An introduction and guide to my series of posts “Corpora and the Second Amendment” is available here. The corpus data that is discussed can be downloaded here. That link will take you to a shared folder in Dropbox. Important: Use the “Download” button at the top right of the screen.
In my last post (longer ago than I care to admit), I offered a very brief introduction to corpus analysis and used corpus data on the word keep as the raw material for a demonstration of corpus analysis in action. One of my reasons for doing that was to talk about the approach to word meaning that I think is appropriate when using corpus linguistics in legal interpretation.
That approach, which is described in my article “A Lawyer’s Introduction to Meaning in the Framework of Corpus Linguistics” (pdf), is based on work in corpus-based lexicography, and it provides a way of dealing with puzzling fact about language. Many words have multiple possible meanings when they are considered in isolation, but when used in a sentence they are typically unambiguous. The basic insight that grew out of lexicographic corpus analysis is that when a word is used in a given context, what is generally thought of as the meaning of the word in that context is often more appropriately regarded as the meaning of a larger unit consisting of the word together with certain elements of the accompanying text. And it turns out that it is often possible to identify the kinds of contextual elements that are associated with particular meanings.
This has implications for the use of corpus linguistics in legal interpretation, because it can provide guidance in querying the corpus and then sifting and analyzing the data. The key is to look for concordance lines in which the relevant word is used in a context that is grammatically and semantically similar to the context in the legal provision at issue. (For an idea of the kind of similarity I’m talking about, see my analysis of Muscarello v. United States and my post “The semantics of sleeping in railway stations.”)
The approach that I’ve described also has implications for this reexamination of District of Columbia v. Heller. Those implications arise from the fact that at a key point in the decision, the Supreme Court relied on the fact that when bear arms is used to denote activities such as serving as a soldier and fighting in a war, its use is idiomatic. And the approach I’m following provides a way of thinking about idiomaticity that differs from the traditional view of the phenomenon—which was the view underlying this aspect of Heller.
What has generally been regarded as the defining characteristic of an idiom is that the meaning of the expression as a whole can’t be derived compositionally. In other words, the expression’s meaning isn’t simply a function of the meaning of its parts and of the way that they are syntactically combined. The classic example of an idiom is kick the bucket ‘die’, which is semantically almost completely opaque. The conventional meaning of kick can’t be combined with conventional the meaning of the bucket in any way that yields the meaning ‘die.’ The meaning of the phrase therefore has to be attributed to the phrase as an indivisible whole.
Idiomaticity and compositionality have traditionally been seen as being mutually exclusive. But over the past several decades, that view has been challenged by the argument (for which there is much evidence) that the relationship between compositionality and idiomaticity is best regarded as a continuum, and that individual constructions can be compositional in some respects and idiomatic in others.
For example, it turns out that even in kick the bucket, the individual meaning of one of the words plays a role in the expression’s semantics. Or at least it does if you share the intuition that it’s more appropriate to use kick the bucket to refer to a death that is sudden rather than to one resulting from a protracted illness. That difference in acceptability can be attributed to the verb kick, which denotes a type of event that is inherently sudden. Thus, while kick the bucket is highly idiomatic, its idiomaticity is diluted with a little bit of compositionality.
At the other end of the continuum are expressions in that are predominantly compositional but reflect a degree of idiomaticity, in that the meaning of a particular word in the expression is determined in part by some other aspect of the expression—other words in the expression, for example, or the expression’s grammatical structure. And it’s that end of the continuum that this post is going to explore. In particular, I’ll be building on my last post by using what we’ve learned from the corpus data about keep to discuss some of the ways in which the meaning of a word in a particular context is affected by that context. And in doing that, I will be laying the groundwork for when I deal with the issue of idiomaticity in the context of Heller, several posts from now.
THE USAGES IN THE CORPUS DATA I’VE DISCUSSED vary in the extent to which the meaning of the individual words is dependent on the context. At the high-compositionality end of the spectrum are constructions following the patterns keep+Verb and keep+Adjective. As I said in my last post, the semantics of these patterns are similar. So in the interest of brevity (which is in short-enough supply as it is), I’ll limit the discussion here to the keep+Verb constructions.
