Jump to ContentJump to Main Navigation
Cognitive GrammarA Basic Introduction$

Ronald Langacker

Print publication date: 2008

Print ISBN-13: 9780195331967

Published to Oxford Scholarship Online: May 2008

DOI: 10.1093/acprof:oso/9780195331967.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2019. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in OSO for personal use.  Subscriber: null; date: 23 October 2019

Nominal Structure

Nominal Structure

(p.310) 10 Nominal Structure
Cognitive Grammar

Ronald W. Langacker

Oxford University Press

Abstract and Keywords

Personal pronouns and proper names are inherently grounded. The structure of other nominals tends to reflect semantic function, with the head noun as core, grounding as the outermost layer, and modifiers in between. Noun modification is varied both semantically and in its structural implementation. When nominal and relational expressions combine grammatically, there is often a discrepancy between the nominal expression's profile and the entity which participates most directly in the relationship—its active zone with respect to that relationship. Noun classes have varying degrees of semantic motivation. Distributional classes, defined by participation in particular patterns, may be semantically arbitrary. While gender-type classes have semantic prototypes, the class as a whole is defined by a consistent set of grammatical behaviors. In a usage based approach, such classes are characterized by families of constructional schemas. Noun classifiers likewise have prototypical values semantically extended to a range of other cases. They represent a distinct kind of nominal structure in which the classifier functions as a schematic head noun. Classifiers are related to quantifier constructions allowing the unitization of a mass. Nouns bear various kinds of grammatical markings. Most intrinsic to nouns are elements deriving them from other classes. There is no sharp distinction between such derivation and noun inflection. Markings of gender and number are intermediate. More extrinsic are markings indicating a nominal's role in higher-level grammatical constructions. These are meaningful in a symbolic account of grammar. Such an account accommodates both agreement—the multiple realization of semantic specifications—and cases where multiple specifications are realized by a single, unanalyzable form. Analyzability is a matter of degree.

Keywords:   active zone, agreement, analyzability, derivation, distributional class, gender class, grammatical marking, grounding, head noun, inflection, noun class, noun classifier, noun modifier, number, personal pronoun, proper name, prototype, quantifier, semantic extension, unitization

The term noun is used in CG for any expression that profiles a thing.1 So defined, it subsumes both lexical nouns and nominal expressions of any size, either fixed or novel. A full nominal expression is one that incorporates grounding and thus singles out a discourse referent. More compactly, it is referred to as a full nominal or just a nominal. In this chapter, I deal with various facets of nominal structure.

10.1 Structure and Function

A nominal corresponds to what linguists often call a “noun phrase” (NP). This term is poorly chosen, since nominals are not always phrases, nor do they always contain a noun (as traditionally understood). An expression does not qualify as a nominal because it exhibits any particular structural configuration. Instead the crucial factors are meaning and function. The schematic characterization of a nominal—that it profiles a grounded instance of a thing type—makes reference to several semantic functions: grounding, instantiation, and type specification. It is due to these semantic properties that a nominal is able to function as it does in larger grammatical structures.

10.1.1 Canonical Structure

There is a natural tendency for the internal structure of nominals to straightforwardly reflect the semantic functions that characterize them. It is normal for the type to be specified by a lexical noun selected from a very large inventory, and for grounding to be indicated by a separate element chosen from a limited set of options. It is not evident that instantiation is ever separately marked, but this too is natural in view of the very slight difference between type and instance conceptions (fig. 9.3). Since they all help specify a single discourse referent, the various components of a (p.311) nominal—grounding element, lexical head, and diverse modifiers—tend to be contiguous and to form a classic constituent (§7.4.1). Moreover, reflecting its status as conceptually the most extrinsic nominal component, grounding is usually the most peripheral component structurally (§9.3.1). It is common, if not typical, for grounding to represent the initial element in terms of linear order and the outermost layer in terms of constituency: (those (two (lazy (cats)))).

Important though they are, factors like these have only a general shaping influence. While they motivate overall tendencies, functional considerations provide no basis for predicting the particular details and full diversity of linguistic structure. Each language develops a broad variety of specific nominal structures serving different purposes and responsive to different functional pressures. Although English conforms to the tendencies just noted, it also has nominals without overt grounding (e.g. nominals without overt grounding) and some that lack a lexical head (e.g. some that lack a lexical head). Other languages differ more fundamentally in their nominal strategies, which, in their own way, are nonetheless quite natural. In some languages, for example, covert grounding is the rule rather than the exception. Not every language requires that a noun and its modifiers all be contiguous. And in languages making systematic use of classifiers (§10.3), a lexical noun is not necessarily the nominal head, being structurally more peripheral (FCG2: §4.3.1).

What to identify as a nominal “head” is a point of some controversy. As is so often the case with theoretical disputes, the issue is terminological rather than empirical. The term “head”, of course, is metaphorical. It suggests that whatever is so identified should be the controlling element or one of chief importance.2 But in a typical nominal, e.g. those lazy cats, there are two components with legitimate claim to this status. The first is the grounding element, which is primary in the sense of being the only one a nominal requires (those can function alone in this capacity). When it combines directly with a modifier, as in those with fleas, the grounding element also exerts control in the sense of imposing its nominal profile on the expression as a whole. The second is the lexical noun (cats), which is centrally important by virtue of providing the most extensive semantic content, thereby establishing what type of thing the nominal designates. Thus it is not a matter of which is right but of how we decide to use the term. In CG, the term “head” is used primarily for the profile determinant at any level of organization (not just with respect to nominals). Following this usage, those would be the head in those with fleas.3 Yet it is also helpful to follow traditional practice by referring to the central component of a nominal—the one providing its type description—as its head noun (or simply its head).

Although the head noun is often called a lexical head, it is not invariably a lexical item. Besides established lexemes, novel expressions of any size can function in this capacity. We can refer not only to a cat or to a cat-lover but also—as need arises—to a cat-lover psychiatric examination, to a cat-lover psychiatric examination manual, to a cat-lover psychiatric examination manual cover designer, and so on indefinitely. (p.312) As these examples indicate, head nouns in English are mainly constructed by means of compounding and morphological derivation. Hence the nouns and verbs they incorporate are usually singular and ungrounded.4 It is only the head noun as a whole that undergoes pluralization: cat-lovers, cat-lover psychiatric examinations, cat-lover psychiatric examination manuals, cat-lover psychiatric examination manual cover designers. While the head overall is the pluralized structure, the plural inflection is realized morphologically just on its final word.

Thus for English, at least, nominal heads are defined by several converging properties. Internally, they are formed by compounding and morphological derivation (cat-lover), in contrast to the syntactic combination of separate words and phrases observed at higher levels of organization (e.g. intelligent cat-lover from Vermont). The head-noun level is also where pluralization occurs to derive a higher-order type (cat-lover > cat-lovers). Moreover, the type defined at this level is the basic type instantiated by the nominal referent. Whatever type is characterized by the head noun overall, an instance of that type is grounded and profiled by the nominal as a whole. The referent of an intelligent cat-lover from Vermont is therefore not a cat but a lover thereof. Likewise, most intelligent cat-lovers from Vermont designates a single (fictive) instance of the plural type cat-lovers.

The global organization of an English nominal thus tends to be as follows:

(1) [Grounding [(Modifiers) [Head Noun] (Modifiers)]]

If not a single morpheme, the head noun is built through a combination of compounding and morphological derivation. Pluralization is an option at the highest level, and if chosen, it is realized on the head's final word. Whatever is profiled by the head is profiled by the nominal overall. However, any number of modifiers may combine syntactically with the head and thus refine (or even substantially alter) the basic type it specifies. Finally, since it offers the most extrinsic characterization of the nominal referent, grounding tends to occupy the outermost structural layer.

Canonically, the minimal components of a nominal are a head noun, providing a type description, together with a separate grounding element. Often, however, these two semantic functions are effected by a single form. In some cases there is no need for a detailed type specification. Prime examples of this are pronouns, whose main import resides in grounding. In other cases any need for separate grounding is obviated by the type description. Here the chief examples are proper names.

10.1.2 Pronouns

Pronouns are so called because they stand in for nouns (pro = ‘for, instead of’). The term is traditionally applied to a substantial variety of forms—properly so, if nouns are broadly defined as in CG. They include both definite and indefinite expressions, as well as those functioning as either full nominals or just head nouns. We might start with a brief look at English one, which has a number of uses and related senses.

(p.313) In expressions like (2)(a), one functions as a head noun. Its type description is so schematic—equivalent to the schema for common count nouns—that the nominal it heads needs outside support for meaningful interpretation. It thus requires the availability, in the previous discourse frame (fig. 9.7), of a more specific type specification with which it can be identified. In (2)(a), its type is equated with that of the previous nominal an expensive car. This is the import of saying that one is a pronoun which “stands for” car. In (2)(b), we observe that the plural form ones is used analogously (standing for computers). Its schematic type description is equivalent to the schema for plural mass nouns. In this construction, one and ones are simply nouns, not full nominals.5

(2) (a) My boss has an expensive car, but I just have a cheap one.

(b) The faculty have fast computers, but the students still have slow ones.

One is also simply a noun when it combines directly with a grounding element: this one, that one, each one, every one, any one. In a number of cases, the two have coalesced into a single word whose type description is slightly more specific, namely ‘person’: someone, everyone, anyone, and no one (also now written noone). These composite expressions, which are full nominals, are commonly referred to as indefinite pronouns. Rather than depending on another nominal for a detailed type specification, they are used precisely on occasions when none is necessary: for either an unidentified person, in the case of someone, or for general statements where ‘person’ is itself the type in question. As is true in many languages, indefinite pronouns enter into (partial) paradigms based on schematic nouns for ‘person’, ‘(non-human) thing’, ‘place’, ‘time’, and ‘manner’: something, everything, anything, nothing; somewhere, everywhere, anywhere, nowhere; sometime, every time, anytime, *notime; somehow, *everyhow, anyhow, ?nohow. These offer a flexible means of referring in general terms to most any sort of virtual entity.

One can also stand alone as a full nominal, with two basic variants. The first, exemplified in (3)(a), is used in making statements pertaining to people in general. It thus resembles someone, its type specification being ‘person’, but unlike someone, its referent is necessarily virtual. As with every, each, and any, the fictive individual it designates is taken as being representative, so that anything ascribed to it is valid for all instances. This representativeness obviates the need for separate grounding. Since the virtual instance is abstracted from actual ones, and immanent in their conception, there is no other instance to distinguish it from. Just evoking the profiled instance is sufficient to single it out.

(3) (a) One can never be too thin, too rich, or too well-connected.

(b) My friends all have yachts. Tom has several, Alice has two, so I at least want one.

One is also used anaphorically, depending on a previous nominal for its interpretation. An example is (3)(b), where its own schematic type is identified more specifically as (p.314) yacht. In this use, one evidently conflates the functions of quantifier and indefinite article. It is a quantifier by virtue of contrasting with other numbers in the counting sequence: one (yacht), two (yachts), three (yachts), etc. At the same time, it is mutually exclusive with the indefinite article a (historically derived from one), which is otherwise required for singular count nouns: the yacht, a yacht, the one yacht, *a one yacht, one yacht. So in this type of use—in a nominal like one yacht or as its anaphoric substitute—one can be described as an indefinite grounding element with an explicit specification of quantity.

First and foremost, the term “pronoun” calls to mind “personal” pronouns. The label reflects the status of their referents with respect to the speech event participants: the speaker (“first person”) and hearer (“second person”), as opposed to others (“third person”). One facet of grounding, that of locating the nominal referent vis-à-vis the ground, is thus intrinsic to personal pronouns. An additional facet, that of singling out the referent, is intrinsic for the first-person and second-person pronouns (I, we, you), since these designate either an interlocutor or a group that includes it. With third-person pronouns (he, she, it, they), the matter is less straightforward. Their minimal type specifications, like ‘animate female’ for she, select open-ended sets of eligible candidates. Quite a number of candidates are likely to be available in a given discourse context. Nevertheless, the third-person forms are full nominals, presumed capable of singling out the intended referent. How do they accomplish this?

Personal pronouns are closely related to definite articles and also to anaphoric demonstratives. Like a definite article (fig. 9.11), they imply that just a single instance of the specified type is readily accessible in the previous discourse frame. She is thus appropriate in (4)(a), where the prior sentence introduces just one eligible candidate, but not in (4)(b), where several are equally available:

(4) (a) I was talking to an interesting woman. She heads a major corporation.

(b) *I was talking to several interesting women. She heads a major corporation.

