Scholars and judges throughout the political spectrum apply the canon that words are afforded their “ordinary meaning,” at least as a default (see, e. g., Eskridge 2016; Slocum 2015; Scalia and Garner 2012; Solan 2010). 1 We devote this article to demonstrating how, in determining a statute’s ordinary meaning, reference to a corpus of historical English can be superior to reliance upon dictionaries or the intuitions of judges as native speakers of English. We further apply a corpus analysis to the facts of Shaw v. United States, a case pending before the Supreme Court of the United States at the time this article is written. An important issue in that case is the meaning of the word “defraud” in a law that makes it a federal crime to “defraud a financial institution,” or to engage in a scheme to do so.
2 Three ways to determine ordinary meaning
The most usual way to determine the ordinary meaning of a word is to rely upon one’s knowledge of the language as a native speaker. We all do this in our daily lives, and we do it so routinely that we do not even notice. Furthermore, we all have a sense of when a person has used a word or phrase in a peculiar way.
Linguists also rely on their intuitions about both grammaticality and meaning in creating theories about the structure and functioning of language. We first comment on this method, then turn to the use of dictionaries by courts, and finally discuss the recent turn to corpora both by linguists and by legal analysts as a tool to determine how words are ordinarily understood.
2.1 Linguistic intuition generating linguistic principles
Traditionally, linguists have used their own judgements about meaning and acceptability to draw generalizations about the structure of language. Just as important as what one can say or mean is what one cannot say or mean. For example, linguists have observed (see, e. g., Freidin 2012) that of the four sentences presented below, the first three are grammatical sentences of English, but the fourth is not:
I hope Bill leaves soon.
I hope that Bill leaves soon.
Who do you hope leaves soon?
*Who do you hope that leaves soon? 2
It appears that we can optionally place “that” between “hope” and the subordinate clause in such sentences, but not in the questioned version. Linguists attempt to explain such facts by creating a theory of syntax that accounts for them. Crucially, the problem occurs only when the subject is questioned, not when an object is questioned:
Who do you hope (that) Bill visits tomorrow?
The data are based on the linguist’s sense of what is grammatical and what is ungrammatical. The intuition is the sense we share that some versions are grammatical, others are not. Work in linguistics by Sprouse et al. (2013) and Sprouse and Almeida (2012) demonstrates a high level of consensus between the intuitions expressed by linguists about the acceptability of sentences with linguistic structures, and those of a large sample of native speakers responding to survey instruments. Subsequent studies have been less optimistic, but confirm that consensus among a small group of individuals (five) predicts conformity with a large percentage of participants in a survey study (Mahowald et al. 2016).
Intuitions, however, have two rather significant limitations, especially when relied upon exclusively in legal contexts, where the stakes are high. First, once an individual takes a position on an issue, it may be difficult to pay adequate attention to examples that may tend to disconfirm that intuition. This is referred to as “the confirmation bias.” Much has been written about this phenomenon, including in legal contexts (see, e. g., Risinger et al. 2002; Nickerson 1998). Secondly, it is always possible that there is significant variation within the population that we have no way of observing in everyday life. It is only natural for individuals to regard their own knowledge of language as reflecting the intuitions of ordinary speakers of the language, even when their intuitions are outliers. This is known as “false consensus bias,” and has also been the subject of study, including in legal contexts (Solan et al. 2008). 3 Consensus is an empirical matter, however, and when there is dispute, some kind of procedure to demonstrate the claim’s validity would seem appropriate.
False consensus bias is especially a concern when the issue is the distribution of understanding. Consistent with Sprouse et al. (2013), we have every reason to believe that “the is table on book the” is not a sentence of English because the words have been scrambled. In contrast, when the issue is whether the “ordinary meaning” of the expression “carry a firearm” 4 as applied to a handgun is limited to carrying it on one’s person, or whether ordinary usage also includes toting it in an automobile, the relevant intuition is not about grammatical judgement, but rather about the distribution of usage over a population. There is less reason to believe that each of us has the same, accurate statistical model of usage in our minds when it comes to such distributional facts.
Taken together, the confirmation bias and the false consensus bias create some risk that intuitions about linguistic facts may not necessarily reflect the ordinary meaning. The risk is all the greater when the intuitions are used in legal contexts, where the intuitions of the lawyers are likely to align with their clients’ interests, and the intuitions of judges may be outcome-determinative.
2.2 Turning to the dictionary
One way to try to avoid the idiosyncratic nature of individual intuition about language is to turn to a neutral source. For many years, judges – Supreme Court justices in particular – have turned to dictionaries as a source of information about word usage. This trend has provoked a robust literature, almost entirely critical (see, e. g., Brudney and Baum 2013; Kirchmeier and Thumma 2010; Solan 2003; Aprill 1998). Judge Posner summarizes some of that disapproval in his opinion in United States v. Costello, 5 as does Utah Associate Chief Justice Lee in his concurring opinion in State v. Rasabout. 6 Disapproval of excessive reliance on dictionaries has often focused on the debates among members of the Court as to which dictionary is the most reliable, when competing dictionary definitions would lead to different legal outcomes.
Linguists have not been so fast to rely on dictionaries as authoritative sources, although linguists frequently participate in the compilation of dictionaries. Definitions often differ from one dictionary to another. Differences arise from the fact that publishers have different policies about how many words to include, which kinds of words to include, how many senses of each word to include, how to split senses that are closely related, how to order the various senses, and how to best exemplify those senses. Some dictionaries also require lexicographers to use a set list of vocabulary when writing definitions, which, when compared to definitions in other dictionaries, may affect nuances in meaning. While publishers set guidelines for such lexicographic decisions, many of these decisions have been left to the individual lexicographers. Oxford Dictionaries (n.d.) makes this statement:
For hundreds of years, including most of the 20th century, lexicographers worked without enough evidence: […] sometimes with no evidence at all except their own intuition. Even when evidence of usage was available, dictionary editors had no means of filtering or sorting large amounts of data efficiently and reliably.