As you’ll recall, those constructions fall into two groups, with some taking the form keep VERBing, and others taking the form keep NOUN VERBing, or its passive-voice counterpart NOUN was kept VERBing. (In the last post, the constructions in these groups are referred to as “intransitive” and “transitive,” respectively, but I’ve realized that those descriptions aren’t appropriate. Therefore, in this post I’ll stick with keepVERBing and keep NOUN VERBing.)
What these two groups of constructions have in common is that their meanings reflect the element of CONTINUATION. In the case of keep VERBing, that element provides pretty much the entirety of what keep contributes to the construction’s meaning. Given that CONTINUATION is as far as I can tell the only semantic element that keep contributes to every construction it appears in, it seems reasonable to assume that it provides the word’s semantic core. And on that assumption, the keep VERBing constructions are for all intents and purposes fully compositional.
In the keep NOUN VERBing constructions, on the other hand, the semantic of keep is more complex, consisting, not of element CONTINUATION, but of the element CAUSATION OF CONTINUATION. To keep stirring a pot is to continue stirring it, but to keep a pot boiling is to cause the pot (more specifically, its contents) to continue boiling.
The correlation between the difference in grammatical structure and the difference in meaning is not coincidental; the grammatical difference reflects a difference in the semantic structure of the event that the sentence denotes.
In Bob kept stirring the pot, the word Bob functions as the explicit grammatical subject of kept, and Bob (the person) is therefore understood to be the actor who is responsible for the continuation of the action of stirring. And at the same time, Bob (the word) functions implicitly as the grammatical subject of stirring, because Bob (the person) is understood to be the actor who is performing the action of stirring. In Bob kept the pot boiling, on the other hand, while Bob is still the subject of keep, the (explicit) subject of boiling is the pot. In addition, it is understood that what is continuing is the boiling, as is shown by the passive-voice version of the sentence, The pot was kept boiling by Bob. But with Bob functioning as the subject of keep, Bob is obviously being credited with having some role in the overall event, and the role of CAUSER is the most obvious candidate.
This difference in meaning isn’t signaled by anything about the verb itself—either way, it’s pronounced and spelled the same. Therefore, what enables the hearer or reader to know which meaning is intended must be the grammatical structure
Linguists have proposed a variety of different ways to think about this phenomenon. One of them, which I’ll call the lexicalist approach, is to say that the syntactic structure of the construction is determined by the semantics of the main verb. An alternative view, which I’ll call the constructionist approach, reverses the attribution of cause, and says that (from the comprehender’s point of view, at least) the difference in meaning results from the difference syntactic structure itself. And while that may strike you as odd, consider the fact that when you encounter an agentless passive (The pot was kept boiling; Mistakes were made), you know that there was somebody who kept the pot boiling and who made the mistakes, even though the only thing informing you of that is the use of the passive voice.
For our purposes here, it’s not necessary to decide between the lexicalist and constructionist approaches. Under both approaches, it is the grammatical structure that enables comprehenders to distinguish constructions expressing the CONTINUATION meaning from those expressing the CAUSATION OF CONTINUATION meaning. And as I’ve previously argued, the appropriate viewpoint for purposes of legal interpretation is that of language comprehension. So under both the lexicalist approach and the constructionalist approach, it is reasonable to conclude that in the keep NOUN VERBing constructions, the semantic element of CAUSATION should be attributed to the construction as a whole rather than to the word keep by itself. Which means that such constructions are compositional in some respects and idiomatic in others.
THE KEEP+NOUN CONSTRUCTIONS (by which I mean simple transitive-verb constructions, and not keep NOUN VERBing or keep NOUN ADJECTIVE constructions) are much more variable in their behavior than the keep+Verb constructions.
As we’ve just seen, the range of variation displayed by the keep+Verb constructions is narrowly limited, with the variation being systematic and predictable. The variability in the keep+Noun constructions doesn’t display the kind of systematicity or predictability. And more to the point, the keep+Noun constructions display a significant range of variation in their degree of compositionality.
