as a construction grammar is its emphasis on symbolic and semantic definitions of theoretical constructs traditionally analyzed as purely syntactic. As noted above, Langacker defines a grammar as a structured inventory of conventional linguistic units. The conventional linguistic units are symbolic units, and their two halves, form and meaning. Cognitive Grammar emphasizes the sym- bolic character of the linguistic sign (to use the Saussurean term). Langacker argues that the properties of constructions, as broadly defined, fall into two categories, which we describe here as form (the signifier) and meaning or function (the sig- nified): the formal properties are syntactic, morphological, and phonological, and the functional properties are semantic, pragmatic, and discourse-functional. A construction is thus a symbolic unit, linking form and function as a symbol or sign. To a large extent, the division between semantics, pragmatics, and discourse is arbitrary. The important distinction is between what is conventionally associated with a construction and what is not conventionally associated with it, but instead conveyed in particular contexts of use. Hence we may group together all functional properties as part of the conventional function of the construction. Langacker de- scribes this structure as the semantic pole of a symbolic unit. The formal properties of a construction also appear to be disparate. Langacker groups them together under the term phonological pole. The term ‘‘phonological pole’’ may sound odd: syntax at least is not ‘‘phonological,’’ particularly with re- spect to schematic constructions. However, Langacker argues that a schema such as Noun in the description of a construction should be thought of as phonolog- ically as well as lexically schematic: the schema ranges over possible nouns, and those nouns are all phonologically contentful, even if their exact phonological form cannot be specified schematically. (Cognitive Grammar and Construction Gram- mar, like Pollard and Sag’s (1993) Head-Driven Phrase Structure Grammar, eschews the use of phonologically ‘‘null’’ or ‘‘empty’’ elements.) In the Cognitive Grammar representation of a construction, the symbolic unit itself must link the two poles of the construction. Langacker describes the link as a symbolic correspondence. Symbolic correspondences are the Cognitive Grammar equivalents to the linking rules between syntactic structures and semantic struc- tures in the componential organization of a grammar described in section 2. There must be a symbolic correspondence that holds between the form (phonological pole) of the construction as a whole and the meaning or function (semantic pole) of the construction as a whole. Recall that a construction such as The X-er, the Y-er has some sort of idiosyncrasy such that its form and meaning are not predictable from more general rules (constructions). Hence, it must be an independent sym- bolic unit in its own right. Cognitive Grammar also has a uniform representation of all grammatical knowledge. Langacker argues that all semantic, pragmatic, and discourse-functional properties are ultimately conceptual, a part of what he calls semantic space, which he describes as ‘‘the multifaceted field of conceptual potential within which thought and conceptualization unfold’’ (Langacker 1987: 76). He argues that phonological space, the space in which linguistic form is defined, is also a subset of semantic 490 william croft space, since in terms of the structure of grammatical knowledge, the formal struc- tures of language are also concepts (76–81). (I) What is the status of the categories of the syntactic elements in construction grammar, given the existence of constructions? Cognitive Grammar argues that fundamental syntactic categories such as Noun, Verb, Subject, and Object are abstract (schematic) semantic construals of the con- ceptual content of their denotations. Thus, fundamental syntactic categories have an essentially semantic basis, but in terms of the construal of experience, not in terms of semantic classes. For example, the category ‘‘Noun’’ represents the construal of an entity as a ‘‘thing’’ (a technical term in Cognitive Grammar; Langacker 1987: 189). That is, the entity is construed as nonrelational and atemporal. This construal is the default one imagined with a prototypical noun in the traditional grammatical sense, such as cat. A cat is an individual that is (as a noun) conceptualized without presup- posing reference to another entity. In this respect, cat contrasts with feline (ad- jective), which construes the entity as a property of another entity (e.g., feline grace), or pounce (verb), which construes the entity as an event which presupposes the existence of participants of that event. A cat (as a noun) is also construed atemporally, that is, as a single Gestalt that does not unfold over mental time (i.e., it is summarily scanned). This construal contrasts with pounce (verb), in which the event is construed as unfolding over mental time (sequentially scanned). The role of conceptualization in the semantics of syntactic categories is dem- onstrated when applied to nonprototypical examples. The event of a pounce (noun) is construed nonrelationally and atemporally. The pounce (noun) profiles just the action; the participant of the action is deprofiled into the base or frame of the concept. Also, the pounce is construed atemporally, as an event that is conceived holistically in the mind, even though it takes place in an interval of external time. Langacker has developed semantic construal analyses of a wide range of syn- tactic categories, including parts of speech, grammatical roles (Subject and Object), the count/massdistinction, variousEnglish