This brief essay was delivered as a response to a paper co-written
by Justice Thomas R. Lee and James Phillips, Data Driven Originalism, at the Originalism Works-in-Progress
Conference held at the University of San Diego on February 16-17, 2018. The desire to generate human meaning by
eliminating from the patterns that convey it all traces of the human is at once
perennial and doomed to be ever unfulfilled.
Back in the seventies there was a fast-growing sub-discipline that
promised to marry linguistics and literary criticism in a way that would
provide an objective basis for the interpretation of texts. It was called
stylistics and I am pleased to say that I pretty much killed it by writing two
essays, “What is Stylistics and Why Are They Saying Such Terrible Things About
It?” and “What Is Stylistics and Why Are They Saying Such Terrible Things About
it? Part Two”. In those essays I said that Stylistics failed in two respects. More
often than not, the mountainous machinery it usually cranks up labors to
produce something less even than a mouse; you wade through a whole lot of
charts, distribution patterns, selection patterns, contiguity patterns and find
waiting for you at the other end
something that would have been obvious from the get-go to a ten year old. And, on the
other hand, if at the end of the whole business there is an interpretive
insight that seems novel and arresting , its relationship to the operations of
the analytical machine is entirely arbitrary. Either you do all that counting
and sorting and come up with a pompously inflated version of what you had at
the beginning, or you do all that
counting and sorting and then attach to the resulting data an interpretation it
does not generate; to be sure, the
interpretation may be one the data can be made to support, but only because the data , just sitting there
in all its empty bulk, can be made to support anything.
Consider, for example the case of Louis Milic, a computational
stylistician, who studied the relative frequency of initial determiners and
initial connectives in the sentences of Swift, Macaulay, Addison, Gibbon and
Samuel Johnson. He discovered that Swift, to a much greater degree than the
other authors, began his sentences with connectives and he concluded that “The
low frequency of initial determiners, taken together with the high frequency of
initial connectives, makes [Swift] a writer who likes transitions and makes
much of connectives”. That’s the banal fruit of the analysis. The arbitrary
fruit comes a bit latter when Milic notes that it was Swift’s habit (as it is
in fact mine) to extend the length of his sentences by producing a series of
appositional phrases that threatens never to end. His conclusion? Swift’s “use of series argues
a fertile and well stocked mind”. Why not say that Swift’s use of series argues
an empirical rather than an abstract cast of mind , or that Swift’s use of
series argues an anal-retentive personality or that Swift’s use of series
argues an unwillingness to let go of a thought. These and a thousand other
interpretive conjectures will fit the data Milic assembles, but there is no
genuinely motivated route from the data to any one of them. As a statistical
output, Swift’s use of series has the status of an ink-blot in a Rorschach
test. The researcher/analyst asks, what might that mean , do you think?, and
the answers are immediate and infinite.
Behind Milic’s efforts is the assumption, difficult to dislodge no
matter how many times it has failed to cash out, that banks of data ,
especially data compiled disinterestedly, can yield interpretive conclusions; and the
further assumption is that the conclusions thus yielded will be more objective,
because less impressionistic, than the conclusions reached by a single interpreter who , because he or she is a
finite, limited creature, can only survey a finite and limited number of texts and will therefore always be working with an
inadequate and distorting sample. This is an important claim, for after all why
go to all the bother of getting the
machine going if there’s no payoff? The payoff has to be some improvement in
interpretive methodology, an improvement that responds to the usual fears of
interpretation’s being biased, partial, manipulative, and willful. Stefan Th.
Gries , a leader in the corpus linguistics movement, puts the matter
forthrightly when he says that the “assumption underlying most corpus-based
analysis is that formal differences reflect or correspond to functional
differences” and adds that “ functional difference is understood …as
anything—be it semantic or discourse pragmatic—that is intended to perform a
particular communicative function”. In short, it should be possible—and this is
the only reason someone other than a descriptive linguist would be interested
in corpus linguistics at all-- to map formal differences onto content
differences, to correlate features of a text described independently of any interpretive hypothesis with a particular
interpretation.
It isn’t possible and it never will be.