However, with the growing access to and functionality of computer technology in the late 1980s and early 1990s, dictionary companies such as Collins Cobuild, Oxford, Cambridge, and Longman began compiling systematic collections of actual language use – corpora – in order to aid in the testing of lexicographic intuitions (see, e. g., Halliday 2004). Subsequent linguistic studies of such corpora have repeatedly demonstrated that speakers of English are frequently unaware of the contextual layers of meaning that accompany words – connotative meaning that is shared by speakers of a language but is not recorded in dictionaries as an integral part of a word’s denotative sense (Hunston and Thompson 2000). As Judge Posner noted, traditional dictionaries simply provide a list of non-contextualized words. This statement is further supported by Yallop (2004: 24):
To do the job of presenting words more or less individually, in an accessible list, the dictionary takes words away from their common use in their customary settings, creating a snapshot that cuts across subtle differences that are context-dependent. While this is in many respects a useful job, the listing of words as a set of isolated items can be highly misleading if used as a basis of theorising about what words and their meanings are.
Judge Posner made a more specific point in United States v. Costello. As detailed by Channel (2000), some words have the tendency to carry a particular “semantic prosody,” i. e., a positive or negative connotation that is not recorded in the dictionary definition of a word. For instance, in her corpus-based examination of the context in which “roam” is used, she found that the word is used primarily in the phrase “roam the streets” and almost exclusively paired with words that carry a negative meaning (e. g., “prostitutes, vagrant children, armed men, mobs”) that is typically not observed in dictionaries. Posner further illustrates this point in noting, based on analysis of a corpus, that the verb “to harbor” connotes an effort to hide an individual, a nuance not mentioned in most dictionaries (see discussion below).
2.3 Turning to corpora
Over the past two decades, linguists have developed a subfield called “corpus linguistics.” Scholars engaged in the endeavor focus their attention on recorded language use (both written and verbal), rather than on intuitions of grammaticality or word meaning drawn from native speaker intuitions (Baker 2006; McEnery et al. 2006; Biber et al. 1998). The inquiry is especially useful when the issue at hand is the distribution of language usage between groups, which is a central inquiry of the field of sociolinguistics (Baker 2010).
As noted, lexicographers have begun to use “big data” to draw conclusions both about what definitions should say, and about the relative prevalence of the different senses of a word in common usage. Judges and legal scholars are accustomed to searching large banks of information for relevant instances of a word’s meaning. That process characterizes the case method generally. What is new is the ability to search electronic corpora for instances of word usage. In the past two decades, large corpora of general American English have been made available to the public as a research tool. The most prominent have been developed by linguistic researchers at Brigham Young University (n.d.). Two well-known examples of their corpora are the Corpus of Contemporary American English (COCA) with its 520 million words from five genres of texts from 1990–2015, and the Corpus of Historical American English (COHA), which is comprised of 400 million words from popular written genres such as magazines, fiction, non-fiction, and news publications from 1810–2000. In these corpora, each genre is balanced decade by decade to “monitor” a representative sample of actual language use over time, 7 making them the largest systematically-collected corpora of American English usage.
Studies performed on such corpora have determined that distributional judgements about word meaning based solely on individual intuition can be unreliable in predicting facts about usage, since people tend to pay attention to unusual features as opposed to usual ones (Biber et al. 1998). Therefore, the purpose of such corpus-based investigations of language use is to test individual intuitions about language use against large quantities of data. From such analyses, conclusions about ordinary meaning and form can be drawn that are not based on what may potentially be a few speakers’ idiosyncrasies.
Corpus linguistics is also useful in legal contexts in which courts must determine whether a possible meaning is within the ordinary understanding of individuals in a speech community, as Mouritsen (2010, 2011) has ably demonstrated. 8 In fact, some judges in recent years have decided to become their own corpus-based lexicographers and have turned to large collections of documented language to investigate meaning and usage (see, e. g., Eskridge 2016: 46–47; Mouritsen 2010, 2011). In his majority opinion in Muscarello v. United States, 9 Justice Breyer relied on Lexis and Westlaw searches of newspaper articles containing the words “weapon,” “vehicle,” and “carry” in close proximity to conclude that a defendant on his way to a drug deal with a gun in one part of the car, the drugs in another, was “carrying a weapon” in the ordinary sense of that expression in violation of a federal statute that makes it a crime to carry a firearm during and in relation to a drug trafficking crime. 10 The defendant had argued the ordinary sense of carry means to carry on one’s person.
Judge Posner, in United States v. Costello, 11 following his criticism of excessive reliance on dictionaries, used a Google News search to determine that a woman whose boyfriend, an undocumented immigrant, was living with her, did not “harbor” him in violation of a federal statute, at least not if the court relies upon the ordinary meaning of the verb “to harbor.”
Posner reported the results of the search he conducted, showing what words most frequently co-occur with “harbor” (called “collocates” in the language of corpus linguistics and lexicography). There were thousands of “hits” for harboring Jews and fugitives, but few or none for harboring victims or guests. Posner concluded:
It is apparent from these results that “harboring,” as the word is actually used, has a connotation – which “sheltering,” and a fortiori “giving a person a place to stay” – does not, of deliberately safeguarding members of a specified group from the authorities, whether through concealment, movement to a safe location, or physical protection. This connotation enables one to see that the emergency staff at the hospital may not be “harboring” an alien when it renders emergency treatment even if he stays in the emergency room overnight, that giving a lift to a gas station to an alien with a flat tire may not be harboring, that driving an alien to the local office of the Department of Homeland Security to apply for an adjustment of status to that of lawful resident may not be harboring, that inviting an alien for a “one night stand” may not be attempted harboring, that placing an illegal alien in a school may not be harboring, and finally that allowing your boyfriend to live with you may not be harboring, even if you know he shouldn’t be in the United States. 12
Two state supreme courts have also turned to corpus analysis. They, however, have done so more directly, recognizing the linguistic subfield itself as important to legal analysis and touting its benefits. In State v. Rasabout, 13 a gang member, Andy Rasabout, fired twelve rounds from a semiautomatic weapon at a house and car as he drove by. Although a Utah jury convicted him of twelve separate offenses, the trial court merged them and limited his conviction to one count of “discharg[ing] a dangerous weapon or firearm from an automobile… without written permission, within 600 feet of a house, dwelling or other building,” in violation of a state criminal statute. 14 The appellate court reversed, reinstating the jury verdict, 15 and the Supreme Court of Utah affirmed the appellate court’s decision.