Tense-Aspectinflections andauxiliaries, the English possessives -‘s and of , ergativity, English complementizers and com- plement types, Cora locatives, and the Yuman auxiliary (see Langacker 1987, 1990, 1991, 1999). In the course of constructing these analyses, Langacker has developed a sophisticated analysis of conceptualization processes, including profiling, scope of predication, active zone, scanning, grounding, reference point, subjectification, the trajector-landmark opposition, and conceptual planes, as well as drawing on other cognitive linguistic conceptual constructs such as mental spaces (Fauconnier 1985), the Figure-Ground distinction, and viewing (Talmy 2000a, 2000b). One question that can be raised about the Cognitive Grammar analysis of grammatical categories is the relationship between the abstract semantic construal definitions and the variation in both formal distribution and semantic polysemy of such categories across languages. It has been suggested that cross-linguistic variation in the universal semantic categories can be accommodated in terms of construction grammar 491 conventionalized construal (e.g., Langacker 1990: 12): the same semantic category is found everywhere, but the construal of specific experiences as belonging to the semantic category is language-specific. But it is not clear whether one can distin- guish conventionalized construal from simple polysemy (that is, semantic varia- tion across languages). (II) What sorts of syntactic relations are posited? Cognitive Grammar takes a more radical departure from the more familiar ana- lyses of relations among parts of a construction (Langacker 1987: chapter 8). The Cognitive Grammar concept of valence, like that of Construction Grammar, is symbolic. Unlike Construction Grammar, however, valence in Cognitive Grammar is gradient. We will begin by looking at a straightforward predicate-argument re- lation, where the Cognitive Grammar and Construction Grammar notions of va- lence coincide, and then examine the extension of valence in Cognitive Grammar to other semantic relations. In the sentence Hannah sings, sings is a predicate because it is relational. The relationality of sings is due to the fact that singing requires a singer. Hence, the semantic structure for sings includes a schematic singer as a substructure. In Hannah sings, Hannah is an argument: it is nonrelational, and it fills the role of the singer for sings. Hannah is nonrelational because the concept of a person does not presuppose another concept. Langacker’s term for an argument filling the role of a predicate is that the argument ‘‘elaborates’’ the relevant substructure of the pred- icate. The substructure that can be elaborated by an argument is an elaboration site (or e-site; Langacker 1987: 304). These relations are illustrated in (31): (31) As we noted above, a unit in a construction may be simultaneously a predicate and an argument, as is read in You should read this. How is this possible? It is because the event of reading elaborates a substructure of the modality expressed by should, and the thing read, this, elaborates a substructure of the event of reading. Hence, predicate and argument status—valence—is relative: they depend on what two se- mantic structures are being compared. Not only is valence relative, it is gradient. In a sentence such as (32), I and what I am reading are traditionally analyzed as complements of read while on the train is an adjunct to read (we ignore the progressive be in this example): (32) I was reading this on the train. Complements are arguments of a predicate: reading inherently involves a reader and a thing read. Adjuncts are predicates and their head is the argument: on the train inherently involves a Figure whose location is described by the spatial rela- tion. Hence, read elaborates a substructure of on the train. But this description is 492 william croft an oversimplification. Reading is a localizable activity: reading takes place in a location, as well as involving a reader and a thing read. This is not true of all predicates; one cannot say, for instance, that *John was widowed on the train. Hence, the location of the reading event is a substructure of the semantic structure of read, and on the train also elaborates that substructure of read. The solution to this apparent paradox is that the substructure of read that is elaborated by on the train in (32) is much less salient in the characterization of the reading event than the substructures of read elaborated by I and this. Conversely, the substructure of on the train that is elaborated by read is highly salient in the characterization of the spatial relation. On the train is more of an adjunct of read than a complement because read elaborates a salient substructure of on the train, while on the train elaborates a not very salient substructure of read. The relative strength of the two relations is illustrated in (33): (33) Langacker adopts the terms ‘‘autonomous’’ and ‘‘dependent’’ to describe the gra- dient reinterpretation of the predicate-argument distinction. The definition of autonomy and dependence is: ‘‘One structure, D, is dependent on the other, A,to the extent that A constitutes an elaboration of a salient substructure within D’’ (Langacker 1987: 300); and conversely A is autonomous relative to D to the extent to which it does not elaborate a salient substructure of D.In(33), on the train is dependent on read because read elaborates the highly salient figure role of the locative relation on the train. Conversely, read is autonomous relative to on the train because on the train elaborates only the not very salient substructure of the location of the reading event. Autonomy and dependence are properties of any pair of