What this means is that while Thomas Lee and James Phillips’s paper
might be interesting for several reasons, one of those reasons won’t be
that its methodology will help us to determine the meaning of legal texts. But
before I explain and defend that large assertion, let me briefly note some of
the problems I have with the paper in general. First, Lee and Phillips fail to
take the measure of the intentionalism they casually dismiss and their failure
takes the form of a misrepresentation of the intentionalist thesis. Lee and Phillips rehearse the now standard
story that intentionalism had been discredited in favor of the doctrine of
ordinary meaning or Original Public Meaning. They explain that “Original Public
Meaning originalists turn the focus away from framers intentions and
toward the public’s understanding of the text.” But this statement
assumes what intentionalists would dispute: that there is a text to be “focused on” independently of the assignment
or assumption of intention. The entire Original Public Meaning project, and
therefore the corpus linguistics project, depends on the possibility of
identifying the text as a text (rather than some abstract scribbles) before any inquiry into the intention
informing it, and that is what intentionalists say cannot be done. That is, for
example, what Larry Alexander says in many places and especially in the essay
Lee and Philips cite, Is that English
You’re Speaking: Why Intention-free Interpretation is An Impossibility. The
subtitle of the paper might be rewritten to read, Why Intention-Free Identification of Something as a Text Isn’t Possible;
for in this and other essays, Alexander, along with other intentionalists,
argues not simply that you can’t interpret the text without recourse to a
governing intention, but there is no text until a specification of an intention
is in place.
In a subsection entitled, Texts
Cannot Declare that they are Texts, Alexander and his co-author, Emily
Sherwin, declare that “one cannot look at the marks on a page and understand
those marks to be a text (i.e. a meaningful writing) without assuming that an
author made those marks intending to convey a meaning by them.” That is, absent
an intention, either assumed or specified, what you will have are black marks
and white spaces. It is only when one thinks that someone meant something by
these marks that one searches them for patterns conveying a message. Just as
when smoke appears on a distant horizon, its forms can only become the object
of interpretation when it is assumed that the serial sequence of puffs is
designed. Once that assumption is in place, then you can ask, what does the
sequence of puffs mean (it will only be a sequence rather than something
random if that condition is met); but without that assumption in place, the
question of meaning does not and cannot arise. The same argument holds for
individual words. If I see these markings – HELP – on a sidewalk and wonder
what they mean, my wonder and my effort at interpretation will both cease if I
discover that the marks were made by dirty rain water dripping from the gutters
of an adjacent building (unless you think that buildings have intentions and
can signal distress at having been allowed to deteriorate). It is not simply
that seeking intention is the best way to interpret a text; rather it is that
if you don’t seek intention (or assume intention) you will not have a text.
Now, one might find Alexander’s argument persuasive or unpersuasive, but it
cannot be characterized as Lee and Phillips characterize it when they say that
it questions “our ability to resolve ambiguities in the meaning of a given text
without resort to authorial intention”. But “given text” – the text just
sitting there detached from authorial intention or institutional history, the
text that might or might not be ambiguous -- is what Alexander denies the
availability of. It is the
intention-free existence of the text, not its being ambiguous or unambiguous,
that Alexander calls into question.
Why do Lee and Phillips misrepresent the intentionalist position?
Because that position, taken seriously, means that textualism in general and
corpus linguistics in particular are nonstarters if the game being played is
the game of interpretation. Lee and Phillips report, quite accurately, that the
majority of citizens in the originalist’s universe are now Original Public Meaning
textualists. This undoubted fact is a statistical measure of what the majority
of those people in the field believe; it has nothing to do with whether that
majority is correct or whether its methodology is coherent or whether it is the
methodology interpreters are actually following. Indeed, intentionalists
characteristically say (of Justice Scalia and others) that textualism is a
method that cannot be followed and that self-identified anti-intentionalists enact
intentionalism even when they rhetorically repudiate it. Lee and Phillips
present an example when they describe Justice Thomas’s dissent in Kelo v. Town of New London as an instance
of a jurist relying on original communicative content -- that is, a jurist
being a textualist. However, they are honest enough to note that Thomas’s cites
“early state practices” that he claims “shed light on the ... meaning of the
... words contained in the Public Use Clause.” What kind of light is that? It’s
certainly not light shed by an examination of the words; it is the light shed
by an understanding of what state legislatures usually had in mind when they
exercised their eminent domain power. The sequence of reasoning is as follows:
here are some words in the Public Use Clause; in order to determine what they
mean let’s look at the way legislatures of the period usually thought about
these issues, what they were concerned to promote or prevent. That’s
intentionalism, pure, simple, and unavoidable. Thomas is not slipping here from
Original Public Meaning textualism to a suspicious intentionalism; he is doing
what everyone must do, whether the text is a clause in the Constitution, or a
grocery list, or a lyric poem, or a word that appears on a sidewalk – trying to
figure out what someone meant by these marks.