At one extreme are keep possession and keep guard, which are only slightly less compositional than the cognate constructions keep possessing and keep guarding. It is easy to separately identify the semantic contribution that each word in the construction makes to the meaning of the construction as a whole. Keep contributes the element of CONTINUATION and the direct objects contribute nominalized representations (i.e., representations in the form of nouns) of the situations denoted by the cognate verbs.
A lower degree of compositionality—and therefore a greater degree of idiomaticity—is displayed by keep watch and keep pace. Although there’s obviously some relationship between the meanings of those constructions and the meanings of the cognate constructions keep watching and keep pacing. But the meanings are by no means identical. Keep watch doesn’t mean ‘continue watching’ and keep pace doesn’t mean ‘continue pacing.’ So even though the nouns watch and pace denote activities, the kinds of activities that are denoted by the use of the nouns in isolation seems to me to differ (in ways that are hard to explain) from the kinds of activities that are denoted when the nouns are used as direct objects of keep. And that makes it hard to separate out what out the semantic contributions of keep on the one hand and the nouns on the other. But whatever it is that keep contributes to these constructions, it’s at least somewhat different from what it contributes to all of the other constructions we’ve looked at so far.
AT THIS POINT, I’m going to digress from discussing keep in order to talk more generally about the phenomenon of which keep watch and keep pace are examples: the fact that the kind of situation denoted by a given verb (in a transitive construction) can vary depending on what the direct object is:
carry a gun, carry a message, carry a state [in an election]
throw a football, throw a football game, throw a party
run a race, run an errand, run an advertisement, run a machine
pull a rope, pull rank, pull a prank, pull a gun, pull a switch
drop a glass, drop a hint
observe a religious holiday, observe something happening
break a window, break a date, break a law
smoke a cigarette, smoke a salmon
Verbs like those above, when considered in isolation, don’t have a single fixed “meaning.” That is to say, they don’t denote any particular type of situation. Instead, they have the potential to mean a variety of different things (or rather, to denote a variety of different types of situation), with the specific content of the denotation being activated (selected? evoked?) only when the verb is used in a particular context.
Since the verb has no determinate meaning until it is used in context, it seems reasonable say that the meaning-bearing linguistic unit is the transitive-verb construction as a whole rather than a single word. Which would mean that constructions like these are not fully compositional. But that is not to say that these constructions all share the same position on the compositionality–idiomaticity continuum. Because they don’t.
Some of the constructions denote relatively simple physical actions (carry a gun, throw a football, drop a glass, break a window). Those actions are probably what would first come to mind for most people upon hearing or reading the verbs in isolation, and these constructions would probably be regarded as expressing the verbs’ literal, basic, or core meanings. And (perhaps for that reason) the semantic content contributed to the constructions by the verbs is prominent and distinct. I would expect that most people have a well-entrenched idea of the nature of the events that are denoted, independently of what object is described as having undergone the action. The kind of carrying that is involved in carrying a gun is similar to the kind of carrying that is involved in carrying a book, a bottle, a hammer, a cell phone, a rock, and so on. And so too for throwing (a baseball, a snowball, a bottle, a hammer, a cell phone, a rock), dropping (ditto, plus a plate, a frying pan, a light bulb, a picture frame, a stapler) and breaking (a plate, a glass, a light bulb, a cell phone, an egg, a wine bottle, a mug). As long as the item that is described as undergoing the action has the appropriate physical characteristics, the verb’s semantic contribution to the construction’s meaning will be pretty consistent from one verb to the next. So these uses of the verbs, these constructions are highly compositional.
In contrast, there are constructions in which the same verbs appear, but are used metaphorically. In these latter constructions, the nature of the situation that the construction denotes is influenced significantly by the semantics of the noun that functions as the direct object. That makes it harder to justify attributing the nature of the situation to the verb alone, and as a result these uses display a greater degree of idiomaticity.
I’ll start with carry a message, which is similar to literal counterparts like carry a gun and carry a bag in denoting a situation in which an agent (the person to whom the subject refers) has personal control of the item to which the direct object refers (a message, a gun, a bag), and causes the item to accompany them as they move from one place to another. However when the message isn’t embodied in a physical object such as a document, the action denoted by carry a message is conceptualized more abstractly than the action denoted by carry a gun. So we can see very distinctly how a semantic characteristic of the direct object affects the nature of the event that the construction denotes.