conceptual structures. Thus, one has unipolar as well as bipolar autonomy/dependence. For example, at the phonological pole a vowel is autonomous—it can occur as the sole member of a syllable—while consonants are dependent—they must be supported by a vowel (Langacker 1987: 298–99). A bipolar autonomy/dependence relation would be the verb-prepositional phrase relation in (32): a circumstantial prepositional phrase like on the train is dependent on read. The Cognitive Grammar analysis of ‘‘head,’’ ‘‘modifier,’’ and so on is both sim- ilar to and different from the analysis in Construction Grammar. In Construction Grammar, the roles represent a relation between the parts of a construction and the whole and are defined syntactically. In Cognitive Grammar, the analogous concepts construction grammar 493 also represent a relation between the parts of a construction and the whole, but they are defined semantically and symbolically. Cognitive Grammar defines a semantic relation between part and whole as the profile determinant: the profile determinant is the part of the construction whose semantic profile the whole construction ‘‘inherits’’ (Langacker 1987: 289). The pro- file is the concept designated by the unit against the background knowledge pre- supposed by that concept (see Langacker 1987: chapter 2). Langacker (1987: 309) combines the concepts of profile determinacy and au- tonomy/dependence to define ‘‘head,’’ ‘‘complement,’’ and ‘‘modifier’’ in the intu- itively expected way: a head is a dependent predication that is a profile determinant; a complement is an autonomous predication that is not a profile determinant; and a modifier is a dependent predication that is not a profile determinant. (III) What sorts of relations are found between constructions? Langacker advocates what he calls a unified approach to categorization (1987:chap- ter 10). A category has a nonclassical structure, in that there is typically a proto- typical member or set of members, and nonprototypical members are categorized by extension from the prototypical members. However, it is also possible for there to exist a schema subsuming both prototype and extension, which has a classical category structure, with necessary and sufficient conditions specifying its instances: (34) Langacker’s model of categorization is of course applied also to constructions. Hence, for Langacker, as for Lakoff and Goldberg, one may have both construction schemas and also nonclassical relations between constructions, such as prototype- extension relations, including metaphorical extensions. (IV) How is information stored in the construction taxonomy? Cognitive Grammar is neither a complete inheritance model nor a full-entry model, in the extreme sense of the latter term as storing information at all levels in the hierarchy. Cognitive Grammar is a usage-based model, in which the establishment of schematic constructions is the result of language use. In particular, one cannot automatically assume that speakers of a language have induced a highly schematic construction, even if linguists can come up with an analysis with such a schema. Nor can one assume that speakers store information only at the most schematic level in the hierarchy. In the usage-based model, the existence of a highly schematic con- struction is ultimately a psychological question. In this respect, Cognitive Grammar differs significantly from Construction Grammar, which does not make any claims for the psychological reality of its complete inheritance model. 494 william croft The principles of the usage-based model governing the storage of grammatical information are based on research on language use, language acquisition, and language change. The usage-based model and evidence supporting it is described in section 6. 5.4. Radical Construction Grammar Radical Construction Grammar (Croft 2001, 2005) was developed to account for typological variation in a construction grammar framework and to address certain issues of syntactic argumentation. Radical Construction Grammar adopts the nonclassical category structure of the Lakoff-Goldberg theory and of Cognitive Grammar and the usage-based model of Cognitive Grammar. The chief innova- tions of Radical Construction Grammar in comparison to the theories of con- struction grammar described above is in the analysis of syntactic categories and syntactic relations. Radical Construction Grammar differs from the preceding the- ories (except, possibly, Cognitive Grammar) in a thoroughly nonreductionist ap- proach to constructions and in rejecting syntactic relations between elements in a construction. (I) What is the status of the categories of the syntactic elements, given the existence of constructions? The standard analysis of meronomic relations between syntactic structures has been adopted by Construction Grammar. In this analysis, a construction such as the intransitive or transitive construction is made up of parts, and those parts are themselves independent constructions. For example, various clausal constructions have verbs, which are analyzed as belonging to the same part of speech, in part because they have the same inflections (present in 3rd-person singular -s and non- 3rd-person singular zero, past in -ed or other allomorphs): (35) Present 3rd-person singular: a. Intransitive: Toni dances. b. Transitive: Toni plays badminton. (36) Present non-3rd-person singular: a. Intransitive: We dance-Ø. b. Transitive: We play-Ø badminton. (37) Past: a. Intransitive: We danced. b. Transitive: We played badminton. In other words, the same units occur as the parts of many different constructions. Ultimately, the decomposition of a construction will lead to a set of basic or primitive elements which cannot be analyzed further and out of which construc- tions are built. These atomic elements include syntactic categories such as Verb or Noun and relations such as Subject or Object, and so on. A model of grammatical construction grammar 495 structure of this type is a reductionist model: more complex structures are treated as built up out of primitive and ultimately atomic units. In the example given here, the atomic units are the basic categories and relations. 