In a majority opinion written by Justice Parrish, the Supreme Court held that “the allowable unit of prosecution for Unlawful Discharge of a Firearm is each discrete shot.” 16 In so holding, the court relied on the morphology of the word discharge (dis+charge), various dictionaries, the statutory definition of “firearm” (“any device… from which is expelled a projectile by action of an explosive”), 17 the purpose of the statute, and the whole act rule. 18
Associate Chief Justice Thomas Lee concurred, applying corpus linguistic methodology to the analysis. 19 He remarked:
I would interpret the terms of the statute by looking for real-world examples of its key words in actual written language in its native context. This sort of analysis has a fancy name – corpus linguistics. But it is hardly unusual. We often resolve problems of ambiguity by thinking of examples of the use of a given word or phrase in a particular linguistic context. I propose to do that (as I have in a couple of prior opinions) on a systematic scale – by computer-aided searches of online databases in an effort to assemble a greater number of examples than I can summon by memory on my own. 20
A former full-time law professor at Brigham Young University, Justice Lee used both Google and COCA to gather reference materials.
The dictionaries that the majority consulted contained two possibly relevant definitions: one, in essence, is “to shoot,” which can easily be understood as meaning shot by shot. But the other is “to empty of cargo” or “to unload.” This definition would tend to support Rasabout’s position that he committed only one offence, since he emptied the weapon only once when he fired the bullets.
Justice Lee’s COCA search found a substantial predominance of examples in which “discharge” was used to mean “to shoot” in the sense of the majority opinion. 21 Justice Lee therefore concluded that the ordinary meaning of the verb is “to shoot,” even though it has other meanings, and concurred in the judgement, based largely on his corpus analysis. In doing so, he noted that the majority had ignored the second definition (which would have assisted the defendant’s case). Rather, the majority had relied on its own intuitions about which meaning is ordinary, without adequate inquiry. Instead, he argued, the court should take on the burden of acknowledging both definitions and coming to a reasoned decision based on empirical investigation as to which one reflects the verb’s most “ordinary” sense. 22
In People v. Harris, 23 the Supreme Court of Michigan also relied on COCA to resolve a statutory dispute. In that case, the seven justices all agreed that corpus linguistics can be a useful tool in statutory interpretation, but they disagreed about how to apply the learning of the field to the case at hand. The case involved an encounter between a driver and three police officers. One of the officers assaulted the driver while the other two stood by and did nothing. The incident was somehow video-recorded. Disciplinary proceedings were brought against all three officers. In the context of the disciplinary proceedings, the officers made false statements that were inconsistent with the facts that became apparent on the video. Subsequently, criminal charges were brought against them for assault and battery, common law felony misconduct in office, and obstruction of justice. 24
A Michigan statute, the Disclosures by Law Enforcement Officers Act (DLEOA), prohibits the use of statements made by law enforcement officers during disciplinary proceedings in subsequent prosecutions:
An involuntary statement made by a law enforcement officer and any information derived from that involuntary statement, shall not be used against the law enforcement officer in any criminal proceeding. 25
“Involuntary statements” include statements made during disciplinary proceedings against the officer. 26
The issue in Harris was whether the false statements provided by the officers during the disciplinary proceedings should be considered “information” under DLEOA, and therefore not admissible in the subsequent proceedings. If so, the obstruction of justice charges would have no factual basis. In other words, does the term “information” encompass both accurate information and false information, or only accurate information?
The majority held that the term does include both true and false statements. Referring to the results of a search in COCA, the court determined that the word “information” occurs with both the adjective “accurate” and the adjective “false” in everyday use. Thus, a statement need not be truthful for it to be considered “information.” Justice Markman’s dissenting opinion disagreed. He had no problem using COCA as a research tool, but believed that the majority had not used it properly. Markman observed that when “information” is not modified by an adjective that describes its accuracy one way or the other, it almost always is used to convey actual facts. Thus, he would have held that the false information conveyed by the officers could have been used to prove obstruction of justice.
The disagreement among the justices on the Supreme Court of Michigan illustrates an important point: A corpus is only a tool. The tool is useful when it is used to answer questions that are legally relevant. In Harris, the justices disagreed about what to ask. The majority asked whether “information” is used in ordinary English to describe misinformation as well as accurate information. The dissent asked whether the word “information” without a modifier is used to describe both accurate and inaccurate statements, or only accurate information. We believe that the dissenters asked the more relevant question. Nonetheless, we further believe that this type of analysis will be relevant to many cases in which (1) the ordinary meaning of a statutory term is either determinative or at least relevant to the construal of that term; and (2) the issue in the legal case is sufficiently specific that dictionaries are not likely to address the legal issue directly. “Ordinary meaning,” especially as applied to particular words and phrases, is a distributional fact. A usage is “ordinary” when it predominates.
3 Case study: Shaw v. United States
In May 2016, the U.S. Supreme Court agreed to review the decision of the United States Court of Appeals for the Ninth Circuit in Shaw v. United States. 27 The problem raised by Shaw is a linguistic one, and the courts of appeals are in disagreement about how to resolve it. By presenting false information, Shaw had convinced a bank to release to him through a phony PayPal account more than $300,000 from the bank account of another depositor. Ultimately, PayPal and the victim of this scheme shared the loss.
