Working toward a synthesis

1 History

A long time ago, in a country far, far away, I ventured a prediction about the future of linguistics (Langacker 1976). It came at a time when generative theory, not that far removed from its triumph over structuralism, was seeming less invincible owing to fragmentation, recognition of its limitations, and questioning of its basic assumptions. The prediction was brief enough to be quoted in full: “The coyotes will inherit the earth”. These metaphorical coyotes were down-to-earth creatures concerned with the messy facts of language (including meaning) in all its complexity, diversity, and idiosyncrasy. I. e., they were language people, while unicorns – the current lords of the earth – were theory people. From their lofty niche, unicorns viewed language as an abstract formal edifice characterized by neatness and elegant simplicity. Despite the sophomoric presentation, this prediction was not entirely erroneous. It might generously be interpreted as foreseeing the emergence (if not yet the triumph) of cognitive-functional linguistics.

The intellectual landscape of linguistics was vastly different four decades ago. Descriptions of meaning were expected to be modeled on formal logic; as a vestige of behaviorism, the possibility of a principled conceptualist semantics was not seriously entertained. Categorization was still based on criterial attributes; while the notion of prototypicality had been introduced (Rosch 1973), it was not yet widely known or accepted. The usage-based perspective lay in the future, the structuralist practice of basing analysis on a corpus having been abandoned in the generative quest for broad generalizations. Linguistic thought and theory were shaped by metaphors emphasizing discreteness, computation, and the tree-like structures posited in generative syntax (Langacker 2016). For many, in fact, the terms syntax and language were virtually synonymous. Finally, widespread factual knowledge of language – what linguists are routinely expected to be familiar with – was extremely limited. For many essential topics (e. g., typology, universals, grammaticization, acquisition, discourse, pragmatics, sociolinguistics), what little I know now represents a vast improvement over what I knew then.

During this era, the largely independent efforts of many scholars produced a wealth of findings and insights that eventually coalesced in the movement now known as cognitive-functional linguistics. It is still quite diverse – as indicated by the hyphenated label – but has an overall coherence for which the shared rejection of generative theory is only incidental. In referring broadly to Cognitive Linguistics, I thus include many linguists who might not categorize themselves with that term. It must be recognized that scholars such as Chafe, Givón, Thompson, Traugott, Du Bois, and Hopper (to name just a few) led the way in what Geeraerts (2010) aptly described as the recontextualization of grammar. If Cognitive Linguistics has indeed undergone a “social turn” and a “quantitative turn”, this was both anticipated and greatly facilitated by their endeavors.

There is no question that Cognitive Linguistics as a whole has taken both a social turn and a quantitative turn when reckoned in terms of the number of scholars and proportion of research. However, for functional linguists these emphases represent a straight line rather than a turn. And for self-styled cognitive linguists these changes in practice do not represent any change in principle; functionalist research was viewed as complementary rather than antagonistic. Certainly my own efforts have always been conceived as meshing with that tradition.

Research leading to what came to be known as Cognitive Grammar (CG) began in 1976. In the effort to provide a viable alternative to generative grammar, it necessarily focused on syntax and inherited certain notions (e. g., strict constituency) that have since been abandoned. It was, however, explicitly and fundamentally antithetical to the generative world view. Among its then radical (and still contentious) proposals were the conceptual nature of linguistic meaning, the meaningfulness of grammar, and the reduction of grammar to symbolic relations between semantic and phonological structures. Instead of modularity, it envisaged a comprehensive and unified account of both structure and use as an integral part of cognition. It was – and Cognitive Linguistics still is – “cognitive” in the sense that, insofar as possible, language is characterized in terms of other, more fundamental phenomena (e. g., memory, perception, attention, imagery).

CG is often criticized as being insular, by virtue of ignoring social factors, and non-rigorous by virtue of lacking empirical, quantitative verification. These characterizations are somewhat uncharitable, in that they fail to distinguish the framework’s essential claims and descriptive potential from the unavoidable limitations of actual practice. Indeed, CG foreshadowed the social turn by claiming that the speaker-hearer interaction is inherent in linguistic units, which are abstracted from usage events. Likewise, it foreshadowed the quantitative turn by its usage-based nature and the view that structure resides in processing activity. The current emphases on social factors and quantitative methods do not require any alteration of CG but conform to the basic vision of its original formulation (Langacker 1987). They represent the serious exploration of vast areas described there in only the sketchiest terms.

