UP UNTIL NOW, the use of corpus linguistics in legal interpretation has gotten almost entirely good press—probably because almost all the press it’s gotten has come from its advocates. That situation has now changed, though, with the posting on SSRN of a paper by UNC law professor Carissa Hessick, who was one of the participants at the BYU law-and-corpus-linguistics symposium this past February. (Hessick has blogged about her paper at Prawfsblawg, here and here.)
The paper, “Corpus Linguistics and the Criminal Law” (pdf), argues that corpus linguistics “is not an appropriate tool” for interpreting statutes. Although it deals specifically with using corpus linguistics in interpreting criminal statutes, and Hessick’s concerns may not be as strong as to other areas of the law, much of her criticism would apply across the board. In this post I am going to discuss some of the issues that the paper raises, and if you’ve followed this blog before, you won’t be surprised to find out that I disagree with Hessick’s conclusion.
HESSICK’S PRIMARY OBJECTION to using corpus linguistics in legal interpretation has to do with how corpus linguistics relates to interpretive theory. In her paper and the first of her Prawfsblawg posts, Hessick regards the law-and-corpus-linguistics “movement” (or should I say “the corpus-linguistic turn in legal interpretation”?) as an effort to radically remake legal interpretation. She situates corpus linguistics within textualism, and therefore associates it with the view that if the statute’s ordinary meaning is clear, nontextual sources should not be considered. And she sees corpus linguistics as reframing, and rigidly constraining, the ordinary-meaning analysis. Hessick attributed to the advocates of corpus linguistics the view (1) that determining ordinary meaning is entirely “an empirical question” that should be resolved solely on the basis of objective data, (2) that the only relevant data is data about how frequently a particular sense of a word appears in the corpus in relation to the word’s other senses, and (3) judges should not draw on their own linguistic intuitions or otherwise exercise their own judgment.
Of the fairly small group of people who are actively interested in using corpus linguistics in legal interpretation, I’m not aware of anyone who holds that set of views. I certainly don’t. And in her more recent blog post, which responds to some of the comments on the first post, Hessick acknowledges that not everyone who supports the use of corpus linguistics subscribes to the views she outlines. She now sees “a rather stark divide in the corpus linguistics community[.]” On one side are those who “are pushing corpus linguistics analysis as a new method of statutory interpretation” (I think approach would be a more appropriate word than method), and on the other are those who “are far more modest about what corpus linguistics can provide.” She puts Brian Slocum and me in the latter category, and Stephen Mouritsen in the former. And although Hessick notes Mouritsen’s comments on the original post, which expressed views that I read as similar to mine, she thinks those comments are in tension with things he had said previously. I’m dubious about that, but I’ll let Mouritsen speak for himself. [UPDATE: He has now done so. Mouritsen has posted a substantial response to Hessick’s second Prawfs post, in the comments to that post. Click here and scroll all the way down.]
In any event, there is no reason to think that the views that Hessick criticizes are somehow inherent in what it means to think like a corpus linguist. I see no inconsistency between believing, on the one hand, that corpus linguistics can be a useful tool in legal interpretation and, on the other hand, that corpus data isn’t necessarily conclusive on issues of either ordinary meaning or legal meaning. As to issues of ordinary meaning, the impact of corpus data is likely to vary from one case to the next, ranging from unimportant at one extreme to damn-near conclusive at the other. And as to the issue of legal meaning, corpus-linguistic methodology can be used by adherents of all interpretive approaches that I’m aware of—intentionalism and purposivism just as much as textualism and originalism. Under all of those approaches, ordinary meaning is relevant, even if not dispositive.
UNDERLYING HESSICK’S OBJECTION to the version of law-and-corpus-linguistics that she ascribes to Mouritsen is a perspective on some purely linguistic issues relating to the use of corpus linguistics. Hessick argues that frequency data is not an adequate basis for determining ordinary meaning—an objection that, if valid, would apply to the use of corpus linguistics in legal interpretation generally, not just in criminal law.
Hessick contends that “the frequency with which a [given sense of a] term is used does not answer the question” of “how an ordinary member of the public would understand the statute[.]” The frequency with which a particular sense of a word appears, she says, “is a separate inquiry from how members of the public would understand that term if used in a statute.”
Despite the seeming clarity of those statements, I think they leave a good deal of uncertainty about the breadth of Hessick’s argument. When she says that frequency data “does not answer the question” of how the statute would be understood by an ordinary person and that frequency “is a separate inquiry” from how we would expect the statute to be understood, does she mean only that such data cannot by itself conclusively resolve those issues? Or is she making the stronger claim that frequency data isn’t even relevant? Or maybe something in between those two poles; for example, that frequency data isn’t necessarily relevant, but might be useful in some cases? Given Hessick’s response to a question I asked in the comments to her second post, it may be that she doesn’t rule out the possibility that corpus analysis could be useful in some cases.
