LINGUIST List 13.1516

Fri May 24 2002

Review: Linguistic Theories: Croft (2001)

Editor for this issue: Naomi Ogasawara <>

What follows is another discussion note contributed to our Book Discussion Forum. We expect these discussions to be informal and interactive; and the author of the book discussed is cordially invited to join in.

If you are interested in leading a book discussion, look for books announced on LINGUIST as "available for discussion." (This means that the publisher has sent us a review copy.) Then contact Simin Karimi at or Terry Langendoen at


  • Cristiano Broccias, Croft (2001) Radical Construction Grammar

    Message 1: Croft (2001) Radical Construction Grammar

    Date: 24 May 2002 13:29:24 -0000
    From: Cristiano Broccias <>
    Subject: Croft (2001) Radical Construction Grammar

    Croft, William (2001) Radical Construction Grammar. Oxford University Press, xxvii+416pp, paperback ISBN 0-19-829954-0, $35.00.

    Book Announcement on Linguist:

    Cristiano Broccias, University of Genoa (Italy) and University of Pavia (Italy) CHAPTER-BY-CHAPTER SUMMARY

    Chapter 1. Syntactic Argumentation and Radical Construction Grammar

    This chapter begins with a brief outline of the main claims made by Radical Construction Grammar (which is defined as a theory of syntax, cf. p.3): -- grammatical categories are not primitives but are construction- specific. Much of modern syntactic analysis is based on a vicious circle: constructions are used to define categories (cf. the distributional method) and then categories are used to define constructions. Croft uses lower-case names to indicate categories/constructions that are non-language specific (e.g. relative clause); capitalised forms are reserved for categories/constructions which are language-specific (e.g. English Relative Clause). -- syntactic relations in constructions need not be posited. Only meronomic or part-whole structures exist; -- constructions are the basic units of syntactic representation; -- constructions are language-specific. The following sections introduce the notion of "construction" (following Langacker 1987) and build on Dryer's (1997) proposal that there is no universal inventory of grammatical categories and relations. The notion of "construction" is applied to any grammatical structure in keeping with the observation that there exists a continuum between (what are traditionally called) syntax and lexicon (cf. Table 1.3. on p.17). Constructions are (partially arbitrary) symbolic units, that is pairings of form (which includes syntactic, morphological, and phonological properties) and meaning (which includes semantic, pragmatic, and discourse-functional properties). Form is referred to as "syntactic structure", meaning as "semantic structure". Constructions form a structured inventory represented in terms of taxonomic hierarchies and may be compatible with a usage-based or redundant model of grammatical representation. Turning to grammatical categories such as Subject and Direct Object, Croft demonstrates that atomic primitives such as Subject and Direct Object cannot be assumed either cross- linguistically or within a specific language. Consequently, grammatical knowledge in Radical Construction Grammar is knowledge of constructions, words, and the many-to-many mapping between words and constructions. Syntactic categories are derivative of such grammatical knowledge. In more detail, they "can be defined in two different ways. They can be defined construction-specifically, as the class of fillers of a particular role in a single construction. They can be defined cross-constructionally, as the class of fillers that has an identical distribution across the relevant roles for all constructions of a language, or at least some specified set of constructions in the language." (p.46) The chapter concludes with a section devoted to Frequently Asked Questions. Croft convincingly defends his proposal that constructions are the primitive units of syntactic representation. Unfortunately, the FAQs do not include what I suspect practitioners of Cognitive Grammar would have very much appreciated, namely a subsection devoted to illustrating the differences between Langacker's Cognitive Grammar and Radical Construction Grammar (but see end of summary of Chapter 2 below). Chapter 2. Parts of Speech

