Adele Goldberg's 2006 book continues some of the topics of her 1995 book on constructions and argument structure. The most important extension of her work since 1995 is the development of a child language and experimental component which explores the way input influences the acquisition of argument structure by the child. While she draws on work in construction theory generally, her focus is primarily on the treatment of argument structure in a constructionist approach.

The subtitle reflects the stated goal of the present book: ‘to investigate the nature of generalization in language: both in adults' knowledge of language and in the child's learning of language’ (p. 3). The importance of this goal cannot be overestimated. While the book itself only opens and sets the stage for such an investigation, it is my view that this is exactly what linguists should have been doing all along – studying the nature of generalizations that real speakers make and trying to establish how children come to these generalizations from the specific material in the input. In this way, then, Constructions at Work sets an important agenda for linguistic and psycholinguistic study. At the same time, some caution must be exercised: the validity of the methods used and the strength of the argumentation offered is somewhat uneven, as we shall see below.

As a general reference work on constructionist approaches to language, the book is valuable in offering a very readable introduction to constructions, the arguments in favor of describing grammar directly in terms of form-function units, the arguments for constructions over derivations from an underlying structure (Chapter 2) and the nature of a usage-based approach to constructions (Chapter 3). In addition, a later chapter addresses the differences among the various constructionists approaches to grammar (Chapter 10). For the child language researcher, these chapters provide valuable orientation and are highly recommended. Even more important for child language researchers, however, are the chapters that directly address how children learn generalizations (Chapter 4), how and why generalizations are constrained (Chapter 5) and how we know that constructions are learned rather than just verb meaning (Chapter 6). In addition in Chapter 9, Goldberg proposes a way of accounting for cross-linguistic patterns of argument structure assignment without resorting to innate universals.

In Chapters 1 to 3, Goldberg lays out the framework for the book, which is a general set of constructionist approaches which have two important properties: they treat grammar in terms of constructions which are taken to be ‘conventionalized pairings of form and function’ and they emphasize that languages are learned or ‘constructed on the basis of input and general cognitive, pragmatic and processing constraints (emphasis in original, p. 3)’. The definition of constructions used encompasses all levels of linguistic analysis from morpheme to clause-level and it is a usage-based one: ‘patterns are stored as constructions even if they are fully predictable as long as they occur with sufficient frequency’ (p. 5). Her approach is usage-based in the sense that it recognizes that specific instances of constructions may be stored in memory along with the generalized or more schematic construction and that cognitive representations are sensitive to frequency in experience.

As noted above, for child language researchers the meat of the book lies in the three chapters of Part 2 which discuss the way that generalizations are learned and constrained in child language. Chapter 4 reviews and discusses the research reported in Goldberg, Casenhiser & Sethuraman (Reference Goldberg, Casenhiser and Sethuraman2004, Reference Goldberg, Casenhiser and Sethuraman2005) and Casenhiser & Goldberg (Reference Goldberg, Casenhiser and Sethuraman2005) in which it is demonstrated that both type and token frequency are instrumental in the acquisition of constructions. Corpus-based research shows that in many cases constructions occur frequently with specific lexical items in addition to occurring with a range of lexical items. Goldberg and colleagues designed an experiment to investigate how constructions are learned. Both children and adults were taught a nonce argument structure construction in English. The meaning of the construction was not one already present in English, and the form was also novel. The construction had a nonce verb (with a suffix in some of the conditions) and the verb appeared finally in the clause. The meaning of the construction was taught through a video presentation that accompanied the linguistic stimuli. In one condition nonce verbs appeared in the stimuli with the same token frequency, while in the other condition the same number of verbs were presented, but one had a higher token frequency than all the others. In the latter condition, learning was more successful. The hypothesis about the facilitation of learning is that the repetition of a particular verb in a particular construction helps to establish the correlation between the meaning of the construction and its formal expression. Goldberg goes on to demonstrate that in category learning in general a centered, or low variance, category is easier to learn. In this case, it was demonstrated that the same generalized category was learned more easily if the input is skewed so that it has a stronger center. Though ancillary to the points just mentioned, it is striking that English-speaking subjects were able to learn and generalize a verb-final construction.

Chapter 5 discusses the factors that determine the range of application or degree of productivity of a construction or generalization under the title ‘How generalizations are constrained’. This addresses the question of why there isn't more overgeneralization in child language. The main factor proposed for constraining generalization is statistical pre-emption, while the factors considered to determine productivity are type frequency and the openness of the pattern.

Chapter 6 discusses the role of constructions in languages such as English in providing the meaning of clauses under the somewhat puzzling title ‘Why generalizations are learned’. The question of the title seems to actually be ‘why are constructions learned rather than just verb meaning’. The answer provided is that in languages such as English, where many verbs go into different constructions, the construction is an equally good predictor of clause meaning as the verb is (indeed a better predictor in some cases). Both corpus evidence and experimental evidence (from Bencini & Goldberg, Reference Bencini and Goldberg2000) are adduced to support this point. Additional motivation for acquiring constructions is said to come from structural priming. The fact that constructions can be primed means ‘that the level of generalization involved in argument structure constructions is a useful one to acquire’ (p. 125), a statement that appears too weak to be convincing.

Part 3 is called ‘Explaining Generalizations’ and takes the form of a more traditional linguistic analysis of so-called ‘island constraints’ in Chapter 7 and Subject–Auxiliary Inversion in Chapter 8. These two chapters are presumably intended to show that constructionist approaches can handle traditional syntactic problems. Chapter 9 argues that argument structure mappings of meaning onto syntax are not universal and not innate. It complements the earlier chapters that have made the point that argument structure relations and constructions are learnable from input.