4 The reductionist model does not capture certain empirical facts about the distribution of words. For example, while many English verbs occur in either the Transitive or Intransitive constructions, many others do not: (38) a. Judith danced. b. Judith danced a kopanica. (39) a. Judith slept. b. *Judith slept bed. (40) a. *Judith found. b. Judith found a 20-dollar bill. One solution is to divide Verbs into Transitive Verbs and Intransitive Verbs. If so, then a decision must be made about verbs such as dance which occur in both constructions: Do they simultaneously belong to both subclasses? Or do they form a third distinct class? One effect of dividing verbs into transitive verbs and intran- sitive verbs is that one essentially defines the categories in terms of the construc- tion(s) they occur in, Transitive or Intransitive. One can deal with such problems in the reductionist model by adding exception features that prevent certain category members from occurring in the unacceptable constructions, as in (39b) and (40a). Again, the effect is that one is introducing a feature that specifies the category in terms of the construction it occurs in/does not occur in. Radical Construction Grammar takes a different approach to the relations of constructions to their parts. It proposes that constructions are the basic or primitive elements of syntactic representation and defines categories in terms of the con- structions they occur in. For example, the elements of the Intransitive construction are defined as Intransitive Subject and Intransitive Verb, and the categories are defined as those words or phrases that occur in the relevant role in the Intransitive construction. Radical Construction Grammar is a nonreductionist model because it takes the whole complex structure as basic and defines the parts in terms of their oc- currence in a role in the complex structure. In effect, Radical Construction Grammar takes to its logical conclusion one of the strategies for handling these problems in reductionist theories, namely the subdividing of classes and the em- ployment of exception features which essentially specify which constructions a par- ticular word or phrase occurs in. Constructions are individuated like any other conceptual object, by catego- rization. Constructions possess formal features, including word order, patterns of contiguity, and specific morphemes (or very small classes of morphemes) in particular roles. Constructions are also symbolic units and possess often discrete meanings. Radical Construction Grammar assumes a nonclassical category model and allows for prototypes and extensions of constructions, as well as the possibility of gradience between construction types. 496 william croft (II) What sorts of syntactic relations are posited? Radical Construction Grammar, like Construction Grammar and Cognitive Grammar, represents the role of a part of a construction in the whole construction. Radical Construction Grammar differs from Construction Grammar in that it defines relations between parts of a construction in purely semantic terms, that is, there are no syntactic relations in Radical Construction Grammar. One motivation for the Radical Construction Grammar analysis is that rela- tions between syntactic elements are not strictly necessary in a construction gram- mar framework from the point of view of language comprehension. Consider the phrase the song, illustrated in (41) with the semantic relation between [def] and [song] indicated by a link (labeled r): (41) If a hearer recognizes the phrase the song as an instance of the construction [[def/ the][thing/Noun]]—that is, can retrieve the semantic structure of the whole construction, can identify the elements of the construction (i.e., the words the and song), and can identify the corresponding components of the semantic pole (i.e., [def] and [thing])—then the hearer can identify the semantic relation r by virtue of the semantic relation between [def] and [thing] in the semantic pole of the construction. Hence, the hearer need not rely on any syntactic relation between the and song. In Radical Construction Grammar, the various morphosyntactic proper- ties that are taken to express syntactic relations in other theories—case marking, agreement, adpositions, word order, contiguity, and so on—are interpreted as expressing the symbolic links from the elements in the phonological pole of the construction to their corresponding components in the semantic pole of the con- struction. The combination of morphosyntactic properties in an utterance taken as a whole aid the hearer in identifying a construction. For example, the combina- tion of auxiliary be, the past participle form of the verb, and the preposition by,in the proper syntactic combination with the subject phrase, the verb, and the oblique phrase, uniquely identify the passive construction, while the individual elements identify the action (verb inflection and position after auxiliary), the agent (by plus oblique phrase), and patient (subject position). (III) What sorts of relations are found between constructions? In Radical Construction Grammar, each part (unit) of a construction constitutes a category whose members are defined solely by their occurrence in that role in the construction. In order to differentiate categories, we