    The chapter begins with a critique of contemporary analyses of parts of speech, which can be divided into two groups. On the one hand, "lumpers" argue that major parts of speech (noun, verb, adjective) can be conflated into one or two broader categories in some languages. On the other hand, "splitters" (on the basis of a detailed examination of distributional data) recognise "major parts of speech in every language, and many minor ones as well." (p.65). In particular, Croft assesses the work of "lumper" Hengeveld (1992). Croft stresses that Hengeveld ignores the importance of semantic shift for lexical items used in two different propositional act functions. For example, English often uses the same lexical item either as Noun or Verb (e.g. "study" in "The child studied at my house" and "I retired to my study"), but we do not usually conclude that Nouns and Verbs are conflated into a broader category in English. This is so because of (systematic) polysemy between, say, a referential meaning ("study" as a Noun) and an activity reading ("study" as a Verb). Further, lumpers often ignore distributional evidence that does show different behaviour for allegedly conflated parts of speech. Splitters, on the other hand, give us no method for "deciding between parts of speech and minor syntactic categories" (p.83); in other words, "[t]here is no way to stop splitting" (p.78). After his critique of lumpers and splitters, Croft moves on to presenting his universal-typological theory of parts of speech. His theory is based on the interaction of semantic classes and discourse functions. Among discourse functions (or pragmatic/communicative functions, see p.66) are the propositional acts of predication, reference, and modification (which are defined on p.66). Predicative, referring, and attributive constructions encode such functions (i.e. they are propositional act constructions or structural coding constructions, which can be divided into overt structural coding constructions and zero structural coding constructions). Croft hypothesizes that the typological prototypes of the referring, attributive, and predicating constructions are the semantic classes of objects, properties, and actions, respectively (as evidenced, for example, by zero coding in those languages which allow it). These semantic classes are only a small subset of the semantic classes of words found in human languages and are defined in terms of four semantic properties (relationality, stativity, transitoriness, gradability). The terms nouns, adjectives, and verbs describe the three discourse function/semantic class pairings alluded to above. Typological evidence for his analysis comes from the theory of typological markedness. Croft contends that his (universal) theory parts of speech can be represented as a (universal) conceptual space, that is "a structured representation of functional structures [i.e. the points in conceptual space] and their relationships to each other [i.e. the connections in conceptual space]" (p.93), onto which a particular language's categories can be mapped (as is illustrated for Japanese in Figure 2.2, page 95) thus obtaining a semantic map (see p.94). The representation of conceptual space is governed by the Semantic Map Connectivity Hypothesis (see p.96), which requires mapping onto a connected region in conceptual space. It is also worth pointing out that the conceptual space for parts of speech implies that properties are intermediate between objects and actions and that modification is intermediate between reference and predication. Croft proposes that the universals of language are found in conceptual structure and in general principles of form- function mapping such as typological markedness (cf. p.105 and p.108 in Chapter 3). Among them are the Structural Coding Map Hypothesis (which requires the marked member to have at least as many morphemes as the unmarked member), the Behavioural Potential Map Hypothesis (i.e. the equivalent of the previous hypothesis although stated for inflectional categories), and the Grammatical Category Structure Hypothesis (i.e. boundaries for parts of speech are language-particular). The last subsection of section 2.4 briefly comments on the relation between Croft's theory of parts of speech and Cognitive Grammar. Croft himself recognises that the two theories are compatible and concludes that "[t]he difference ... is chiefly a matter of emphasis". Chapter 3. Syntactic Categories and Semantic Relativity

    Croft dismisses radical semantic relativism (i.e. syntactic structure determines semantic structure and there is no universal conceptual structure), by exposing its fallacious hidden assumptions. "The first assumption is that syntactic differences across languages imply semantic differences" (p.129). The fact, for example, that English speakers say "I'm cold" and French speakers use the structure "J'ai froid" (i.e. the verb "have" is used) does not necessarily imply that a difference in conceptualisation (of the sensation of coldness) holds. This might be the case if both structures were available to a speaker in the same language. The second assumption is that there is a one-to- one relation between form and meaning (i.e. monosemy). For example, the fact that properties and bodily expressions can be coded by using the same construction in English ("I am American" and "I am cold") does not imply that they are conceptualised in the same way. First, their syntactic behaviour may be different (cf. "I am always cold" vs. "*I'm always America"), thus pointing at differences in semantics. Second, polysemy is to be preferred to monosemy (and homonymy), see also summary of Chapter 2 above. The third assumption is that "linguistic analysis should minimize syntagmatic redundancy" (p.120), whereas redundancy seems to be the normal case in grammar. The fourth hidden assumption is that the meaning of inflections, particles, etc. is universal, but "there is no a priori reason to assume that the grammatical elements always are universal" (p.125). Croft concludes the chapter by proposing a conventional universalist position, that is an interplay between universality (having to do with our being affected by common cognitive principles) and relativity (categorisation can be affected by conventionalised expressions of a language, as in the case of diverging spatial coding for English and Korean children): "speaker behavior is affected both by the universal conceptual space and the language-specific maps on conceptual space" (p.131). Chapter 4. Clausal Syntactic Roles ("Grammatical Relations")