Despite the uneven coverage of self-styled CG research, the framework is compatible with a broad array of other research efforts and can readily incorporate their findings. A guiding principle has been the unification of seemingly disparate areas, with an overall synthesis as the ultimate objective. In earlier years, the emphasis was on articulating a unified account of lexicon, morphology, and syntax as meaningful symbolic structures. In later years, emphasis has shifted to the integration of structure, processing, and discourse (e. g., Langacker 2001a, 2008, 2016b). This does not represent any fundamental change, but is just a matter of bringing essential phenomena into the realm of more explicit consideration.

2 Social turn

Language being multifaceted and immensely complex, linguistics is a vast enterprise involving many interests, methods, and kinds of expertise. This diversity brings with it the challenge of avoiding false choices, bogus issues, and presumptions of privileged status. For example, it is pointless to argue whether language is a mental or a social phenomenon, for obviously it is both, and a characterization from either perspective is incomplete without the other. Linguistic investigation cannot be based exclusively on data obtained by introspection, from corpora, or through experiment, for all have their uses and limitations.

Cognitive Linguistics was so called in order to highlight its opposition to modularity and to an objectivist view of meaning. The label, however, has given rise to spurious criticisms and gratuitous assumptions about its view of language. Most egregiously, cognitive linguists have sometimes been accused of solipsism. More commonly, the notion that language has an essential cognitive basis is wrongly taken as indicating that it dwells exclusively in the minds of individual speakers. Accordingly, Cognitive Linguistics is often challenged based on points that its practitioners actually subscribe to: that no one individual controls “a language” as a whole; that aspects of linguistic organization have to be studied at the population level; and that instead of being invariant, linguistic structures are created anew in usage, hence subject to contextual influence.

Cognitive Linguistics has never been conceived as an alternative to a linguistics grounded in social interaction: its social basis is simply taken as a given. A key point is that cognition itself is inherently interactive, consisting in processing activity through which we engage the world. Sensory and motor activity, which directly engage the physical world, provide the basis for many levels of mental construction, which largely constitute the world we engage by thinking and talking about it. At all stages and levels this activity is socially guided in a cultural context, language being both a product and a primary instrument of this social influence. Owing to the vicissitudes of individual experience, this influence falls short of full determination, with the consequence that no two speakers have precisely the same linguistic repertoire; all that really matters is that there be enough overlap for communication to succeed. So taken as a whole, what we call “a language” is distributed over a population. Still, a given speaker controls an immense array of linguistic units, i. e., established patterns of activity.

As defined in CG, the units of a language are both cognitive and social: the former as an aspect of neural processing, and the latter in a number of respects. For one thing, a unit is part of a language only to the extent of being conventional in a certain speech community. Also, units emerge from usage events – instances of language use in the full detail of their contextual apprehension – by the reinforcement of recurring commonalities. One recurring feature is the very fact that the speaker and hearer are interacting by using the language in question. Hence the ground (the interlocutors, their interaction, and its circumstances) figures at least peripherally in the import of every unit. Indeed, abstracted units can incorporate any facet of the speech situation common to the usage events giving rise to them, such as the following: age, gender, and status of the interlocutors; their social relationship; nature of the occasion; degree of formality; attitudinal, emotive, and affective factors; and the language (or conceived linguistic variety) employed.

In the CG account, such specifications are inherent aspects of linguistic units, arising in the same way as “structure” in the narrow sense. The need for a cognitive sociolinguistics is thus implicit in the framework, and to the extent that sociolinguistic patterns are abstracted by individuals, it provides a way of incorporating them as an integral part of language structure. To be sure, this treatment of sociolinguistic factors amounts to little more than a placeholder. Calls for a social turn (Croft 2009; Harder 2010) reflect the necessity of doing the actual work of sociolinguistic investigation, which of course is now well underway (Geeraerts et al. 2010).

