More on the relevance of frequency data: Responding to Steinberg

In a comment on one of Carissa Hessick’s posts about corpus linguistics at Prawfsblawg, Asher Steinberg expressed the view that relying on frequency data in deciding issues of ordinary meaning is misguided. (Steinberg blogs at The Narrowest Grounds, where he frequently writes intelligently about statutory interpretation.) Shortly after that, I posted Meaning in the framework of corpus linguistics here, in which I explained why I believe that frequency data can in fact be relevant in doing legal interpretation. And that post prompted a long comment by Steinberg, elaborating on his objection to using frequency data in legal interpretation.

Steinberg fears that if the courts were to draw on corpus linguistics in the way I that I advocate, statutory interpretation would “fall into fundamental error[.]”His point of departure is my analysis of the corpus data regarding the issue raised by Muscarello v. United States—whether driving somewhere with a gun in the trunk or glove compartment counts as carrying a firearm. (My conclusions are briefly summarized in the post Steinberg comments on; for the full analysis, see my forthcoming article A Lawyer’s Introduction to Meaning in the Framework of Corpus Linguistics (henceforth, A Lawyer’s Introduction)) Steinberg argues that frequency data—or at least the kind of frequency data that my analysis is based on— is inherently unreliable as evidence of ordinary meaning.

I beg to differ.

Steinberg starts with my finding that in the corpus data I reviewed, the pattern [human] carry [object] (as in A human was carrying an object) is not used as a way of denoting the carrying of an object in a vehicle “without the text or utterance containing some kind of overt indication to that effect.” That finding suggested to me that the pattern has a default meaning—that without some indication to the contrary in the text, the pattern is used and understood to refer to in which the object is being held by the person doing the carrying (who I’ll call the agent) in their hands or arms, is slung over their shoulder, etc. (I’ll call this type of carrying prototypical carrying)

But Steinberg demurs: “Well of course that’s what you found; you would have no way of concluding that the people in your corpora were carrying objects in vehicles unless a vehicle was mentioned or implied by context.” Judging from that statement, Steinberg accepts my conclusion about the default meaning of sentences such as A human was carrying an object. But he goes on to argue that the existence of a default meaning is irrelevant:

[S]uppose it were true that when people talk about carrying objects in vehicles they usually mention or strongly imply the vehicle….So what? They are usually describing particular instances of carrying in which the information that the object is in the vehicle or some particular part of it is relevant, not talking about carriage in the abstract, which the statute is and which your examples are not. That’s why they mention the vehicle, not because one has to mention a vehicle to negate a conventional meaning of on-person carriage, but because they want to make the fact and mode of vehicle carriage clear for some particular reason.

I demur from Steinberg’s demurral; I think that he’s the one whose argument is misdirected. Steinberg talks about the reasons that (he assumes) motivated the speakers and authors in question to phrase their statements utterances and texts in the ways that they did. But the treatment of ordinary meaning in legal interpretation is essentially reader-oriented rather than writer-oriented. In other words, it deals mainly with how the text is likely to be understood, rather than what the writer intended. Of course, the reader’s understanding may be influenced by what the reader knows or assumes about the writer’s intentions and motivations (see Comprehension, ordinary meaning, and linguistics), but even then, it seems to me, what matters it is the reader’s likely understanding of those intentions and motivations. However, when Steinberg talks about the speakers’ and authors’ motivations, he says nothing about how those motivations might influence how the language in question would be understood. As a result, he has stopped short of saying anything relevant.

THIS BRINGS US to Steinberg’s most consequential argument, namely that frequency data has nothing to contribute to legal interpretation—or at least nothing to contribute to deciding whether a given meaning represents a word’s ordinary meaning in the relevant context, as opposed to deciding whether a particular reading of the  text is reasonable. Steinberg’s basic point is that the frequency difference that is summarized in my previous post (and discussed in more detail in A Lawyer’s Introduction) reflects a difference in the frequency of occurrences in the real world, not a difference in meaning:

Maybe it’s true that when people talk about people carrying objects, more often than not the people carrying them have them on their person outside of a vehicle. Here too, so what? We simply have more occasion to talk about people carrying objects outside of vehicles than inside them. For one thing, people spend much more time outside of vehicles than they do in them; for another, there aren’t that many occasions on which we talk about the things we passively have with us in our cars. I think what I would be more interested in is: when we do talk about transporting things with us in our cars, is it unusual to say we carry them, or is that a perfectly normal way to talk about transporting things with you in a car? If so, I think transporting objects with you in a car is simply a subset, if not a frequently *discussed* one, of the acts encompassed by “human carrying object.”

