In my post Three syntactic canons, I discussed the three canons of interpretation in Scalia and Garner’s Reading Law that deal with syntactic ambiguities:
Last-Antecedent Canon. A pronoun, relative pronoun, or demonstrative adjective generally refers to the nearest reasonable antecedent.
Series-Qualifier Canon. When there is a straightforward, parallel construction that involves all nouns or verbs in a series, a prepositive or postpositive modifier normally applies to the entire series.
Nearest-Reasonable-Referent Canon. When the syntax involves something other than a parallel series of nouns or verbs, a prepositive or postpositive modifier normally applies only to the nearest reasonable referent.
At the end of the post, I raised the question whether the preferences stated in these canons can be justified on purely linguistic grounds—i.e., whether they represent “valid generalization[s] about how expressions in English are understood[.]” I’m going to try to answer that question here.
This post is a long one, so let me say up front that the verdict on these canons is a mixed one. The good news is that in some categories of cases where these canons apply, there is in fact a linguistic justification for applying them, by which I mean that in those applications the canons more or less parallel one of the recognized tendencies in how people process and comprehend sentences. (The same parallel applies to the rule of the last antecedent as it’s generally applied by the courts, which differs somewhat from the Scalia–Garner version).
Now the bad news. First, the processing tendency that I’ve referred to is just one of the factors that influences how a given sentence is understood, and other factors can exert an influence in a different direction. So even in cases where the canons are consistent with this processing tendency, it’s hard to say in advance whether applying the canons would result in a linguistically justified interpretation.
Second, in the canons’ other applications, I don’t think there is the same parallel between the canons and the kinds of processing tendencies that I’ve referred to above. And more than that, in a large subset of those applications, the canons are actually inconsistent with what the literature on processing tendencies would predict. In those applications, therefore, there is reason to think that applying the canons would result in interpretations that are linguistically unjustified.
On to the details.
The rule of the last antecedent (non-Scalia/Garner version). As I’ve said previously, the Last-Antecedent Canon in Reading Law differs in its coverage from the rule of the last antecedent as it is generally applied today. So before discussing the Scalia/Garner version, I’m going to deal with the rule of the last antecedent as generally applied. But in doing so, I’m going to rephrase the rule in terms that are more precise than the formulations that the courts typically use. My reformulation covers what I think are the prototypical applications of the rule of the last antecedent; while there are undoubtedly some cases that apply the rule but don’t fit the description, they are outliers.)
1. The rule of the last antecedent applies to the interpretation of phrases in which a modifier appears after the word(s) or phrase(s) that it modifies (men and women in glass houses as opposed to transparently-housed men and women).
2. When the rule of the last antecedent applies, it states a preference for reading the modifier as modifying the item that immediately precedes it (the first diagram below), but that preference may be overridden by other other factors.
Preferred by the rule of the last antecedent:
Dispreferred by the rule of the last antecedent:
So is this default rule of interpretation linguistically justifiable? The answer is a qualified “yes.”
Before I explain that answer, a note is in order about the terminology I’ll be using. When linguists use tree diagrams like the ones above, they typically describe the relationship of the various parts of a phrase in terms of how the part of the diagram representing the phrase is “attached” to the portion of the tree that precedes it. When there are multiple potential attachment sites, the expression is syntactically ambiguous, with each possible structure typically corresponding to a different interpretation. It is common to distinguish between the potential attachment sites in terms of whether they are high or low on the tree diagram Thus, in diagram (a), the phrase in glass houses attaches low in the diagram to the noun women, while in (b) it attaches higher in the diagram to the larger phrase men and women. Low attachment corresponds to the interpretation called for by the rule of the last antecedent. This can be seen by the fact that the modifier in glass houses combines with women to form the noun phrase women in glass houses, which then combines with men to form the larger noun phrase [[men] and [women in glass houses].