Although the use of carry in carry a message retains the semantic element of accompanied motion, it can be regarded as a metaphoric extension of the literal sense of the word. There are also uses that are much more clearly metaphorical, such as carry a state [in an election] and carry a penalty. And turning to other verbs, there are metaphorical uses such as throw a football game, run an advertisement, pull rank, and break a law.
All of these constructions have several things in common beside being metaphorical. They all denote situations that are more complex in their structure than those denoted by carry a gun and carry a message. The situations that they denote arise out of patterns of behavior and interaction involving multiple people, rather than the physical actions of a single person. And although the behaviors in question are ultimately based on physical actions, the situations play out primarily in the domain of social interaction. As a result, the constructions can be understood only against a background of familiarity with the relevant social and cultural practices and institutions. For example, you can’t know what it means to carry a state in an election unless you know what states and elections are, how elections are conducted, what consequences flow from “carrying” a state, and so on.
That background knowledge is referred to as a frame. In some of the examples I’ve given, the construction explicitly evokes the relevant frame (throw a football game—but note that the football-game frame is itself embedded in other frames, such as sports and competition, and the phrase throw a football game may well prompt associations with the frames gambling and bribery). In other instances (e.g., carry a state) the relevant frame isn’t evoked explicitly; instead, an association with the relevant frame (here, the election frame), has to be recognized by the reader or hearer.
Either way, the frame associated with the direct object contributes in a major way to delineating the situation that the construction denotes. For each of the constructions we’ve been looking at, the denoted situation represents one of the many types of situations that the frame encompasses. And the effect of the construction is to focus attention on one of those situation-types (or on a specific instance of such a type). Constructions like these therefore present powerful evidence for the inadequacy of the traditional view of word meaning.
Although the examples I’ve been discussing haven’t come from the corpus data, that data reveals a group of constructions in which the construction’s meaning is due in large part to a frame that is explicitly evoked by the construction or for which the construction as a whole prompts an association:
keep tavern (managing a tavern frame)
keep school, etc. (operating a school frame)
keep house (domestic life frame)
keep sheep, keep cattle, keep dogs, keep hogs, etc. (raising farm animals frame)
keep prisoners (prison/imprisonment frame)
keep concubines (concubinage frame)
keep slaves, keep negroes (slavery frame)
Many of the uses of these constructions refer generally to the activity that is the subject of the frame:
Our landlord, who kept tavern at the sign of the black horse, at Charing-cross, furnished us with every requisite
I had several offers to go into the country to keep school in credible families, where I had a prospect of wanting for nothing of the necessaries of this
However, other uses highlight a particular aspect of the activity. In the case of keep house and keep sheep/cattle/etc., the construction sometimes referrs to the activities of the head of the household or owner of the animals, and sometimes to the activities of the servants who preformed the gruntwork:
Thus with 7 servants and hireing a chore woman upon occasion of company, we may possibly make out to keep house; with less we should be hooted at as ridiculous and could not entertain any company.
Their way of living is by husbandry, and keeping cattle and swine; wherein they do as well, or rather better, than any other Indians
What think you of your daughters comeing to keep House for you?
Sir Willam Phips was a New-England man, born at Pemaquid in 1650, where he kept sheep until he was 18 years old, then was an apprentice to a ship carpenter.
This pattern of variation is consistent with the notion that when a sentence containing the construction is heard or read, the construction itself acts merely to evoke or prompt the relevant frame in the listener’s mind, and that if any particular aspect of the frame is highlighted (as in the examples above), that is due to an inference from the context.
HOW CAN WE ACCOUNT for the range of behavior that we have seen displayed by transitive-verb constructions?
On the traditional view that words have stable, independent meanings, we would say that the choice of the direct object disambiguates the verb—in effect selecting the appropriate “meaning” from the set of all possibilities. Such a process would be analogous to what we’ve seen in connection with the keep+Verb constructions, where the grammatical structure determined whether keep’s contribution to the construction’s meaning was the element of CONTINUATION or of CAUSATION OF CONTINUATION.
