One hallmark of collected volumes seems to be that they are often heterogeneous regarding the objects of study covered as well as the approaches to and theoretical outlooks on them. The present book, originally published as a special issue of Studies in Language 36(3), is no exception, which is certainly in part because cognitive linguistics, though in existence for more than a third of a century, still is rather a bundle of related theories and approaches which ‘are not (yet) brought together under the common rule of a well-defined theory’ (Geeraerts 2006: 2). The present volume is not only a testament to this theoretical heterogeneity still being a conspicuous characteristic of cognitive linguistics in general. It furthermore demonstrates that the most important aspect in which various flavours of cognitive linguistics differ is the role ascribed to data, or, more precisely, empiricism, and how this evidence from natural(istic) language is weighed against purely theory-internal considerations.

In their introduction to the volume, Nikolas Gisborne & Willem Hollmann note that ‘the divide between generative and cognitive approaches to language is intimately connected to the kinds of data drawn upon . . . and how this data is gathered’ (p. 3). That this divide is as real as it is wide is probably one of the very few notions linguists of all persuasions can readily agree upon. In light of this, the question arises whether bridging this gap is at all possible, and, if so, whether attempting to do so is in any way useful or desirable.

Much of Gisborne & Hollmann's introduction revolves around the question in what way cognitive linguistics ‘could provide coherent and empirically sound accounts of . . . theoretically uncovered data’ (p. 3), which is their term for the outcomes of chiefly or exclusively theory-driven introspection. This stance could not contrast more sharply with one found in the volume's first chapter, by Stefan Gries, who describes the generative enterprise as reliant on a ‘methodology largely based on made-up judgements of made-up (often context-free) sentences’ while characterising cognitive/functional linguistics as displaying ‘an (increasing) reliance on various sorts of both experimental and observational data’ (p. 16). On this basis Gries assesses the latter framework to have cast serious doubt on generativism's two most central assumptions, i.e. a highly modular linguistic system and largely innate language structure, as well as on its predominant method of introspection-driven acceptability judgements. Thus, while Gisborne & Hollmann see both the possibility and desirability of bridging the rift between cognitive (though not necessarily usage-based) linguistics and formal-generativist frameworks, and thus ask to what extent ‘cognitive linguistics offered an alternative paradigm to the formalist traditions’ and in how far ‘cognitive linguists spoke to linguists from formalist traditions’ (p. 3), Gries (p. 16) notes in no uncertain terms that the evidence amassed over the last thirty years or more renders the most central presumptions of generativism, viz. the poverty of the stimulus and a modular grammar, by and large falsified, and that the reliability of introspective acceptability judgements has been found wanting ever since the early 1970s. The outlooks and objectives hence are very dissimilar: Gisborne & Hollmann's apparent attempt to mediate between and reconcile the respective proponents of two vastly different approaches to and conceptualisations of language, in the hope of bridging the chasm between them, contrasts strongly with Gries’ straightforward question as to which theoretical paradigm and which methodology produces more valid, reliable, verifiable and representative results in describing and/or predicting variation and change in real language. The eight chapters in this volume reflect the breadth of outlooks on and approaches to language within cognitive linguistics, ranging, for example, from inductive empirical analyses (e.g. Jóhanna Barðdal et al.) via studies equally devoted to empirical data and theoretical issues (e.g. Graeme Trousdale) to ones in which real-language data has no more than a walk-on part (e.g. Gisborne).

Turning now to the individual contributions, the volume's first chapter focuses on methodology. It provides an overview of different methods for assessing frequencies and co-occurrence probabilities and their respective differences, and refutes criticism issued against one particularly well-tried-and-tested method. Gries first summarises collostructional analysis and its underlying logic, then reiterates criticism of perceived shortcomings of this method, and afterwards clarifies misapprehensions and refutes criticisms based thereon. Finally, he sketches different approaches to measuring co-occurrence and broadly outlines how these work, concluding that collostructional analysis is a method well suited for ascertaining co-occurrence frequencies and thus constructions, even though it is not the most elaborate one, and that raw frequencies are ‘too crude a tool to . . . [understand] the cognitive and statistical properties of language acquisition, processing, usage, and change’ (p. 45). The chapter is a useful summary of collostructional analysis and other co-occurrence measures as well as criticisms (mis)directed at these, though of limited utility as a tutorial for applying collostructional analyses. The latter purpose is better suited by Stefanowitsch & Gries’ (2003) original work.