In the course of comprehending an ambiguous expression, the brain has to somehow settle on one of the possible structures as the right one, thereby resolving the ambiguity. The process by which that happens has been the subject of extensive research in psycholinguistics; a fairly recent overview of this work can be found here. It appears from this research that in resolving syntactic ambiguities, there is a tendency toward low attachment. And as was first noted by Larry Solan in his book The Language of Judges, this tendency parallels the rule of the last antecedent.
I discussed this parallel in my amicus brief in United States v. Hayes. The brief described the rule of the last antecedent as “an example of a principle of legal interpretation that has a solid linguistic basis” and as “an instance in which the law anticipated later developments in linguistics.” But on reflection, and after further reading in the literature, that strikes me as having been a overly exuberant. Because, here’s the thing: The low-attachment tendency is only one of the factors that influences the way sentences are understood. Other factors are at work as well, such as the context in which the sentence appears, the semantic properties of the words in the sentence, and how the sentence is punctuated. In any given case, those factors may push toward a different interpretation, and in some case they will overcome the tendency toward low attachment.
In trying to figure out whether the rule of the last antecedent makes linguistic sense, it seems to me now that it’s not enough simply to point to the parallel with the low-attachment tendency. I think it’s necessary to ask whether the application of the rule will result in the most natural interpretation in a majority of cases. In other words, is the rule a more accurate guide for decisionmaking than the alternatives?
That’s a question that I can’t even begin to answer given what I currently know. It would be necessary to select a sufficiently large sample of statutes that involve the appropriate kind of syntactic ambiguity (how big a sample? beats me) and then analyze each statute in an attempt to arrive at the interpretation that seemed the most natural. The amount of work required would be enormous, and it’s not at all clear that the results of the analysis would be reliable.
So I’m afraid that the linguistic validity of the rule of the last antecedent will have to remain shrouded in uncertainty. The most that I’m prepared to say is that the reliability of the rule in any given case is likely to increase in proportion to the distance between the modifier and the high-attachment site. The longer that distance, the greater the likelihood of a low-attachment interpretation.
Reading Law: The Last-Antecedent Canon
A pronoun, relative pronoun, or demonstrative adjective generally refers to the nearest reasonable antecedent.
Although this canon is Scalia and Garner’s version of the rule of the last antecedent, I can’t simply say, “See the discussion above.” That’s because the coverage of this canon differs from that of the rule of the last antecedent. The rule of the last antecedent (as I’ve formulated it) covers cases involving the semantic relation of modification, while the coverage of the Scalia/Garner canon is defined with reference to specified lexical categories (i.e., parts of speech). That’s significant because expressions using those categories—pronouns, relative pronouns, or demonstrative adjectives—don’t necessarily involve modification, and therefore don’t necessarily present the kind of syntactic ambiguity that implicates the tendency toward low attachment.
Consider, for example, the following provision of the Constitution, which Scalia and Garner use to illustrate this canon:
In Case of the Removal of the President from Office, or of his Death, Resignation or Inability to discharge the Duties of the said Office, the Same shall devolve on the Vice President.
When President William Henry Harrison died in office, the question arose whether his vice president, John Tyler, became the new president or instead merely assumed the president’s duties, while remaining vice president. The interpretive question has to do with the clause, “the Same shall devolve on the Vice President.” As Scalia and Garner put it, “[W]hat is the antecedent of the legalistic pronoun same?”
That issue has nothing to do with modification or with tendencies either toward or away from low attachment. The uncertainty about the antecedent of “the Same” has no implications for the sentence’s syntactic structure, which is the same under either interpretation. The low-attachment tendency that’s discussed above is therefore irrelevant; the process by which pronouns are matched up with their antecedents are different than those by which grammatical structure is parsed.
Where does that leave us? I’m less familiar with the research into the resolution of ambiguities in pronoun reference than I am with the research about attachment ambiguities. However, my impression is that the picture regarding pronoun-reference ambiguities is even muddier than it is regarding attachment ambiguities. This is an area in which interpretations are highly dependent on context. And although the recency of a potential antecedent may play a role, other factors are probably more important, such as the relative prominence of each potential antecedent in the text.