Unlike articles, pronouns stand alone as nominals and are thus dependent on their own schematic type specification to select the pool of eligible candidates. Also unlike articles, they imply that the referent has already been singled out for joint attention in the previous frame, as shown in figure 10.1. In this respect they resemble anaphoric demonstratives (fig. 9.10). They differ from such demonstratives by lacking directive force, as well as the proximal/distal distinction.

Instead of proximity, a third-person pronoun relies on contextual prominence to identify its referent. It presupposes that a particular instance of its type has not only been singled out in the previous discourse frame, but is salient enough to be the sole instance that counts for anaphoric purposes. In (5)(a), for example, both the yacht and the car single out referents that satisfy the type specification of it (‘inanimate thing’). As clausal subject, however, the former has sufficient prominence to eclipse the latter and establish itself as the pronoun's antecedent.6 It thus takes its reference from the yacht—they are interpreted as being coreferential. (p.315)

Nominal Structure

Figure 10.1

(5) (a) The yacht is more impressive than the car, but I probably can't afford it.

(b) That yacht is certainly impressive, but I can't afford even this cheap one.

By comparison, the anaphoric use of one in (5)(b) depends on the antecedent nominal only for its type specification, not its reference: whereas it and the yacht are construed as designating the same instance of this type, one and the yacht profile distinct instances. The basis for the contrast is that the anaphoric elements represent different levels of functional organization. One in (5)(b) is just a head noun, so only type is relevant for its interpretation. On the other hand, personal pronouns are full nominals, so reference too is at issue.

These two kinds of anaphora are sketched in figure 10.2, where t indicates a specific type and ellipses (…) a more schematic type (cf. fig. 9.3). With a personal pronoun, both the antecedent and the anaphor are full nominals, so each profiles an instance of its type. Effecting their anaphoric relationship is a correspondence serving to identify these profiled instances. The identification of their referents implies that the antecedent's type description is also applicable to the referent of the pronoun. In the other kind of anaphora, the antecedent and the anaphor are simply nouns, the heads within their respective nominals. A correspondence therefore identifies their types but gives no indication concerning the nominal referents. Thus in (5)(b) we know that this cheap one is a yacht, albeit not the instance profiled by the subject nominal.

Nominal Structure

Figure 10.2

(p.316) Recall that a nominal grounding element is itself a schematic nominal, for it profiles a thing and relates it to the ground (fig. 9.4). When it stands alone as a nominal, a grounding element is even more schematic than a pronoun in regard to type. In such cases, the type is often determined anaphorically. Thus we know in (6) that those refers to steaks, most to students, and each to a (fictive) witness:

(6) (a) I left some steaks out to thaw. You can have those for dinner.

(b) We admitted a thousand new students last fall. Most were deficient in reading and math.

(c) Numerous witnesses claimed to have seen the robber's face. Each gave a different description, however.

In effect, these elements are functioning here as pronouns. Note that those could be replaced by them. And just as in figure 10.2(a), those and some steaks are related by reference anaphora—they are coreferential. With most and each the anaphoric relationship is slightly different. Since these quantifiers designate virtual entities, strictly speaking they are not coreferential to the antecedent nominals, a thousand new students and numerous witnesses, whose referents are interpreted as being actual. Instead these nominals indicate the group with respect to which most and each specify a quantity. More technically, they identify and contextually delimit the maximal extension (Et) in terms of which the quantifiers are characterized (figs. 9.12 and 9.13).

10.1.3 Proper Names

Proper names are often claimed to not even have a type description. Indeed, they are often considered meaningless, their sole import residing in their reference to something in the world. This classic view cannot be sustained, however. Many proper names are conventionally employed for particular types of entities (e.g. Jack for ‘human male’, Jill for ‘human female’). Others evoke substantial bodies of information that is widely shared within a speech community. The name George Washington does not just refer to a person, but tends to activate a conventional array of encyclopedic knowledge (army general, first American president, wife named Martha, thought to be honest, and so on). Likewise, Chicago does not simply name a city, but evokes an array of widely known properties and associations. To the extent that they are entrenched and conventional, these specifications have to be included in the meanings of such expressions.

The distinguishing feature of proper names is not that they are meaningless, but is rather to be found in the nature of their meanings. As one component of its meaning (one domain in its matrix), a proper name incorporates a cognitive model pertaining to how the form is used in the relevant social group.7 According to this idealized model, each member of the group has a distinct name, with the consequence that the name itself is sufficient to identify it. The name Jack, for example, carries with it the supposition that within the relevant group (e.g. a family) there is just one person (p.317) referred to in this manner. The name can thus be thought of as defining a type—the type ‘person named Jack’—which (in the context of the group) the model specifies as having just one instance. Since the name itself singles out the only instance, there is no need for separate grounding.

This idealized cognitive model implies that the name itself—that is, the expression's phonological pole—figures in its type description at the semantic pole. Central to the meaning of Jack is the specification ‘person named Jack’. But this property is not limited to proper nouns. Central to the meaning of a common noun, e.g. yacht, is the specification that many distinct entities bear this label. One domain in its matrix is the very knowledge that members of the speech community conventionally refer to such entities in this manner. Hence the type description for yacht includes the specification ‘thing called yacht’.8

To make these notions explicit, let τ represent a noun's phonological pole, and t its semantic specifications. Its overall type description can thus be given as t/τ: that is, it includes the very fact that things of this sort are symbolized by τ. This is shown in figure 10.3 (a refinement of fig. 9.3). A common noun is one whose type has multiple instances, each of which can also be labeled τ. On the other hand, a proper noun labels just a single entity, so there is no basis for abstracting a separate type. We can either say that the type/instance distinction is neutralized or, equivalently, that the type has just one instance. In the case of proper names, this uniqueness follows from the idealized cognitive model for naming within a social group.9 And since the name itself is their central defining property, the remaining semantic specifications are often quite schematic (e.g. ‘human male’ for Jack), as shown by using ellipses in lieu of t.

It sometimes happens, though, that the expectations of an idealized cognitive model fail to be satisfied in actual practice. In particular, it is not uncommon for multiple individuals to have the same name within the relevant social group. We thus find expressions like the following:

Nominal Structure

Figure 10.3

Nominal Structure

Figure 10.4

(7) (a) There were four Davids on the soccer team I coached.

(b) Are you the Hank Barnes who owns the liquor store, or the one who ran for mayor?

In such examples, a proper name functions grammatically as a common noun—it can pluralize, occur with overt grounding, take a restrictive relative clause, participate in type anaphora, and so on. English handles this discrepancy between idealized model and actual situation by suspending the model's requirement that a name be assigned to just one person. A particular name (τ) can thus be used to designate an individual who shares that name with others, as shown in figure 10.4(a). Observe, now, that this situation supports the abstraction of a type, roughly ‘person named τ’ (…/τ), having each such individual as an instance. The resulting configuration, shown in diagram (b), fully conforms to the characterization of a common count noun (fig. 10.3(a)). The proper names in (7) behave as common nouns because they are common nouns.

10.2 Noun Modifiers

Nominals can be of any size and are structurally quite diverse. In large measure this is due to modifiers, which are many and varied and occur in different combinations. Here we can manage just a brief look at noun modifiers and modifying constructions.

10.2.1 Semantic and Formal Variety

Modifiers vary in their size, their grammatical category, and the nature of their semantic contribution. With respect to size, they run the full gamut: from words (Brazilian parrot), to phrases (parrot with a lisp), to clauses (parrot who kept us awake last night). And since phrasal and clausal modifiers can themselves incorporate nominals with modifiers, modifying expressions of any length can be constructed: that parrot who kept us awake last night > that parrot who kept us awake last night with a constant stream of obscenities > that parrot who kept us awake last night with a constant stream of obscenities that really shocked us > that parrot who kept us awake last night with a constant stream of obscenities that really shocked us by their level of vulgarity

(p.319) Expressions representing a number of different categories can be used to modify nouns: adjectives, prepositional phrases, present participles, past participles (both stative and passive), and infinitives. From the standpoint of CG, these categories all have something in common (§4.3.3). We can make the generalization that noun modifiers profile nonprocessual relationships.10 The exclusion of verbs, which profile processes, is a natural consequence of modifying relationships being construed holistically. A modifier is only apprehended in relation to the modified noun, which—as the profile determinant—imposes its summary view on the composite expression. Verbs are poor candidates to directly modify nouns because the sequential scanning characteristic of a process cannot be manifested in a summary view. Of course, finite relative clauses do modify nouns, despite their processual profile. Unlike other modifiers, a finite clause is grounded. And since the ground is the vantage point for apprehending the clausal content, the profiled process is to some extent viewed independently (not solely in relation to the head). The separate access afforded by grounding makes possible the sequentiality of a process conception.

In English, whether a modifier precedes or follows the head noun is partly determined by its degree of internal complexity. Adjectives, present participles, and stative past participles often consist of just a single word, in which case they precede the noun they modify: anxious woman, sleeping child, disfigured statue. They still precede the noun when they themselves are preceded by a modifying adverb: very anxious woman, soundly sleeping child, completely disfigured statue. But when a noun modifier incorporates more elaborate material, which follows it and would therefore separate it from the head noun, the entire complex modifier comes after the noun instead: woman anxious about her children, child sleeping too soundly to wake up, statue disfigured beyond recognition. Also following the head are prepositional phrases, passive participial phrases, infinitives, and finite relative clauses. These are always multiword expressions: house with a view, fire started by vagrants, person to watch out for, lawyer who has never lost a case.

At least as important as complexity, in determining a modifier's placement, is the kind of semantic contribution it makes. There is a definite tendency in English for a modifier that directly precedes the head to specify an intrinsic or permanent property, whereas post-head modifiers tend to be used for properties of a contingent or temporary character. For example, an anxious woman is probably anxious by nature, as a stable personality trait, while a woman anxious about her children may simply be waiting for the school bus to arrive. A stative participle describes a property (the “state” resulting from a change) that is likely to endure: a broken watch, toasted almonds, his sullied reputation. By contrast, a passive participle profiles an event, often with no persisting result. We can thus refer to leaves rustled by the wind, but not to *rustled leaves; to an error caught by the proofreader, but not to *a caught error; to that fire started by vagrants, but not to *that started fire. More generally, adjectives with the strongest claim to prototypicality are those pertaining to inherent (p.320) characteristics of indefinite duration: big, red, strong, flat, smart, etc. Post-head modifiers are most typically used for contingent circumstances, such as locations (e.g. the flowers on her desk), specific events (a man I insulted), and temporary situations (that spider climbing up your leg).

Even confining our attention to modifiers traditionally classed as adjectives, we observe considerable semantic diversity and numerous departures from the prototype of describing inherent properties. For example, some adjectives specify position in a sequence or location in time: my first teacher, our next president, a prior commitment, future events, a former girlfriend. Others assess the validity of the nominal type specification: genuine leather, fake Rolex, putative expert, real gold, counterfeit tickets, true patriot. These shade into adjectives indicating the referent's status with respect to a category: typical doctor, perfect circle, complete idiot, canonical example, ordinary member, representative instance. Rather than intrinsic properties, many adjectives describe how a thing is experienced by others: comfortable chair, scary movie, offensive statement, pleasant evening, welcome break, unsatisfactory answer. These in turn shade into evaluative assessments whose basis may be entirely subjective: marvelous report, charming couple, wonderful vacation, darling restaurant, horrible person. Instead of a property, certain adjectives specify which domain a thing pertains to: electrical engineer, mental hospital, corporate executive, medical textbook, culinary institute. Still other adjectives relate to quantity: abundant resources, rare coins, countless opportunities, infinite patience, meager allowance. In fact, absolute quantifiers (many, few, much, little, several, nine, etc.) qualify as adjectives from both a semantic and a grammatical standpoint.

Permissible combinations of adjectives and their sequencing are a complex matter about which I can offer only some fragmentary observations. There is arguably an overall tendency for proximity to the head to correlate with intrinsicness of the property specified. Quantifiers are always farthest from the head: nine black cats, *black nine cats, several important visitors, *important several visitors. Closest to the head are adjectives that directly pertain to type. Domain adjectives have to be adjacent to the head: excellent culinary institute, *culinary excellent institute, young electrical engineer, *electrical young engineer. Indeed, since they also resist predicative use (e.g. *The engineer is electrical), they might best be analyzed as part of the head. Also close to the head are modifiers that assess a type specification's validity: large fake diamond, *fake large diamond, cheap imitation leather, *imitation cheap leather. A number of specific patterns are well established. For example, adjectives of nationality follow those assessing validity but precede domain adjectives: true American patriot, fake Moroccan leather, British mental hospital, German corporate executive, genuine French culinary institute. Modifiers describing size, color, and material normally occur in that order: large black woolen coat, small red cardboard box, big blue wooden sign (but not *blue big wooden sign, *wooden big blue sign, or *blue wooden big sign).