Chapters 7 and 8 fall back on generativist means of argumentation and the structuralist assumption that the more general a rule, constraint or principle is, the more valuable it is. It is here that the text seems to drift away from its focus on generalizations that speakers make into the dubious territory of generalizations that linguists come up with.

Chapter 7 attempts a pragmatic account to answer the question of why odd sentences result when one tries to move an NP out of certain structures, as in ‘Who did she see the report that was about?’ (cf. ‘She saw the report that was about x’). Goldberg's proposal is that backgrounded constructions are ‘islands’, i.e. do not allow extraction of an NP since extracted NPs are focused. In other words, there is a pragmatic clash when an extracted NP, which is brought into focus, comes from a backgrounded part of the clause. While the intuition behind this proposal may apply in some cases, there are weaknesses in the argument and a disharmony between the theoretical and methodological approach taken here and that followed in the rest of the book. Of course, the first problem is that no successful characterization of a discourse notion ‘backgrounding’ has been found. Goldberg therefore equates backgrounding with presupposition and uses the test of constancy under negation to identify presupposed clauses. The problem here is that presupposition is a logico-semantic aspect of clauses which is viable only at the sentence level and therefore does not necessarily relate to the discourse level concepts such as topic and focus to which Goldberg also appeals.

More serious, however, is the fact that presupposition does not in fact account for all the data: as Goldberg herself notes, lexical factors come into play, as ‘who did he say she was dating’ seems natural, but ‘who did he whisper she was dating’ is decidedly odd. Verhagen (Reference Verhagen, Angeliki, Costas and Bert2006) studies such examples in a large corpus and finds that certain specific, formulaic sequences allow long-distance movement. In his approach, all subordinate clauses are not equally subordinate (see also Thompson, Reference Thompson2002, and for child language Dabrowska, 2004, Chapter 9). In addition, the proposal that ‘islands’ are all backgrounded runs into trouble with non-restrictive relative clauses, which behave like assertions (new information) in discourse. Since non-restrictive relatives are intonationally separate and present new information, it seems that a more appropriate explanation for the difficulty of extracting NPs from them to place at the front of the main clause is that they are separate assertions and more like conjoined structures, i.e. not truly subordinate to the main clause. Given the specificity inherent to construction grammar it does not seem imperative to come up with a single generalization that predicts all constraints on extraction.

In terms of method, it is surprising to find that Goldberg uses made-up sentences out of context to argue her point. This is especially problematic in the absence of a characterization of what grammaticality judgments are based on. Under the usage-based notion that lack of grammaticality is lack of familiarity, the oddness of these sentences can be said to be in part due to the fact that one rarely hears such combinations of structures. This could be for different reasons in different cases, and indeed, Goldberg's proposal of a pragmatic clash may come into play in some cases.

In Chapter 8, Goldberg tries to find a single function for English Subject–Auxiliary Inversion (SAI) under what I take to be the mistaken impression that a cognitive/functionalist approach means that there are no purely arbitrary aspects of grammar and that every aspect of form must have a one-to-one mapping to meaning. Goldberg proposes that the various constructions in which SAI occurs, such as questions, exclamations, clauses with fronted negative adverbs, counter-factual conditionals, etc. constitute a radial category whose dominant attribute is the feature ‘non-positive’; that is, each construction has ‘non-positive’ as part of its meaning or is related to a construction that has this feature. I find this proposal less than convincing for the following reasons. A pattern such as SAI is such an old, entrenched pattern in the English language that one could hardly expect it to have a single function. Thus it is not surprising that assigning it a function, or even a cluster of related functions, as Goldberg does, requires the postulation of features so abstract as to be virtually meaningless. One wishes here for some experimental evidence such as that provided in other chapters to show that speakers can come up with such abstract categorization.

In Chapter 9, Goldberg proposes cognitive and discourse principles to account for the cross-linguistic tendencies that others have ascribed to innate universal linking rules (e.g. Pinker, Reference Pinker1989). She argues from general cognitive evidence that actors and undergoers are the most salient participants in events and proposes a generalization by which salient participants are expressed in prominent slots in the clause. She notes that any such generalization must be expressed as a tendency since within every language there will be particular constructions that violate it. Similarly, the proposal by Lidz, Gleitman & Gleitman (Reference Lidz, Gleitman and Gleitman2003) that the number of overtly expressed complements equals the number of semantic participants has many exception across and within languages since it is quite common to omit mention of some participants (e.g. ‘The package was delivered yesterday’). Rather than an innate universal, Goldberg argues that the determinants of the arguments that are expressed in any given clause are a function of cognitive, semantic and pragmatic factors, such as recoverability and relevance.

As Tomasello, Lieven and colleagues have shown (e.g. Lieven, Pine & Baldwin, Reference Lieven, Pine and Baldwin1997; Tomasello, Reference Tomasello2003), usage-based construction grammar provides an excellent framework for tracking the development from specific to general patterns in child language. Goldberg's book is an accessible demonstration of how constructionist approaches, combined with certain cognitive, semantic and pragmatic factors, provide a viable means of describing both adult and child language without innate universals and abstract underlying structures. The argumentation is strongest and most interesting where a range of evidence from cognitive, psycholinguistic and corpus studies is cited; it is less successful where it is based on judgments of grammaticality.