What Shaw did was surely illegal. In fact, 18 U.S.C. § 1344(2) specifically makes it a crime to:
knowingly execute or attempt to execute, a scheme or artifice – (2) to obtain any of the moneys, funds, credits, assets, securities, or other property owned by, or under the custody or control of, a financial institution, by means of false or fraudulent pretenses, representations, or promises. 28
However, Shaw was not charged with violating that law. Instead, he was charged with violating 18 U.S.C. § 1344(1). That provision makes it a crime for a person to:
knowingly execute, or attempt to execute, a scheme or artifice – (1) to defraud a financial institution. 29
The bank lost nothing in this scheme, and proof was not offered to demonstrate that Shaw intended for the bank to lose anything. In these circumstances, did Shaw defraud the bank, or did Shaw merely use the bank as a tool in defrauding his victim? The circuit courts are divided on this issue, nine holding the statute does not apply absent a showing that a financial institution was either defrauded out of its property, or there was a scheme intended to accomplish this goal, 30 and three, including the Ninth Circuit, which affirmed Shaw’s conviction, holding that it does apply without such proof of intent to deprive a financial institution of its property. 31
In affirming Shaw’s conviction, the circuit court relied on its earlier decisions holding that § 1344(1) does not require proof that the bank was an intended victim of the fraud, 32 on the Supreme Court’s decision in Loughrin v. United States, holding that no such requirement is present in § 1344(2), 33 and on the following characterization of Shaw’s argument concerning the statute’s language:
Shaw thus seeks to characterize the difference between the two clauses as involving the intended financial victim of the fraud, i.e., the intended bearer of the loss. The language of neither clause of the statute, however, refers to monetary loss or to the risk of such loss. The statutory language focuses on the intended victim of the deception, not the intended bearer of the loss. Section 1344(1) requires the intent to deceive the bank. Section 1344(2) requires false or fraudulent representations or pretenses to third parties. The Supreme Court made this point in Loughrin when it noted that the second clause was intended to broaden the scope of bank fraud to include schemes that did not involve deception of the bank directly, such as schemes to use stolen credit cards. See 134 S. Ct. at 2391–92. Section 1344(1) thus covers schemes to deceive the bank directly. Neither clause requires the government to establish the defendant intended the bank to suffer a financial loss. 34
The Ninth Circuit correctly identified the issue in the case as depending in large part on the meaning of defraud. However, the court’s statement assumes the answer to the ultimate question without analyzing the statutory language. If the direct object of defraud is understood as “the intended bearer of the loss,” then the Ninth Circuit was wrong. If, instead, “the statutory language focuses on the intended victim of the deception, not the intended bearer of the loss,” then the Ninth Circuit was correct.
In its brief before the Supreme Court, 35 the government devotes itself not to supporting the Ninth Circuit’s analysis, but rather to demonstrating that one can defraud another by depriving the victim of property interests other than pure ownership interests, including the possessory interest that a bank has on the funds of depositors. We believe that the government is correct up to that point, but begs the ultimate question by stopping its analysis short of addressing the actual issue: whether the bank bears the loss in the end. For example, when a bank lends money to a mortgager, the funds it lends may be those of depositors. A borrower who obtains the loan by means of a fraudulent device has surely committed bank fraud. But it is the bank that will suffer the financial loss if the loan is not repaid. The loss will not be distributed among depositors. The same is true for fraud concerning a leasehold, which is a possessory interest in land, short of ownership. If A defrauds B out of a leasehold, the result is B suffering a loss – not some other individual C suffering the loss. The problem with the Ninth Circuit’s analysis, then, is not that the funds taken were possessed by, but not owned by, the bank, but rather that the bank was not victimized by the scheme.
The government cites Loughrin for the proposition that the word defraud should be understood in accordance with the statute’s “plain text.” 36 We agree. Our intuitive understandings of the meaning of defraud in this context are that the Ninth Circuit got it wrong, at least to the extent that the court relied on the meaning of the verb in the statute: Unless Shaw engaged in a scheme to defraud the bank of property, he did not violate the statute.
With this as background, let us identify the question in Shaw as this: When a direct object is specified (“financial institution”) in a statute making it a crime to “defraud” the party conveyed by the direct object, is the direct object ordinarily intended as the property loser?
Our findings, as set forth below, are consistent with analyses of grammaticality by linguists, the definitions provided by major dictionaries of American usage, and by examples of large quantities of actual language use over time. In particular, the corpus analysis demonstrates that over the long history that begins with the enactment of predecessor statutes from which the language of § 1341 was derived in the nineteenth century, and the enactment of the current statute in 1984, defraud is used almost exclusively in situations where the victim of the fraud and the party being deprived of property are one and the same. Such a reading is consistent with a substantial majority of circuit courts that have opined on the issue.
4 Analysis of the ordinary meaning of defraud
The following analysis examines the three ways of determining ordinary meaning: by examining linguistic intuition, by consulting dictionary meaning, and by analyzing a corpus of large quantities of actual language use.
4.1 Linguistic intuition of defraud
Shaw is about the grammatical context in which the verb defraud may be used. English verbs differ from one another in what kinds of objects they require or allow. Some verbs require only direct objects, or both direct and indirect objects.
Frank robbed a bank.
Frank gave his book collection to the public library.
Sometimes, objects that are optionally expressed are implied even when they are not present. We know that if Frank robbed a bank he must have taken something from the bank or he could not have robbed it (see, e. g., Fillmore 1968). 37 Linguists have studied English verbs to determine the syntactic contexts in which they appear. In her book on English verb classes, Levin (1993: 129) refers to defraud as a verb of “possessional deprivation.” 38 Defraud requires that there be an object that is defrauded, whether expressed or implied, and a loss of property.
Bill defrauded the company, and was sued for it.
?Bill defrauded and was sued for it.
The second sentence is not nearly as good as the first. If it is grammatical at all, we understand it with an implied object and further understand that Bill came away with some ill-gotten gains. Defraud may also appear with a description of the property that was transferred as a result of the fraud:
Bill defrauded the company (out) of $4 million, and was sued for it.
It may not appear with a description of the property without the object, however:
*Bill defrauded out of $4 million.
The statute does not actually require that the defendant defraud the victim. Rather, it requires that the defendant “execute a scheme… to defraud a financial institution.” This covers inchoate crimes as well as executed ones. When defraud is embedded in the “scheme” phrase, the direct object becomes optional, although one is implied even when not expressed. Consider the following sentence:
By arranging to transfer money from the victim’s account, Bill engaged in a scheme to defraud.
The absence of an explicit direct object and description of what was lost does not mean that these are conceptually absent, however. We understand that Bill must have schemed to defraud someone of something for this sentence to be true. 39
The issue in Shaw is very specific. The statute makes it a crime to engage in a scheme or artifice to defraud a financial institution. The question is this: In ordinary usage, is the one who is intended to lose property as the result of the fraud be the same individual as the party being defrauded? That is, in ordinary speech, would one say that someone has defrauded an individual by tricking that individual into doing something in which another is intended to suffer a loss?