This is all to say that except in narrow and carefully defined circumstances
(which I will describe later) Original Public Meaning or ordinary meaning
doesn’t get you anywhere. I am not denying either that Original Public Meaning
is a thing or that specifying it, by whatever means, might be occasionally and
limitedly helpful. I am only saying that it is not a magic key to
interpretation and indeed is not a key at all.
What then is it? It is in fact a construct, a put together set of
mark – meaning correlations that are statistically predominate in a particular
population of speakers at a particular time. But, these correlations, in and of
themselves, are of no more interpretive help than the raw data of frequencies
and distributions. You can know the shape of public meaning, just as you can
know that Swift often begins his sentences with connectives and still be unable
to proceed to any legitimate interpretive conclusion. Original public meaning
is interpretively inert; the mere fact of it doesn’t get you anywhere, just as
the mere fact that Swift makes extensive use of appositional series doesn’t get
you anywhere. In both contexts what is missing and required is a determination
of what an agent intended to signify by those patterns. Let’s say that in the
lexicon of Original Public Meaning a certain word is associated with a certain
meaning; but having specified that, nothing interpretive follows unless an
audience knows that Original Public Meaning is the code the author is
deploying. The phrase “Original Public Meaning” is deceptive because the word
“meaning” suggests that meaning is what it delivers. But what it delivers is
data; what it captures are some observed regularities, but those regularities
are of no more interpretive significance than randomly occurring puffs of
smoke. It is only if you know that the author has tied himself to the code of
Ordinary Public Meaning that knowledge of that code would be interpretively
relevant.
But how do you know that? The question is urgent because as
Alexander often says, the text won’t tell you what language it is written in. That knowledge, in the absence of which
interpretation cannot begin, must come from the outside, from some persuasive
indication—non-textual—of what the author had in mind, of what he or she or
they intended. Data analysis might be able to tell you whether x is or was the
ordinary meaning of y, but you’d still be faced with the question was it
ordinary meaning the writer was deploying. The bottom line is that while corpus
linguistics can help specify what the Ordinary Public Meaning is or was, it
cannot help you take the next step, the step of interpreting; for that you need
the intentionalism corpus linguistics dismisses. If it is somehow determined
that the speaker/writer intended to deploy the code of Original Public Meaning,
Original Public Meaning can help guide interpretation. If that intention is not
in place, the fact that there is something called Original Public Meaning --
again, I stipulate to that fact -- is no more interpretively interesting than the fact that the speaker/writer is
six-feet tall or the fact that he or she likes coffee ice-cream. Lee and
Phillips say that “by tabulating the relative frequencies of different senses
of a word or phrase within a corpus, a linguist can . . . discern the more
common sense of a given term in a given linguistic context.” But there is
nothing a linguist can do with what he has discerned unless the intention of
the author to tie himself to the common sense of a word has been established by
nonlinguistic means. Original Public Meaning is a resource, not a constraint; it
is an option speakers and writers are not required to choose; the constraints
come in, if they do, when the intentionalist question—Is that Original Public
Meaning you’re speaking?—has been answered. In and of itself corpus linguistics is an
interpretive dead-end.
Let’s test this out by considering Lee and Phillips’s examples.
Lee and Phillips’s first example is the phrase “domestic violence”
as it appears in Article IV of the Constitution where the government is
assigned the task and duty of protecting the republic from foreign invasion and
“domestic violence.” The corpus linguistics question is how do you know the
“domestic violence” here means violence committed by militia men like Timothy
McVeigh rather than the violence committed by one spouse or partner against the
other. The obvious answer is that anyone who has gone beyond the sixth grade
knows that; but that answer will not be accepted by corpus linguists because it
does not have any statistical or numerical backup. So, Lee and Phillips proceed
to their data mining operation and, after a lot of work, say triumphantly that
their method has “confirmed the intuition” that they and we had in the first
place. “Our data show that domestic
violence today is almost always used in reference to an assault on a member
of a person’s household, but was a reference to an insurrection or rebellion in
the late 18th century.” Shades of saying that Swift’s preference for
connective transitions over determiners means that he is an author who likes
transitions.