The second chapter, by Jóhanna Barðdal, Thomas Smitherman, Valgerður Bjarnadóttir, Serena Danesi, Gard B. Jenset & Barbara McGillivray, attempts to empirically reconstruct the constructional semantics of the dative-subject construction in an earlier stage of Indo-European by analysing such constructions in Old Norse/Icelandic, Classical Latin, Ancient Greek, Old Russian and Old Lithuanian. The dative-subject construction describes non-canonical sentence structures in which the subject-like element, representing the most agentive participant in the event structure, is in dative case, not in the canonical nominative. The study is based on 835 types of this construction, almost half of which stem from Old Norse/Icelandic, while only about 5 per cent are from Old Lithuanian. The remaining types are roughly evenly distributed across the other three historical languages. By means of an elaborate methodology and multivariate statistics, Barðdal et al. show the dative-subject construction in said languages to involve well-nigh exclusively verbs of happenstance and experience, i.e. denoting activities ranking low in agentivity (e.g. like, need, lack, know, happen). As is often the case with pioneering work, some points of detail invite comment and discussion, which the present review regrettably cannot do justice to. The authors conclude that the dative-subject construction must be monogenetic in origin, i.e. in all cases inherited from (western) Proto-Indoeuropean, yet they might be somewhat rash in entirely dismissing the possibility that dative subjects could also have developed separately in the languages concerned, through an inherited, hence very similar, case machinery in which datives tend to encode experiencers, and the cognitive commonality of weakly agentive subjects in these constructions. This methodologically innovative and stimulating study also concludes that a formal, non-constructional approach to the issue is a ‘theoretical impossibility’ (p. 73) in view of the generativist doctrine that syntax is independent of – and therefore cannot have – meaning.

In the third chapter Amanda Patten compares a constructional account of it-cleft development to a generative one. She contrasts two of her own publications from the 2010s with two generativist studies from the early 1990s, and finds the it-cleft to originate in Old English (OE), and not to emerge only in Middle English, as was claimed about twenty years earlier. She notes that the purported ancestors of the it-cleft were very rare and confined to close Latin translations, while the constructions she identifies as the origins of the it-cleft were rather more frequent and widespread. Her analysis of the OE data relies on gender agreement of demonstrative þæt in relative function as a diagnostic tool, citing Mitchell (1985: 102, 108) in support, yet not heeding Mitchell's (1985: 101) caution that ‘[i]n matters of concord, þæt appears to be the most consistent offender’, an assessment fully borne out in a recent large-scale investigation of mediaeval English gender agreement (see Dolberg 2015). This caveat does not, however, threaten the overall validity of Patten's analysis. In conclusion, she notes that the two pairs of studies considered have very different outcomes, due to dissimilar theoretical assumptions, and also because these in turn largely determine ‘which evidence counts as relevant’ (p. 109). Even so, Patten shows that her recent constructional account involving a piecemeal development is more coherent and fits the available evidence better than the generative model from the early 1990s invoking abrupt changes.

Trousdale provides the fourth chapter, devoted to the development of the what with-construction, arguing that it manifests grammatical constructionalisation. As in Patten's chapter, a constructional account of the phenomenon at hand is compared to a previous generative (minimalist in this case) approach, discussing their diverging theoretical implications. Subsequent corpus analyses involve seventy instances from eighteenth-to-early-twentieth-century English and a random sample of 500 items from the COCA corpus, representing present-day American, leading to a convincing in-depth discussion of why and how the rise of the construction constitutes a case of grammatical constructionalisation, involving a series of ‘micro-steps, piecemeal realignments in signs’ (p. 138). In conclusion, Trousdale further notes that a focus on these micro-steps ‘is an area of considerable convergence between generative and cognitive approaches to grammatical change’ (p. 138).