The various patterns and tendencies noted are neither exceptionless nor even close to being exhaustive of English nominal structure. But while the facts of noun modification are quite complex, they also show a great deal of systematicity. On the face of it, their description requires a flexible, usage-based model of the sort outlined in chapter 8. A large inventory of constructions and constructional schemas, (p.321) characterized at appropriate levels of specificity, will serve to capture both local and global generalizations. Moreover, since units vary in degree of entrenchment and ease of activation for the sanction of new expressions, the model accommodates both exceptions and regularities that are less than categorical, representing tendencies of different strengths.

10.2.2 Canonical Constructions

The traditional distinction between complements and modifiers depends on their direction of elaboration vis-à-vis the constructional head (i.e. the profile determinant): a complement elaborates a salient substructure of the head, while a modifier contains a salient substructure elaborated by the head (fig. 7.14). Given the diversity of modifiers, it is noteworthy that this substructure (the e-site) is almost always the trajector.11 In table near the door, for example, table corresponds to near's trajector rather than its landmark. The head also elaborates the trajector of a modifying adjective or participle: small table, table sitting by the door, broken table, table polished every morning.12

A canonical modifying construction is represented at the left in figure 10.5. The noun profiles a thing and describes its basic type (X). The modifier profiles an atemporal (i.e. nonprocessual) relationship; typically, it ascribes some property (y) to its trajector. Effecting their integration is a correspondence between this schematic trajector and the profile of the noun, which thus elaborates it. The noun is the constructional head, its profile being inherited at the composite-structure level. And since it too profiles a thing, the composite expression is itself a complex noun.

Nominal Structure

Figure 10.5

(p.322) A modifier allows the profiled entity to be described in greater detail. Starting from the basic type X, modification produces a more elaborate type description consisting of X augmented by property y. Let us call this Xy. Associated with the type at each level is a mental construction referred to here as the maximal extension of that type, i.e. the set of all (contextually relevant) instances. The two maximal extensions, EX and EXy, are shown in figure 10.5 on the right. Clearly, not every instance of X is also an instance of Xy—being more detailed, Xy selects a smaller pool of eligible candidates. Further modification will shrink the pool still more.

When there is just one modifier, the same noun functions as both constructional head and head noun. Table has this dual role in a table near the door: it is both the constructional head in its combination with the prepositional phrase and the head noun for the overall nominal. The grounded structure (i.e. all but the indefinite article) is sketched in figure 10.6(a).13 With respect to the modifying construction, table is the head because it imposes its profile at the composite-structure level. With respect to the nominal as a whole, it is the head noun because its profile corresponds to the grounded entity, the composite-structure profile (grounded by a at the highest level of organization). These are distinct characterizations, even though they identify the same element as head in certain cases.

The distinction between constructional head and head noun becomes evident in nominals with two levels of modification, e.g. a small table near the door. The grounded structure is diagrammed in figure 10.6(b). At the first level of organization, small combines with table to form small table. Since the composite expression profiles the table, the constructional head is table. At the second level, small table combines with near the door to form small table near the door. The composite

Nominal Structure

Figure 10.6

(p.323) expression once more designates the table (not the locative relationship), so at this level the constructional head is small table. The constructional heads are different (table vs. small table), being locally defined in the context of each construction. By contrast, the head noun is globally defined for the nominal as a whole. The head noun specifies the nominal's basic type, an instance of which is singled out for grounding, and is the lowest-level structure whose profile corresponds to the nominal profile. In a small table near the door, the head noun is table, an instance of which is grounded by the indefinite article at the highest level of organization.

The nouns at each successive level—table, small table, and small table near the door—characterize the profiled thing in progressively greater detail. If we say (for sake of uniformity) that each describes a type of thing, the word “type” must be flexibly interpreted. It cannot be limited to the standard, culturally recognized types invoked by lexical nouns, or even to what are called types in ordinary usage: while a table is certainly a type of thing, a small table would not ordinarily be considered a type of table, and a small table near the door would never be. We resort to modifiers precisely because standard, lexically coded types are inadequate for the task at hand, and often the needed description makes reference to nonstandard properties or wholly contingent circumstances.

Furthermore, modifiers contribute semantically in a variety of ways, not merely by adding a property to those defining the basic type (Sweetser 1999). They sometimes have the more drastic effect of suspending aspects of the basic type, commenting on its validity, or indicating a restricted scope of application: flightless bird, nonalcoholic beer, fake diamond, so-called conservative, imaginary kingdom, hypothetical situation, future president. In this case the basic type is construed in relation to a broader scenario that permits such assessments. Fake, for example, invokes the cognitive model of people creating objects whose appearance is intended to bring about their erroneous categorization as instances of the type in question. But despite such qualifications, the nominal referent is still treated linguistically as an instance of the basic type. Sentences like the following are quite acceptable:

(8) (a) These fake diamonds are the only diamonds I have.

(b) Nonalcoholic beer is better than no beer at all.

10.2.3 Nominal Constituents

When there is more than one modifier, the problem of constituency rears its ugly head. How do we know, for example, that small table near the door has the constituency shown in figure 10.6(b)? Instead of ((small table) (near the door)), why not ((small) (table near the door)), or even ((small) (table) (near the door))? There is no consensus about the internal constituency of nominals, due in part to the matter being quite complex. But let me suggest a more basic reason: that there is no definite constituency. As viewed in CG, constituency is neither essential nor fundamental to grammar (§7.4). While certain hierarchical arrangements are fixed and well established, constituency groupings are often flexible and variable, if not just indeterminate. We have no reason to think that the structures constituting a symbolic assembly (p.324) all have to be arranged in strictly hierarchical fashion, nor does any single hierarchy capture all aspects of grammatical organization.

From the CG standpoint, questions of constituency have to be addressed as part of a broader consideration of symbolic assemblies. A symbolic assembly consists of semantic structures, phonological structures, and symbolic links between the two. Semantic and phonological structures can be of any size. At either pole, complex structures arise through composition, where simpler (component) structures are grouped and integrated to form more elaborate (composite) structures. A symbolic structure consists in the linkage of a semantic structure and a phonological structure (regardless of their size). It cannot be assumed that the groupings at either pole are fully nested (rather than cross-cutting), hence all arranged in a single consistent hierarchy, nor that every grouping participates in a symbolic relationship.14

With respect to this general scheme, we can characterize grammatical constituency as a matter of grouping being coordinated at the semantic and phonological poles. Suppose we have three symbolic structures: [[A]/[a]], [[B]/[b]], and [[C]/[c]]. Claiming constituency for [[A]/[a]] and [[B]/[b]] implies both that [A] and [B] are grouped (to the exclusion of [C]) at the semantic pole and also that [a] and [b] are grouped (excluding [c]) at the phonological pole. The respective outcomes of these groupings are the composite conception [AB] and the composite form [ab], whose coordination thus produces the higher-order symbolic structure [[AB]/[ab]]. In assessing the claim of constituency, the critical factor is therefore the status of [AB] and [ab]. We can justify positing these structures by showing that they actually have some linguistic manifestation. If not directly observable, they may be manifested indirectly through the necessity of referring to them for particular descriptive purposes.

Phonological grouping is more accessible to direct observation. To support the constituency shown in figure 10.6(b), we can note the possibility of a slight pause (/) at the putative constituent boundary: small table / near the door. At least for this purpose, therefore, small table constitutes a phonological grouping. Does it also represent a semantic grouping? Here we can note its role in type anaphora:

(9) Do you prefer the small table near the door, or the one next to the kitchen?

In (9) we can interpret one as referring to either table or small table. The latter option indicates the conceptual grouping of small and table, since the type they jointly define is accessible for anaphoric purposes. If small table represents both a semantic and a phonological grouping, together these amount to a symbolic grouping, i.e. a grammatical constituent.

Matters are not quite this simple, however. A case can also be made for the “flat” structure ((small) (table) (near the door)), where neither modifier combines exclusively with the head. Phonologically, an absence of intonational grouping is (p.325) characteristic of slow, deliberate pronunciation: small / table / near the door. At the semantic pole, one refers only to table in (10)(a):

(10) (a) Do you prefer the small table near the door, or the big one next to the kitchen?

(b) A SMALL table near the door is preferable to a BIG one.

Likewise, the form and meaning of (10)(b) point to the grouping ((small) (table near the door)). It is possible here to interpret one as referring to table near the door. And while it is not very natural to effect this grouping by means of a pause (?small / table near the door), it is manifested phonologically in another dimension, namely stress. In a discourse context where (10)(b) is appropriate, it is also appropriate for the nominals to be pronounced with reduced stress on all but the adjectives.15 The grouping of table and near the door is thus effected on the basis of their diminished accentual prominence.

These conflicting alternatives pose no problem for CG, which eschews the standard assumption of a single, fixed constituency. Rather than being fundamental, constituents emerge within symbolic assemblies as a special case (albeit a typical one) of the configurations their elements can assume. It is only to be expected that the same symbolic components might sometimes be grouped in alternate ways, without significantly affecting the ultimate composite structure. It may also happen that grammatical constituents do not emerge at all. Given three component elements, like small, table, and near the door, nothing prevents them from combining at a single level of organization, with no internal grouping: ((small) (table) (near the door)). And should grouping occur at one pole, there is no necessity that it be concordant with grouping at the other. In (9), for example, one can be interpreted as referring to either table or small table regardless of whether the intonation is small / table / near the door or small table / near the door. If small and table are grouped at only one pole, there is no clear basis for positing a grammatical constituent. Grammar being bipolar in nature, a claim of constituency implies that grouping occurs in parallel at the two poles.

In the absence of clear-cut evidence, the proper analysis may simply be indeterminate—certainly for the analyst, and very possibly even for speakers. Consider the sequence frisky young horse. Are there grounds for claiming that young and horse form a constituent, with frisky then combining with young horse at a higher level of organization? We have some indication of conceptual grouping in (11), where young horse is a possible antecedent of one:

(11) She wants to ride a frisky young horse, not a lazy one.

Moreover, we can perfectly well say that young and horse are grouped phonologically on the basis of adjacency in the stream of speech. Yet the case is rather tenuous. The evidence for semantic grouping disappears when one is interpreted as referring just to horse, or in uses where frisky young horse has no anaphoric connection. And (p.326) while adjacency is consistent with young and horse being a constituent, it is also consistent with the tripartite constituency ((frisky) (young) (horse)). Perhaps, then, the proper analysis is not to impose a particular analysis at all.

On the basis of more general considerations, however, I do incline toward treating young horse as a constituent in frisky young horse. Grouping is such a natural and pervasive phenomenon, and temporal adjacency such a strong grouping factor, that binary structures are often (and not unreasonably) considered the default.16 Given a modifier sequence, the head and a modifier adjacent to it would thus have a strong tendency to form a constituent, which would then combine—also on the basis of adjacency—with the other modifier. This layering of modifiers is sometimes clearly required on semantic grounds. One example is the contrasting pair counterfeit American money vs. American counterfeit money. If there were no layering, i.e. if both adjectives modified the head noun directly, we would expect the two expressions to be semantically equivalent. In fact, though, they have different meanings reflecting alternate conceptual groupings. On the one hand, counterfeit American money groups American and money, since American money represents an essential conceptual component (the item being imitated). On the other hand, American counterfeit money evokes the conceptual grouping counterfeit money, describing it as an American product (perhaps an imitation of Japanese currency).

In the case of frisky young horse, it makes no evident difference whether frisky modifies horse directly (implying a tripartite structure) or instead modifies young horse. Either compositional path yields the same composite meaning, in which the basic type specification (horse) is augmented by two distinct properties (frisky and young). Here, though, we find possible evidence for constituency by observing a correlation between intonation and word order. Frisky young horse would normally be pronounced without any pause, and without one the order of adjectives is fairly rigid—??young frisky horse is marginal at best. The reversed order is perfectly natural, however, when pauses are added: young/frisky/horse.17 This intonation suggests a flat structure, with no internal grouping, as shown in figure 10.7.

Nominal Structure

Figure 10.7

(p.327) We can therefore posit a distinct modifying construction, where pauses indicate that multiple adjectives directly modify the head at a single level of grammatical organization. If this construction indicates the absence of layering, we can plausibly suppose that the alternative pattern (e.g. frisky young horse) results from successive levels of modification.