However, I’m less interested in how broadly Hessick’s objection would apply than I am in her understanding of the relationship between frequency data on the one hand and determinations of ordinary meaning on the other. Or maybe I should say her understanding of how advocates of using corpus linguistics see that relationship; I’m not entirely sure.
Hessick says, “Corpus linguists believe that the frequency with which [particular senses of] words are used in various non-legal publications can tell us how a member of the public would understand those words when they appear in statutes.” I think that’s broadly correct; although there are some details I would take issue with, I agree that the ultimate goal is to determine how members of the public would be likely to understand the statute’s language. And although I’m hesitant to speak for anyone else, my guess would be that most if not all of those who support using corpus linguistics would agree.
However, I think Hessick is going in the wrong direction when she gets more specific about what is involved in the inquiry into how the statute would be understood. She appears to regard that inquiry as being comparable to asking members of the public what they think the statute means, as if they were making the same kind of ordinary-meaning determination as a judge would make in interpreting the statute.
For example, she says, “It is true that judges will not always be able to accurately determine what the majority of Americans believe is the linguistic prototype of a statutory term.” Hessick had previously described the word prototype as the term used by linguists to refer to “how an ordinary citizen would understand [a] word[.]” That’s not really correct as a matter of linguistic terminology, but I’m not going to get into that here. What matters for purposes of this discussion is that she describes the relevant inquiry as focusing on what people “believe” about the statute’s meaning. That suggests to me a situation in which those people are in effect engaging in the kind of analysis that is part of statutory interpretation, rather than merely reading and comprehending the statute.
You’ll note that I am distinguishing between interpretation and comprehension. That’s something that I have discussed in previous posts (here, here, and here (under the heading “Interpretive rules”)), and the reason for the distinction is that the words interpret and interpretation are polysemous. They can be used to refer (1) to the conscious process of deliberation that underlies legal interpretation, and (2) to the automatic and effortless cognitive processes that underlie the comprehension of utterances and texts. Those are different processes, and I want to avoid the ambiguity that would result from calling them both by the same name. So I use interpret and interpretation only with respect to legal interpretation, and comprehend and comprehension with respect to the cognitive processes by which utterances and texts are understood.
I will explain why this distinction is important, but first I want to call your attention to language from Hessick’s paper that I read as confirming my understanding that she is talking about interpretation rather than comprehension. The language that I’m referring to is part of Hessick’s argument that using corpus linguistics is inappropriate because “[m]embers of the public can hardly be expected to perform their own corpus searches and analyses.” She notes that a possible response to this concern would be to say “that members of the general public need not conduct their own corpus analyses because the frequency of usage reflected in the corpus database is likely to correspond to their own intuitions about the meaning of language.” But she sees such a response as being “at odds with a fundamental premise of corpus linguistics—that judicial intuition ought to be replaced with corpus analyses precisely because that intuition is unreliable.” As shown by the footnote to that statement (citing Mouritsen’s article Hard Cases and Hard Data, pages 175-78), the reference to “intuition” is shorthand for the use by judges of their own linguistic intuitions—their sprachgefühl—in determining an expression’s ordinary meaning. That falls on the “interpretation” side of the distinction between interpretation and comprehension. And I assume that when Hessick talks about laymen’s “intuitions about the meaning of language,” she is using intuition in the same way.
In the portion of his article that Hessick cites, Mouritsen explain why he thinks (as I do) that the kind of intuition Hessick refers to is unreliable: “[J]udges are subject to the same linguistic limitations as the rest of us, which limitations include the inability to intuit which features of the language are common or ordinary and which are unusual. A great deal of information regarding language use, including information about one’s own language use, is ‘not susceptible to recovery via introspection.’” [quoting this book] The information that Mouritsen is referring to is the tacit knowledge that enables us to understand and use language. That knowledge plays a part in the process of comprehension, which happens below the level of conscious awareness. When we hear an utterance or read a text, we understand it effortlessly, with the same apparent lack of intervening thought that characterizes our ability to see in 3D even though the images on our retinas are in 2D. In both cases, we feel like we are experiencing unmediated reality, but the fact is that both 3D vision and understanding language depend on a lot of subsurface cognitive processing.
As Mouritsen notes, one kind of information that we can’t consciously access is the kind of frequency data that we are talking about here. Of course, we have access to such information at the extreme ends of the bell curve: it’s a safe bet that carry in all of its grammatical forms occurs more frequently than deontology (at least outside a philosophy department). But at a more fine-grained level, we aren’t able to reliably make judgments of relative frequency.