append the name of construction grammar 497 the construction to the labels for each unit in the construction. A representation of the intransitive and transitive constructions is given in (42): (42) The establishment of a category Verb is a linguistic generalization over the cate- gories Intransitive Verb and Transitive Verb. This generalization is thus a taxo- nomic relationship, with Verb superordinate to Intransitive Verb and Transitive Verb. However, one cannot posit a superordinate category such as Verb, or any linguistic category, without linguistic motivation. The basis of the linguistic gen- eralization of a superordinate category such as Verb must be its occurrence as the category in some other construction. The standard basis for positing a category Verb is the ability of its members to be inflected with the tense/agreement suffixes. In a construction grammar, this linguistic fact is essentially another construction, the morphological construction [MVerb-TnsAgr]. We use the label MVerb (Mor- phological Verb) to emphasize that this category is defined by a morphological construction, namely its occurrence with the Tense/Agreement suffixes (abbrevi- ated TnsAgr). This additional fact is represented, as in (43), with the irrelevant argument elements suppressed for clarity: (43) Radical Construction Grammar essentially hypothesizes that meronomic relations between a constructional whole and its parts are solely internal to the construction. In other words, Radical Construction Grammar rejects meronomic links between constructions and replaces them with taxonomic links between parts of different constructions. By using taxonomic links for parts of a construction, Radical Con- struction Grammar explicitly represents the process of analyzing parts of a con- struction as an abstraction or schematization process. (IV) How is information stored in the construction taxonomy? In Radical Construction Grammar, as in Cognitive Grammar, it is assumed that information may be stored redundantly in the construction taxonomy and that the principles governing the level at which information is stored redundantly are determined by the usage-based model. The next section describes that model and its relationship to the dynamic aspects of language: use, acquisition, and change. 498 william croft 6. Construction Grammar and the Usage-Based Model The usage-based model is a model of grammatical representation in which lan- guage use determines grammatical representation. Specifically, frequency of use and similarity of form and meaning are the determining factors for the structure of grammatical knowledge in the mind. The basic principles of the dynamic usage- based model have been developed largely in the area of morphology (see Bybee 1985, 1995, 2001, and references cited therein). The basic principles may be for- mulated in the following four hypotheses. Hypothesis 1: The storage of a word form, regular or irregular, is a function of its token frequency. Token frequency is the frequency of occurrence in language use of individual tokens of a grammatical type, such as the English regular past-tense forms. The usage-based model predicts that the degree of entrenchment of a form in a speaker’s mind (Bybee’s 1985 notion of autonomy) is a function of its token fre- quency, hence the concentration of irregular word forms in high-frequency items. There is also some evidence for the independent storage of high-frequency indi- vidual word forms even when those word forms are fully regular. Hypothesis 2: The productivity of a schema is a function of the type frequency of the instances of the schema. Type frequency is the frequency of word types that conform to a schema. For example, the type frequency of the English regular past-tense inflection is the frequency of all the different verbs that use the regular past-tense inflection. Bybee (1985) argues that type frequency determines productivity. One consequence of this hypothesis is that productivity is predicted to come in degrees: schemas with a low type frequency will have a limited degree of productivity. This appears to be the case: for example, the English irregular past with [¼()(g/k)] is slightly pro- ductive (compare colloquial or dialectal sneak/snuck, bring/brung). Hypothesis 3: In addition to source-oriented morphological rules/schemas, there also exist product-oriented schemas, which cannot be easily represented by rules. Many traditional, structuralist and generative theories of morphology assume the existence of rules that derive one word from another, such as the past verb form from the present verb form. In those cases where such a rule is possible, Bybee (1985) speaks of a source-oriented schema, that is, a schema for a word form that can be formulated in terms of a single simple morphological operation on the alleged source form. However, there is a class of schemas which Bybee calls product-oriented schemas, in which no simple process derives the alleged product form from the alleged source form. For example, the English past schema [¼()(g/k)] is a phonologically coherent construction grammar 499 . examples. The event of a pounce (noun) is construed nonrelationally and atemporally. The pounce (noun) profiles just the action; the participant of the action is deprofiled into the base or frame of the concept certain issues of syntactic argumentation. Radical Construction Grammar adopts the nonclassical category structure of the Lakoff-Goldberg theory and of Cognitive Grammar and the usage-based model of Cognitive. recognizes the phrase the song as an instance of the construction [[def/ the] [thing/Noun]]—that is, can retrieve the semantic structure of the whole construction, can identify the elements of the construction