Many fraud statutes are written without specifying a victim. The federal mail fraud statute, for example, states:
Whoever, having devised or intending to devise any scheme or artifice to defraud… places in any post office or authorized depository for mail matter, any matter or thing whatever to be sent or delivered by the Postal Service, or deposits or causes to be deposited any matter or thing whatever to be sent or delivered by any private or commercial interstate carrier… shall [be punished]. 40
Similarities between the bank fraud statute and the mail fraud statute are especially significant because the bank fraud statute, enacted in 1984, was modeled on the mail fraud statute. 41 The operative “defraud” language in the mail fraud statute has been in place since the nineteenth century. Section 301 of the 1872 Postal Act read:
That if any person having devised or intending to devise any scheme or artifice to defraud, or be effected by either opening or intending to open correspondence or communication with any other person (whether resident within or outside the United States), by means of the post-office establishment of the United States or by citing such other person to open communication with the person so devising or intending, shall, in and for executing such scheme or artifice (or attempting so to do), place any letter or packet in any post-office of the United States, or take or receive any therefrom, such person, so misusing the post office establishment, shall be guilty of a misdemeanor, and shall be punished with a fine of not more than five hundred dollars, with or without such imprisonment, as the court shall direct, not exceeding eighteen calendar months. 42
By not specifying either the nature of the direct object or the nature of the property that is the subject of the fraud, and by embedding the entire crime as engaging in a scheme or artifice, Congress has created maximum flexibility in mail fraud prosecutions. Any scheme to defraud anyone out of anything by using the mail is a crime. The absence of specificity in the statute, however, does not alter the fact that one must defraud someone of something to succeed. Rather, it both criminalizes inchoate crimes (even unsuccessful schemes are crimes) and applies to frauds committed against any potential victim.
4.2 Dictionary definitions of defraud
Of additional import to the interpretation of the bank fraud statute in Shaw is whether defraud communicates only “deceit” or whether “depravation” is also a core part of the verb’s meaning, and what kind of argument structure the verb supports. Two leading dictionaries of American usage define the verb consistent with our own sense of the word and consistent with the linguistic literature discussed above. Webster’s Third New International Dictionary (1961: 593) defines the word: “vt: to take or withhold (from one) some possession, right, or interest by calculated misstatement or pervasion of truth, trickery, or other deception <~the heirs of their bequests> <citizens ~ed of their voting rights>.” The verb may also be used intransitively to mean, “to engage in fraud,” as we saw in the mail fraud statute discussed above. The American Heritage Dictionary (2016: n.p.) defines the word similarly: “To take something from by fraud; swindle: defrauded the immigrants by selling them worthless land deeds.” These definitions demonstrate that the verb can be used both transitively (with an explicit object) and intransitively (with an implicit object). They contrast with the Ninth Circuit’s position that the verb is about deceit, not about deprivation of property. The examples all involve a victim of the fraud as the ultimate loser of the property interest.
Yet, given the caution that needs to be taken when relying on intuitions or dictionary definitions, such intuitions and definitions should be examined in larger quantities of actual language use to determine how “ordinary” they actually are. Below, we examine a large set of data in which defraud occurs, in order to test whether our intuitions, those of the linguists who have written about this issue and those of the lexicographers, reflect general usage.
4.3 Corpus analysis of defraud over time
We examined instances of defraud (and its verb-based variants such as “defrauds,” “defrauding,” “defrauded,” etc.) in COHA from the decades of 1870–2000. As noted, 18 U.S.C. § 1344 developed from the mail fraud statute, first enacted in 1877. The current version of the law has been in place, with some minor amendments, since 1984. Thus, to determine whether the meaning of defraud is ordinarily used as the government claims or as Shaw maintains, we reviewed – decade by decade – the use of the word defraud. In all, we found 647 instances in COHA during the period from 1870 through 1999. We sampled half of them in each decade (the even numbered occurrences in each decade) for a total of 321 instances over the thirteen decades. 43 Across all decades, our overall analysis finds: “defraud” virtually always takes a direct object that is the same as the individual (i. e., the victim) suffering a loss of property. The results are summarized below in Figure 1.
As seen in Figure 1, in 98 % of the examples in COHA, the object of fraud was also the victim of fraud. In one instance, the object of the fraud was NOT the victim of the fraud, and in 4 instances, there was not enough contextual information to determine if the object and victim of fraud were the same.
To illustrate, in the following instances, the object and victim are represented by “the other” (God’s children), “a child,” “the rest of the company,” “the bank,” “many banks,” “our fleet,” “the program,” and “any of his wives,” respectively. Dates reflect the decade in which the quoted text occurred:
1840s: But Luque came into the arrangement with more difficulty. “God grant, my children,” exclaimed the ecclesiastic, “that one of you may not defraud the other of his blessing!”
1860s: No miserly parent, no hard master, no careless guardian, can now defraud a child of his right to so much instruction as will make it easy for him to go on instructing himself all his life.
1880s: Betsey prided herself upon her good and proper manners; but the rest of the company, who would have liked to hear the bit of morning news, were now defrauded of that pleasure.
1900s: “Do I understand, sir,” said the officer, taking out his notebook, “that you confess to defrauding the bank of seven thousand dollars?”
1920s: Many banks have been defrauded by the destruction of a canceled check and the submission of an affidavit by the depositor stating no such check was ever issued.
1940s: The good Admiral Vernon, who suffered in the reign of George II for being the sailors’ outspoken friend, declared that our fleets are defrauded by injustice, manned by violence and maintained by cruelty.
1960s: Four administration employes [sic] were arrested on Sept. 13, along with two outsiders, on charges of conspiring to defraud the program of about $240,000 through the issuance of checks to fictitious persons.
1980s: Vigliotto, who is in his early 50s, faces up to 34 years in prison. Although he conceded making the multiple marriages (and adopting some 50 aliases), he denied defrauding any of his wives.
