A theory of interpretation

If innovative denominal verbs are contextuals, they cannot be accounted for by the traditional theories that assume fixed sense and denotation. They require a theory of what a speaker means in uttering such a verb on particular occasions — a theory of interpretation. Note that the indexical he, which has shifting reference, requires a theory that characterizes how its referent is determined for each utterance. The theory must specify when a speaker has good grounds for believing that the listener can, on the basis of their mutual knowledge, identify its referent uniquely. The same goes for innovative denominal verbs, which have shifting sense and denotation. They require a theory that characterizes how their senses are determined in each utterance. Our first task is to outline such a theory.

3.1. interpreting innovations. At the heart of this theory is a convention, in Lewis’s sense, about the use of language. The idea is this. Infusing an expression sincerely, the speaker intends the listener to come to a unique interpretation of what he has said — not from the meanings, of the words alone, but also on the assumption that the speaker has good grounds for thinking that the listener can come to that interpretation uniquely on the basis of what they mutually know. This convention is obviously akin to Grice’s cooperative principle. For innovative denominal verbs, the convention takes this particular form:

(23) The innovative denominal verb convention. In using an innovative denominal verb sincerely, the speaker means to denote:

(a) the kind of situation;

(b) that he has good reason to believe;

(c) that on this occasion the listener can readily compute;

(d) uniquely;

(e) on the basis of their mutual knowledge;

(f) in such a way that the parent noun denotes one role in the situation, and the remaining surface arguments of the denominal verb denote other roles in the situation.

‘Situation’ is being used here as a cover term for states, events, and processes.

This convention, in effect, has two parts. Conditions 23a — e, or something like them, appear to apply to all contextuals. The condition specific to denominal verbs is 23f, which refers to the syntactic structure of denominal verbs as opposed to compound nouns, shorthand expressions, or other contextual expressions. The importance of these conditions will become clear as we proceed.

To see how this convention applies, imagine a news agent one day insisting to us that The boy porched the newspaper. By the convention, the news agent had in mind a kind of situation he felt we would be able to identify uniquely from our mutual knowledge ofporches, their relation to newspapers, paper boys, and the topic of conversation — the boy’s delivery of the newspaper. To be so confident, he must have judged that this kind of situation would be salient — conspicuously unique, given our mutual knowledge or beliefs. What could be so salient? A distinguishing characteristic of porches is that they are shelters adjacent to the main door into a house. They are associated with a state that can, for convenience, be expressed as the propositional function On (x, a porch) — ‘x is on a porch’, where x is ordinarily something susceptible of being sheltered. The direct object of porched, namely the newspaper, refers to an entity that fits x ’s specifications, so we have On (the newspaper, a porch). To use up the surface subject the boy, we can best view this state as the consequence of the boy’s action, adding the inchoative Come-about (x), the causative Cause (x, y), and the act Do (x, y) to give 24 and its paraphrase 25:

(24) Cause (Do (the boy, something), Come-about (On (the newspaper, a porch)))

(25) The boy did something to cause it to come about that [the newspaper was on a porch].

As part of this reasoning, we also realize that the news agent’s topic of conversation was newspaper deliveries; and that he mentioned the paper boy, the newspaper, and a porch. On these grounds alone, we could infer that he very likely intended porch to denote the act of the boy’s delivering the newspaper onto the porch. That agrees with 24 to give us more confidence in our inductive inference.

This, however, isn’t enough. The propositions in 24 express only the bare bones of what the news agent meant. From The boy porched the newspaper, we wouldn’t infer that the boy had pinned the newspaper page by page to the inside of the porch. The news agent wouldn’t have had good reason to think we could arrive at that interpretation uniquely. From our mutual knowledge, we are warranted in inferring only the ordinary manner of delivery. The kind of situation denoted has to be the most salient one under the circumstances; and the ordinary manner is the most salient unless there is good reason to think otherwise.

This convention, therefore, relies critically on a theory of what people know about concrete objects. Although such a theory is not available, there are strong suggestions about what it might look like from the work of Berlin, Breedlove & Raven 1968, 1973; Brown 1976; Brown et al. 1976; Hampton 1976; Smith, Shoben & Rips 1974; Rips, Shoben & Smith 1973; Smith, Rips & Shoben 1974; Smith 1978; and Rosch and her colleagues. We will begin by outlining a suitably-framed theory suggested by this work. The aim, we emphasize, is not to establish a theory of real-world knowledge, but to outline an empirical enterprise that we claim must be worked out before one can have an adequate explanation of innovative denominal verbs.