The fifth chapter is long and entirely theoretical. Gisborne discusses the semantics and grammaticalisation of the, beginning by comparing two approaches to definiteness: one is functional and hinges on familiarity of the addressee with the referent (i.e. definite reference conveys hearer-old, discourse-given information), the other, so-called uniqueness approach understands the as sharing characteristics with both the existential and the universal quantifier in formal logic, thus likening the to proportional quantifiers such as all, some or most. Understanding these two approaches as mutually exclusive, Gisborne adopts the latter, formal approach, because it provides (formal) solutions to (formal) problems such as specificational sentences or scope and definiteness effects (by means of introspective acceptability judgements of made-up sentences). Utilising set-theoretic semantics, he models this formal quantifier approach within Word Grammar, a cognitive theory of language structure. Turning to grammaticalisation, Gisborne adopts the known, though not uncontested view that the developed/grammaticalised from the OE demonstrative se/seo/þæt, referring to Mitchell (1985), and that it did so by means of losing its deictic meaning component, retaining only definiteness. He concludes that either theory can easily handle the grammaticalisation of the by invoking a lexical split se ➔ the vs that, without categorial change or postulation of a determiner category. He also concludes that the set-theoretic quantifier account outperforms the familiarity approach, because the latter would require definiteness to be heterogeneous, effected by, for example, contextual identifiability, shared world knowledge, etc., possibly organised by means of family resemblance, while the former captures definiteness as a clear-cut unitary category, and merely requires ‘the loss of the few network links that make the meaning of se or that deictic’ (p. 180).

Gisborne clearly has a point, at the very least from the perspective of Occam's Razor; nonetheless, one or two issues might be raised. First, there is the unstated premise that OE demonstratives in pre-nominal position were always and necessarily deictic (i.e. like today's that, but not the), which is not only very hard to demonstrate empirically, but also unlikely, given that the distal demonstrative se and its forms vastly outnumber its proximal counterpart þis and its forms. Hence, the proximal þis and its forms are marked, while the distal se and its forms are unmarked. Unless Anglo-Saxons were in the strange habit of chiefly talking about distant things as opposed to ones close by, the difference in frequency and thus markedness is more parsimoniously motivated by surmising that the OE distal demonstrative in pre-nominal position was in fact polysemous, that is, always definite, but not always deictic, thus doing the double duty of today's the and that. If so, Gisborne's main argument against the familiarity account, viz. that it ‘requires us to assume that a new relation comes into play, which allows the speaker/hearer to identify a referential token in the (discourse) environment’ (pp. 179–80) no longer holds, as said relation would not be new, but present all along. Second, Gisborne himself notes that ‘cognitive and functional theories are interested most in language use’ (p. 180, original emphasis). So, even if the familiarity account requires us to make an additional assumption, it allows, or even requires categories involving overlaps, fuzzy boundaries, and in-between cases, with the various categories and their instances being cognitively organised according to, for instance, family resemblance, exemplars and/or prototypes. Thus, the familiarity account not only captures the development and semantics of the in terms of definiteness and deixis, but at the same time it also provides a plausible account of how all this is mentally represented, invoking mechanisms known to be involved in other aspects of cognition. In short, the familiarity account may well be psychologically real, while the set-theoretic one is unlikely to be. Ultimately, however, which account is given preference will follow from the purpose at hand, and there may be worthwhile avenues of research that do not require accurate depiction of psychological reality, but rather benefit from the leanest possible theory.

In chapter 6, Sonia Cristofaro discusses the role of synchronic vs diachronic evidence in modelling mental representations and their underlying psychological mechanisms, drawing on typological data. She claims that synchronic, distributional data is not per se evidence for mental representations and the psychological mechanisms effecting these, but converging diachronic evidence elucidating how these distributions came about is required before claims of psychological reality should be made. Conversely, instances of language change should not be taken as directly reflecting psychological mechanisms if no corroborating synchronic, distributional evidence is obtained. Cristofaro demonstrates that typologically parallel alignment configurations in different languages can be effected by very dissimilar developments and reanalyses, to the effect that the mechanisms uncovered may ‘account for the evolution of specific constructions, rather than the relevant alignment systems in general’ (p. 196). Turning to prototype effects in word classes, she focuses on morphemes that cause verbs to be profiled as objects, not actions (e.g. Reading is fun.), which bring about reified actions and hence less prototypical verbs. She posits that usage of these verbs does not evidence prototypicality effects, but a ‘rather general principle whereby complex meanings are conveyed by a combination of different elements’ (p. 204).

Cristofaro concludes that synchronic distributions may result from developments unrelated to psychological principles ostensibly effecting these distributions, and that parallel synchronic distributions can be brought about by ‘different diachronic processes in different cases’ (p. 205). This chapter is commendable in cautioning against the epistemologically questionable practice of interpreting synchronic distributions as directly reflecting cognitive architecture and/or psychological principles, but likewise overemphasises the purportedly greater utility of diachronic data for getting at speakers’ mental organisation, because both types of data ultimately treat language as a product. Psycholinguistic experimentation by contrast, a type of synchronic evidence given somewhat short shrift in this article, necessarily treats language as procedural and is thus better suited to uncover mental organisation and operations driving actual language use. However, psycholinguistic experiments do not directly show mental processes either, so that a reliable account of psychological principles and cognitive mechanisms underlying language use should incorporate converging, independent evidence from diachronic developments, synchronic distributions as well as experiments.