In a second example utilizing eye boggling charts and tables, Lee
and Phillips perform the same operation, in which something obvious is put
through a machine that huffs and puffs and delivers exactly what it began with.
It’s like the transporter device in Star Trek that breaks down your molecules and
then reassembles them in the exactly the same form on the other side. (The
protagonist of the movie “The Fly” was not so fortunate.) In this second case, the object of analytical
attention is the word “commerce” as it appears in article 2 of the constitution,
where Congress is given the power to “regulate commerce.” The question is, what
does “commerce” mean? You might think that commerce means trade, although, of
course, “trade” itself can be understood in both narrow and expansive ways. Lee
and Phillips also think that commerce in this context means “trade” but they
want to support that intuition by lexical analysis performed by a powerful
search engine on a massive database. They proceed by thinking up some of the
other things that “trade” might have meant or could possibly mean, and then
they search the frequency with which the word “trade” is collocated with some of
those other things. So they pair off or face off “trade” and “manufacturing” as
two potential meanings of “commerce” and discover that “trade,” “shares six top
30 collocates with commerce” (don’t ask!), while “manufacturing” “shares just
two.” Their conclusion: “the fact that commerce and trade have more overlap in
their collocate network than “commerce” and “manufacturing” do is evidence that
the “trade” sense is likely more common than the “manufacturing” sense.” Louis
Milic lives. Lee and Phillips then say, in a truly terrifying statement, that
“further research could be done,” for, “the value of a corpus is the ability to
slice and dice context to get to the most relevant semantic context.” I would
ask relevant to what? As far as I can tell, there is no semantic context in
their analysis -- all there is is the tabulation of frequencies -- and no way
of getting from their analysis to a semantic context.
It is important to understand what Lee and Phillips’s analyses prove
and do not prove. They do prove (or strongly suggest) that in the code of
ordinary meaning as it existed in the late eighteenth century “commerce” in the
phrase “regulate commerce” meant “trade.” They do not prove that this is the
meaning the framers chose, for it remains possible that they chose a meaning
for “commerce” that departed from the statistical regularities Lee and Phillips
uncover; chose a meaning, that is, which departs from Ordinary Public Meaning.
They were certainly free to do so – it is always the speaker who nominates
mark-meaning correlations – and evidence that they did or did not will not be
provided by the data Lee and Phillips have so laboriously assembled. In short,
the fact that there is an Original Public Meaning doesn’t compel a
speaker/writer to use it and the determination of whether he did or not, cannot
be made by consulting the text.
The third example cited by Lee and Phillips is the phrase “public
use” as found in the Constitution’s Taking Clause. The question is, does
“public use” mean use related to public government purposes (military,
economic, education, etc.) or can it be interpreted, as the majority in Kelo does, to mean private uses of which
the public government approves? Lee and Philips are on the side of Justice
Thomas, who believes that the state errs in adopting the more expansive
definition of public use, and I agree with them. Lee and Phillips dutifully
fire up their machine again and find that “the direct sense that Justice Thomas
argued for is much more common than the broader, indirect sense that the Kelo majority adopted.” But again, the
fact that the direct sense is statistically predominate in the period says
absolutely nothing about what sense the founders chose. Information about that
must come from elsewhere, perhaps from an argument that foregrounds the
founders concern to separate the public and the private, a concern in the light
of which you might conclude that what they had in mind -- what they intended --
when they wrote the taking clause was the maintenance of the public/private
distinction.
At one point in their analysis of the word “commerce”, Lee and
Phillips wonder if perhaps “commerce” was the wrong linguistic unit to begin
with. “Maybe we need to search for institutions of the regulation of commerce
by the government.” They withdraw from their own suggestion because it “demands
too much semantic context.” They realize that in order for their methodology to
work, it must begin (as Chomsky tried years ago to begin) with no hostages to
semantics at all; it must be purely formal, a matter of assembling the data
independently of semantic contexts or suppositions and then proceeding from
there to a genuinely motivated interpretation. In short, they know at some
level that beginning with semantics will render their countings nugatory
because the interpretive conclusion those countings supposedly generate have
already been put in at the beginning. Here then, is the corpus linguist’s
choice. Either stick to your method and have nowhere legitimately to go after
the data has been assembled or compromise the method by grounding it in a
semantic hypothesis and end up with a product that is nothing more than your
original input tricked up in numbers.