    Chapter 4 discusses "subject" and "object" as clausal syntactic roles (i.e. fillers in a Clause construction). It offers a detailed critique of the two roles as uniform categories both cross-linguistically (by examining "accusative" and "ergative" languages) and within single languages; it proposes that syntactic roles are construction-specific and language-specific (as was already pointed out in Chapter 1). The lack of universal syntactic roles does not prevent Radical Construction Grammar from finding universals of syntactic roles (such as the participant role hierarchies, see below). In order to do so, Croft identifies the well-known labels A (the "subject" of transitive verbs), P (the "object" of transitive verbs), and S (the "subject" of intransitive verbs) with semantic roles (or, more accurately, participant role clusters). Croft also defines groupings (or sets) of participant roles: A+S as nominative, S+P as absolutive, P as accusative, and A as ergative. We can thus define a conceptual space having (in)transitive events and participant roles as dimensions. Case marking systems can be regarded as semantic maps for that conceptual space (in accordance with the Semantic Map Connectivity Hypothesis). Croft goes on to propose the following implicational universals: "If a language has overt coding of the nominative set of participant roles, then it has overt coding of the accusative set of participant roles." (p.139), i.e. nominative < accusative ... "If a language has overt coding of absolutive roles, then it has overt coding of ergative roles." (p.141), i.e. absolutive < ergative ... The two hierarchies are not only relevant for structural coding of participant roles (i.e. case marking), but also for behavioural potential (i.e. agreement). Croft also analyses ditransitive verbs. He recognises the following semantic roles (which are polysemous categories, that is participant role clusters): P (the "object" of monotransitive verbs, see above), T (the "theme" object of ditransitive verbs), and G (the "goal" object of ditransitive verb). Finally, he defines the following sets of participant roles: P+T as direct object, P+G as primary object, G as indirect object, and T as secondary object. He can then move on to propose the following implicational universals: direct object < indirect object (p.144) primary object < secondary object (p.145) As was the case above, such hierarchies are relevant for both structural coding and agreement. Syntactic roles are not a unitary category within a language either, as is demonstrated in section 4.3.1 through a critique of Anderson (1976). Again, this does not prevent Croft from finding universals of encoding of participant roles. In particular he proposes the Subject Construction Hierarchy (coordination < purposive < relativization < verb agreement < case marking), which can be represented in conceptual space (in keeping with the Semantic Map Connectivity Hypothesis) by using the two dimensions of foregrounding and topicality. Chapter 5. Dependency, Constituency, and Linear Order

    Chapter 5 begins the investigation of the internal structure of constructions. Radical Construction Grammar dispenses with syntactic relations (i.e. coded dependencies and collocational dependencies) by showing that (a) collocational dependencies (such as "burst into bloom" in "The trees burst into bloom") are the overt manifestation of semantic relations and (b) coded dependencies (i.e. syntactic relations evidenced, for example, by agreement) and constituency cannot be defined unambiguously. The notion of constituency is ultimately based on semantic relations and Croft, following Langacker (1997), prefers to use the term (formal) grouping. Formal grouping can be examined in terms of contiguity/separation of elements in an utterance (e.g. "A guy [who I hadn't seen since high school] came in" vs. "A guy came in [who I hadn't seen since high school]"), occurrence of a grammatical unit in a single prosodic unit, and unit of first-position self-repair (Fox and Jasperson 1995). The conclusion is that Radical Construction Grammar does away with a "unique abstract constituent structure" (p.195). Overtly coded dependencies are also problematic because, for example, they imply the unambiguous identification of heads and dependents, but the notion of syntactic "head" is also problematic, as is demonstrated in Chapter 7. Croft proposes to classify overtly coded dependencies semantically, dividing them into relational coded dependencies (i.e. "relational morphemes encode the semantic relation between the denotations of the elements they relate", cf. p.199) and indexical coded dependencies (i.e. "indexical morphemes index the referent of one of the members of a dependency", cf. p.199). Chapter 6. A Radical Approach to Syntactic Relations