But there is an important difference. In the case of the keep+Verb constructions, the variation is limited to the presence or absence of a single semantic feature, and is linked to a single syntactic difference. The variation in meaning can therefore be easily explained by an overarching generalization (or “rule,” if you insist). In contrast, that kind of generalization won’t account for the far wider range of variation that characterizes the meanings are associated with specific pairings of verbs and nouns.
This raises questions relating to the mental processes by which fluent speakers of English produce and understand sentences. As with the keep+Verb constructions, I’ll focus on those questions from the perspective of language comprehension. Based on the fact that language users can generally understand one another, we can infer that they are generally successful in assigning appropriate meanings to what they hear and read, and in particular that they are successful in assigning appropriate meanings to transitive-verb constructions. Since we’re starting from the assumption that the this process includes selecting the right sense of the verb and plugging it a matching slot in the sentence, it follows that there is some mental mechanism that enables the correct selection to be made. And since the appropriate meaning varies depending on which noun fills the direct-object slot, the mental mechanism must include some process that matches meanings to Verb+Noun combinations.
The need for such a function is something that the traditional approach to word-meaning shares with the newer approach that is the subject of this post. But under the newer approach, there doesn’t exist any independent word sense that’s ready to be plugged into the sentence. Instead, a variety of possible mechanisms have been proposed that would render such independent word senses unnecessary. For verb–noun combinations that are sufficiently frequent, the appropriate meaning might simply be mentally associated with the construction as a whole, much the same way that the meaning ‘die’ is mentally associated with the construction kick the bucket. In other words, such constructions would be functionally indistinguishable from idioms.
Another possible mechanism is that what is stored mentally is a pattern, in which each of the potential meanings in which a given verb can conventionally participate is associated with a set of semantic constraints that the direct object must satisfy in order for that meaning to be communicated. In the case of throw NOUN, in order to activate the meaning ‘propel-X-by-arm-movement,’ the direct object has to be a noun that denotes a tangible object of an appropriate size and weight: a baseball, a book, a lamp, a frying pan, a pie. To activate the meaning ‘deliberately lose X’ (in the sense of ‘fail to win X’), the direct object has to denote some kind of competitive event. And so on.
Yet another possibility is suggested by the discussion above of frames. Maybe, with at least some constructions, the direct object evokes or otherwise prompts a mental association with a particular frame—possibly with the relevant frame being determined by some contextual factor—and the specific situation that is evoked is the result of a further cognitive process that might itself be driven at least in part by the context. Under approaches like that, meaning is regarded, not as being extracted from the utterance or text by decoding, but as being constructed in and by the mind of the comprehender.
IN SHORT, WHAT ARE TRADITIONALLY REGARDED as issues of word-meaning can be analyzed without assuming that there actually are such things as free-standing word meanings. Instead of invariably regarding words as the basic unit of meaning, it can make sense to attribute that role to larger constructions of the kind I’ve been discussing. (If you scoff at that idea, you will probably be surprised to learn that among the people who advocate this approach are lexicographers. See, for example, “Do Word Meanings Exist?” by Patrick Hanks. The discussion above about patterns is extrapolated from an approach developed by Hanks called Corpus Pattern Analysis [link, link, link].)
If you’re someone who looks to dictionaries as authorities about word meaning, what I’ve said will probably have struck you as radical, and you may well be skeptical. But hopefully I’ve given you food for thought.
And in closing, here’s a question and a challenge: Given that even the independent-meaning approach requires meanings to be mapped to constructions, what analytical work is done by the idea of independent meanings? If both approaches are adequate to account for how constructions such as keep+V/Adj/N are understood, maybe the default assumption should be to follow the constructionalist approach, unless the lexicalist approach earns its keep by, e.g., providing a better way to explain the data.
I’m not saying that such a showing couldn’t be made, or that lexicalist approaches can’t provide useful insights. But maybe if you’re inclined to reject my argument as a nonstarter, you should try to articulate an argument explaining why my argument is unreasonable.
[Cross-posted on Language Log.]