Epistemological challenges in corpus analyses of alternating constructions

Over the last 20 years quantitative corpus studies have yielded many insights into the regularities involved in numerous alternations in various languages (cf. Colleman 2009, De Vaere et al. 2018, Gries 2003, Gries & Stefanowitsch 2004, Geleyn 2017, Wolk et al. 2013, among others). However, such corpus-based studies of alternating constructions face a number of epistemological challenges. This paper identifies three epistemological challenges for the corpus-based study of alternating constructions and outlines a proposal to deal with these challenges. First, corpus studies face the difficulty to distinguish between normative language rules and regularities in language use. While both normative rules and regularities in language use can be determined from a quantitative and experimentally controlled perspective (Cowart 1997, Gibson & Fedorenko 2013, Schutze 1996, Wasow & Arnold 2005) and while normative rules emerge diachronically out of regularities in language use, rules and regularities have a different ontological, epistemic and psychological status and therefore require to be studied in their own right (Coseriu 1985, Itkonen 1978, 1997, 2006, 2016, Newmeyer 2003, Willems 2012). Second, because correlation does not equal causation (cf. Lass 1980), caution is required in the use of causal language when describing alternations that are studied by means of a purely observational corpus approach. While correlations between constructions and parameters can be described in terms of predictions in statistical models, this does not entail that one has identified the factors that causally govern speakers’ choices. Third, linguists sometimes pay little attention to the difference between functional causes of behavior and mediating mental mechanisms (cf. De Houwer 2011, De Houwer et al. 2017, Hughes et al. 2016). The finding that there is a functional causal relation between a parameter and the choice for a construction does not directly bear on the nature of the mediating mental mechanisms responsible for this finding (cf. Bechtel 2008, Bechtel & Wright 2009). Corpus findings constitute therefore no direct evidence for postulating underlying mediating mechanisms on the subpersonal cognitive level responsible for certain outcomes (cf. Arppe et al. 2010, Sandra 1998, Stefanowitsch 2011). These three epistemological challenges can be confronted in various ways. The paper argues that corpus analyses of alternating constructions should take into account the difference between regularities and rules and study both by means of an adequate methodology, viz. corpus research as such and a combination of corpus research and experimentally controlled introspective intuition-based judgments tasks respectively. Second, while correlational corpus findings cannot be used to make claims about causality, they can be used to generate hypotheses for subsequent behavioral experiments that can shed some light on the functional causal relationships between specific parameters and speakers’ choices. Third, corpus findings can shed light on underlying mental mechanisms, if used to support or falsify independently established and explicit psycholinguistic hypotheses (cf. Gries 2003, 2017, Gilquin & Gries 2009, Stefanowitsch 2011).