It is certainly true that there are cases in which frequency data wouldn’t be relevant and in which relying on such data would  raise precisely the kind of concern that Steinberg raises here. Steinberg gives an example of such a case: if the corpus data does not reflect the use of the phrase carrying a lot of stuff to refer to carrying toasters, that wouldn’t mean that toasters can’t be referred to as stuff. But that example is extreme to the point of being specious. The word stuff is in effect a universal variable; it can stand for pretty much anything. And even if the corpus includes no examples of stuff being used to refer to toasters, a big enough corpus will contain enough examples stuff in use to establish its virtually limitless breadth.

A less extreme example was given in the first case ever in which a judge explicitly invoked corpus linguistics as an interpretive tool. The judge was Utah Supreme Court Justice Thomas Lee, and the invocation of corpus linguistics occurred in a separate opinion he wrote. None of the other justices joined that opinion, and the majority raised an objection to using corpus linguistics that is similar to Steinberg’s: “If the word ‘car’ is ten times more likely to co-occur with the word ‘red’ than with the word ‘purple,’ it would be ludicrous to conclude from this data that a purple car is not a ‘car.’” The majority was of course right about that, and on that narrow point, Justice Lee was in agreement: “A car’s purpleness does not detract from its carness any more than its redness does.”

In the hypotheticals I’ve just discussed, the difference in relative frequencies of what the words referred to (purple cars as opposed to red cars, toasters as opposed to other stuff) did not correspond to differences in what the words themselves are understood to mean. That is the reason that the issues raised by the hypotheticals wouldn’t be good candidates for using frequency data. But the fact that frequency analysis wouldn’t be useful in those cases doesn’t mean that it will never be appropriate. And it seems pretty clear that the issue in Muscarello is one for which frequency data is genuinely relevant.

THERE IS A VERY REAL SEMANTIC DIFFERENCE between, on the one hand, using the word carry to describe the conduct that the Court in Muscarello held to constitute carrying a firearm (driving a car with a gun in the trunk or glove compartment), and on the other hand, the kinds of activities that constitute prototypical carrying of a firearm (e.g., walking while holding the gun in your hand or tucked into your waistband). Although both kinds of activities involve movement by the agent while causing the weapon to move along with them, that doesn’t distinguish them from activities that we wouldn’t regard as carrying, such as dribbling a basketball or pushing a shopping cart.

This semantic difference manifests itself in the differing types of contexts in which the two meanings appear. I described this in A Lawyer’s Introduction:

With respect to the both the human carry object and the human carry ob­ject in vehicle categories, I looked to see how often the manner of carrying was ex­plicitly encoded. Each category’s results were virtually the mirror-image of the other’s. Within the human carry object category, the great majority of lines did not encode the manner in which the object was carried, leaving that fact to be inferred. The opposite was true of the category human carry object in vehicle; most of the uses in that category did explicitly encode the fact that the object was being carried in a vehicle, as in the following examples:

He dug out the flashlight he carried in the glove box and clicked it on.

… no reason to keep carrying stuff in our trunk

Two were charged with stripping parts from a parked car and three with carrying a sawed-off shotgun and drugs in a car, police said.

In the front seat, I’m carrying a gadget slightly larger than an electric shaver.

In all but one of the remaining lines, the manner-of-carrying information was partially encoded, by which I mean that there were explicit references to vehicles (under­lined in the examples below), which prompted inferences that the object was carried in that vehicle:

Shagren said what sparked the proclamation was concern over truck drivers carrying dairy products not being able to drive more than 12 hours a day…

He was thrown into the back of a deuce-and-a-half truck and carried south in a pile of wounded and dying men…

We returned with the truck to carry loot for the soldiers, mainly furniture.

Thus, when comparing the relative frequencies of the two kinds of uses, we are comparing the frequencies of distinct types of linguistic patterns, not merely of the kinds of events that are referred to. This is significant because one of the things that has been shown by work in corpus linguistics and corpus lexicography is that the different senses of a given word are often associated with distinct textual patterns. And in this case, the differing textual patterns are important indicators of meaning. That is shown by the fact that when carry appears without any contextual support for a vehicular-carry reading, it is understood to refer to prototypical carrying.