    In this chapter, Croft shows that "the only syntactic structure in constructions is the part-whole relation between the constructions and its elements" (p. 203). He begins by arguing that a model without syntactic relations can account for effective communication. He then moves on to demonstrate that "symbolic relations in a construction grammar are not a notational variant of syntactic relations in a componential model" (p.206). Even if one posited linking rules between syntactic and semantic structures (i.e. no symbolic relations), one would have to posit different linking rules for each construction, thus ending up with a notational variant of construction grammar. Further, syntactic structure does not always reflect semantic structure iconically (as is shown, for example, for Possessor Ascension, Quantifier Float, and Adverb Agreement); as a matter of fact, syntactic structure can also reflect information structure. In the following part of the chapter, Croft argues that syntactic roles and syntactic relations are not notational variants. This is so because syntactic relations "impose more structure on constructions than syntactic roles. But there is strong evidence that the additional structure is problematic, or - fatally - is simply not there" (p.221). For example, if agreement markers indicate syntactic relations, how can there be a syntactic relation if one of the elements is not present (e.g. Spanish "esa moderna", "that modern one", which contains the feminine singular agreement marker "-a" and where the intended feminine singular controller "casa", "house", is not present)? Croft rejects all possible (syntactic) solutions and concludes that agreement can be treated as expressing a symbolic relation (that of indexing the referent) rather than a syntactic relation. In the last part of the chapter, Croft contends that "what appears to be the coding of syntactic relations is in fact scaffolding [as is defined by Langacker 1987: 452, 461] to help the hearer to identify which element of the construction symbolizes which component in the semantic structure of the construction." (p.238). Chapter 7. Heads, Arguments, and Adjuncts

    Chapter 7 discusses the notions of head and dependent (i.e. argument and adjunct). In the first part of the chapter (Sections 7.1 to 7.5), Croft proposes that "head" is not a syntactic notion, but must be interpreted as a symbolic relation between a syntactic role and a semantic component. In order to do so, he reviews criteria proposed for headhood (in particular Zwicky 1985, 1993) and show that they are problematic. Still, he argues that it is possible to define a "dominant" element in a construction provided that we adopt a semantic analysis. In particular, he builds on Zwicky's definition of semantic head and Langacker's definition of profile determinant and proposes the notion of profile equivalent: In a combination X + Y, X is the profile equivalent if X profiles/describes a kind of the thing profiled/described by X + Y. (p.257) >From his analysis follows, for example, that "phrases with articles have two profile equivalents, the article and the noun" (p.258) and that "clauses with auxiliaries have two profile equivalents, the auxiliary and the verb" (p.258). Croft also suggests that "[i]f inflectional categories relevant to a phrase or clause exist in a language, then they are expressed on at least one of the profile equivalents of the phrase/clause" (p.258). In order to account for the fact that one of the two profile equivalents may seem to be dominant, he proposes the following notion of semantic head: A (semantic) head is the profile equivalent that is the primary information-bearing unit [PIBU], that is, the most contentful item that most closely profiles the same kind of thing that the whole constituent profiles. (p.259) Croft argues that grammaticalisation is sensitive to PIBU status in that "[I]f there are two candidates for headhood [...], the one that is not the PIBU will undergo grammaticalisation" (p.259), as is illustrated in Section 7.5 by examining auxiliaries, articles, numerals, quantifiers, classifiers, adpositions, complementizers, and copula. In Section 7.6, Croft proposes that the morphological distinction between root and affix involves the notion of PIBU, but not the notion of profile equivalent. In particular, the root is the PIBU. For example, if we consider agent nominalizations, the agent-nominalizing affix is the profile equivalent; however, in order to claim that the root is the PIBU (which was defined above for those cases in which elements share profiles), Croft proposes that "[w]hichever form is in paradigmatic contrast with more elements is the primary information- bearing unit" (p.270). It seems that non- alignment of profile equivalence and PIBU status is the rule below the word level, contrary to what happens above the word level. The last section, Section 7.6, discusses the distinction between argument and adjunct. Following Langacker (1987), Croft argues that semantically such a distinction is a matter of degree. Further, he shows that the syntactic criteria of obligatoriness and latency (which are usually invoked to support the argument-adjunct distinction and which he describes by using the single dimension of "instantiation") are general properties of constructions, not of words that fill roles in constructions. Chapter 8. The Voice Continuum