10.2.4 Noncanonical Constructions

By permitting multiple modifiers at a single level of grammatical organization, the configuration in figure 10.7 is noncanonical for modifying constructions. You can probably guess (given the title of this section) that such constructions can deviate in other ways as well from the typical arrangement (fig. 10.5).

A seemingly drastic departure from the canon are modifying expressions where there is no head noun. Though more limited than in many languages, the phenomenon does occur in English. Apart from certain fixed expressions (e.g. the poor), it is largely restricted to complex modifiers occurring with demonstratives (especially those) and certain quantifiers: all who qualified, any with valid complaints, that which he fears the most, those ready to leave, those arriving late, those offended by my remarks. Our only concern here is with the general nature of such constructions. Why are they possible at all? How can a noun modifier modify a noun that isn't there?

In principle these constructions are unproblematic in CG, since nominals are characterized semantically, not in terms of any particular structural configuration. As long as an expression profiles a grounded instance of a thing type (even a highly schematic type), it counts as a nominal. We have seen that a nominal grounding element is itself a schematic nominal by this definition. And because it profiles a thing, there is no inherent reason why a grounding element cannot itself take a noun modifier. As shown in figure 10.8, we need only specify that the grounded thing instance corresponds to the trajector of the modifier, and that the grounding element functions as profile determinant. In the absence of a lexical head noun, the basic type instantiated by the grounded instance remains schematic. The overall type description is however augmented through the content of the modifier.

Another departure from the canon involves modifiers that play no role in identifying the nominal referent. A well-known case is the contrast between “restrictive” and “nonrestrictive” relative clauses. A restrictive relative clause serves to limit the

Nominal Structure

Figure 10.8

(p.328) pool of eligible candidates, restricting it to a subset of the basic type's maximal extension. In (12)(a)—where these candidates really are candidates—the specified property (really deserving to win) limits the pool to a single candidate, as required by the definite article:

(12) (a) The candidate who really deserves to win ran a positive campaign.

(b) The candidate, who really deserves to win, ran a positive campaign.

The information supplied by a nonrestrictive clause fails to be exploited in this manner. In (12)(b), the profiled instance of candidate is contextually identified independently of deserving to win (rather than on the basis of that property).

To represent the distinction, restrictive relative clauses are usually analyzed as being part of the nominal in question, and nonrestrictive clauses as being external to it. Supporting this analysis are the pauses (“comma intonation”) associated with nonrestrictive relatives. In and of itself, however, this structural difference is not sufficient to account for the semantic contrast. This is evident from the fact that the same contrast is observed with modifying adjectives, where there is no difference in constituency. As part of the tiny mouse, for example, tiny is used restrictively in (13)(a) and nonrestrictively in (13)(b). Only in the former does it help to identify the nominal referent. But in both uses, tiny is clearly internal to the nominal.

(13) (a) In the cage she saw a big mouse and a tiny mouse. The tiny mouse was shaking.

(b) In the cage she saw a mouse. The tiny mouse was shaking.

A general account of the restrictive/nonrestrictive contrast must therefore be independent of the structural difference observed with relative clauses. In the case of adjectives, it turns out that the contrast does not reside in the modifying construction itself, but rather in how the symbolic assembly constituting it is accessed for higher-level purposes. Whether the tiny mouse is interpreted restrictively or nonrestrictively in (13), tiny itself has the same meaning. The same is true for the composite expression tiny mouse. The semantic contrast is, instead, a matter of how this composite expression is integrated with the definite article at a higher level of grammatical organization. The article profiles an instance of some type and indicates its discourse status. The question is, which type? As shown in figure 10.5, two thing types figure in the characterization of a phrase like tiny mouse, one specified by the head (mouse) and the other by the overall expression (tiny mouse). The difference between a restrictive and a nonrestrictive interpretation depends on which of these, in the context of the discourse, is invoked in the higher-level construction.

Let us see in detail how this works. In figure 10.9, diagrams (a) and (b) show how the combines with tiny mouse in (13)(a) and (13)(b), respectively. The composite structures are omitted because the crucial difference resides in how the component structures are integrated. The internal structure of tiny mouse, which by now should be self-explanatory, is the same in both diagrams. Also the same in both is the definite article, which profiles a thing singled out by virtue of being the only instance of its type accessible in the current discourse frame. The type it schematically invokes (p.329)

Nominal Structure

Figure 10.9

is represented at the top. For sake of clarity, separate indications are given of the total type description (rectangle) and the profiled entity (circle).18

Under either interpretation, tiny mouse is grounded by the to form a nominal. As is usual for grounding constructions, their integration involves a correspondence between the profiles of the grounding element and the grounded structure. This basic correspondence—the same in both diagrams—is labeled (i). The difference between a restrictive and a nonrestrictive interpretation stems from an additional correspondence, labeled (ii). This second correspondence identifies the article's schematic type with a more specific type inherent in the grounded structure. In diagram (a), that type is tiny mouse. This yields the restrictive interpretation, since the adjectival property figures in the referent's identification. Alternatively, in diagram (b), the type is simply mouse. This yields the nonrestrictive interpretation, since the adjectival property does not figure in the grounding.19

Though both are often possible, a restrictive interpretation is certainly more usual than a nonrestrictive one. The reason is apparent from the diagrams. In diagram (a), correspondence (ii) holds between elements of the two component structures that directly combine: the and tiny mouse. In contrast, diagram (b) has the noncanonical feature that one corresponding element is found at a lower level of constituency. That is, the combines with tiny mouse partly on the basis of a correspondence that connects it, not with tiny mouse as a composite whole, but rather with one of its own component structures (mouse). While atypical, this is not unique or in any way problematic in CG—it is just one configuration that symbolic assemblies can assume.20

(p.330) We observe a comparable configuration in a modifying construction that is noncanonical in another way as well—namely, a relative clause that is nonadjacent to the modified noun. A previous example, the package arrived that I was expecting, was diagrammed in figure 7.19(b). The essentials of this construction (omitting the overall composite structure) are shown once more in figure 10.10. At issue is how the main clause (the package arrived) and the relative clause (that I was expecting) are integrated. Semantically, the relative clause modifies package, whose profile corresponds to its landmark. Grammatically, though, it combines with the main clause as a whole, for which the package functions as one component structure. The earlier diagram represented their integration by means of correspondence (i), which identifies the schematic landmark with the main clause trajector. This is noncanonical because the element elaborated usually corresponds to the elaborating structure's entire profile (not just a subpart). An elaborating structure canonical in this respect is, however, available at a lower level of organization—namely, the package (or even just package, at a still lower level). We can thus posit correspondence (ii) as the basis for integrating the main and subordinate clauses. This is noncanonical because the elaborating structure is a lower-level component. The two analyses are consistent and effectively equivalent, so there is no point in trying to choose between them.

With either correspondence, the construction is noncanonical in several further respects: because the modifier is external to the nominal, because the two are nonadjacent, and because the pivotal correspondence involves the modifier's landmark (instead of its trajector). Can we still legitimately describe this as a modifying construction? Perhaps not under a narrow definition (fig. 7.14(b) and fig. 10.5), but certainly in a looser sense emphasizing semantic relationships. Despite their nonadjacency, it seems quite evident that the package and that I was expecting constitute a conceptual grouping. As one aspect of the expression's overall meaning, these two elements are specifically construed in relation to one another in order to identify the nominal referent. But since this conceptual grouping fails to be symbolized by any phonological grouping, the nominal and the relative clause are not a grammatical constituent.

It is worth reiterating that the structures in a symbolic assembly are only partially organized into hierarchies based on coordinated grouping at the semantic and phonological poles. No single constituency hierarchy exhausts the structure of complex

Nominal Structure

Figure 10.10

(p.331) expressions, nor is every important conceptual grouping directly and individually symbolized by a phonological grouping (let alone one based on linear adjacency). An assembly may thus incorporate an unsymbolized conceptual constituent, with its own internal configuration, irrespective of how its individually symbolized conceptual components are arranged hierarchically in grammatical constituents of the classic sort. Though it is not explicitly shown, the assembly in figure 10.10 incorporates a conceptual constituent established by correspondence (ii), comprising the phonologically discontinuous elements the package and that I was expecting.21 It is due to the internal configuration of this unsymbolized constituent (analogous to fig. 10.9(a)) that the relative clause is interpreted restrictively.

10.2.5 Active Zones

We have seen that noun modification involves considerably more than meets the eye or ear. Even seemingly straightforward cases, like the direct, restrictive modification of a noun by an adjective, prove subtle and varied when examined in detail. A lot of this hidden complexity pertains to conceptual integration. We can profit by looking more closely at the basic generalization that the noun's profile corresponds to the adjective's trajector. While this is perfectly valid as a coarse-grained description, there is much to learn from the fine-grained details of particular examples.

Consider first the conceptual integration in phrases like reluctant agreement, informed consent, and conscious awareness. Since the adjective describes an attitude or mental state, one would expect its trajector to be a person (or at least a sentient creature). But in these expressions the modified noun profiles instead an abstract entity that is not itself capable of mental experience. The semantic characterizations of the adjectival trajector and the nominal profile are therefore incompatible.22 Yet we do not perceive these expressions as being semantically anomalous—their meanings are perfectly coherent. How do they escape anomaly? The answer lies in the specific details of the component meanings and their conceptual integration.

As shown on the right in figure 10.11(a), the modified nouns are nominalizations. They are based on verbal and adjectival stems (agree, consent, and aware) describing

Nominal Structure

Figure 10.11

(p.332) occurrences that are largely mental in nature. Their trajector is thus a sentient individual engaged in mental activity (represented by a dashed arrow). Semantically, the nominalization consists in the conceptual reification of this activity, producing an abstract thing (bold ellipse) which is profiled by the noun. In this semantic characterization we find a coherent basis for the noun's integration with the adjective. As the diagram indicates, the adjective's trajector is not identified with the reified activity as an undifferentiated whole, but specifically with the sentient individual engaged in it. We interpret reluctant agreement as ascribing reluctance to the person who agrees, not to agreement per se. Likewise, informed consent is consent on the part of an informed individual, and conscious awareness is the awareness of a conscious individual. Marked by shading, the individual in question is the noun's active zone with respect to its integration with the adjective. The active zone is the entity that anchors the correspondence with the adjective's trajector and directly manifests the property it specifies.

We thus observe a discrepancy between the noun's profile and its active zone for combining with the adjective. It is not evident when the construction is viewed at low resolution, where only focused elements (like profile and trajector) rise to the level of awareness. So in a coarse-grained description, it is quite correct to posit a correspondence between the adjectival trajector and the nominal profile. The discrepancy reveals itself when we look at the fine-grained details. At a higher resolution, we find that the correspondence is anchored by a particular conceptual element, the noun's active zone, which is important to its characterization without however being its referent. The active zone therefore mediates the profiled entity's participation in the adjectival relationship. For example, it is only in relation to the person who agrees that an agreement is said to be reluctant.

Further illustrating profile/active-zone discrepancy are combinations like fast car, loud parrot, and unhealthy diet. In these examples we do conceptualize the noun's referent as exhibiting the adjectival property.23 So even in a fine-grained view, it is not inaccurate to describe the adjectival trajector as corresponding to the nominal profile. Nevertheless, their conceptual integration depends on other elements that are left unexpressed. The adjectives are “scalar”. Respectively, fast, loud, and unhealthy locate their trajector—and hence the nominal profile—on scales of rate, amplitude, and health. But in each case, as shown in figure 10.11(b), its placement on the scale is mediated by another entity (represented by a shaded box). This entity is what the scale directly measures: an activity for fast, a noise for loud, and a person's physical state for unhealthy. Yet these entities are not themselves profiled by the modified noun. A car is not an activity, a parrot is not a noise, and a diet is not a person's physical state. These unprofiled entities are accessible as part of the noun's encyclopedic meaning, however: a car moves at a certain rate, a parrot makes a noise, and diet determines a person's physical state. These tacit entities are thus invoked as active zones for the noun's participation in the adjectival relationship.

