« Mila Sohoni: The Power to Vacate a Rule
Michael Ramsey
| Main | Originalism and Secession
Michael Ramsey »


A Response to Ryan Nees on Textualism and Title VII [Updated] [Further Updated]
James Phillips

[Ed.: James Phillips is a Nonresident fellow with the Constitutional Law Center at the Stanford Law School.]

My piece shows that this is a mistake—that the textualist arguments for the plaintiffs in the Title VII cases actually overlook a plainly linguistic, and thus textualist, consideration: the principle of compositionality.  I explain and then apply this principle to the linguistic data available in both dictionaries and the largest available database of English texts from the time of Title VII’s enactment (the Corpus of Historical American English, or “COHA”). 

Doing so shows that on the most faithful reading of the text of Title VII, its ban on sex discrimination applies only to differential treatment based on “unfair beliefs or attitudes directed at some or all men in particular, or at some or all women in particular—whether the beliefs be outright misconceptions or just unduly rough or weak generalizations; and whether the attitudes be indifference, discounting of interests, distaste, or outright antipathy” (6). (I also show that this reading fits well with the Supreme Court’s precedents on Title VII.) The plaintiffs’ reading of Title VII contradicts this reading, so it “produces a demonstrably inferior reading of the text on purely linguistic grounds, prior to any appeal to subjective intent, purpose, policy concerns, or other modes of legal argument” (3).

Rising Stanford 3L Ryan Nees has posted a thoughtful reply to my essay on the American Constitution Society’s blog. I write with a few reactions. Given the limited nature of a blog post, I won’t address every concern I have with Mr. Nees’s post, but just focus on some points related to corpus analysis and the statute’s BFOQs.

Mr. Nees alleges that I have “effectively incorporate[d] purpose into the very meaning of the words of the statute.” But nowhere does my analysis delve into congressional intent or analyze the type of evidence one would to divine such (legislative history, etc.). Instead, it merely takes the operative words at the statute and looks at how they were used by ordinary Americans and defined at the time Title VII was enacted. It’s hard to think of a more textualist way to approach interpretation. If that’s purposivism, then everything is.

As to my reliance on the uses of “discriminate against” in the database of English texts, Mr. Nees contends that my analysis goes awry because the phrase “discriminate against” appears only 125 times out of the 48 million words in the databases’ texts from the 1950s and 60s. But here Mr. Nees is focused on the wrong denominator. If we’re trying to see if “discriminate” had taken on the kind of pejorative sense described in my piece, the absolute number of times that the phrase “discriminate against” appears is irrelevant. The question is how often “discriminate” is paired with “against” as opposed to other words. And it turns out that this pairing occurs over half the time the word discriminate is used and is five times more likely the next most frequent pairing.

More to the point, my submission is not just that “discriminate against” should be analyzed under the notion of compositionality (a notion Mr. Nees seems to confuse with the related notion of an idiom), but that it’s the whole operative language of Title VII that should be examined, to the extent possible. That language is “discriminate against any individual . . . because of such individual’s  . . . sex” So Mr. Nees’s ability to find cases where “discriminate against” did not connote prejudice, when used in reference to things like commerce, has no bearing on the meaning of “discriminat[ion] against” someone because of a suspect classification or characteristic. So Mr. Nees’s focus on “discriminate against” in isolation from a protected classification falls into the same error as the petitioners—trying to slice and dice the language contrary to the principle of compositionality.

In other words, no one denies that“discriminate against” can be used in other contexts involving no prejudice. But in focusing on this point, Mr. Nees’s analysis overlooks the usefulness of a corpus: the ability to drill down on context. Title VII doesn’t prohibit discriminating against just anything. It prohibits discriminating against an individual because of that individual’s enumerated group characteristic. So that is the only context that matters. Likewise, if I’m trying to understand what it means to “read a book,” it would make little sense to look in a corpus at instances of “reading the sky” or “reading body language.” It’s just not the relevant linguistic context.

Mr. Nees also turns to a different corpus—one of Supreme Court opinions—and argues that I overlook the possibility that “the phrase’s usage in the legal context may have differed from popular usage.” But no one in the briefing or arguments has ever argued that Title VII is using a legal term of art. Besides, if we were to pursue this suggestion seriously, looking at a corpus of Supreme Court opinions alone wouldn’t do the trick. We would need to look at a corpus of legislation and other legal materials as well.

Moreover, Mr. Nees’s findings about the corpus of Supreme Court opinions don’t get him where he appears to want to go. He notes that “discriminate against” appears three times as often in the Supreme Court corpus than in COHA. The fact that something appears more often in a legal corpus than an ordinary corpus does not mean that it has a specialized legal meaning. Take the noun “court.” In COHA, the word appears 237 times per million words. In the Supreme Court corpus Mr. Nees used, “court” appears 80,447 times per million words, or 339 times more frequently than in the ordinary corpus. But no one would think this shows that “court” has a special legal meaning that diverges from the ordinary meaning.