These examples demonstrate that both animate and inanimate objects can be the victim of fraud. In the case of inanimate objects, such as “the bank,” “the fleet,” or “the program,” it is understood that the entities either represent a group of people (“the fleet” is a metonym for the group of people who make up the fleet) or an entity that has legal status (“the bank,” “the program”).
There were four examples that did not contain enough information to determine if the object of the fraud was also the victim of the fraud. In these cases, while an object and victim of the fraud are implied, it is not clear who or what that entity would be. For example:
1880s: Men loved women in those days, as now. God knows how they dared be fathers, but they had babies as sweet, no doubt, to them as ours to us, whom they must feed, clothe, educate. The gentlest creatures are fierce when they have young to provide for, and in that wolfish society the struggle for bread borrowed a peculiar desperation from the tenderest sentiments. For the sake of those dependent on him, a man might not choose, but must plunge into the foul fight, – cheat, overreach, supplant, defraud, buy below worth and sell above, break down the business by which his neighbor fed his young ones, tempt men to buy what they ought not and to sell what they should not, grind his laborers, sweat his debtors, cozen his creditors.
1910s: Then the Cella Commission Company, the biggest bucket-shop chain ever conducted, E. S. Boggs &; Co. of New York and Price &; Co. of Baltimore, had offices in the District of Columbia. The managers of these offices were arrested for violation of the Federal law, and their principals and associates outside the District were charged with conspiracy to violate that law. In the present proceedings the Federal authorities rely on charges of using the mails to defraud and conspiracy to do so. According to the New York Stock Exchange authorities, there are now in operation only two bucket shops east of the Mississippi – one in Pennsylvania and the other in West Virginia.
1910s: also he had figured out to his own satisfaction the exact method by which the land-grabber was enabled to grab; or, provided the grabber did not care to retain his grab, how he could nevertheless derive tremendous profits from his control of certain officials in the State Land Office. Therefore, after his day spent in the public law library in San Francisco, Bob’s brain was primed with every detail of the land laws, and had confirmed his original interpretation of the land-grabbers’ clever schemes to defraud. However, not satisfied with his own opinion, he decided to seek a little expert advice on the subject, and to that end he went the following morning to his father’s old friend and his own former employer, Homer Dunstan, the corporation attorney, whom he knew to be an authority on land law.
1970s: Why, do you suppose? Because he is ourselves. We are sorry for him. (It is terrfying [sic] but true. In the world of the artists lies the answer to this paradox also.) We are starved, we are thirsty for that which we can not attain. We know money is not the thing, we have been told so until we want to puke at mention of it again. But what else have we to strive for? So we still defraud and murder, to say nothing of slaving our hearts out in the name of “holy” work? for money!? thinking it may yet buy us what we want. (But we do very little to supply by other means the things money can not buy. In the world of the artist lies the answer.) Do you blame the artist, then, if he is arrogant? in the manner of Rembrandt? It is because he knows.
Finally, there was one example in which the object of the fraud was not the victim of the fraud. In this case, it is the “laws” and not an entity that represents or is run by animate objects – those that can be affected by fraudulent behavior – that is defrauded.
1900s: The meaner subordinates of Fritz Braun’s crime were all easily disposed of, for both Lilienthal and Timmins were now serving long sentences for defrauding the United States customs laws.
This provides the only example in COHA where the object of the fraud and the victim of the fraud are two separate entities.
Although we suggested that caution be used when relying on intuition and dictionary definitions to determine subtle nuances of meaning in legal contexts, in this instance all three methods lead to the same conclusion: the verb defraud is almost always used to describe a situation in which the object of the fraud suffered the loss in the end. While both the linguists and the dictionaries refer to a loss of “possessory deprivation” or “possession,” leaving some room for ambiguity, all of the illustrations are of instances in which this deprivation caused the object to bear the loss ultimately.
The corpus analysis is most telling: defraud is primarily used (in 98 % of the instances) with an object that is the same as the victim of the fraud. However, given the fact that some examples include explicit references to the object/victim vs. implicit references, inferences must be drawn in the latter cases that may result in some confusion as to who the actual victim of the fraud is. This may affect individual intuition about the way in which defraud can be used.
Of the 282 examples in COHA where the object of the fraud was the same as the victim of the fraud, 89 % (252 total) explicitly named the object/victim of the fraud within the surrounding context. In 11 % of the instances (30 total), however, the object/victim of the fraud was implicitly present. These examples are different than those presented earlier that have unclear objects/victims based on the given context. For example:
1950s: In 86 minutes and twelve arrests they cracked the international racket that, by securing advance answers to the contests, swindled U.S. newspapers for more than a year. The transcontinental swoop bagged two key figures in Detroit: Walter Rex Johnston, 30, part-time car salesman whom the FBI identified as chief architect and brains of the swindle ring, and a key Johnston lieutenant, Harry H. Balk, 33, theatrical booking agent. Two Canadians who managed the flow of puzzle information were accused of using the mails and long-distance telephone to defraud, but were not arrested; the crimes are not extraditable.
1960s: According to the Department of Health, Education and Welfare, both Burton and Belin Colleges are “degree mills,” organizations that award degrees without requiring students to meet the standards necessary to win a degree at an accredited institution. In 1959, Dr. Clyde Belin, the head of the latter college, was sentenced to a one-year term in federal prison on six counts of using the mails to defraud.
1970s: The wheel locked, and the we skidded for nearly 1,500 feet before corn; to a halt. The plane was jacked up and the [wh]eel removed. The fused parts within the brake [had] to be pried apart. [T]hat evening I left work early and went to see [my] attorney. After I told him the story, he a[d]ded that, while I was probably not actually [guilty of fraud, I was certainly part of a conspiracy to defraud.
In these examples, because the objects and victims of the fraud are not explicitly stated, they need to be inferred by the reader. Here, they are implied to be U.S. newspapers and their readers, students of said colleges, and passengers of the airplane, respectively. However, as previously noted, since speakers of a language tend to notice unusual features rather than the usual ones (Biber et al. 1998) it is possible that the implications drawn by some speakers when presented with implicit objects is that the object and the victim may be different.
