Theory and data in cognitive linguistics
Nikolas B. Gisborne and Willem B. Hollmann
University of Edinburgh / Lancaster University
The genesis of this volume was a workshop at the Societas Lingüistica European conference in Vilnius in 2010, organised by Hollmann and Gisborne, on the question of what specific contribution cognitive linguistics has made to the gathering and analysis of linguistic data. The workshop, Theory and data in cognitive linguistics, included the chapters in this volume by Barðdal et al., Cristofaro, Gisborne, Hollmann, and Trousdale; Patten's chapter should have been presented at the conference, but the author was unable to attend; and the chapters by Gries and Matlock et al. were specially commissioned for this volume.
One of the motivations for the workshop was the observation that cognitive linguistics has an honourable tradition of paying respect to naturally occurring language data. In a tradition which typically describes itself as "usage based", it makes sense to consider linguistic data in terms of what can actually be found. Therefore, for example, there have been fruitful interactions between corpus data and aspects of linguistic structure and meaning: Gries & Stefanowitsch (2006) have proposed a particular research strategy couched in construction grammar; Cuyckens et al. (2003) explored corpus data in the analysis of lexical semantics; and Bybee (2001) presented a usage-based approach to phonology. The usage-based research strategy has also led to an exploration of dialect data, sociolinguistic data collection methods, and sociolinguistic theoretical concepts in the light of the cognitive stance on language (Hollmann & Siewierska 2006, 2007, 2011; Clark and Watson 2011). In addition, there has been an increase in several kinds of experimental work seeking to link cognitive linguistics to current research in psycholinguistics (Bergen et al. 2007, Dabrowska 2008), as well as in attempts to combine evidence from various sources, as in Gries et al. (2005).
However, not all linguistic data is simply naturally occurring or derived from experiments with statistically robust samples of speakers. Other traditions, especially the generative tradition, have fruitfully used introspection and questions about the grammaticality of different strings to uncover patterns which might otherwise have gone unnoticed. The difference between "control" and "raising" patterns (Rosenbaum 1967), the distributional constraints on reflexive pronouns (Reinhart 1983), the island constraints (Ross 1967), and that-trace effects (Chomsky & Lasnik 1977) are all important linguistic phenomena which were uncovered by the research methodologies of the generative tradition. In relation to that-trace effects it is also worth mentioning the work of Cowart (1997) and Featherston (2005a, 2005b). Both of these generative grammarians have done important pioneering work on turning acceptability judgments into a methodology that is considerably more robust than it had been in what Cowart first referred to as the "Hey Sally method" (1997: 2), where the researcher collects a few informal judgments at the most in addition to their own.
Some of the data sets considered by generative linguists have been subject to cognitive analyses: Van Hoek (1997) wrote on anaphora, while island constraints were discussed by Deane (1992), who attempted to locate the constraints in a view of language that was consistently embedded within a larger theory of human cognition. More recent cognitive work has addressed other topics which have been important in generative and formal theorizing. For example, Barðdal (2008) offered a cognitive perspective on "quirky" case marking, and Israel (2004) presented a cognitive analysis of polarity, a topic that has received extensive discussion in the generative and formal semantic literature. We might call the data sets analysed in these studies "theoretically uncovered" data, i.e. facts about language that linguists have discovered as a result of their particular theoretical tools.
There is also a diachronic tradition which, by necessity, must pay attention to real language data (because it could not otherwise describe phenomena) but where the degree to which data are (necessarily) idealized is contingent on whether language change is viewed as sudden and abrupt (Lightfoot 1979) or gradual and incremental (Croft 2000, Hopper & Traugott 2003). The assumption that language change is catastrophic was first developed by Lightfoot in the generative tradition. The gradualist tradition is in line with cognitive assumptions about the organization of lexical and grammatical categories, and although it is usually not explicit, most (functionalist) work in grammaticalization is consistent with many of the research results of the cognitive tradition. Indeed, there is now an emerging body of work in diachronic construction grammar, such as Gisborne and Patten (2011), Israel (1996), Patten (2010), Traugott (2003), Trousdale (2008), and Patten (2010), which brings the functionalist and cognitive traditions together in exploring language change.
However, there are areas where cognitive linguists have neglected some of the data sets which have been important in generative theorising. For example, if we take language change, although there is work where the grammaticalization tradition has examined the same data sets as generative linguists, such as the emergence of the English modals, there are areas where the two research paradigms have not converged. There are several generative studies of changes in word order, just as there are studies of changes in patterns of negation (van Kemenade 1999, Pintzuk 1999, Koopman 2005) but we are not aware of cognitive research into these topics.