3.2. world knowledge can be divided roughly into two parts. Generic knowledge is what people tacitly know about space and time, the basic physical laws, natural kinds, manufactured artifacts and their functions, and so on. People normally assume that generic knowledge doesn’t vary much from person to person; they believe that a large core of it is shared by friend and stranger alike. Particular knowledge, however, is what people tacitly know about particular or individual entities — particular objects, events, states, and processes. Particular knowledge depends critically on a person’s history. The particulars that one person knows — his parents, his experiences yesterday, and the person to whom he has just talked — won’t necessarily be particulars that the next person knows. The commonest denominal verbs, both idiomatic and innovative, depend mainly on generic knowledge about concrete objects; and so it is important to understand what this knowledge might be like.

Our first premise is that people have generic theories about concrete objects, theories they use for categorizing objects. These theories specify three basic aspects of an object: its physical characteristics, its ontogeny, and its potential roles. The theory for ordinary bricks, for example, specifies (a) the normal range of their physical characteristics, e. g. their color, shape, weight, and breakability; (b) their normal ontogeny, e. g. that they are molded from clay, baked in ovens, and sold by building-supply firms; and (c)their potential roles, e. g. that they are ordinarily cemented with mortar in horizontal rows to form walls, are sometimes used as doorstops, and can be used as riot missiles. These theories, we assume, have evolved to be conceptually optimal; in these three respects, the objects within a category are as similar as possible to each other, and as different as possible from objects in neighboring categories at the same level of abstraction. This assumption has empirical support in Rosch & Mervis.

These theories are essential in order for people to deal effectively with the world around them. If something looks like a brick, people must be able to infer that it probably has certain other physical characteristics, the normal brick ontogeny, and the potential to play the normal brick roles. Without such a theory, each new brick would have to be treated as novel and without predictable properties. Animals must also have such theories, of course; the ability to categorize isn’t an exclusively human prerogative.

Because of these theories, some objects are viewed as more central to (or typical of) a category than others. Red bricks, for example, are probably viewed as more typical of the category ‘brick’ than gold bricks, wooden bricks, glass bricks, bricks of cheese, or bricks of ice cream: red bricks fit people’s theory for bricks best. As has been shown by Rosch & Mervis and by Hampton, the more properties an object shares with other objects in a category, the more typical of that category it is judged to be.

The properties within each theory, however, do not carry equal weight; some are more central to the characterization of the category than others. The most central of these we will call predominant features. Thus predominant features of bricks seem to be their box shape and child’s-shoebox size. The brick’s other physical characteristics, ontogeny, and potential roles seem generally less central, although not equally so. The predominant feature of orphans, in contrast, is a fact about their ontogeny: they are people whose parents died before they were raised. The predominant feature of vehicles is the fact about their potential role that they are used for transportation.

What exactly are predominant features? Our hypothesis is that they can be derived from notions of ‘cue validity’. According to work by Rosch and her colleagues and Tversky, the categories that people prefer for natural objects and human artifacts are those that maximize both the similarity between any two members of the same category and the dissimilarity between any two members of different categories. That is, the categories maximize ‘cue validity’: the more that cues or features are associated with the members of a category, and not with the members of other categories at the same level, the better that category is. As for any particular cue, the more it distinguishes the members of the category from the members of other categories, the more valid it is said to be. Formally, cue validity can be defined very precisely.

Since there has been little discussion of the practical identification of predominant features, we will offer several tentative procedures. A predominant feature of a category is one that tends to hold for most of its members — especially its typical members — but not for members of neighboring categories. So a predominant feature of a widow is that she is a woman whose current social status is the result of her husband’s being deceased. Being human, adult, or female are not by themselves predominant features of widows — since these do not distinguish widows from wives, spinsters, husbands, and widowers. When a predominant feature is relational, its relation to a second category tends to be asymmetric; thus a predominant feature of quivers is that they are for holding arrows. If arrows didn’t exist, neither would quivers. But it is not a predominant feature of arrows that they can go in quivers, since arrows can exist on their own. Not all asymmetric relations take this form: parts tend to be related to their wholes, not vice versa. It is a predominant feature of arms that they are related to the whole body, but not vice versa. Note that a category may have more than one predominant feature, since it may be distinguished from different kinds of neighbors in different respects.