Mr. Nees performs collocate analysis on the term “discriminate against” in the Supreme Court corpus. But that is just not the relevant corpus for understanding the ordinary meaning of the non-technical phrase as it appears in Title VII. So it tells us nothing of value. With a corpus, if you ask the wrong questions you’ll get the wrong answers. Likewise if you use the wrong corpus.

Finally, Mr. Nees notes that after forbidding “discriminat[ion] against” individuals “because of” their “sex,” Title VII makes an exception for cases where sex is “a bona fide occupational qualification reasonably necessary to the normal operation of that particular business or enterprise.” But this exception would be redundant, Mr. Nees argues, if Title VII’s main operative language only ever covered employment actions based on prejudice, since actions based on BFOQs are not rooted in prejudice. So the main operative language in Title VII must sweep more broadly. It must include more than employment decisions rooted in a “prejudicial motive.” And this, Mr. Nees supposes, refutes my linguistic argument.

But I did not suggest that Title VII’s main operative language (“discriminate against an individual . . . because of such individual’s. . . sex”) is limited to employment policies rooted in irrational hostility (such that employment policies based on BFOQs would clearly fall outside the ban from the get-go). Just the opposite: I wrote that the main  operative language would cover employment policies based not only on “outright misconceptions” about the sexes, but also on “unduly rough or weak generalizations” based on sex (6). Given this reading, the BFOQ exception has a critical job to do: it’s there to tell us when a particular generalization about one of the sexes is compelling enough to provide a lawful basis for a given employment action.  

For example, a refusal to hire women as close-contact guards at a maximum-security prison for men obviously rests on some generalizations about women as compared to men. Those generalizations might be “unduly rough or weak” for some purposes and not for others. How to tell which they are here? By asking if this is the particular kind of work for which maleness is a BFOQ. (The Supreme Court held that it was in Dothard v. Rawlinson, 433 U.S. 321 (1977).) This example proves that the BFOQ provision is anything but redundant under my reading of the main operative language of Title VII. 

Further, it is common for laws to provide clarity as to where the line is (as Pam Karlan noted in her briefing before the Court, “Congress is free to take a ‘belt-and-suspenders’ approach”). The classic historical example of this is the Bill of Rights. The Federalists said a bill of rights was not necessary since the federal government could only exercise limited, enumerated powers anyway, and those limited powers did not allow the federal government to infringe rights. The Antifederalists, however, were presciently worried that the federal government’s powers might be construed more broadly. So they demanded explicit constitutional protection of certain rights. By Mr. Nees’s logic, the Bill of Rights is redundant. Does anyone today think so? But in the end, it is only the text that matters. We look at the words and give them their best reading. And that reading of Title VII, in the context of sex, requires a prejudice lacking in the cases before the Court, despite Mr. Nees’s best efforts to contend otherwise.

UPDATE (by Michel Ramsey):  Ryan Nees responds:

I’d just add a few sentences by way of rejoinder. The major conceptual move Phillips makes is his insistence on focusing on what he calls the “relevant linguistic context.” But doing so stacks the deck: if Phillips gets to decide which uses of the phrase are “relevant” and which are not, it will of course take on whatever meaning he wants it to, even where (as here) most of the usages cut the other way. To be sure, we shouldn’t derive ordinary meaning from frequency alone, and Phillips’s argument now implicitly acknowledges the “faulty frequency hypothesis” that corpus linguistics advocates once made. But what he’s replaced it with shows the risk in the corpus linguistics enterprise: it permits cherry-picking, just as some judges used to do with legislative history.

FURTHER UPDATE: James Phillips answers:

Two quick thoughts on Mr. Nees’s sur-reply:

1.  Relevance and cherry-picking:  Nees charges that I've "stacked the deck" by allowing myself to decide what the "relevant" phrase is.  But doing that is step one (or step zero) of any interpretive argument.  The key is to let the statute and the interpretive question determine what is "relevant."  Here I took my cue entirely from the Title VII language on which all agree that these cases turn: “discrimination against an individual … because of such individual’s … sex.”  I looked for structural parallels of that language in a corpus of enactment-era texts.  And I used a corpus of ordinary texts because the key statutory language is non-technical.  It is Nees's decision to turn to a corpus of Supreme Court opinions that is cherry-picking, because it has no discernible motivation.  While the corpus of judicial  opinions would be relevant for determining the sense of a term of art, no one (including Nees) has even argued that Title VII's key language here contains any term of art. Likewise is it cherry-picking to rely on portions of the operative language in irrelevant contexts, such as discrimination against commerce.

2.  Frequency: Mr. Nees appears to have confused the “faulty frequency hypothesis” with my acceptance of a much simpler and uncontroversial proposition. That proposition is this:  relying on the frequency of the wrong linguistic phenomena is faulty, but relying on the frequency of the right linguistic phenomena can be highly probative evidence in the quest to operationalize ordinary meaning.  As for what is the "right" linguistic phenomena, see point 1.