In this article we have attempted first to demonstrate that there is no agreed upon method for determining the “ordinary meaning” of a statutory term for purposes of statutory interpretation. Rather, courts use three different approaches: intuition, dictionary definitions, and corpus analysis, sometimes leading to different results depending upon the method on which a particular court relies. For that matter, because both intuitions and dictionary definitions differ from one another at the margins, the courts fail to achieve unanimity even within a particular approach to discovering ordinary meaning. We put aside the issue of the defeasibility of the ordinary meaning approach to statutory interpretation, recognizing that it is likely to be part of the analysis in most cases, even when courts turn to different evidence.
Because ordinary meaning is a fact about the distribution of usage, we have focused much of this article on the use of corpora, since they provide empirical evidence of distributional facts. Yet, we express some caution even in this realm because the small number of cases in which courts have actually turned to the use of corpora demonstrates that corpora provide detailed data, but do not replace the legal analyst’s obligation to ask the right questions in exploiting that data. Thus, courts are not always in agreement about the proper outcome of a case even after analyzing the ordinary meaning of language using a corpus.
Finally, as a case study, we applied all three methods to analyze a case currently pending before the Supreme Court of the United States – Shaw v. United States. Our analysis confirms the conclusions about the object of defraud from the previously cited linguistic intuitions about grammaticality judgements and reputable dictionary definitions about nuances in meaning and allowable argument structure. Our COHA concordance results demonstrate that the object of defraud should be interpreted not only as the party that was tricked, but also as the victim of the fraud. Given such strong evidence of common usage, that is how the Supreme Court should interpret the statute when the opportunity arises.
The authors wish to thank James Brudney, William Eskridge, and Gordon Smith for their valuable comments on an earlier draft and to express their gratitude to Elyssa Abuhoff and Benjamin Levites for their excellent work as research assistants on this project.
The American Heritage Dictionary of the English Language. 2016. Houghton Mifflin Harcourt. https://ahdictionary.com/word/search.html?q=defraud (accessed 9 August 2016).
Aprill, E. P. 1998. The law of the word: Dictionary shopping in the Supreme Court. Arizona State Law Journal 30. 275–306. Google Scholar
Baker, P. 2006. Using corpora in discourse analysis. London: Continuum. Google Scholar
Baker, P. 2010. Sociolinguistics and corpus linguistics. Edinburgh: Edinburgh University Press. Google Scholar
Biber, D, S. Conrad & R. Reppen. 1998. Corpus linguistics: Investigating language, structure, and use. Cambridge: Cambridge University Press. Google Scholar
Brudney, J. & L. Baum. 2013. Oasis or mirage: The supreme court’s thirst for dictionaries in the Rehnquist and Roberts eras. William and Mary Law Review 55. 483–580. Google Scholar
Channell, J. 2000. Corpus-based analysis of evaluative lexis. In S. Hunston & G. Thompson (eds.), Evaluation in text: Authorial stance and the construction of discourse, 38–55. Oxford: Oxford University Press. Google Scholar
Eskridge, W. N., Jr. 2016. Interpreting law: A primer on how to read statutes and the constitution. St. Paul, MN: Foundation Press. Google Scholar
Fillmore, C. 1968. Lexical entries for verbs. Foundations of Language 4(4). 373–393. Google Scholar
Freidin, R. 2012. Syntax: Basic concepts and applications. Cambridge: Cambridge University Press. Google Scholar
Gove P. B. (ed.). 1961. Webster’s third new international dictionary. Springfield, MA: Merriam-Webster. Google Scholar
Halliday, M. A. K. 2004. Lexicology. In M. A. K. Halliday, W. Teubert, C. Yallop & A. Čermáková (eds.), Lexicology and corpus linguistics: An introduction, 1–22. New York: Continuum. Google Scholar
Hunston, S. & G. Thompson. 2000. Corpus-based analysis of evaluative lexis: Editors’ introduction. In S. Hunston & G. Thompson (eds.), Evaluation in text: Authorial stance and the construction of discourse, 38–39. Oxford: Oxford University Press. Google Scholar
Kirchmeier, J. & S. Thumma. 2010. Scaling the lexicon fortress: The United States supreme court’s use of dictionaries in the twenty-first century. Marquette Law Review 94. 77–259. Google Scholar
Levin, B. 1993. English verb classes and alternations: A preliminary investigation. Chicago: University of Chicago Press. Google Scholar
Levin, B. & M. R. Hovav. 1991. Wiping the slate clean: A lexical semantic exploration. Cognition 41. 123–151. Google Scholar
Mahowald, K., J. Hartman, P. Graff & E. Gibson. 2016. SNAP judgements: A small N acceptability paradigm (SNAP) for linguistic acceptability judgements, Language 92. 619–635. Google Scholar
McEnery, T., R. Xiao & Y. Tono. 2006. Corpus-based language studies: An advanced resource book. New York: Routledge. Google Scholar
Mouritsen, S. C. 2010. Note, the dictionary is not a fortress: Definitional fallacies and a corpus-based approach to plain meaning. Brigham Young University Law Review 2010. 1915–1978. Google Scholar
Mouritsen, S. C. 2011. Hard cases and hard data: Assessing corpus linguistics as an empirical path to plain meaning. Columbia Science and Technology Law Review 13. 156–205. Google Scholar
Nickerson, R. S. 1998. Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology 2(2). 175–220. Google Scholar
Oxford Dictionaries. (n.d.). Using the corpus. Oxford: Oxford University Press. http://www.oxforddictionaries.com/us/words/using-the-corpus (accessed 9 August 2016).