How, then, do concrete nouns work? Our assumption is that, in using a concrete noun, a speaker intends to denote objects by virtue of their membership in the category defined by the appropriate generic theory. In using brick, a speaker intends to denote the kind of object that fits his theory for bricks. For this to succeed, speakers and listeners must share roughly the same generic theory for bricks. The work on categories shows that this is a reasonable assumption — at least for the most prominent real-world categories, those named most simply within languages.

Under this view, concrete nouns are related in meaning to the extent that the theories conventionally associated with them are related. One way in which two theories are related is by predominant features. ‘Ball’ and ‘brick’ form one class of theories, because both have predominant features that specify size and shape. ‘Widow’ and ‘orphan’ form another, because both have predominant features that concern ontogeny. And ‘tool’ and ‘vehicle’ form a third class, because both have predominant features that specify potential roles. These classes, of course, can be further subdivided according to the kinds of physical characteristics, ontogeny, and potential roles that are referred to in the predominant features.

When the parent nouns of denominal verbs are classified in this way, we argue, they fall into classes and subclasses that correspond closely (if not exactly) with the classes and subclasses that we arrived at in our analysis of denominal verbs. Briefly, the parent nouns can be classified according to their predominant features roughly as follows:

(a) placeables. The parent nouns of locatum verbs denote placeables — things whose conventional role is to be placed with respect to other objects. A predominant feature of carpets, for example, is that they potentially go on floors. Note that carpets depend for their characterization on floors, not the reverse. So the right characterization of carpets is as placeables (carpets go on floors), not as places (floors go under carpets).

(b) places. For location verb, the parent nouns denote places — things with respect to which other objects are conventionally placed. Thus a predominant feature of kennels is that they are places where one ordinarily keeps dogs. Note that kennels rely for their characterization on dogs, whereas dogs can exist without kennels.

(c) time intervals. The parent nouns of duration verbs denote time intervals — temporal ‘places’ in which events and processes can be located. Thus summers consist of June, July, and August, aspecific time interval.

(d) agents. The parent nouns of agent verbs denote agents, things whose predominant feature is that they do certain things. Butchers cut meat professionally; companions accompany people; and tailors make clothes professionally.

(e) receivers. The parent nouns of experiencer verbs denote things picked out for their role in receiving or experiencing things, e. g. witnesses.

(f) results. With goal verbs, ontogeny is important. Their parent nouns denote results, entities whose predominant feature is that they are end-products of some action or transformation. Thus widows form a category because they are a social product caused by the loss of their husbands. For many results, like braids, powder, and sandwiches, physical characteristics are also important: the end-product is distinguished not just by the action or transformation carried out, but also by the physical characteristics that result.

(g) antecedents. For source verbs, ontogeny is also important, but the parent nouns denote antecedents — the beginnings, not the final states — of some actions or transformations. A predominant feature of some types of pieces, for example, is that they are things out of which some products can be made.

(h) instruments. The things denoted by the parent nouns of instrument verbs are picked out for their potential roles as instruments. One of their predominant features is that they must be physically present for certain actions to take place, or for certain results to be accomplished. It is a predominant feature of ambulances that they are instruments for transporting the sick or wounded; a predominant feature of glue is that it is an instrument for attaching one object to another. These eight categories, of course, do not exhaust the way predominant features can be classified. The miscellaneous denominal verbs have special predominant features; six of the eight categories are susceptible to a finer analysis; and some predominant features can be cross-classified — e. g., those of both places and placeables concern location. At this time, more detail would help very little.

The eight major types of predominant features can be represented, for convenience, as propositional functions. For example, a predominant feature of carpets is that they are located on floors: On (carpets, floors). More generally, placeables like carpets fit the broad locative proposition Loc (e, x) ‘eis located with respect to x’ (in which e denotes the entity in the category, and x denotes the class of things with respect to which it can be located). The propositional functions for these eight predominant features are listed in Table 1. We turn next to the role these predominant features play in the interpretation of innovative verbs, and to other empirical consequences of the innovative denominal verbconvention.

Table 1. Principal categories and their predominant features