    In Chapter 8 (and Chapter 9) Croft details the claim that constructions are language-specific and that constructions can be compared cross- linguistically in terms of their function. The latter observation is what allows typologists to formulate language universals. He illustrates this in Chapter 8 by analysing voice (Active vs. Passive, Direct vs. Inverse) in different languages. He considers the region of conceptual space (involving two-participant events) with a partial ordering of the speech act participant (SAP) status of A and P participants in the event (i.e. 1, 2 < 3; 1=speaker, 2=addressee, 3=other), along the horizontal and vertical dimension, respectively. First, he gives a structural definition of the passive type (cf. (3) on p.285) and the inverse type (cf. (6) on p.286). Then he discusses basic and non-basic voice constructions for various languages (Section 8.3) and draws semantic maps for them. He shows that such constructions can vary with respect to every structural property of the passive type and the inverse type; further, they also interact with the SAP hierarchy. In sum, voice constructions seem to vary along a continuum, from the active/direct type through the indirect type to the passive type (see Table 8.4). We can compare voice constructions cross-linguistically by considering a multidimensional space defined by the functionally grounded syntactic categories of agreement, case marking, coding of A(gent) and P(atient), and morphological marking of V(erb). We thus obtain a syntactic space for voice constructions, which Croft illustrates (see Figure 8.13) by using only the two dimensions of A coding and P coding. Croft observes that the syntactic space for voice constructions can be mapped onto the conceptual space for voice (see Figure 8.1). He proposes that the maps conform to the following language universal: "If there is a contrast between a basic and non-basic voice [...], then the semantic map of the basic voice will include the upper right corner of the conceptual space in Figure 8.1, and the semantic map of the nonbasic voice will include the lower left end corner of the conceptual space in Figure 8.1" (p.315). Croft points out that the SAP hierarchy stands for topicality (or salience) and extends the newly defined A and P dimensions so as to include intransitive situation types (Figure 8.16). Finally, he contends that the typologically least marked SAP alignment is 1/2 ? 3 and "hence is the typological prototype for the transitive (active/direct) construction." (p.318). The typologically most marked SAP alignment is 3 ? 1/2. Chapter 9. The Coordination-Subordination Continuum

    Chapter 9 analyses the form and function of complex sentences. They can contain verbs that are either identical to main clause verbs (the relevant clauses are called "balanced") or verbs that are different from main clause verbs (they are "deranked"). He argues that complex sentence types can be described by resorting to a conceptual space which is illustrated as a rectangular having as its points (starting from the upper left corner and moving clockwise) complex figure, figure-ground, figure-ground and e-site elaboration, and e-site elaboration. Croft argues, building on the work of Talmy (e.g. Talmy 1978) and Reinhart (1984), that the perceptual distinction between figure and ground characterises adverbial subordination (e.g. "After Tom resigned, all hell broke loose"). He points out (p.333) that figure does not correspond to "important information" or "focus of attention" and, similarly, ground does not stand for "given" or "presupposed". On the other hand, coordinate constructions (e.g. "The vase fell and broke", which is an example of a C(onsecutive)-chain, and "The sun was shining and the birds were singing", which is an example of a S(imultaneous)- chain), as is implicit in Wierzbicka's (1980) analysis, are based on the construal of two events as a single whole. Of course, this is possible if some link or common denominator (e.g. a causal or temporal one) exists between the two (in accordance with the principle of continuation in Gestalt psychology). Still, it is possible that either the figure-ground or complex construal becomes conventionalised in a language (as is cross- linguistically common for conditionals and comparatives). The important point here is that "the conventional expression may not share the construal of its source construction" (p.341). This is also the case when coordinate and adverbial subordinate clauses grammaticalize into deranked structures (see p.346). Langacker's notion of e-site elaboration - a dependent concept has a salient substructure that is elaborated by another concept; such a substructure is called e-site and the elaborating concept is the autonomous concept in the relation - is also important for complex sentences. "Complements are [...] examples of e-site elaborations by the "subordinate" clause" (p.346) and "a relative clause elaborates an e-site of the main clause event because it describes the referent more specifically than the head noun does" (p.347). This characterisation captures the similarity between complements and relative clauses. Further, Croft proposes that relative clauses are figure-ground constructions (see Section 9.3.2). On the other hand, he points out that complements are neither figure-ground configurations nor complex figure constructions (see Section 9.3.3). After having shown that complex sentences exhibit a continuum between coordination and subordination (because of the connections that can be established in the relevant conceptual space), Croft shows that "the domain of complex sentences also represents a continuum in syntactic space, in at least three dimensions: deranking, embedding, and clause fusion" (p.354). The deranking syntactic space can be reduced to a single dimension, the Deranking Hierarchy. Further, Cristofaro's (1998) Subordinate Clause Deranking Hierarchy (i.e. (97) on p.359) allows us to link syntactic space with conceptual space (in similar fashion to the syntactic-conceptual mapping for voice in the preceding chapter): deranking can be seen as radiating outwards from the typologically unmarked main clause type (see Figure 9.7). The Subordinate Clause Deranking Hierarchy has probably to do with semantic integration (i.e. the degree of interconnectedness of two situations), as is also the case with embedding and clause fusion. Chapter 10. Syntactic Theory and the Theory of Language