Profile/active-zone discrepancy is neither unusual nor in any way problematic. It is, in fact, the usual case: an efficient way of accommodating both the multifaceted (p.333)

Nominal Structure

Figure 10.12

complexity of linguistic meanings and the special cognitive salience of particular elements. While focused elements (like profile and trajector) are the ones we primarily want to talk about, they are conceived and characterized in relation to any number of associated entities, each providing a potential basis for integration. Figure 10.12 can thus be offered as a general description of adjectival modification. In a coarse-grained view, where only focused elements are clearly evident, the adjective and noun are seen as being integrated by a correspondence between the former's trajector and the latter's profile. But at a higher resolution, where details start to emerge, we find that the focused elements merely deliver us to the relevant conceptual neighborhood, not a specific address. That is, the trajector and profile evoke arrays of associated entities—shown as ellipses—any one of which can be invoked as the specific point of connection. Only as a special case do these points of connection (the active zones) precisely coincide with the focused elements through which they are accessed.

This kind of discrepancy is not a peculiarity of adjectives or noun modifiers, but is typical of grammatical constructions in general. When two salient entities correspond, each provides mental access to an array of associated elements, any one of which can function as its active zone for this purpose. We see this in figure 10.13, where dashed arrows represent a conceptualizer's path of mental access. This sequenced mental access can be recognized as a special case of reference-point organization (fig. 3.14): by directing attention to a salient reference point (R), the conceptualizer can readily access anything in the reference point's dominion (D), one such element being the target (T). This natural and efficient strategy is a basic feature of cognitive processing, evident in numerous aspects of linguistic structure (ch. 14).

A particular sort of profile/active-zone discrepancy occurs in the modification of plural nouns. By their very nature, certain adjectives occur primarily with plurals: equal portions, parallel lines, identical descriptions, adjacent lots, various possibilities, similar faces, numerous commentators. Since they specify properties requiring multiple entities for their manifestation, these adjectives have a multiplex trajector comprising a set of constitutive elements. Their integration with a plural noun is thus straightforward. It is represented in figure 10.14(a).

Nominal Structure

Figure 10.13


Nominal Structure

Figure 10.14

But plural nouns are also modified by adjectives that specify properties manifested individually: sharp knives, even numbers, intelligent women, long novels, ripe bananas, single mothers, abstract concepts, frisky horses, white kittens, flimsy houses. How is this possible? If the adjective's trajector is uniplex, and the noun's profile is multiplex, why does their correspondence not result in semantic inconsistency? The answer, it should be clear, is that a global correspondence between focused elements leaves open the specific means of their conceptual integration. Rather than corresponding directly to the adjectival trajector, the noun's profile may simply provide mental access to what does—its active zone with respect to their integration. In the case at hand, the point of connection resides in the individual elements that constitute the plural mass.24 As shown in figure 10.14(b), each constitutive element is an active zone identified with the adjective's trajector. In the composite conception, the adjectival property is thus ascribed to each of these elements individually, not the nominal referent as a whole.

10.3 Classification and Quantification

Two aspects of nominal organization that deserve a closer look are noun classes and quantifier constructions. Their intimate relationship is evident in noun classifiers, a major phenomenon in many languages.

10.3.1 Noun Classes

Grammatical classes have varying degrees of semantic motivation. At one extreme lie fundamental and universal categories, notably noun and verb, which are claimed to have a fully consistent (albeit schematic) conceptual basis. At the opposite extreme are classes defined solely by occurrence in a particular grammatical construction with no possibility of semantic characterization. Prime examples here are elements that exhibit some morphological peculiarity, such as nouns where final f changes to v in the plural: wife/wives, leaf/leaves, etc. Most classes lie somewhere in between.

The extreme cases reflect different rationales for classification. Classes like noun and verb are based on fundamental cognitive abilities (like grouping and sequential scanning) inherent in the conception of experientially grounded archetypes (objects (p.335) and events). While these categories are available for grammatical exploitation, they are not defined in terms of any particular grammatical behavior, and there need not be any single construction in which all members of a class participate. We cannot, for example, define a noun as an element that occurs with articles, since many nouns do not (e.g. proper names and pronouns). Nevertheless, reference to a noun is part of the characterization of innumerable grammatical patterns, even if that alone is insufficient to delimit the set of participating elements. It is clear, moreover, that much of linguistic structure subserves the need for expressions that profile things, ranging from lexical nouns to full nominals (also classed as nouns in CG).

At the other extreme, classes with arbitrary membership arise through accidents of history and the vicissitudes of usage. They reflect the brute-force fact that particular ways of talking have been conventionally established and have to be learned in the acquisition process. For contemporary speakers, the reason wife has wives as its plural rather than *wifes is simply that people talk that way. They similarly have to learn—essentially by rote—that a handful of other nouns pattern like wife (including leaf, knife, thief, calf, life, loaf, and half) but that many others do not (e.g. fife, puff, reef, safe, belief, whiff, cuff, cliff, staff, waif, chief, plaintiff). This minor variant of the plural construction, where f alternates with v, implicitly defines a class consisting of those nouns which participate in it. The members of this class cannot be predicted on the basis of their meaning. They have no uniform semantic characterization distinguishing them from other nouns ending in f.

Every construction defines a class consisting of the elements that appear in it. Since they pertain to the problem of distribution—that of specifying which elements occur in which patterns—the classes defined in this manner are called distributional classes. Knowledge of these classes is clearly vital to speaking a language properly. How, then, is this knowledge represented? In gaining control of this aspect of linguistic convention, what does a speaker specifically have to learn? For the two extremes, simple answers can be imagined. It may well be, however, that neither extreme situation is ever realized in unadulterated form.

Suppose, first, that a distributional class coincides exactly with a semantically defined category (like noun, inanimate noun, or perfective verb). The elements that occur in the construction can then be specified just by invoking this category. What this amounts to, in CG terms, is that the schema for the construction incorporates the schema defining the category as one of its component structures. In this way, the constructional schema correctly describes the distributional facts and captures the appropriate generalization. Even so, it is probably not exhaustive of a speaker's distributional knowledge. We saw in chapter 8 that constructional schemas describing general patterns coexist with more specific structures reflecting their conventional implementation. It is often these lower-level constructional schemas, even those incorporating particular lexical items, that are most important in determining conventional usage.

Suppose, on the other hand, that a distributional class is completely random: there is no semantic (or other) basis for even partially predicting its membership. In this case its members simply have to be learned individually, as an arbitrary list. What this amounts to, in CG terms, is that each element's occurrence in the pattern is specifically learned as a distinct conventional unit. For example, expressions like wives, leaves, knives, and so on have to be learned as such, in addition to the (p.336) constructional subschema representing their abstracted commonality. But seldom, if ever, is a class completely random. Although the list of members may be arbitrary, so that each has to be specifically learned as such, they still tend to cluster in certain regions of semantic and phonological space. Certain members may be similar enough, semantically or phonologically, that their inclusion in the pattern is mutually reinforcing. Schemas can then emerge to capture these local regularities (schematization being merely the reinforcement of recurring commonalities). If so, there is more to a class than just a list of members. A class exhibits some measure of coherence and organization.

Even the class of f/v nouns amounts to more than just an unstructured list. While it is hard to make a case for any semantic grouping, in phonological space its members are far from being randomly distributed. First, they are all monosyllabic. Beyond this, most are subsumed by a number of intersecting clusters: (i) wife, life, knife; (ii) leaf, thief, sheaf; (iii) life, leaf, loaf; (iv) calf, half. Clusters (i) and (ii) are based on the vocalic nucleus. A member of each is also part of cluster (iii), based on the initial consonant. As for cluster (iv), calf and half share not only their vowel but also the orthographic peculiarity of silent l. In no way do these local regularities obviate the need to specifically learn each plural form. Such regularities do, however, facilitate the acquisition process, giving rise to low-level schemas that help maintain a class and sometimes even attract new members into it.

Distributional classes are quite varied as to the nature and extent of regularities in their membership. These can be based on either semantic or phonological properties, or a combination of the two. A single generalization can be extracted, or any number of local ones. Generalizations differ in their level of specificity, the proportion of the total membership they subsume, and their accessibility for the sanction of new expressions. Despite their variety, CG accommodates distributional classes straightforwardly and in a unified manner—by positing appropriately configured networks of constructional schemas (§8.3). The schemas comprising a network are characterized at various levels of abstraction and further differ in degree of entrenchment and ease of activation. Each captures a local or global generalization concerning the elements permitted in the construction. Whether a given generalization extends to new cases depends on whether the schema that embodies it is able to win the competition for selection as categorizing structure (§8.2.1).

In this usage-based approach, membership in a distributional class does not require separate listing. Rather, it is inherent in a full description of the construction in terms of which the class is defined. Grammatical patterns are not learned in isolation, but are abstracted from expressions consisting of specific elements in every position. In the network describing a pattern, each constructional schema makes some specification concerning participating elements. The lowest-level schemas incorporate particular lexical items that have been conventionally established in the pattern. It is through this combination of regularity and idiosyncrasy (in any proportion) that the proper elements are specified as occurring in it.

Conversely, lexical items are learned in the context of larger expressions containing them. Part of a lexeme's characterization is thus a set of structural frames representing the constructions it occurs in. These frames are nothing other than the lowest-level constructional schemas for the patterns in question (fig. 8.13). In a (p.337) usage-based perspective, therefore, a lexeme's membership in distributional classes does not require separate specification but is inherent in its full description. For instance, the constructional subschema [send NML NML] provides the information that send occurs in the ditransitive construction (§8.3.2)—no special marking is required. Likewise, the very existence of the conventional unit wives indicates the participation of wife in the pattern of f/v alternation.25

Important though they are, distributional classes do not exist in their own right as explicit objects of awareness or even as distinct linguistic entities. In a CG account, they are seen instead as being intrinsic to the full, proper characterization of constructions and lexical items. Much the same is true for classes defined semantically. What makes something a noun, for example, is not that it bears any special label, or is found on a list of category members that have to be learned as such. What makes it a noun is rather an intrinsic aspect of its meaning: the fact that it profiles a thing. In similar fashion, the count/mass distinction depends on whether this thing is construed as being bounded, and the common/proper distinction depends on whether the specified type is conceived as having multiple instances (fig. 10.3). When we talk about the schema for such a class, or categorizing relationships between the schema and its members, it is not implied that these are separate and distinct. On the contrary, schemas are immanent in their instantiations and thus inherent in their conception.

We have been considering two basic kinds of classes: distributional classes, and those defined in terms of meaning. A distributional class comprises the elements that occur in a particular pattern or construction. Its members are determined by this single property, irrespective of any semantic regularities they might exhibit. On the other hand, fundamental categories like noun and verb are claimed in CG to have a semantic definition. Although they are central to grammatical organization, and their members participate in many constructions, no particular construction is invoked for their characterization. These two basic sorts of classes are not the only possibilities. Languages also present us with intermediate cases, classes that are partially but not exclusively semantic, and are recognized through occurrence not just in one but in a number of constructions.

I have in mind “gender” classes, so called because the prime examples are noun classes bearing traditional labels like “masculine”, “feminine”, and “neuter”. Linguists are fond of pointing out the inadequacy, if not the downright folly, of such labels. On what rational basis can one say, for instance, that German Löffel ‘spoon’ is masculine, Gabel ‘fork’ is feminine, and Messer ‘knife’ is neuter? These classes are posited not because their members exhibit any consistent meaning, but rather because they pattern alike grammatically, in terms of their inflectional endings and the forms of cooccurring elements (like articles, demonstratives, and adjectives). Still, the traditional labels were not chosen arbitrarily. They are indeed semantically appropriate for a substantial range of vocabulary, where they do reflect biological gender. For instance, Mann ‘man’ functions grammatically as a masculine noun, Frau ‘woman’ is feminine, and Kind ‘child’ is neuter (since a child can be either male or female).26

(p.338) It is quite common for the nouns of a language to be divided into categories of this general sort. Languages vary in the number of classes they exhibit, their degree of semantic coherence, and the semantic properties they are partially based on. In addition to gender, classes can be anchored by a wide range of culturally salient notions: ‘person’, ‘animal’, ‘deity’, ‘artifact’, ‘instrument’, ‘tree’, ‘plant’, ‘fruit’, ‘collection’, ‘liquid’, ‘food’, and so on. Languages also differ in the array of grammatical constructions on the basis of which the classes are posited. While descriptively quite complex (from the standpoint of membership, grammatical ramifications, and historical development), these classes can be seen as a natural outcome of general processes. First, they reflect the emergence of complex categories (describable as networks) by extension from a prototype. Second, lexical items are learned from their occurrence in particular structural frames, which are thus retained as an aspect of their characterization.

