Introduction to the workshop ‘What counts as an alternation in usage-based construction grammar?’

An important part of the empirical foundations of construction grammar is formed by alternation studies. In alternation studies, both within and outside the framework of construction grammar, researchers investigate the choice between two or more linguistic variants (e.g. Grondelaers 2000; Colleman 2009; Szmrecsanyi et al. 2016; Perek and Goldberg 2017). However, these studies only rarely explicate their view of what exactly constitutes an alternation. In Labovian sociolinguistics, alternations were originally defined as ‘alternate ways of saying the same thing’ (Labov 1972: 188). This definition is hardly tenable for alternation studies in construction grammar, however, as much of the relevant research, including our own work, has focused on discovering meaning differences between the variants, i.e. has shown that the alternating constructions are not saying the same thing (e.g. Levshina, Geeraerts and Speelman 2013; Perek 2015; Krawczak and Glynn 2015).
There are then several possible ways to redefine an alternation: for instance, (i) as a practical research setup created by the researcher to test more general hypotheses (Arppe et al. 2010: 13–15); (ii) as two or more forms that compete for the same function in a community of language users (Van de Velde 2014, 2017); (iii) as a choice point of the individual language user (Bresnan et al. 2007); or (iv) as various constructions that have a special relation to one another in the constructicon, e.g. as allostructions (Cappelle 2006; Perek 2012). This choice of definition has direct practical consequences when carrying out an alternation study. The present workshop is intended to discuss such issues, some of which are listed below, as well as their wider theoretical implications.

• What are the variants of the alternation?

Psych verbs exhibit variation in argument structure in various languages, including Dutch (Pijpops and Speelman 2017). Here, psych verbs can be used in a transitive construction, e.g. dat interesseert mij ‘that interests me’, or a reflexive construction, as in ik interesseer me daarvoor, lit: ‘I interest me for that’. Of course, these two variants are not the only ways of expressing the same proposition: other options include dat is interessant voor mij ‘that’s interesting for me’, dat vind ik interesssant ‘I find that interesting’, ik ben daarin geïnteresseerd ‘I’m interested in that’, etc. What criteria should we use to determine which constructions to consider as variants in an alternation study: should this be determined by the hypothesis at issue? Should we look at the function that each construction fulfills in the community of language users? Or should we rather look at the individual language user, and whether each construction presents an option for them? Or, finally, should the variants have some paradigmatic link to one another?

• Which instances should be taken up in an alternation study?

Alternation studies often restrict themselves to instances that are deemed interchangeable. Is this a valid choice, and if so, what constitutes interchangeability? This question has mostly been discussed with reference to semantic factors and has been dubbed the problem of semantic equivalence (Lavandera 1978; Geeraerts, Kristiansen and Peirsman 2010: 7–9). This problem asks whether we should allow differences in terms of construal or even referential meaning into our dataset, such as the English conative alternation displays in e.g. she struck her legs vs. she struck at her legs (Perek 2015: 92–93). Or what about lexical senses that only seem to be expressed by one of the variants: should their instances be excluded a priori?
The question not only arises with regard to semantic factors, though, but also with regard to processing effects and lectal differences. For example, an important predictor in the alternation between regular and irregular Germanic past tense forms, e.g. sneaked vs. snuck, is the token frequency of the verb: more frequent verbs are more often conjugated irregularly. As such, extremely frequent verbs invariably exhibit the irregular inflection. But where to draw the line between verbs that rarely exhibit the regular inflection and those that categorically do not? Many verbs that were originally thought to be conjugated in a categorical fashion, have in fact been observed to show occasional variation (van Santen 1997; Knooihuizen and Strik 2014).
As for lectal factors, the verb maken ‘make’ is invariably conjugated with a suffix in standard Dutch, as maakte ‘made’. In some dialects, however, it can also employ a vowel change, as miek ‘made’. But how to distinguish between speakers or contexts where miek ‘made’ is improbable and those where it is impossible?

• What is the relative status of lectal, semantic, lexical and processing-related alternation factors?

Do lectal factors correspond to differences between grammars, with e.g. the grammar of one variety functioning differently than the grammar of another (Szmrecsanyi et al. 2016), or are they part of grammar itself (Geeraerts 2010; Höder 2014)? Are lexical biases mostly caused by differences in meaning (Gries and Stefanowitsch 2004), processing pressures (De Smet and Van de Velde 2019), or lectal distinctions (Pijpops and Van de Velde 2018), or can they randomly arise? What is the role of semantic and processing-related factors in not only explaining, but also causing variation (Franco 2017)?
On a practical level, how should we deal with these various types of independent variables? Should we build separate models for separate varieties, or take up lectal factors as predictors in a unified model? Can lexemes be implemented as random effects in mixed models, even though their preferences may be structured?

• Are alternation factors cognitively real?

Alternation studies typically rely on offline corpus evidence and as a result, we only have indirect access to the cognitive processes of language users. Can we then assume that alternation factors are represented in the minds of language users (Divjak, Dąbrowska and Arppe 2016; Klavan and Divjak 2016)? Does corpus evidence suffice, or is it high time to invest in multi-methodological research (e.g. Grondelaers et al. 2009)?