    The last chapter summarises the main points argued in the book. It also states that Radical Construction Grammar is a theory of syntax, not a theory of language. Croft devotes much of the chapter to summarising the theory of language he has put forward in his book "Explaining Language Change: An Evolutionary Approach". GENERAL DISCUSSION

    This is a very important book. It gathers intra- and inter-linguistic evidence that point at the centrality and relativity of the notion of construction; it presents an in-depth discussion of grammatical relations thus revealing problematic hidden assumptions; it argues for universality on the basis of our common cognitive processes but also recognises the importance of language as a system capable of influencing grammatical codification; it shifts the focus of attention to the notion of function. Moreover, Croft weaves his ideas and critical analyses into a coherent picture that exposes the contradictory foundations of much of modern linguistic theory (viz. the circularity of syntactic argumentation). For all these reasons, Croft's book is a must-read for all those interested in grammatical theory and typology. There are, however, some points I would like to comment on. The length of the following subsection, compared to the previous paragraph, does not imply by any means that I am retracting from my very positive evaluation of the book. Rather, I am going to dwell on some issues which I hope might be of some use in a new edition of this important book. 1) Compatibility and lexical semantics

    I think that an explicit comparison between Croft's Radical Construction Grammar and Langacker's Cognitive Grammar would be much needed. To be sure, Croft's syntactic theory is radical in the sense that it questions the basic assumptions of formal analyses and rejects its primitives by showing that syntactic categories are not unitary either cross-linguistically or inter-linguistically. Nevertheless, it seems that Croft's theory is compatible with Langacker's Cognitive Grammar (see end of p.6 and Section 2.4 for example). If the two theories are indeed compatible, could we not regard Radical Construction Grammar as an application of Cognitive Grammar to typological analysis? In other words, is Radical Construction Grammar a "new" theory with respect to Cognitive Grammar? Of course, differences between the two Grammars do exist, but it remains to be shown whether they are substantial. Consider the notion of "subject" for instance . In Cognitive Grammar, "subject" is regarded as a universal notion ultimately based on the (perceptual) distinction between trajector and landmark by being defined as the clause-level trajector (i.e. the schematic definition of subject). On the other hand, Croft denies the existence of the notion of "subject" as a universal concept. This however has to do with the fact that Croft approaches the notion "subject" by using a bottom-up approach (i.e. distributional facts) and does not rely on a (schematic) semantic characterisation. Hence, we can conclude that the same label "subject" is used to refer to different notions by the two authors. Similarly, it is not clear to me if Croft's definitions of profile equivalent and semantic head (see Chapter 7) imply a "radically" different view from Langacker's profile determinant other than a difference in emphasis. One may also add the observation that Croft's definitions do not rest on explicitly articulated notions (e.g. what is the difference between the notions "to profile" and "to describe"?, what does "a kind of thing" mean technically? what is a "constituent" in the definition of semantic head?). In some cases, genuine contrasts seem to exist. For instance, Langacker seems to rely much more on "lexical" polysemy than Croft. Just to make an example, Langacker (1999) analyses cases such as "I washed the car" vs. "I washed the mud off the car" as involving two lexical variants of "wash", thus proposing that "wash" is polysemous. However, other cases similar to the latter example seem to involve constructional rather than lexical semantics (thus possibly lending support to Croft's proposal concerning the primacy of constructions). Consider "She frightened an admission out of him" and "She kissed the anxiety away from him", where the "removal" interpretation probably does not depend on the verbs' being polysemous but on the type of construction into which such verbs are inserted (i.e. a "removal" construction). Hence, the interaction between lexical semantics and constructional semantics should also be worth addressing within Radical Construction Grammar. In sum, as I pointed out in my summary of Chapter 1 above, a FAQ along the lines of "What are the main differences between Radical Construction Grammar and Langacker's Cognitive Grammar?" would be a useful addition (especially for those familiar with the latter approach). 2) Meaning: Why not a radical semantics?