For convenient illustration, we can take a quick look at Spanish nouns, which clearly divide into two broad categories. The semantic distinction which anchors them is gender: nouns like hombre ‘man’, hijo ‘son’, and tío ‘uncle’ are masculine, whereas mujer ‘woman’, hija ‘daughter’, and tía ‘aunt’ are feminine.27 However, this category distinction extends to all nouns in the lexicon, for most of which the notions ‘male’ and ‘female’ are irrelevant. For instance, tenedor ‘fork’, mes ‘month’, and techo ‘roof’ are masculine, whereas cuchara ‘spoon’, semana ‘week’, and casa ‘house’ are feminine. In general, therefore, the basis for categorization is grammatical rather than semantic. Masculine and feminine nouns are distinguished by a whole series of grammatical properties, of which just three will be mentioned: they differ in the form of the definite article (el hombre ‘the man’ vs. la mujer ‘the woman’), the indefinite article (un hombre ‘a man’ vs. una mujer ‘a woman’), and certain modifying adjectives (hombre simpático ‘nice man’ vs. mujer simpática ‘nice woman’).

Given this array of data, what sorts of linguistic units can we posit, in accordance with the content requirement? First, particular expressions can coalesce as units if they occur with any frequency. This is especially likely in the case of articles, so we can reasonably posit a large number of conventional units such as the following: [un hombre], [la mujer], [un tenedor], [el techo], [una semana], etc. Also permitted by the content requirement are schematizations of occurring expressions. On the basis of expressions like el hombre, el hijo, and el tío, we can therefore posit the constructional schema [el Nm], where Nm indicates a noun referring to a male. This schema represents an important generalization concerning the use of el. But it does not tell the whole story. Since el is further used with nouns like tenedor, mes, techo, and countless others, the highest-level schema simply specifies its occurrence with a noun: [el N]. Analogously, expressions like la mujer, la hija, and la tía give rise to the constructional schema [la Nf] (where Nf is a noun referring to a female), while the further use of la with cuchara, semana, casa, etc. supports the higher-level schema [la N].

Likewise, schemas emerge at different levels of abstraction from expressions involving the indefinite article or a modifying adjective. Schemas incorporating the (p.339)

Nominal Structure

Figure 10.15

indefinite article include [un Nm], [un N], [una Nf], and [una N]. For adjectives we can posit [Nm …o], [N …o], [Nf …a], and [N … a] (where …o and … a represent adjectives ending in o and a). Of course, these various schemas do not exist in isolation from one another. Connecting them are categorizing relationships (also permitted by the content requirement), such as [[el N] → [el Nm]], [[el Nm] → [el hombre]], and [[una Nf] → [una mujer]]. In this way, constructional schemas and instantiating expressions are organized into networks representing both general grammatical patterns and their specific implementation in conventional usage. Fragments of two such networks are shown in figure 10.15.

The networks for constructions are themselves connected in various ways. In particular, a lexical item appearing in multiple constructions provides a point of overlap among them. A frequent noun like hombre, for example, is no doubt well established in a number of structural frames: [el hombre], [un hombre], [hombre…o], etc. Each of these complex units is part of the network describing a general pattern. The frames are not disjoint, for they share the component hombre. And since the frames intersect in this manner, so do the networks they represent. In fact, each lexical item that appears in all three patterns is a point of intersection for them.

That hombre appears separately in each formula is solely due to the limitations of this notational format. This should not obscure its essential identity in the three frames, or the fact that it ties them together to form a complex symbolic assembly in the paradigmatic plane. To show its identity more directly, we might adopt an alternate notation: [hombre {[el X], [un X], [X…o]}]. This is meant to indicate that hombre functions in the X slot of all three frames. The outer brackets represent the overall paradigmatic assembly they constitute. They can also be taken as indicating the full characterization of hombre—not just its basic form and meaning but also the structural frames in which it figures.

(14) [[el hombre] [un hombre] [hombre …o]] = [hombre {[el X], [un X], [X …o]}]

Of course, other lexical items give rise to similar paradigmatic assemblies: [hijo {[el X], [un X], [X…o ]}], [techo {[ el X], [un X], [X…o]}], etc. And to the extent that these assemblies are analogous, they are themselves susceptible to schematization. From animate nouns we thus obtain the schematic assembly [Nm {[el X], [un X], [X… o]}]. From the full array of nouns that pattern in this manner, we obtain (p.340)

Nominal Structure

Figure 10.16

the still more schematic assembly [N {[el X], [un X], [X …o]}]. Naturally, the same developments occur with feminine nouns, resulting in the schematic assemblies [Nf {[la X], [una X], [X …a]}] and [N {[la X], [una X], [X …a]}]. This is sketched in figure 10.16.

The networks comprising these schemas and their instantiations constitute a description of masculine and feminine nouns in Spanish.28 The highest-level schemas, [N {[el X], [un X], [X …o]}] and [N {[la X], [una X], [X … a]}], describe the categories just in terms of their grammatical behavior. The lower-level schemas [Nm {[el X], [un X], [X … o]}] and [Nf {[la X], [una X], [X … a]}], referring specifically to the semantic properties ‘male’ and ‘female’, represent the category prototypes. In this way we succeed in capturing the semantic anchoring of the categories, as well as their basically grammatical nature. As outlined in chapter 8, these networks constrain the conventional behavior of familiar nouns and are readily invoked as models for new ones. And although the classification of many nouns is purely arbitrary (a matter of historical accident), a proper description of the categories requires nothing more than assemblies of symbolic structures.

10.3.2 Quantifier Constructions

When the nouns of a language divide into semantically anchored classes, these are often manifested grammatically in the form of classifiers. Noun classifiers, which range in number from a handful to several score, are themselves schematic nouns. Their meanings are schematic with respect to the nouns they classify, or at least those representing the category prototype.29

Grammatically, noun classifiers are closely tied to quantification, grounding, and anaphoric reference. They typically combine directly with a quantifier, demonstrative, or possessive to form a schematic nominal. This nominal then combines with a lexical noun, deriving a higher-level nominal more specific in type. Here are some examples from Thai: (p.341)

(15) (a) khruu lâaj khon (b) mǎa tua nán (c) sôm hâa lûuk

teacher three person dog body that orange five fruit

‘three teachers’ ‘that dog’ ‘five oranges’

In (15)(b), for example, the classifier tua combines with the demonstrative nán to form the schematic nominal tua nán ‘that body’. The schematic type ‘body’ can then be elaborated by a noun like mǎa ‘dog’ at a higher level of structure. Alternatively, the schematic nominal can stand alone and be used anaphorically. In this case tua nán is comparable to that one, except that tua indicates the referent's general category.

Compared with the canonical structure described in §10.1.1, nominals based on classifiers have a fundamentally different organization. In a language like English, it is typically a lexical noun that lies at the core of a nominal, with quantification and grounding representing the outermost structural layers: (those (five (rotten (oranges)))). By contrast, in expressions like (15) the lexical noun is peripheral and even optional. The structural core consists of a grounded or quantified classifier, such as tua nán, to which a lexical noun may then be added: (mǎa (tua nán)). This organization is shown in figure 10.17, where b is the schematic type (glossed here as ‘body’) specified by the classifier, and D the specific type ‘dog’. Observe that the classifier is actually the grounded noun. It is thus an instance of tua ‘body’ that the demonstrative singles out as nominal referent. The lexical noun mǎa ‘dog’ is not directly grounded but simply elaborates the grounded entity's type. Indirectly, of course, an instance of the lexically specified type is singled out and grounded by the nominal overall.

The same lexical noun is often able to occur with alternate classifiers, each of which imposes a different construal on its content. In Mandarin, for example, the noun shéngzi ‘rope’ occurs with either tiáo ‘long, thin object’, juǎn ‘roll’, or duàn ‘segment’, thereby portraying the referent just as a rope, as a coil of rope, or as a piece of rope: yī-tiáo shéngzi ‘one rope’, liǎng-juǎn shéngzi ‘two coils of rope’, zhè-duàn

Nominal Structure

Figure 10.17

(p.342) shéngzi ‘this piece of rope’. With the second two examples, we begin to see a transition between simple classification and a related phenomenon that is commonly associated with the same grammatical form. It makes perfect sense to say that a rope belongs to the class of long, thing objects. But can we say that a coiled rope belongs to the class of rolls? Or a short one to the class of segments? It is instructive here to compare the Mandarin expressions with their English equivalents. We use coil and piece to translate juǎn and duàn, but nothing to translate tiáo: a coil of rope, a piece of rope, a rope.

This other phenomenon, closely associated with classifiers cross-linguistically, is the unitization of a mass.30 Semantically, it consists in some portion of a mass being conceived as a discrete, bounded unit. Though composed of the “substance” constituting the mass, this unit exists in its own right as a distinct and separate entity. A coil, for example, is not just rope or wire, but something we can recognize by its shape irrespective of its substance. Similarly, a stack of plates is not just plates—it is also a stack, a higher-order entity with its own form and function. Grammatically, unitization is effected by count nouns that profile collections, configurations, constitutive elements, or containers: flock, pack, cluster, pile, grain, speck, drop, slice, chunk, cup, bottle, bag, etc. Unitization reflects our propensity to conceptualize the world in terms of discrete objects that we can apprehend as wholes and deal with individually. It allows the application to masses of the semantic and grammatical apparatus based on count nouns.

A unit and the portion of the overall mass it delimits are essentially coextensive. In terms of their real-world reference, a drop of water is the same as the water constituting it, and a flock of geese consists of nothing more than geese. A container is more distinct from its contents—if I remove the wine from a bottle of wine, the bottle still exists. Still, the volume a container encloses is spatially coextensive with a mass that fills it. Due to their coincidence, it may be indeterminate whether the composite expression refers to the unit or just to the mass it delimits. The point is often moot: if I see a flock of geese, I see both the flock (which consists of geese) and the geese (which constitute the flock). Yet the distinction is sometimes indicated linguistically. In (16)(a), the relative salience of unit and delimited mass is reflected in the choice of anaphoric pronoun (it vs. they), as well as number marking on the verb (was vs. were). With container nouns, the composite expression refers either to the combination of container and content, as in (b), or else to just the content, as in (c). What is not permitted is for the expression to refer exclusively to the container, as in (d).

(16) (a) I saw a flock of geese. {It was/They were} clearly visible against the blue sky.

(b) She stacked three bags of mulch in the wheelbarrow.

(c) She spread three bags of mulch around the roses.

(d) *The bags of mulch were plastic.

(p.343) The structure of these expressions, exemplified by flock of geese, is sketched in figure 10.18. A key factor is the meaning of the preposition of.31 Described schematically, of profiles an intrinsic relationship between two things. Prototypically, for example, its trajector is an intrinsic subpart of its landmark (e.g. the tip of my finger).32 The phrase of geese thus designates an intrinsic relationship (represented by a double line) that its trajector bears to a mass identified as geese. In accordance with the general pattern for noun modification, its trajector corresponds to the thing profiled by flock. A unit noun like flock also invokes an intrinsic relationship: one of coextension between the profiled unit (outer circle) and a replicate mass (inner circle). Further integration of the two component structures is thus effected through identification of the intrinsic relationships each is based on. This is reflected diagrammatically in a second correspondence line, connecting the prepositional landmark with the mass unitized by flock.

The basic schema for noun modification identifies the modified noun as constructional head (profile determinant). Accordingly, the composite expression flock of geese is expected to profile the bounded unit flock rather than the mass of geese it delimits. We have seen, however, that either construal is possible. The two interpretations are related metonymically, a matter of alternate profiles on the same conceptual base. The possibility of profiling either the unit or the coextensive mass instantiates a general metonymic pattern of English.

This unitizing construction is the source of more grammaticized expressions which are becoming central to the English quantifier system. In particular, a lot of and a bunch of are well-entrenched alternatives to the absolute quantifiers many and much:

Nominal Structure

Figure 10.18


(17) (a) We invited {many / a lot of / a bunch of} people to the party.

(b) They drank {??much/ a lot of} beer.

Pivotal to their grammaticization is the fading from awareness of the original concrete sense of lot and bunch, where lot designates a collection of objects (especially for auction) and bunch a cluster of objects bound together (e.g. bunch of grapes). This loss of content has nearly reached the point where lot and bunch are pure indications of quantity, denoting a certain extension along a quantity scale. All that remains of their unit sense is the function of delimiting a mass in quantitative terms. And since there is no longer any concrete unit to refer to, the metonymic shift in figure 10.18 becomes obligatory. As an inherent aspect of their characterization, composite expressions of the form a {lot/bunch} of N can only profile the quantified mass (N), not the quantifying unit (lot or bunch). This is evident from both anaphoric reference and number marking on the verb, as seen by comparing (18) with (16)(a):

(18) I saw a {lot / bunch} of geese. {*It was / They were} clearly visible against the blue sky.