Linguistic typology has not received as much attention from cognitive linguists as one would perhaps expect, given that crosslinguistic facts may give clues to cognitive structure (Croft 1998). Nevertheless, there has been enough work to see how a cognitive approach might differ from a generative one. Generative work (e.g. Baker 2009) is generally motivated by a search for Universal Grammar and tends to be based on relatively small language samples. Cognitive work on linguistic typology, by contrast, continues the legacy of the Greenbergian approach by using relatively large language samples, and by arguing that what is universal does not lie in language structure as such, but in speakers' conceptual spaces and in the constraints on the mappings between functions and structures (Croft 2001, Cristofaro 2003).
Summing up the observations made thus far, we can see that the divide between generative and cognitive approaches to language is intimately connected to the kinds of data drawn upon, the way in which generalisations are derived from these data, and how data is gathered. The divide is wide, but we note that there have been attempts to bridge it, to some extent on the cognitive side (Croft 1999), but more clearly by generative linguists such as Culicover & Jackendoff (2005). Some cognitive work sets out explicitly to address data and research questions that have been discovered and addressed in the generative tradition. Hudson (2003) is an exploration of gerunds within a cognitive theory of linguistic categories, and Gisborne (2010) includes a discussion of the perception verb complements which were a motivation for situation semantics (Barwise & Perry 1983) and event semantics (Higginbotham 1985).
In selecting the chapters for this volume, and indeed for the earlier workshop, we thought it was important to address the questions of the extent to which cognitive linguistics offered an alternative paradigm to the formalist traditions and, indeed, to what extent cognitive linguists spoke to linguists from formalist traditions. As well as developing the research traditions that have emerged out of cognitive linguistics' early focus on lexical semantics, and a respect for naturally occurring data, we thought that it was important to see how cognitive approaches could provide coherent and empirically sound accounts of the kinds of theoretically uncovered data that we referred to above. If cognitive approaches can provide robust accounts of those data, then they (should) have an advantage with regard to their compatibility with findings in cognitive psychology.
In addition, we took the view that cognitive linguists must critically evaluate their growing set of theoretical constructs and assumptions, some of which may not make clear contributions to our understanding of language (Broccias & Hollmann 2007). Essentially, against a background of how we gather and analyse our data, we wanted to offer a forum where cognitive linguists could critically evaluate their research practice and engage with a range of research questions, not only those which naturally arise within the cognitive research tradition.
As a result of the thoughts and ideas we describe above, we asked our contributors to address some specific questions and these questions frame the contents of the chapters which we have gathered in this volume. Although we describe each chapter in turn below, this is an opportune point to explain how the chapters generally fit the volume. The first question was whether cognitive linguistic approaches could uncover generalisations that are not so easily uncovered in other approaches. Four chapters treat this question in various ways, those by Bar5dal et al., Gisborne, Patten, and Trousdale, which all have a historical focus as well, as does the chapter by Gries, which does not have a diachronic dimension. Another question was about the reliability status of theoretically uncovered data and generalisations: This is addressed in the chapters by Gries and Hollmann. We also wondered whether there were assumptions in cognitive theories which do not have obvious benefits in the analysis of linguistic data, and the chapters by Cristofaro, Gries, and Hollmann address this. Finally, we were interested in how a theory's stance on the mind (is it embodied or encapsulated?) affects research results, and this question is addressed in the chapter by Matlock et al.
The contributions to this volume overlap in terms of the themes they address but, on top of that, they also fall into three "clusters". A number of them seek to address the questions from a (partly) diachronic perspective: Patten, Trousdale, and Bar5dal et al. But Bar5dal et alls main aim is to provide a blueprint for the reconstruction of constructional semantics, and its emphasis is thus really methodological. Gries's chapter addresses methodology as well. Given the primacy of methodology in any field of research, these two chapters start off the volume, with Bar5dal et al. forming a natural bridge to the diachronic section. The remaining contributions by Gisborne, Cristofaro, Hollmann, and Matlock et al. primarily address theoretical issues. Gisborne's and Cristofaro's chapters also discuss historical evidence and the nature of theoretical questions and thus figure here as a link to the final, theoretical section of the volume. Matlock et alls chapter engages with general real world concerns as well, and language and power in particular. For this reason, it seemed fitting for their study to conclude this volume.
We now turn to discussing the chapters in more detail.
-  In a slightly different turn, some data which have been discussed within the generative tradition have been subject to challenges by cognitive linguists as to their reliability, as in Dbrowska's (1997) study on WH-clauses.