    The second point (which is related to the first) involves the role of meaning in Croft's theory. He explicitly states (see end of p.62) that "one of the reasons why this book is half the length of its nearest counterparts ... is that the latter volumes (especially Langacker's) discuss semantic issues at greater length than I do here. Nevertheless, it possible to discuss - and resolve- fundamental issues in syntactic theory without presenting a full-fledged semantic theory...". Although Croft's theory is a syntactic theory (cf. p.3), meaning is what it ultimately rests upon. Hence, one would expect a more detailed explanation of the underlying semantic assumptions despite his disclaimer. This is especially so because some assumptions may not be clear to the reader. A couple of examples will suffice. On p.87, Croft states that the semantic classes of objects, properties and actions "are defined in terms of four semantic properties", namely relationality, stativity, transitoriness, gradability. Does this imply that such semantic properties can be defined or assumed as unitary concepts (i.e. they don't exhibit a prototype structure for example)? Why do these properties suffice? In other words, why does a radical investigation have to involve syntax only and not semantics as well? Although the distinction between the act of predication and that of modification (see p.66) is intuitively clear, how can we be sure that it is always possible to neatly distinguish between the two (and the same holds good of the act of denotation vs. the act of modification)? Consider the sentence "John lives poorly". Which is the predicating element in this sentence? Does the verb "live" ascribes something to the referent? Intuitively, what is predicated of John here is that he seems to be in a state of poverty. Further, Croft states that in Hengeveld's analysis (see Chapter 2), "the semantic difference between "big" [property] and "big one" [object] is considered to be irrelevant by Hengeveld". Still, is such conceptual difference psychologically relevant or salient to the conceptualiser? What does such a difference consist of in more detail? Similarly, is it always possible to distinguish between an object and a property? For example, "happiness" is regarded as the property word "happy" being used as a referring expression. Would this line of reasoning mean that "beauty" (vs. "beautiful") is an object word? And if this is so, why is the notion of "beauty" primarily an object and "happiness" primarily a property? Developing his universal-typological theory of parts of speech, Croft states (p.88) that object and property words can be used as predicates as in "That is a cypress" and "That cypress is big". Does this mean that the words in question bear the whole burden of the semantics of the construction(s)? Is it not the case that such words are used not as predicates, but in a predicate-construction (i.e. their function is always the same, but such function does not correspond to that of the whole sentence)? In sum, since semantic analysis and conceptualisation are crucial in Radical Construction Grammar, perhaps some notions should be discussed in much more detail. This is also related to the need to show what cognitive processes are relevant to grammatical organization and whether such processes might also be relative (i.e. some languages use only a subset of possible processes). 3) Typological data

    Not being a typologist, I am not in a position to judge the extraordinary wealth of data Croft presents. Still, I have some doubts about the examples presented in Section 6.2.1, which aim to show that elements linked by a so-called syntactic relation - such a notion is rejected by Croft - are not necessarily linked semantically. Croft reproduces an example from Tzotzil whose English gloss reads "I'll turn its [the soup's] pot face down. [i.e. the pot that the soup was cooked in]" (see example (2) on p.209); the important point is that the original verb's object is "the soup" and not "the pot". Croft writes (p.209) that "Since the soup is already cooked and out of the pot, it is implausible to assume that the soup is affected by someone turning face down the pot it was cooked in." However, more examples like this would be needed to conclude that this is indeed the case. Couldn't this be a case of what Langacker calls profile/active zone asymmetry (i.e. the expression we use actually refers to something else, e.g. "I'm in the telephone book" talking, of course, of my telephone number)? Similarly, an example from Tsakhur poses a very interesting problem (see (7) on p.211). The relevant bit is glossed as "Ibrahim-Pasha quietly took the handbag and went away". The point is that the Adverb "quietly" agrees with the Absolutive NP "handbag", which does not make much sense semantically. Nevertheless, this seems to be the case if we rely on the English translation "quietly" since handbags cannot be "quiet". But couldn't one think of a scenario where the handbag (by being moved) produces some kind of noise (e.g. by coming into contact with the surrounding furniture) and hence describe the scenario in which no noise occurs as involving a "quiet" handbag? To put it differently, a more detailed semantic analysis of the language(s) in question is needed which does not simply rely on English translations. 4) Syntax and other definitions

    Croft describes his theory as a theory of syntax (see p.3). Later on, however, (see p.17) he writes "[C]onstruction grammar's great attraction as a theory of grammar-not just syntax..." and on p.26 he states, for example, that the Ditransitive construction [SBJ DITRVERB OBJ1 OBJ2] does not specify the order of elements. An explicit definition of what he regards as "syntax" might be needed. It seems to me that such a term is used in more than one sense, having to do both with word combination (cf. Table 1.3) and, more generally, with form (as opposed to meaning) (cf. p.3, where syntax is taken as a synonym for "grammatical structures that are assumed to be represented in the mind of a speaker"). In sum, a glossary of many of the concepts introduced throughout the book (especially because of its length and complexity) might be useful to the reader (cf. the very useful glossary in Langacker's "Foundations of Cognitive Grammar"). 5) Psychological evidence