Originally the CG treatment of discourse was also little more than a placeholder, but it is gradually becoming more substantive. Although the specifics are naturally different, the framework makes no fundamental distinction between (parts of) sentences and larger structures (“sentence” is not even adopted as a basic notion). By the same token, the speaker-hearer interaction extends through discourse, so that neither an expression nor its interactive basis can be fully characterized in isolation. CG thus follows the functionalist tradition by envisaging a seamless account of grammar and discourse. Indeed, grammar in the standard, narrow sense can be seen as consisting in local, more codified discursive patterns. Spelling this out in explicit detail is essential for achieving an overall synthesis.

Certain aspects of the integration of grammar and discourse can be briefly noted. In addition to their descriptive content, lexical and grammatical elements incorporate expressive, interactive, and discursive components of meaning. Also, various constructs needed for the characterization of grammar and discourse are the same or directly analogous, e. g., topic as a discursive analog of subject (Langacker 2001b) and profiling (conceptual reference) as the grammatical analog of focus of interest in discourse (Thompson 2002; Langacker 2015). More generally, CG takes grammar as consisting in the implementation of semantic functions (cf. Croft 2007), which all contribute to the global function of meaningful interaction. Profiling, for example, serves the intersubjective function of conceptual alignment, such that the interlocutors, by symbolic means, momentarily direct attention to the same conceived entity. Finally, since constructions reside in symbolic assemblies of any size, they are not limited to a single sentence or a single turn (Langacker 2012, 2016b). The framework can thus accommodate the sorts of patterns uncovered in Dialogic Syntax (Du Bois 2014). Moreover, discourse genres can be characterized by collections of highly schematic constructions (Langacker 2008: § 13.3).

3 Quantitative turn

For the sake of discussion, we can distinguish “qualitative” and “quantitative” investigation even though they are non-dichotomous and the terms are less than adequate. The latter applies to research based on such methods as corpus study, psycholinguistic experiment, computer modeling, neural imaging, as well as the collection and quantitative analysis of naturally occurring speech (including child language). This now being the default, the word qualitative offers itself as a contrastive label for “old-style” analysis and description: attempts at analyzing and describing linguistic structures without employing such methods. That Cognitive Linguistics has indeed undergone a quantitative term is evident from the sheer prevalence of articles and talks that do employ them – even, it seems, to the near exclusion of old-style research. Papers presenting the statistical analysis of corpus data or experimental findings are now quite common if not prototypical.

The quantitative turn is welcome and of fundamental significance. Quantitative methods are important not only because they deal with essential matters outside the scope of old-style research, but also as independent sources of evidence bearing on qualitative claims and descriptions. That I do not myself employ such methods is attributable to training, inertia, personal choice, and the recognition that others are better equipped to exploit them. While they would have been helpful, over the years they have not been truly necessary for conducting my (old-style) research agenda, which is more than enough to occupy my remaining years.

This leads back to my central point that linguistic investigation is a highly complex and multifaceted enterprise requiring many kinds of methods and expertise. They are all essential, and mutually supportive, in arriving at an accurate understanding of language in all its aspects. If purely qualitative concerns originally predominated in Cognitive Linguistics, the quantitative turn is now correcting the imbalance. Like any swing of the pendulum, this one brings with it the potential for unfortunate consequences. One is the temptation of doing “numbers just for numbers’ sake”, which is harmless but non-revelatory. More serious would be the emergence of a “more rigorous than thou” attitude, the notion that quantitative approaches are inherently “better” than others by virtue of being “scientific”. This could lead to the marginalization of other sorts of investigation which are necessary to the overall enterprise. For instance, with the current emphasis on experimentation, corpus study, and statistical methods, one wonders whether students receive sufficient training in basic qualitative analysis (e. g., the explicit description of grammatical constructions).