Toward the end of their paper, Lee and Phillips display an
attractive modesty. Although, they remain confident that “the use of corpus
analysis” can be “a central element of the first stage of any originalist
inquiry,” they do acknowledge that it is not quite clear what happens after
this first stage. “We are less sure of the precise role [corpus analysis]
should play.” In a recent paper, Stefan Th. Gries and Brian G. Slocum (two
leaders in the field) display an even more robust modesty: “corpus analysis
cannot by itself provide conclusive meanings to legal text” and again “corpus
analysis can provide valuable insights about language usage but cannot by itself
resolve normative issues.” That is it cannot direct us to what words and
phrases mean. But I thought that was why we were going to all that trouble in
the first place.
Can corpus analysis do anything? Yes it can. It can reveal, verbal
patterns the naked eye, or the merely intuitive reader, would never see. And
those patterns, often deeply embedded in an author’s stylistic presentation,
can serve as a means of identifying him, of fingerprinting as it were. Larry
Alexander and I have been talking about collaborating on an op-ed about current
campus protests, a subject on which we have similarly Neanderthal views. Now
let’s suppose Larry and I do manage to write this piece, and for some
unfathomable reason an analyst wants to figure out which of the sentences in it
were written by me and which were written by Larry. No doubt both Larry and I
have verbal ticks of which we are largely unaware, ticks that have no
particular meaning except the meaning that it is we who perform them.
Corpus analysis would reveal those ticks and therefore would be able to
distinguish Larry sentences from Stanley sentences. But that’s all it could do;
as for contributing to a determination of what Larry or Stanley means by those
sentences, forget about it.
To be sure, there are more serious uses to which a fingerprinting
analysis might be put. It has been speculated that Shakespeare had a
collaborator on some of his later plays. Did he, and if so who was it? There
are poems whose authorship is in dispute between John Donne and Ben Jonson. Who
really wrote them? This is a serious question in literary history and it is one
corpus linguistics could help answer, and if it did there would be an
interpretive pay off, for you would now approach the poem and receive its words
and phrases within the knowledge that
it was Donne or Jonson, each with his distinctive concerns, obsessions, and
visions who had designed it. Real interpretation could then begin and corpus
linguistics would have brought you to that moment, but it would itself have done
no interpretive work; it would have just cleared the ground in a way that
allowed interpretive work to proceed. That’s not nothing, but it isn’t a
contribution to interpretation, just as a modern tennis racquet will help you
to hit a winner -- it puts you in a position to do the job -- but doesn’t
itself do the job. So there’s a use, and for some a significant one, of corpus
linguistic analysis. And as I have already acknowledged corpus linguistic
analysis can also help us to determine what the contours of Original Public Meaning
or ordinary meaning were at a particular time. But beyond that, I don’t see its
utility and therefore, I don’t see why the enormous expense and Casaubon-like
erudition involved in learning how to do it is worth the candle.
Lee and Phillips and Gries and Slocum might respond by saying two
things: (1) they might say that although the way we humans produce and receive
meanings is quite different from the digitizing and retrieval processes of
corpus linguistics, there is no reason why corpus linguistic methods cannot
elucidate meanings produced by other means. But that’s like saying one could
observe and transcribe the physical patterns of movement enacted by players in
a football game without knowing that the moves you are describing are moves within that game, and have something
interesting to say. If all you have is the catalog of physical actions
performed in a sequence by 22 persons you can never get from that catalog to
any statement of what those actions mean. Once you detach patterns from the
intentional context in which they have significance, you can’t get the
significance back; and (2) they might say as the stylisticians of the 1970s
said and the digital humanists of the present day are saying, our discipline is
new; don’t expect us to have all the answers at the beginning: but in time,
trust us, this methodology will deliver what we seek, a legitimate relationship
between the amassing of data and interpretive conclusions. To which I would
respond by invoking one of my favorite lines from Shakespeare’s King Lear:
“Never, never, never, never, never.”
Stanley Fish is the Davidson-Kahn Distinguished University Professor of Humanities and Law at Florida International University. You can reach him by e-mail at fishs at fiu.edu