Also occurring in an of-construction are the quantifiers discussed in chapter 9: all of those geese, some of the geese, many of these geese, three of his geese, etc. One difference from expressions with unit nouns is that the prepositional object is overtly grounded and definite. The landmark is thus a contextually delimited mass established independently of the quantification. Another difference is that the noun modified by the of-phrase is fully grammaticized as a quantifier,33 with no vestige of a unit sense analogous to flock, drop, or bunch. Consequently, reference to the quantified mass does not result from metonymic shift but directly reflects the quantifier's meaning. The overall expression therefore designates a quantified portion of the mass singled out by the prepositional object. The relation this subpart bears to the whole is identified as the intrinsic relationship profiled by of.

The case of some is diagrammed in figure 10.19. As a grounding quantifier, some singles out a profiled thing instance via the quantificational strategy (fig. 9.6). Specifically, its profile represents a non-empty proportion of the maximal extension of a type, Et (fig. 9.12(c)). In some of the geese, what counts as the maximal extension of geese (Eg) is the contextually delimited mass singled out by the. The portion profiled by some corresponds to of's trajector, and Eg to its landmark. The composite expression therefore designates a non-empty proportion of the mass identified in the discourse as the geese.

The solid arrow in figure 10.19 indicates that the profiled mass represents a limited portion of the maximal extension. In this construction, the relation the profile bears to Et is identified with the intrinsic relationship evoked more schematically by of. Among the grammaticized quantifiers, all stands out as having the special property that the profiled mass and the maximal extension are identical. If we think of a quantifier as singling out the profile by restricting the maximal extension, the restriction imposed by all is zero. All is therefore vacuous with respect to a prototypical of (p.345)

Nominal Structure

Figure 10.19

relationship, where the trajector is an intrinsic subpart of the landmark. This special semantic property is reflected in a special grammatical behavior. Alone among the quantifiers, all occurs in an alternate construction lacking the preposition whose meaning it subverts: we can say either all of the geese or just all the geese (but not *{most/some/any/each/many/three} the geese).34 This construction is diagrammed in figure 10.20(a). It is simply a matter of all combining directly with the geese based on the same correspondence as the one connecting Et with the prepositional landmark in figure 10.19.
Nominal Structure

Figure 10.20

(p.346) In all the geese, the grounding quantifier combines with a component structure which is itself a grounded nominal. The composite expression is thus a higher-order nominal that profiles a mass coextensive with one independently singled out as a discourse referent. In this respect all the geese contrasts with the simple grounding expression all geese, shown in figure 10.20(b). The only difference in constructions (a) and (b) is that geese is ungrounded in the latter. Consequently, the default interpretation of all geese is that the mass it designates is coextensive with geese in general, rather than any particular subset.

10.4 Inflection and Agreement

A final dimension of nominal structure, quite extensive in many languages, is the marking of nouns to indicate their category, semantic properties, or relationship to other elements. These markings vary as to how internal or intrinsic they are to a noun, with respect to both their semantic import and their formal manifestation.

10.4.1 What's in a Noun?

The most internal markings are those deriving nouns from other categories, exemplified by endings like -ness, -er, -ion, and -ity: sadness, firmness, emptiness; driver, boiler, teacher; demonstration, persuasion, digression; laxity, diversity, stativity. These are internal to a noun in the strong sense that it is only through their effect that a noun exists at all. Semantically, they create a noun by shifting the profile to a thing associated with the relationship designated by the verb or adjective they attach to. They are thus internal to a noun morphologically because only the composite expression they derive is so categorized: [[sad] ADJ -ness] N .

Markings that derive nouns from other categories are traditionally labeled derivational. Of more immediate concern are nonderivational markings, usually referred to as inflectional. Not a little ink has been spilled over the issue of how and where to draw the line, but from a CG perspective such discussion is largely beside the point. The very notion that there is a specific line of demarcation rests on theoretical assumptions (e.g. a categorical distinction between lexicon and grammar) viewed in CG as being both gratuitous and empirically problematic. Imposing a strict dichotomy is less than helpful in the case of number and gender, which are usually regarded as inflectional but also function in the derivation of nouns.

Consider pluralization. It is generally thought of as inflectional because it applies to a noun rather than deriving it, tends to be marked by stem-external means (e.g. the suffix -s), and often participates in “agreement” phenomena (a plural noun requiring a plural verb or adjective). Yet these points are anything but definitive. While pluralization does apply to a noun, it also derives one—a higher-order noun that specifies a distinct type representing a different category (mass instead of count). Morphologically, its position vis-à-vis the stem is not exclusively either internal or external, for it is often marked by ablaut (e.g. man vs. men), reduplication (Hopi saaqa ‘ladder’ vs. saasaqa ‘ladders’), or even full stem suppletion (Hopi wuùti ‘woman’ vs. momoyam ‘women’). Nor, in CG, is agreement viewed as the (p.347) “copying” of inflectional features or as having any particular diagnostic value. It is simply a matter of the same information being symbolized in multiple places. As such, it is just a special case of conceptual overlap, which is characteristic of all grammatical constructions.

Also considered inflectional, primarily because they participate in agreement phenomena, are markings for gender and similar categories. With respect to a noun, however, these are often internal and derivational even in the strong sense of being responsible for its categorization as such. Let us return for a moment to Spanish gender (§10.3.1), this time focusing on the endings -o and -a, the general markings for “masculine” and “feminine”. For animate nouns, those labels are semantically appropriate: hijo ‘son’ vs. hija ‘daughter’; amigo ‘male friend’ vs. amiga ‘female friend’; gato ‘male cat’ vs. gata ‘female cat’; etc. Through standard morphological analysis, these can be divided into a noun stem that specifies a basic type (hij- ‘child’; amig- ‘friend’; gat- ‘cat’) and a gender-marking suffix. Yet this is not viable for the large numbers of inanimate nouns where gender is likewise marked by -o and -a. For instance, techo ‘roof’, vaso ‘glass’, and año ‘year’ function grammatically as masculine nouns, and casa ‘house’, mesa ‘table’, and semana ‘week’ as feminine nouns. But with these there is no separate stem to which the ending attaches. Only as whole does techo mean ‘roof’, casa ‘house’, and so on. In such cases, we have to say that the gender marking is part of the noun itself.

How might these endings be analyzed? They are problematic in classic morphological analysis, where morphemes are construed metaphorically as building blocks. One entailment of this metaphor is that a word should be exhaustively divisible into discrete component morphemes. Consequently, the endings -o and -a in nouns like techo and casa cannot be recognized as morphemes, for this would leave a nonmorphemic residue (tech, cas). In CG, however, morphological description eschews the building-block metaphor. It is based instead on symbolic assemblies, in which composite structures are entities in their own right, motivated by component structures without being literally constructed out of them. It is therefore unproblematic for a composite expression to incorporate material not inherited from any component, and even for a construction to be defective in the sense of having only one component, corresponding to just a portion of the composite structure. Nouns like techo and casa can thus be described as shown in figure 10.21. In contrast to animate nouns like gato/gata, they comprise an assembly with a composite structure and only one symbolic component. Though noncanonical, this is one configuration that symbolic assemblies can assume.

Nominal Structure

Figure 10.21

(p.348) What about the meanings of -o and -a? With animate nouns they contribute the meanings ‘male’ and ‘female’, which anchor the gender categories. Whether they can be ascribed a specific meaning in cases like techo and casa is a matter for careful investigation, but let us assume they cannot. In this event -o and -a are still considered meaningful in a CG perspective. Their meanings are simply quite schematic, probably to be identified with that of nouns as a class (i.e. they profile things). If so, they represent a linguistic phenomenon attested in numerous languages: the existence of morphological elements serving to mark nouns as such without having any additional content or function.35 Of course, their status as purely grammatical markers is wholly consistent in CG with their treatment as meaningful symbolic elements.

In terms of being internal or external to a noun, number and gender can thus be seen as intermediate. More clearly external are markings whose semantic and grammatical import specifically pertains to relationships with other elements in a larger configuration. With respect to a lexical head, grounding has this character: semantically, it specifies how the nominal referent relates to the ground; grammatically, it is both external to the head noun and tends to be manifested at the outermost layer of nominal organization.36 More external still are markings that indicate a nominal's syntactic role. The main examples are possessive markers, case inflections, and adpositions. These pertain to an entire grounded nominal (not just a head noun) and specify how it is connected with other elements in larger symbolic assemblies.

A possessive nominal functions as a grounding element with respect to a higher-order nominal (ch. 14). For example, the man's serves to ground the head noun dog in forming the higher-order nominal the man's dog. While internal to the latter, the possessive marker 's is clearly external to the lower-level nominal, the man. Whereas the plural of man is marked internally by ablaut (men), the possessive is formed by attaching 's to the nominal as a whole.37 Indeed, it is not even a suffix, in the narrowest sense, but a clitic that attaches to the nominal's final word. We observe the contrast with a complex expression like the king of Denmark: plural the kings of Denmark vs. possessive the king of Denmark's (fig. 6.13(a)). Semantically as well, possession involves the entire nominal, not just the head noun. To serve its grounding function, the possessor itself has to be singled out as a grounded instance of its type.

Just as possessive marking indicates a nominal's role in a higher-order nominal, other markers specify a nominal's role in a clause. They can mark its grammatical status as clausal subject or object. In CG, of course, these are meaningful notions, a matter of primary vs. secondary focal prominence (trajector vs. landmark). Other roles have more tangible conceptual import. It might be specified, for example, that a nominal referent functions as an agent, instrument, patient, recipient, beneficiary, or location with respect to the clausal process. Formally, these roles can marked by inflection, affixation, or a separate word or particle. In unipolar terms their placement (p.349) varies. A preposition, for instance, would normally precede the entire nominal, whereas case inflection appears inside it, on the head noun or on multiple nominal components. Still, in terms of bipolar composition (§6.3) these elements combine with the nominal as a whole.

Semantically, these markers differ in the extent to which they invoke a relationship distinguishable from the profiled clausal process. Toward one extreme lie the prepositions in (19), which mark the knife as an instrument in the process of cutting, Sarah as the agent, and her brother as the beneficiary. As in other uses, these prepositions profile relationships and thus have their own trajector/landmark organization. They serve to specify the clausal role of the participant introduced as their landmark (the prepositional object). In this grammatical use, the relationships they designate are precisely those which hold between a clausal process (their trajector) and a participant in it. These markers tend to be used for participants whose involvement is more peripheral, and are said to introduce them “periphrastically”.

(19) The meat was cut with a knife by Sarah for her little brother.

At the other extreme are markers that do not invoke any relationship other than the process itself, but simply register the status of a focal participant as such. A simple example is the object-marking suffix -i of Luiseño, as in (20). It does not profile a relationship, have its own trajector/landmark organization, or introduce a participant beyond those invoked by the verb—it merely identifies ’awaal ‘dog’ as the clausal landmark. ’awaal is said to be a “direct” participant of ’ari ‘kick’ because it is not introduced periphrastically, by means of a separately coded relationship.38

(20) Nawitmal=upil ’awaal-i ’ar-ax. ‘The girl kicked the dog.’

girl=3s:PAST dog-OBJ kick-PAST

If it does not profile a relationship, what does a marker like -i designate? Since it does not alter the nominal character of the element it attaches to, it may itself profile a thing. This thing is specified only as being the landmark of some process, as shown in figure 10.22 (where a schematic process is represented by an arrow with ellipses). A correspondence equates the profiles of the noun and the suffix, so the composite expression ’awaali designates a dog with the role of processual landmark. At a higher level of organization, the schematic process evoked by -i is identified with the specific process profiled by the verb, in this case ’ari ‘kick’.39 In this way the noun is explicitly marked as object of the verb and the clause it heads. (p.350)

Nominal Structure

Figure 10.22

10.4.2 Morphological Realization

Luiseño -i exemplifies what is traditionally referred to as a case marker: one that profiles a thing, combines with a nominal as a whole, and specifies its syntactic role in larger structures. Of course, this is not a standard definition, since case is usually considered meaningless (purely grammatical in nature). More pertinent here, though, is the question of whether it really combines with a nominal as a whole. From the standpoint of form, the matter is not all that obvious. Often case is marked on the head, which may be internal to the nominal, or on multiple components, including modifiers and especially a grounding element. For Luiseño, which allows covert grounding, ’awaali is indeed a full nominal in (20). But a modifying adjective is also marked for object case—for example, yot-i ’awaal-i ‘big dog OBJ’—and so is a demonstrative: wunaal-i yot-i ’awaal-i ‘that big dog OBJ’. Yet we do not want to claim, semantically or grammatically, that these elements function individually as objects. Here traditional accounts would seem to be on the right track in saying that case is marked redundantly. This “agreement” in case helps identify the marked elements as all belonging to the same nominal constituent.

