airplane does not count as a vehicle (or rather, as an “ailac”), contrary to the folks who were asked to rely on their own sense of the word. Originalist fans of dictionary definitions shouldn’t take any comfort in the study, either: Those relying on dictionary definitions tended to have an especially broad sense of the tested word, saying, for instance, that a shoulder baby carrier is a vehicle.
In general, Tobia writes, corpus analysis points users to “prototypical” uses of a given word, while dictionaries provide definitions broad enough to cover uses that qualify only “technically.” Certainly, it can be useful for a judge to see a range of plausible options: Sometimes context will make clear that the word is being used in a narrow or broad way, for instance, and under the “rule of lenity,” ambiguous laws are supposed to be read in whatever way favors a criminal defendant. But corpus analysis does not reveal some hidden, true ordinary meaning of words that we’ve lacked until now; it just highlights some typical uses.
To this point, you may notice, I’ve focused on problems with interpreting the results of a corpus analysis, implicitly assuming that the analysis itself is sound. But lots can also go wrong in the process that culminates in a computer spitting out results and a researcher counting them up.
Obviously, no corpus is comprehensive. Those from the Founding era necessarily include only what was written down, as recording technology did not exist, and are biased in favor of the elite, male, and white Americans whose writings are most likely to have survived. In addition, corpus analysis tends to give all uses of a term equal weight, rather than giving special consideration to the most on-point examples. If we’re analyzing a constitutional provision, for instance, we might want to pay extra attention to uses of the term in that particular context.
And any attempt to “query” these databases and “code” the results requires a series of judgment calls. Researchers need to pick out which sets of words they’re looking for, which can include numerous variants of the term used in the law they’re studying, and they have the option of searching for examples of a word only when it’s used in close proximity to others (as a way of looking within a specific context, or of finding variants of multiple-word terms). They can also exclude uses they think are irrelevant to the question they’re trying to answer. Then they have to decide which of the results they kept count as examples of the disputed use, and which don’t.
To be fair, researchers can make all of these decisions in a very transparent manner, publishing their exact queries, results, coding, and so on. But this is all an art, not a science.
Interpreting the Second Amendment in Particular
With all that in mind, let’s review what corpus analysis has actually revealed about the term “bear arms.”
Several analyses suggest that “bear arms” was used overwhelmingly — though not exclusively — in the military context in and around the Founding era. In a 2018 Washington Post article, one linguist announced that only a “handful” out of 1,500 uses he’d found “don’t refer to war, soldiering or organized, armed action.” A 2019 article in BYU Law Review concluded that just 4 percent of uses were of the “non-militia/private use” variety. Yet another 2019 article, published on the SSRN platform, found that “the corpus data for bear arms was overwhelmingly dominated by uses of the phrase in its idiomatic military sense.”
For the reasons noted above, it’s not even clear how devastating or consequential we’re supposed to find this. No one ever denied that “bearing arms” frequently had a military connotation, and at any rate, the Second Amendment itself contains that pesky reference to the “militia.” The question is whether the term, as used, is limited to the military sense. Scalia even discussed this issue briefly in Heller itself, though he was more dismissive of some empirical findings than, in retrospect, he should have been:
Justice Stevens points to a study by amici supposedly showing that the phrase “bear arms” was most frequently [i.e., in 110 out of 115 examples] used in the military context. . . . Of course, as we have said, the fact that the phrase was commonly used in a particular context does not show that it is limited to that context, and, in any event, we have given many sources where the phrase was used in nonmilitary contexts.
Other results reveal additional nuance. A paper by Josh Jones, for example, points out that some earlier research excluded all uses of “bear arms” in contexts directly related to the Second Amendment, which are actually the most relevant uses. Jones further notes that the preponderance of military uses isn’t surprising in a violent era such as the Founding and that when the term is used in the military context, it can still be used literally: to refer to carrying weapons, rather than as an idiom that denotes military activity in general. Jones provides several examples where “bearing arms” was discussed as just one of several aspects of military service. When he includes all uses of the term and divides them into literal vs. idiomatic senses, he finds that between one-fifth and two-fifths of uses were literal.
Phillips and Blackman have also previewed some results of a corpus analysis, though their final paper is still forthcoming. On the overall question of who got Heller right, they are “more convinced by Scalia’s majority opinion than Stevens’s dissent.” They do find that about 90 percent of uses of “bear arms” and its variants were military in nature, but they rightly caution that “whether these results show that the Second Amendment language precludes an individual right is a more complicated question.” They further point out that when the word “arms” was used in the key context of rights, individual uses were hardly rare: “About 40 percent of the results had a militia sense, about 25 percent used an individual sense, and about 30 percent referred to both militia and individual senses.”
To go even further into the relevant context, one can focus specifically on Second Amendment analogues — bearing-arms protections in state declarations of rights, draft versions of the Second Amendment itself, etc. Some of these provisions protected the people’s right to bear arms in defense of themselves and the state; one proposal mentioned “the purpose of killing game”; and even more restrictive formulations, such as “bear arms for the common defense,” imply that the term could have a broader meaning when not so restricted. A brief in the current Supreme Court case by longtime Second Amendment lawyer David T. Hardy argues that “whatever the most common use of ‘bear arms’ was in ordinary speech, in constitutional enactments, ‘bear arms’ was universally understood as protecting defensive carry of ordinary arms.”
But once we get to that point, we’re back where we started, with the extensive record that informed the decisions and dissents in Heller and McDonald. As those cases (and the copious briefs in the current case) make obvious, there’s a whole lot of history there that doesn’t take a corpus to analyze. This starts with the English rights that formed the backdrop to American law. It includes those analogues to the Second Amendment. It runs through the plain language of the amendment itself and contemporary dictionaries’ definitions of “bear.” It continues through court decisions and other legal writings in the early decades of the country, and through events that unfolded well after the Founding: It was the 14th Amendment, enacted after the Civil War, that applied federal constitutional protections to the states, and black Americans’ right to bear arms was frequently discussed in that context.
I won’t wade through all of that again here; there’s no point when so many others on both sides of the debate already have. I will just say that if you studied this issue ten years ago and came to have a strong opinion, the results of the emerging field of corpus analysis should at most mildly color your view of the topic.