In many cases, for example, there appears to be a tendency to interpret pronouns as referring to potential antecedents that appear as the subject of the sentence or (what is often the same thing) that appear in the sentence before the other potential antecedents. These tendencies will generally favor an interpretation inconsistent with what would be favored by the Last-Antecedent Canon.
Interestingly, Scalia and Garner recognize this tendency: “The last-antecedent canon may be superseded by another grammatical convention: A pronoun that is the subject of a sentence and does not have an antecedent in that sentence ordinarily refers to the subject of the preceding sentence.” (146.) Yet that tendency operates not only across sentence boundaries but also within a single sentence.
One last point before moving on. Scalia and Garner argue that the Last-Antecedent Canon is “the legal expression of a commonsense principle of grammar,” which they describe as follows (quoting from Robert Burchfield’s revision of Fowler’s Modern English Usage):
“It is clearly desirable that an anaphoric (backward-looking) or cataphoric (forward-looking) pronoun should be placed as near as the construction allows to the noun or noun phrase to which it refers, and in such a manner that there is no risk of ambiguity.”
But that’s not a “principle of grammar,” it’s a piece of advice about how to write. Such advice, as I’ve said before, isn’t a reliable source of guidance on how to go about the process of interpretation. (And by the way, Burchfield’s statement refutes itself. It contains an anaphoric pronoun (to which it refers), the antecedent of which is the word pronoun. If Burchfield had followed his own advice, he would have written, [a] pronoun should be placed as near to the noun or noun phrase to which it refers as the construction allows to the noun or noun phrase to which it refers.)
Reading Law: The Series-Qualifier Canon and the Nearest-Reasonable Referent Canon
Series-Qualifier Canon. When there is a straightforward, parallel construction that involves all nouns or verbs in a series, a prepositive or postpositive modifier normally applies to the entire series.
Nearest-Reasonable-Referent Canon. When the syntax involves something other than a parallel series of nouns or verbs, a prepositive or postpositive modifier normally applies only to the nearest reasonable referent.
I’m going to discuss these canons together because they are in a sense mirror-images of each other. On the one hand, each one applies to both premodification (where the modifier precedes the item it modifies) and postmodification (where the modifier follows the modified item). But on the other hand, they differ in the scope that they impute to the modifier: the Series-Qualifier Canon says that the modifier “normally applies to the entire series,” while the Nearest-Reasonable-Referent says that it applies “only to the nearest reasonable referent.”
Let’s start with the Nearest-Reasonable-Referent. With respect to postmodification, this canon parallels the tendency toward low attachment that I talked about in connection with the rule of the last antecedent. This is hardly surprising. The Nearest-Reasonable-Referent Canon as formulated by Scalia and Garner was carved out from the rule of the last antecedent. And in a tree diagram, the way to show that a word or phrase modifies only “the nearest reasonable [preceding] referent” is to show it as attaching lower in the tree than if it modified all the reasonable preceding referents.
To illustrate this canon in action, Scalia and Garner discuss a statute providing that a debtor in a Chapter 13 bankruptcy was not entitled to have his debts discharged if he had “received a discharge…in a case filed under Chapter 7…during the 4 year period preceding” the filing of the Chapter 13 petition. This language is ambiguous as to whether the Chapter 13 discharge was barred (1) if the debtor had received a Chapter 7 discharge within the four-year period, or instead (2) if the Chapter 7 case in which he had received a discharge had been filed within the four-year period.
The two interpretations correspond to the following tree diagrams, the second one of which features low attachment of the phrase during the 4-year period…, and is the interpretation favored by the Nearest-Reasonable-Referent canon.
High attachment; interpretation 1 (received discharge during the 4-year period). Disfavored by the Nearest-Reasonable Referent Canon; inconsistent with processing tendency toward low attachment:
Low attachment; interpretation 2 (Ch. 7 filed during the 4-year period). Favored by the Nearest-Reasonable Referent Canon; consistent with processing tendency toward low attachment:
So that’s how the Nearest-Reasonable-Referent Canon works with respect to postmodification. What about cases involving premodification?