Let us see how this works for yoti ’awaali ‘big dog OBJ’. Figure 10.23 shows yot ‘big’ as first combining with ’awaal ‘dog’ in the normal construction for adjectival modification.40 With respect to bipolar composition, it is the composite expression yot ’awaal that is marked for object case. Semantically this is quite straightforward: the thing profiled by yot ’awaal is identified by correspondence (i) with the one specified by the object marker as being a processual landmark. What makes the construction noncanonical is the nature of their phonological integration. For one thing, it is atypical because, in unipolar terms, the suffix is doubly manifested, appearing on each nominal component rather than in just one place. It is further atypical because the structures it combines with morphologically are found at a lower level of (p.351)

Nominal Structure

Figure 10.23

organization (cf. fig. 10.10). The thing which -i marks as landmark figures semantically not just as the profile of yot ’awaal but also as the trajector of yot and the profile of ’awaal. These lower-level component structures, identified by correspondences (ii) and (iii), are the ones to which -i attaches phonologically. Its parallel phonological integration with each of two components jointly and redundantly symbolizes its semantic integration with the higher-level structure yot ’awaal.

If the same semantic element sometimes has multiple phonological manifestations, the converse also happens: multiple semantic notions may have a single phonological realization. Rather than being marked by clearly distinct affixes, two or more categories (e.g. gender, number, case) might be realized together through a single, unanalyzable inflection. A simple example is the marking of gender and number in Italian, as compared with Spanish. The core system in Spanish is transparent: gender is marked by adding -o and -a to the noun stem, and plural by suffixing -s to the result. For Italian, on the other hand, there are four distinct endings, each marking a particular combination of gender and number: -o ‘masculine singular’, -a ‘feminine singular’, -i ‘masculine plural’, -e ‘feminine plural’. There is no evident way to decompose these into a part indicating gender and another indicating number.

(21) (a) Spanish: tío ‘uncle’, tía ‘aunt’, tíos ‘uncles’, tías ‘aunts’

(b) Italian: zio ‘uncle’, zia ‘aunt’, zii ‘uncles’, zie ‘aunts’

From a symbolic perspective, there is no inherent reason why a single marker should not make multiple semantic specifications. This is different only in degree from a lexical item invoking multiple cognitive domains as the basis for its meaning. To be sure, specifications like gender and number have systematic grammatical relevance (notably in “agreement” phenomena). They do not, however, have to be symbolized individually to be available for grammatical purposes.41 Ultimately, a (p.352)

Nominal Structure

Figure 10.24

composite expression provides the same information whether the specifications are made individually at successive levels of composition or simultaneously at a single level. This is shown for Spanish tíos and Italian zii in figure 10.24.

Decomposability into distinct, individually recognized morphemes is of course a matter of degree. Its graded nature is problematic for the building-block metaphor but easily handled with symbolic assemblies. Two features of such assemblies make its treatment quite straightforward. First, composite structures are entities in their own right, neither limited to what the component structures contribute nor constrained to mirror them faithfully. Second, composite expressions vary in their analyzability (§3.2.2)—that is, the likelihood or extent of components being activated by way of apprehending the composite whole.

Three degrees of decomposability and analyzability (which tend to correlate) are exemplified in figure 10.25. Presuming it to be novel, the possessive nominal Beverly's is fully analyzable into the component morphemes Beverly and 's, each of which is reflected without distortion in the composite expression. Each component structure categorizes the corresponding portion of the composite whole, and in this case both categorizations result in full recognition (solid arrows). By contrast, the possessor pronouns his and my are well-entrenched units, learned and used as wholes. Decomposability into recognizable morphological components is not required for this purpose, and to the extent that it occurs at all, the expressions are only partly analyzable. In the case of his, the components would be he and 's, the latter being mirrored faithfully, the former only with distortion (dashed arrow) in the quality of the vowel. Finally, although my invites categorization by me, only the

Nominal Structure

Figure 10.25

(p.353) consonants match. The construction is also defective, in that nothing at the phonological pole individually symbolizes the notion of possession.

In their specific detail, noun inflection and agreement are sometimes quite complex. That complexity arises, however, from the proliferation and interaction of factors that individually are fairly natural and easy to grasp. We have seen how a variety of basic phenomena can be dealt with in CG. There seems little doubt that actual inflectional systems are in principle fully describable by means of symbolic assemblies.


(1) This definition avoids arbitrary distinctions and bogus theoretical issues. Since the adjectival form of noun is nominal, a noun is also described as a nominal structure or nominal expression.

(2) Pardon the etymological pun. (If you missed it, look up chief in the dictionary.)

(3) By analogy, those would also be the head in those lazy cats. Recall, however, that in canonical expressions the grounding element has the same profile as the grounded noun (fig. 9.4). In such cases neither is considered a profile determinant (in accordance with another terminological decision).

(4) There are some exceptions, such as arms dealer and Clinton hater.

(5) For the case of nonplural masses, no single form is fully established in this construction. Sometimes zero suffices: She bought some wine, and I bought more. Sometimes stuff is pressed into service: She bought good wine, and I bought cheap stuff. But there are also cases where neither works: *They don't want just partial freedom, they're holding out for total {stuff/Ø}.

(6) As described in ch. 14, pronominal anaphora is a reference point phenomenon: the yacht functions as reference point for interpreting the target pronoun it (fig. 3.14). Being determined by a variety of interacting factors (van Hoek 1997), the choice of reference point is sometimes not unique. In the following, for instance, she might refer to either Jill or the sister: When I saw Jill talking to my sister, she was quite agitated.

(7) This group can be of any size: a family, the students in a class, an occupational group, the members of a culture, and so on. Its specific identification depends on the discourse context.

(8) More generally, an expression's phonological pole is viewed in CG as one aspect of its global meaning (FCG1: §2.2.1). This is by no means problematic, but a key to understanding various phenomena (e.g. onomatopoeia).

(9) This uniqueness can have other sources as well. Within the calendrical cycle, for example, the term for each month (January, February,…) labels just a single entity. Likewise, the terms for basic colors (yellow, red, blue, etc.) designate unique regions in color space (fig. 4.4(a)). While these are not proper names (in a narrow sense), they occur without separate grounding and can thus be considered proper nouns (cf. Coates 2006).

(10) An apparent exception are the initial nouns in expressions like stone wall, tile floor, and paper bag. Possibly they profile relationships in this construction. Suggesting a relational construal are phrases like completely tile floor, where tile is modified by an adverb, as well as their use as clausal predicates, e.g. The floor is tile.

(11) Apart from finite relative clauses (special due to their independent grounding), the one exception in English consists of infinitival modifiers marked by to. With these, the head noun corresponds to either the trajector (the first person to arrive), the landmark (a woman to admire), or even the landmark of a preposition (something to stir the soup with).

(12) As evidence for the e-site's trajector status, observe that the same participant is coded by the subject when the modifier functions as a clausal predicate: The table is {near the door / small / sitting by the door / broken / polished every morning}.

(13) Despite its schematized format—obviously more practical—fig. 10.6(a) is equivalent to the pictorial representation in fig. 7.13. T and D represent the basic types table and door, while G indicates that door is grounded. In diagram (b), s represents the property specified by small.

(14) Structures defined in unipolar terms (§6.3) do not per se participate in symbolic relationships. Also, the grouping of structures defined in bipolar terms sometimes yields a higher-order structure that does not itself achieve symbolic linkage. Recall a previous example: The package arrived that I was expecting (fig. 7.19(b)). Although the package and that I was expecting form a natural semantic grouping, the composite conception remains unsymbolized.

(15) Small caps indicate their unreduced stress. (Boldface, of course, is not being used for stress but for anaphoric relationships.)

(16) Temporal adjacency is criterial for the “classic” conception of constituency (§7.4). The characterization given above is more inclusive, as it does not specify any particular basis for grouping.

(17) This is written young, frisky horse. While the comma represents the first pause, standard orthographic practice neglects the comparable pause between frisky and horse.

(18) In previous diagrams, these have been conflated. This representation of the is intended as being equivalent to the one in fig. 9.11 but differs in its details because other factors are now more relevant. In addition to showing the type explicitly, the present diagrams more simply depict the ground and omit the previous discourse frame.

(19) Should correspondence (i) likewise equate the's profile with that of mouse? It really makes no difference, since the profiles of mouse and tiny mouse correspond in any case. The correspondence shown is the one expected on the basis of the general constructional schema for nominal grounding.

(20) Its degree of departure from the canon should not be exaggerated. Since mouse is schematic with respect to tiny mouse—hence immanent in its conception—we could also say that the's schematic type corresponds to this immanent substructure. The two characterizations are fully equivalent.

(21) Presumably, the conceptual constituent's internal structure conforms to the semantic pole of constructional schemas describing well-behaved (i.e. continuous) nominals like the package that I was expecting.

(22) It is odd at best to say ??The agreement was reluctant, ?*The consent was informed, or *The awareness is conscious.

(23) The following are thus quite natural: That car is fast; The parrot is very loud; His diet is unhealthy.

(24) Alternatively, we could take the active zone to be the type that these elements all instantiate. Since a type is immanent in the conception of its instances, the two alternatives are equivalent.

(25) Note that wives represents an entire symbolic assembly. The component structures are wife and a special variant of the plural morpheme: [[…f] —> […vz]] (cf. fig. 6.11).

(26) The correlation with biological gender is not exceptionless. Nor is the assignment of gender in other cases wholly arbitrary—indeed, it exhibits a great deal of systematicity (Zubin and Köpcke 1986).

(27) The masculine and feminine categories are strongly associated with the endings -o and -a (FCG2: §4.4), but to keep things simple we will basically ignore these for now (see §10.4). The points at issue can be made regardless of whether there is any such marking.

(28) The descriptions are comparable to that of a verb conjugation class (§8.4.2).

(29) Classifiers are often polysemous, reflecting the diversity of their associated categories. For example, the Thai classifier tua, roughly glossed as ‘body’, is used for animals, furniture, and clothing. (For a useful overview of classifiers, see Allan 1977.)

(30) The mass can be continuous, particulate, or replicate (i.e., plural). Unitization is basically the inverse of pluralization: instead of replicating a discrete entity to create a mass, it creates a discrete entity by bounding a mass.

(31) Naturally, CG rejects the prevalent view that of is meaningless (a purely “grammatical” element). Its meaning is simply abstract (GC: ch. 3).

(32) Compare this with the splinter in my finger (not *the splinter of my finger). Unlike its tip, a splinter is quite extrinsic to a finger. Of is also used to indicate substance (a ring of fire), instantiation (the month of August), and the relationship inherent in the meaning of certain nouns, such as kin terms (a descendant of Abraham Lincoln).

(33) These quantifiers can function as nominals independently of this construction: {All/Some/Many/Three} are obviously overfed.

(34) Predictably, all shares this behavior with both: both of the geese; both the geese. Like all, both has the semantic property that the mass it profiles coincides with the contextually relevant maximal extension, which is further specified as consisting of just two elements.

(35) Luiseño has a series of noun-marking endings of this sort (ch. 8: nn. 19 and 26).

(36) Grounding is obviously not external with respect to inherently grounded nouns (like pronouns and proper names). And for other nouns, manifestation in the outermost structural layer is typical but not invariant. Grounding is more internal, for example, in a classifier construction (fig. 10.17).

(37) Possessor pronouns (my, her, etc.) are of course different in this respect.

(38) There are intermediate cases, and whether a role marker profiles a relationship is sometimes hard to determine. The general correlation of periphrasis with separate words, and affixation or inflection with direct participants, is far from exceptionless.

(39) Recall the notational practice of indicating the identity of two relationships by showing their participants as corresponding.

(40) There is some possibility that an element like yot should actually be analyzed as a noun rather than an adjective in Luiseño (so that yot would mean something like ‘big one’ rather than just ‘big’). If so, this is just a matter of it profiling the thing instead of the defining relationship, which does not affect the basic analysis.

(41) For example, since number is inherent in each of the Italian forms, they can be assessed for conformity to a constructional schema requiring that a subject and verb “agree” in number.