So the issue in Muscarello is different from those raised by hypotheticals involving toasters and purple cars. There is good reason to think that the frequency differences regarding carry correspond to differences in meaning, but no such reason exists regarding the hypotheticals.

(There is more that could be said as to why Muscarello is an appropriate case for using corpus analysis and why the hypotheticals are not, and more generally about how to figure out whether corpus analysis would or would not be appropriate in a given case. However, I will save those points for a later post.)

In any event, given the ways in which the issue in Muscarello differs from the issues in the two hypotheticals, I don’t think there is any need to figure out why there is a difference in the frequencies of the two meanings of carry. In trying to determine the ordinary meaning of language in a statute, the object (to state the obvious) is to identify the meaning that is ordinarily associated with the language that is at issue. Considering the relative frequencies of the potentially relevant meanings is unavoidably part of investigating how the linguistic pattern in question is ordinarily understood.

As I’ve said, the relevant pattern here consists of using the verb carry to denote (1) the carrying of a physical object (2) by a human being, (3) where there is nothing in the surrounding text giving any reason to think that the act of carrying consisted of driving a vehicle that contained the object. The corpus evidence I reviewed shows that this pattern is overwhelmingly associated with prototypical carrying. While one might argue over the weight that should be given to that evidence, the conclusion that the evidence is at least relevant to the issue of ordinary meaning seems to me to be unavoidable.

I’M NOT GOING TO DISCUSS everything Steinberg says, but I do want to talk about one additional issue that he raises. What interests him, he says, is not so much what people mean when they use the linguistic pattern that is at issue in Muscarello, but rather, this: “When we do talk about transporting things with us in our cars, is it unusual to say we carry them, or is that a perfectly normal way to talk about transporting things with you in a car?” And as to that question, he says, “If you can’t show me that it’s unusual to describe the act of driving around with a gun in the car as ‘carrying,’ you haven’t shown me, I don’t think, anything.”

While I don’t agree that that such an analysis is necessary, I do think that it would be interesting to find out the answer. However, I’d modify the question: rather than look simply at whether it is unusual “to describe the act of driving around with a gun in the car as ‘carrying,’” I’d want to find out whether it is unusual to use such a description without providing information sufficient to indicate the vehicular nature of the carrying.

But there is a problem: finding out the answer to either of these questions would probably be virtually impossible. You would have to start out with a sample of instances in which people drove a car or truck somewhere while having a gun somewhere in the vehicle. You’d then have to narrow that sample to instances in which someone made a statement referring to the gun being transported in the vehicle or being present in the vehicle while the vehicle was being driven. And then you’d need to know exactly what was said. Good luck with that.

Nevertheless, the corpus data makes it possible to undertake an analysis that is in some ways roughly analogous. That analysis takes advantage of the fact that for each of my corpus searches, I identified the concordance lines in which carry was used in referring to a physical object being transported in a vehicle, and assigned each such line to one of two categories, which together comprise the overall category I’ll call Carry in/by vehicle:

  • Human carry in vehicle: The object was referred to as having been carried by a human being (as in “He dug out the flashlight he carried in the glove box and clicked it on”).
  • Vehicle carry: The object was referred to as having been carried by the vehicle itself (as in “There is a distinct possibility that the truck was carrying a back-up tactical rocket”).

Starting from these categorizations, it is a simple matter to compute the percentage of concordance lines in the overall Carry in/by vehicle category that were attributable to the lines coded as Human carry in vehicle. And the results showed that overall, Human carry in vehicle accounted for less than a third of the Carry in/by vehicle lines:

  • No collocates specified: 7.7% (2 out of 26).
  • Vehicle-related collocates (car, truck, seat, backseat, frontseat, trunk, glovebox, glove, and compartment): 36.4% (44 out of 121).
  • Firearms-related collocates (gun(s), pistol(s), rifle(s), shotgun(s)): 33.3% (2 out of 6). Note that this result includes only those lines in which the object that was transported was a firearm.
  • Total: 31.4% (48 out of 153).

So to the extent that the corpus data can provide the kind of information that Steinberg wants to see, it suggests that it is somewhat unusual to refer to the transportation of objects in vehicles by using the pattern Human carry in vehicle. And although I don’t think that is especially relevant to the issue in Muscarello, it obviously doesn’t detract from the conclusions that I do think are relevant.


Leave a Reply