In such cases, the effect of applying the canon is precisely the opposite of its effect with respect to postmodfication. Rather than preferring low-attachment interpretations, it prefers interpretations corresponding to high attachment—a preference that runs counter to the processing tendency toward low attachment.
This can be seen in statutory language at issue in Liparota v. United States, which I’ve discussed previously: knowingly uses…food stamps in a manner not authorized by law. The premodifier is knowingly, and the question is whether it modifies only uses food stamps or instead modifies the larger phrase uses food stamps in a manner not authorized by law.
High attachment; knowingly modifies uses…food stamps. Favored by Nearest-Reasonable Referent Canon. Inconsistent with processing tendency toward low attachment:
Low attachment; knowingly modifies uses…food stamps in a manner not authorized by law. Disfavored by Nearest-Reasonable Referent Canon. Consistent with processing tendency toward low attachment:
“Wait a minute,” you may be thinking, “In the first diagram, where only the nearest referent is modified, the initial modifier (knowingly) is lower in the tree than in the second diagram, where both referents are modified.” And that’s correct, but the question isn’t how knowingly is attached to the tree, it’s how the phrase in a manner not authorized by law is attached. This is because the focus in dealing with syntactic ambiguities is always on the right edge of the tree.
Here’s why. When we read or hear a sentence, the input comes one word at a time. But the mental process of interpreting the sentence isn’t delayed until the sentence is complete. Rather, it starts immediately, and an interpretation is built up piece by piece. As a result, each incoming word has to be integrated into a partial interpretation that has already been formed. In the context here, with different possible interpretations being represented by tree diagrams, this is like saying that a tree diagram is being constructed a word at a time, and that each new word has to be attached somehow to the diagram. And because English is written from left to right, incoming items appear on the right edge of the tree. Thus, the right edge is where the action is, and—as I said above—the question for the phrase old men and women is how the word women attaches to the tree, not the word old.
As applied to premodification, therefore, the Nearest-Reasonable-Referent Canon states a preference for high attachment—a preference that runs counter to the mental tendency toward low attachment. Since I’m agnostic about how much that tendency supports the rule of the last antecedent and the NRR Canon as applied to postmodification, I’m not about to point to the tendency as a reason to say the NRR canon is unjustified as to premodification. But if you do want to rely on the low-attachment tendency as justifying the canon with respect to premodification, you’re stuck with having it undermines the canon’s validity as to premodification.
Having slogged through that long discussion of the Nearest-Reasonable-Referent Canon, we can deal with the Series-Qualifier Canon much more quickly. Everything just said about the former applies equally to the latter, except that it applies backwards.
Whereas the Nearest-Reasonable-Referent Canon is consistent with the low-attachment tendency in cases when the canon is applied to postmodification but not pre-, the Series-Qualifer Canon is consistent with that tendency when it is applied to premodification but not post-. This is because the two canons differ in the scope that they assign to the modifier. NRR assigns the modifier a narrow scope, interpreting it as applying only to the nearest referent, while SQR assigns the modifier a wide scope, interpreting as applying to the entire series of potential referents. As a result, the structural effect of applying the canon is reversed:
Premodification
forcibly modifies assault. Later items in the series are attached high. Disfavored by Series-Qualifier Canon. Inconsistent with processing tendency toward low attachment:
forcibly modifies assault, resist, or impede. All items in the series are attached low. Favored by Series-Qualifier Canon. Consistent with processing tendency toward low attachment:
Postmodification
registered in Delaware modifies corporations and partnerships. High attachment. Favored by Series-Qualifier Canon. Inconsistent with processing tendency toward low attachment.
registered in Delaware modifies partnerships. Low attachment.
Disfavored by Series-Qualifier Canon. Consistent with processing tendency toward low attachment:
Thus, the Series-Qualifier-Canon is consistent with the tendency toward low attachment when the canon is applied to cases involving premodification, but inconsistent with that tendency when it’s applied to cases involving postmodification.
Pingback: Koncision » With Syntactic Ambiguity, Avoiding the Accident Spares You the Autopsy