The seventh chapter concerns word classes, specifically nouns and verbs. Hollmann summarises structuralist/generative, cognitive-linguistic and psycholinguistic takes on the topic: the first school of thought differentiates parts of speech by means of their distribution, while cognitive-linguistic theories take word classes either as largely semantically (and maybe also phonologically and/or distributionally) determined, or as classified according to an admixture of pragmatics and semantics. Psycholinguistics has empirically ascertained the importance of distributional, semantic, as well as phonological properties of word classes; however, Hollmann observes that this rich body of evidence is essentially ignored by generative and cognitive theorists alike. He notices a gap in previous psycholinguistic research, which utilised corpus data, artificial language learning algorithms, and comprehension experiments with human participants, though no production experiments, and sets out to fill this gap by asking seventy-two native-speaker undergrads from his university to concoct three nonce nouns and verbs each and to write down sentences involving these. This data he analysed according to phonological and distributional predictors and scoring procedures taken from the literature, and tested these predictors for significance, though not effect sizes, using series of univariate statistics. This choice of statistical tools is unfortunate, because using univariate tests in sequence on the same data drastically increases the risk of false positives, and also does not allow testing for correlation between predictors (see, e.g., Field et al. 2012: 274–6, 399–400). Multivariate statistics can remedy these problems, and additionally allow assessing and comparing the impact of each predictor. Most phonological parameters display influence in the same direction as in previous research. The distributional parameters showed that nonce nouns occur predominantly in subject or object position, even more so if preceded by a determiner, while nonce verbs prefer finite clauses. Hollmann convincingly concludes that cognitive-linguistic models of word classes need to include phonological and distributional cues in addition to semantic(-pragmatic) criteria.

In the final chapter, Teenie Matlock, David Sparks, Justin L. Matthews, Jeremy Hunter & Stephanie Huette experimentally ascertain how (im)perfective aspect influences the description (including gestures), and thus construal, of witnessed events. The authors required twenty-two Anglophone undergrads from their university to watch muted videos of road-accidents. After each clip half of the students were asked “What happened?” (perfective), the other half “What was happening?” (imperfective). In both conditions, participants produced descriptions of similar length, featuring similar ratios of perfective and imperfective verb phrases, and produced a similar number of gestures. However, the imperfective condition elicited significantly more motion verbs and fewer non-motion verbs than the perfective condition and prompted significantly more phrases indicating dangerous driving. Iconic gestures also precipitated significantly more often in the imperfective condition, while ones carrying ‘no obvious semantic meaning’ (p. 251) were significantly fewer. As in Hollmann's contribution, statistical significance was assessed by means of series of univariate tests, so the criticism issued above applies also here. In keeping with much previous research in various disciplines, the authors interpret their results as evidencing embodied perceptual simulations, which reactivate ‘patterns . . . anchored in past perceptual and motor experience’ (p. 252), thus providing yet further evidence against an encapsulated, autonomous language module.

What then is the common denominator of these eight chapters, what renders them a book? One leitmotif is that many of the studies directly or indirectly compare a cognitive approach to a formal/generative one, finding that the former outdoes the latter in every case. Another commonality is already suggested in the title, i.e. the relation and relative weight of empirical data vs theoretical considerations in cognitive linguistics. A majority of the chapters is empirical, and thus in accordance with Hollmann's astute verdict that ‘further research is not well served by rigid theoretical assumptions or distinctions, but should be driven first and foremost by empirical evidence’ (p. 234). However, not all the chapters included in this volume are primarily empirical, and this primacy of empirical data is still not universally acknowledged (let alone practised) within cognitive linguistics. This lucidly demonstrates the continued need to keep advocating as well as employing empiricism in order to expedite the paradigm shift towards (cognitive) linguistics as a more empirical science. In sum, then, the present book does not quite fulfil its potential in providing a decisive contribution to overcoming the theoretical heterogeneity that has impeded cognitive linguistics for decades, however it does successfully bundle a selection of insightful research papers highly relevant to this issue and thus constitutes a much-needed step towards rendering cognitive linguistics more consistent in theory and more empirical in methodology. Therefore, it can be recommended to linguists of any persuasion.