Risinger, D. M., M. J. Saks, W. C. Thompson & R. Rosenthal. 2002. The Daubert/Kumho implications of observer effects in forensic science: Hidden problems of expectation and suggestion. California Law Review 90(1). 1–56. Google Scholar
Scalia, A. & B. Garner. 2012. Reading law: The interpretation of legal texts. St Paul, MN: Thomson/West. Google Scholar
Slocum, B. G. 2015. Ordinary meaning: A theory of the most fundamental principle of legal interpretation. Chicago: University of Chicago Press. Google Scholar
Solan, L. M. 2003. Finding ordinary meaning in the dictionary. In M. Robinson (ed.), Language and the law: Proceedings of a conference, 255–278. Buffalo, NY: William S. Hein & Co. Google Scholar
Solan, L. M. 2010. The language of statutes: Laws and their interpretation. Chicago: University of Chicago Press. Google Scholar
Solan, L. M., T. Rosenblatt & D. Osherson. 2008. False consensus bias in contract interpretation. Columbia Law Review 108. 1268–1300. Google Scholar
Sprouse, J. & D. Almeida. 2012. Assessing the reliability of textbook data in syntax: Adger’s core syntax. Journal of Linguistics 48. 609–652. Google Scholar
Sprouse, J., C. T. Schütze & D. Almeida. 2013. A comparison of informal and formal acceptability judgements using a random sample from linguistic inquiry 2001–2010. Lingua 134. 219–248. Google Scholar
Yallop, C. 2004. Words and meaning. In M. A. K. Halliday, W. Teubert, C. Yallop & A. Čermáková (eds.), Lexicology and corpus linguistics: An introduction, 23–72. New York: Continuum. Google Scholar
For a recent example of the application of this principle by the U.S Supreme court, see Bond v. United States, 134 S. Ct. 2007 (2014); there are disagreements, however, over how defeasible ordinary meaning analysis is in light of specific evidence of a broader or narrower intended meaning. See, e. g., Chisom v. Roemer, 501 U.S. 380 (1991) (majority holding that the word “representative” in the Voting Rights Act applies to the election of judges as well as legislators based on history and purpose of the statute, Justice Scalia dissenting based on ordinary meaning rule). These are important issues, but we do not pursue them further here.
We use the linguistic convention of placing a “*” before a sentence to reflect that it is not grammatical and a “?” before a sentence to reflect that it is possibly grammatical to some speakers, but doubtfully so.
666 F.3d 1040, 1044 (T.O 2012) (summarizing literature critical of judicial reliance on dictionaries to ascertain ordinary meaning, focusing on the gap between the context-sensitive use of words, and the acontextual nature of dictionary definitions).
Many earlier corpora were synchronic, meaning that they only collected samples of language from one time period. Such corpora are considered to be static (i. e., representative of only that one time period). Diachronic corpora, such as COCA and COHA, are considered monitor corpora, which sample the same genres over time in order to provide examples of how language may change over time.
For excellent discussion of the advantages of corpus analysis over dictionaries in determining ordinary meaning in statutory interpretation, see Mouritsen (2011) (crediting amicus brief filed by Neal Goldfarb on behalf of the Project for Government Oversight for influencing the Supreme Court’s decision in FCC v. AT&T, Inc., 562 U.S. 397 (2011) based on corpus analysis of the word “personal,” demonstrating that it does not apply to corporations in ordinary speech); Mouritsen (2010) (discussing a more sophisticated corpus-based approach to the issue in Muscarello v. United States).
Id. at 129–30. The Court described the results: “We found thousands of such sentences, and random sampling suggests that many, perhaps more than one-third, are sentences used to convey the meaning at issue here, i. e., the carrying of guns in a car.” We do not endorse Justice Breyer’s “perhaps more than one-third” standard.
Justice Lee’s analysis provoked a response from the majority, criticizing him for engaging in expert analysis outside the bounds of the adversarial system. The issue was not the legitimacy of his argument, but the legitimacy of judges taking the initiative to engage in scientific investigation without the parties being afforded the opportunity to cross-examine or otherwise challenge the analysis. Id. at 1264–66. Because COCA is publicly available and the results of searches transparent, we are not as bothered by Justice Lee’s foray as was the majority. This is particularly the case, because if Justice Lee had limited his inquiry to a Google search there would have been little or no controversy, even though the analysis would be less transparent and less likely to reflect the reality of ordinary usage. Nonetheless, our point in this paper goes to the merits of the methodology rather than to such questions of procedure.
United States v. Morganfield, F. 3d 453 (5th Cir. 2007); United States v. Leahy, 445 F.3d 634 (3d Cir. 2006); United States v. Brandon, 298 F. 3d 307 (4th Cir. 2002); United States v. Thomas, 315 F.3d 190 (3d Cir. 2002); United States v. Moran, 312 F. 3d 480 (1st Cir. 2002); United States v. De La Mata, 266 F. 3d 1275 (11th Cir. 2001); United States v. Higgins, 270 F. 3d 1070 (7th Cir. 2001); United States v. Rodriguez, 140 F. 3d 163 (2nd Cir. 1998); United States v. Sapp, 53 F. 3d 1100 (10th Cir. 1995); United States v. Brandon, 17 F. 3d 409 (1st Cir. 1994); United States v. Young, 952 F. 2d 1252 (10th Cir. 1991); United States v. Briggs, 930 F. 2d 222 (5th Cir. 1991); United States v. Hooten, 933 F.2d 293 (5th Cir. 1991); United States v. Hubbard, 889 F. 2d 277 (D.C. Cir. 1989).
United States v. Wolfswinkel, 44 F.3d 782, 786 (9th Cir. 1995) (holding that the bank’s administrative expenses in completing the transaction were intended and adequate to meet the statute’s requirements). See United States v. Bonallo, 858 F.2d 1427, 1429–30, 1430 n.2 (9th Cir. 1988).
Fillmore (1968) provides a discussion of the argument structure of “rob” and “steal” in this light.
In earlier work, Levin and Hovav (1991) refer to “defraud” as a “deprive” verb. One can “wipe the chalk from the blackboard” or “wipe the blackboard clean of chalk.” “Defraud” and related verbs have no such alternation, however. You cannot “defraud $40,000 from the bank” although you can “defraud the bank of $40,000.”
The same holds true in English for generic statements, which are expressed in the simple present tense. One can say “Bill steals,” which means that Bill does so habitually. Regardless of the absence of any object or a description of anything stolen, we understand these to remain part of the concept.
Two research assistants coded the examples separately. They then compared their results and determined whether they could reach agreement on the disputed ones. Of the 321 instances in COHA that were examined, after discussion, there were 287 that were agreed upon in terms of coding, resulting in an interrater reliability of 89 % between two coders. Only those examples that were agreed upon are included in the results in Figure 1.
About the article
Published Online: 2016-12-08
Published in Print: 2016-12-01