Granted the vital role of quantitative methods, it is worth enumerating other components of the overall endeavor. (i) At the top of the list is basic data gathering, notably documentation of the world’s languages before this diversity disappears. Suppose an undescribed language has only one remaining speaker, with only a few months to live. Who should be sent to record it before it is lost forever, an old-style field linguist or a specialist in computer modeling? (ii) Even the best documented languages are only partially and imperfectly described, leaving ample scope for qualitative analysis and description. Tasks like the following are not only unfinished but open-ended: describing lexical meanings as reflected in grammatical distribution; working out the specifics of grammatical constructions and how they relate to one another; analyzing cases of conceptual blending in terms of mental spaces and the connections among them. (iii) Another qualitative task is theoretical synthesis pertaining to such analysis. One facet is the postulation of specific constructs in terms of which adequate and revealing descriptions can be formulated (notions like profile, mental space, or image schema). Also involved are theoretical issues pertaining to the organization of linguistic structure (e. g., the continuum of lexicon and grammar, or constructions [vs. rules] as the primary objects of description). (iv) One should not underestimate the importance (even for theoretical purposes) of the practical application of linguistic notions. A case in point is showing and then exploiting the efficacy for language teaching of particular analyses or a general descriptive framework (Achard 2008). (v) Yet another component is using linguistic concepts to illuminate problems in other disciplines. Examples include the application of CG notions to the analysis of literature (Harrison et al. 2014), as well as the insight brought by conceptual metaphor theory to politics, philosophy, and other areas (Lakoff 1996; Lakoff and Johnson 1999).

At the same time, quantitative methods are essential for numerous reasons, among which are the following. (i) Corpus studies reveal phenomena that would otherwise not be noticed (e. g., Du Bois 2014; Pascual 2014). (ii) They allow the correction of biases in data obtained through introspection (Dąbrowska 2016). (iii) Experimental methods are a potential means of validating proposed descriptive constructs, e. g., the notion trajector in CG (Tomlin 1995; Ibbotson et al. 2013). (iv) Acquisition studies bear on claims about the nature of linguistic structure, e. g., the usage-based model (Tomasello 1992; Dąbrowska 2000). (v) A realistic account of how the structures described by linguists are actually implemented has to be inspired and supported by neurological investigation (Feldman 2006; Pulvermüller et al. 2013).

Thus many approaches, quantitative and qualitative, all contribute to the scientific enterprise of understanding language. It is important to resist any tendency to impose a dull conformity in the name of “rigor”; one should not expect every linguistic presentation to be based on corpus data, employ statistical methods, or follow the typical format used in describing experiments. In particular, while corpus data is essential for many purposes (e. g., when frequency is pivotal to an issue, or in documenting the progression of a change), for others it is otiose. If the goal is simply to characterize the grammatical organization of a basic construction (e. g., a prepositional phrase), the citation of quotidian examples without a corpus count or specific attestations does not imply that the work is “unscientific”. Although the term has negative connotations, “introspection” is a vital source of data; in particular, the controlled elicitation of introspective judgments is a basic method of field linguistics. Such judgments cannot always be taken at face value, but that is true of any kind of data. While the recording of spontaneous speech uncovers important patterns that might not otherwise be noted, it also includes extraneous factors (like coughs) as well as simple errors that speakers can recognize as such.

Basic qualitative description remains essential, even granting (as I do) that the structures described ultimately consist in processing activity which has to be investigated and characterized by other means. A purely quantitative approach would overlook something vital: the experiential basis of language structure (Zlatev 2016). Moreover, qualitative descriptions provide the basis for quantitative methods such as experiment, neural imaging, and computer modeling – they suggest what to look for and allow the interpretation of results. In that sense description is prior; if linguists limited themselves to what had already been securely established on quantitative grounds, there would essentially be no linguistics or detailed knowledge of language structure. Many crucial aspects of language are not yet directly accessible to experiment (or other quantitative methods), or were initially apprehended without them (e. g., the details of any grammatical construction). Moreover, quantitative findings are subject to interpretation and are seldom definitive (Blumenthal-Dramé 2016; Dąbrowska 2016). They do not themselves constitute scientific characterizations or establish their validity, but simply bear on them, often very indirectly.

Qualitative and quantitative research are complementary and synergistic, as the structures described in qualitative terms ultimately consist in neural processing. There is no real inconsistency in the fact that old-style descriptions tend to be discrete, in contrast to the intrinsic variability and continuity of neural activity. Despite its non-discrete aspects, the activity in question is highly organized (often effectively categorical), and this organization constitutes (or at least provides the basis for) linguistic structure. Conversely, discrete characterizations of language structures are not intended by cognitive linguists as a serious claim about their actual nature. A basic feature of the usage-based perspective (Barlow and Kemmer 2000) is that the abstraction of units, as well as their subsequent activation, are unavoidably influenced by their context. As a consequence, most units have an array of established variants, and a given unit or variant is never precisely identical in all of its occurrences.