For example an earlier article by Mouritsen discusses corpus data showing something that many people would probably find to be surprising: more than half the uses of carry did not involve the action of physically carrying an object. Instead, they involved uses in idioms and phrasal verbs (carry on, get carried away, etc.) and nonphysical uses such as these:
The term gender while often used synonymously for women did not carry the same political weight.
The Golf Channel now will carry early rounds and some tournaments.
Our legacy copper wiring just can’t carry the data to support HD-video streaming.
(I’ve independently done my own corpus analysis of carry, and the results were consistent with Mouritsen’s.)
What we can see with regard to carry is not an isolated example. One of the things that corpus linguistics has shown is that what we regard as the most basic meaning of a word is not necessarily the most frequent. Consider this statement by John Sinclair, who was a pioneer in the use of corpus data to study word meaning and who was editor-in-chief of the first dictionary based entirely on corpus data: “[W]e think of verbs like see, give, keep, as having each a basic meaning; we would probably expect those meanings to be the commonest. However, the database tells us that see is commonest in uses like I see, you see, give in uses like give a talk and keep in uses like keep warm.” (From Looking Up: An account of the COBUILD Project in lexical computing (1987)) And more broadly, Patrick Hanks, who is probably the most important lexicographer of the past 50 years, had this to say about the impact of corpora on the field: “At last lexicographers have sufficient evidence to make the generalizations that they need to make with reasonable confidence. We can now see that pre-corpus lexicography was little more than a series of stabs in the dark….” (link)
So when Hessick says that the there’s no reason to think that intuitions of members of the public are more reliable than those of judges, she’s right. Given the kind of intuitions she is talking about, both groups’ intuitions are equally bad. But that doesn’t undercut the case for using corpus linguistics (on the contrary, it bolsters it), because, here’s the thing: nobody is suggesting that corpus linguistics be used for investigating those kinds of intuitions. The whole purpose of using corpus linguistics is to learn things that aren’t accessible to intuition.
* * *
My intention when I began writing this post was for it to include, in addition to my response to Hessick, a discussion making the affirmative case for the relevance of frequency data in determining ordinary meaning. However, the post is long enough as it is, so I will save that for later. Stay tuned.
I wonder if a better way of thinking about the use of corpora in non-linguistics contexts is as a heuristic rather than method or approach. And it is a heuristic for checking pseudo-intuitions about language use. The sort of things that feel like intuitions but are in fact empirical statements based on made up data.
Your example with ‘carry’ is a case in point. It does not really say anything about what ‘carry’ means but about what it does NOT mean in most of its contexts of usage. I’ve made a similar analysis of ‘literally’ here for a bit of fun: http://metaphorhacker.net/2011/02/literally-triumph-of-pet-peeve-over-matter.
But in fact, there is a more generalized heuristic of pseudo intuition checking for which corpus is only one tool. Not everything can be easily searched for in a corpus such as figurative language. This is what I wrote about how metonymy can be confused for real statements: http://metaphorhacker.net/2013/12/binders-full-of-women-with-mighty-pens-what-is-metonymy. A more traditional linguistic analysis made the point just as easily as a corpus, here.
I do have some sympathy for Hessick’s 3 points, though (as a complete outsider to law). I have seen similar sentiments underlying many discussions in corpus-based sociolinguistics at conferences, although everybody would deny making such claims explicitly. It is a sort of rhetorical plausible deniability and I could easily see how a data-based approach could turn into a sort of mechanical positivistic frequentism down the line if the legal profession is not careful in the same way other quantitative approaches such as forensics have become at its extremes. (CSI as Corpus Science Investigations …).
Thanks so much for blogging about my paper and more generally for engaging with my about these ideas. I know that you are still working on a post that will talk about why frequency analysis is relevant for ordinary meaning. But in the meantime, I wanted to ask you about this paragraph:
“However, I think Hessick is going in the wrong direction when she gets more specific about what is involved in the inquiry into how the statute would be understood. She appears to regard that inquiry as being comparable to asking members of the public what they think the statute means, as if they were making the same kind of ordinary-meaning determination as a judge would make in interpreting the statute.”
Leaving aside what the ordinary/plain meaning inquiry requires from textualists, I’m having a hard time understanding why, as a matter of due process, asking members of the public what they think a statute means is “going in the wrong direction.” We obviously can’t conduct those sorts of polls for all statutes or during litigation (or at least, we don’t). But it seems to me that the answer to that question maps nicely onto the Court’s discussions about the notice that is required by the Due Process Clauses.