    The final point I would like to raise has to do with psychological evidence. Croft explicitly claims that he is dealing with "grammatical structures that are assumed to be represented in the mind of a speaker". Hence, psychological evidence for this claim should be included in the book. In some cases, such evidence is strongly needed. For example, Croft argues that his grammatical model (which dispenses with syntactic relations in the sense of formal analyses) is compatible with successful communication (see Chapter 6). However, his arguments are not based on actual experiments carried out with speakers but rest on conceptual plausibility. TYPOS I conclude my review by listing some typos. p.13 description of example (7): "Null Noun Phrase in Infinitive Complement "controlled" by Main Clause Subject". The null NP is controlled by the Main Clause Object. p.27, tenth line from the bottom, 'gammar' should be 'grammar' p.49, first line, "a set of features each of which in essence indexed whether ..." should read "a set of features each of which is in essence indexed whether ..." p.74, tenth line from the bottom: (31), not (85) p.91 (definition of behavioural potential), no. (85), second line: "that it is found..." for "then it is found ..." p.129, penultimate line of section 3.3.: 'unversality' p.139 brackets missing around 11 and 12 in the line separating the actual examples p.153 comma not full stop after "Wardaman", second line above example (56) p.163, (88) to be replaced with (87) p.184 "English predicated adjective construction", capitals needed, second line above (16) p.191 (37a) not (36a) p.209 the abbreviation PF is not included in the abbreviations list p.214 "given", not "give", second line above (14) p.217 (22a-d), not (21a-d), last line of the page and third line from the bottom of the page p. 274 right bracket needed for (see Section 9.2.2 p.290 explication, not expli-cating, penultime line of Section 8.2 p.357 (87)-(89), not (86)-(89) p.368 (References) Aissen ... "... structure", not "structurei" REFERENCES

    Anderson, S. 1976. "On the Notion of Subject in Ergative Languages". In: Li, C. (ed.). Subject and Topic. New York: Academic Press. 1-24. Cristofaro, S. 1998. Subordination Strategies: A Typological Study. Revised version of Ph.D. Thesis, University of Pavia. Dryer, M. 1997. "Are Grammatical Relations Universal?" In: Bybee, J. et al. (eds.). Essays on Language Function and Language Type. Amsterdam: John Benjamins. 343-366. Fox, B. and Jasperson, R. 1995. "A Syntactic Exploration of Repair in English Conversation". In: Davis, P. (ed.). Alternative Linguistics: Descriptive and Theoretical Modes. Amsterdam: John Benjamins. 77-134. Hengeveld, K. 1992. Non-verbal Predication: Theory, Typology, Diachrony. Berlin: Mouton de Gryuter. Langacker, R. 1987. Foundations of Cognitive Grammar. Vol. 1: Theoretical Prerequisites. Stanford: Stanford University Press. Langacker, R. 1999. Grammar and Conceptualization. Berlin: Mouton de Gruyter. Reinhart, T. 1984. "Principles of Gestalt Perception in the Temporal Organization of Narrative Texts". Linguistics 22: 779-809. Talmy, L. 1978. "Figure and Ground in Complex Sentences". In: Greenberg, J. et al. (eds.). Universals of Human Language. Vol. 4. Syntax. Stanford: Stanford University Press. 625-52. Wierzbicka, A. 1980. Lingua mentalis: The Semantics of Natural Language. New York: Academic Press. Zwicky, A. 1985. "Heads". Journal of Linguistics 21: 1-29. Zwicky, A. 1993. "Heads, Bases and Functors". In: Corbett, G. et al. (eds.). Heads in Grammatical Theory. Cambridge: Cambridge University Press. 292-315. ABOUT THE REVIEWER

    Cristiano Broccias holds a Ph.D. in Linguistics from the University of Pavia (Italy). This academic year he has been teaching English Language at the University of Genoa (Italy) and Pavia. He is currently revising his Ph.D. thesis "The English Change Network" and looking for a post for next year. His research interests include Cognitive Linguistics and the grammar of contemporary English (especially "change constructions").