The discreteness of qualitative descriptions vs. the continuity of processing activity has been a frequent source of pointless discussion. Any number of contentious issues prove to be bogus when the positions are reasonably interpreted and one looks beyond the metaphors employed to the actual claims being made (Langacker 2006, 2016a). Here are a few examples: (i) Can we speak of a language or a dialect given that there are no clear boundaries and no two individuals are identical in their speech? (ii) Do linguistic patterns reside in schemas or in clouds of stored exemplars? (iii) Should lexical polysemy be conceived in terms of a network of senses or a continuous field of semantic potential?

Such issues evaporate when it is realized, on the one hand, that discreteness is a matter of degree consisting in non-random distribution, and on the other hand, that this very non-randomness represents a measure of discreteness of the sort that we recognize as “structure”. We can liken such structures to waves, which cannot be precisely delimited and have no independent existence – they are nothing more than water distributed in a certain fashion. Yet waves are real phenomena with causal influence in the world (e. g., they can capsize a boat) and can be characterized with greater or lesser accuracy; to ignore their existence would be a gross distortion of reality. So properly understood, it makes no real difference, for example, whether we speak of lexical senses or posit a continuous field of semantic potential certain portions of which are favored owing to prior activation (Allwood 2003; Zlatev 2003). These are alternative perspectives on the same basic story: that processing history leaves traces which influence subsequent processing.

My own work has mostly been concerned with description and theoretical synthesis. Not being quantitative, it has sometimes been criticized as lacking rigor, not being scientific, or being non-empirical (not based on evidence, but just a matter of how I feel like describing things). I would argue, however, that qualitative research, when properly conducted, can itself be empirical: a source of evidence, not just a matter of speculation or personal preference. Among the factors that make it empirical are the following: (i) being based on a wide array of data from diverse languages; (ii) being constrained in what it posits to account for this data – a limited set of descriptive constructs, each with many applications; (iii) the plausibility of these constructs, as manifestations of well-established cognitive phenomena; (iv) the coherence of the overall descriptive framework, affording a unified account of the different facets of language structure; and (v) the resolution of classic problems, e. g., “raising” being just a special case of active zones, in turn just a kind of metonymy (Langacker 1995). In general, certain analyses are simply better than others, in terms of coverage, simplicity, elegance, and invoking only phenomena justified on independent grounds (Nesset 2016). These considerations hold not just for individual analyses, but for an overall descriptive framework.

Taken together, such factors afford a principled basis for adopting particular qualitative descriptions as likely candidates for being valid. Naturally, one hopes for other, external sources of validation, but even without them one can build a strong case. It is analogous to putting together a large jigsaw puzzle and seeing that it represents (say) a flower: everything fits together coherently. In the absence of external evidence (like the picture on the box), one might argue that this need not be correct – perhaps, when the pieces are properly assembled, it really depicts a locomotive. Myself, I would go with the flower.

In short, the issue of whether research can be characterized as “scientific” or “empirical” is distinct from the issue of whether it is quantitative.

4 Projection

I believe that coyotes are indeed in the process of inheriting the earth, but it takes a lot of coyotes, with different skills and interests, and with enough mutual awareness for their efforts to be synergistic. Language is such that linguistic research is necessarily multifaceted, as will be an eventual overall synthesis. Even for a single aspect of language structure (narrowly conceived), an adequate account cannot be based on just one kind of data, analytical focus, or descriptive format. As a source of metaphors for language, a biological organism is much more appropriate than algorithmic computation. An organism’s full characterization requires many kinds of investigation (e. g., anatomical, physiological; systemic, cellular, molecular; genetic, developmental, behavioral), each with its own methods, findings, and types of representation. These different perspectives are complementary and mutually informative.

Because the same holds for language, we can reasonably project that Cognitive Linguistics will continue to evolve in this direction. It will then be indistinguishable from linguistics, broadly defined as the investigation of language in all its aspects. Despite many sorts of disagreement (real and apparent), the general outlines of a comprehensive synthesis are gradually emerging. Filling in the myriad details will, of course, be a long-term, open-ended project.