Implicit Causality and Explicit Consequentiality

Oliver Bott

Universität Bielefeld, Department of Linguistics

In the study of sentence processing, Implicit Causality (IC) and Consequentiality (ICons) biases have enjoyed significant attention as expectation-based phenomena (a.o. Garvey & Caramazza 1974, Brown & Fish 1983, Koornneef & van Berkum 2006, Pykkönen & Järvikivi 2010, Bott & Solstad 2014, Garnham et al. 2020). Language production research has found a ‘mirror bias’ for a number of the involved verb classes, e.g. stimulus-experiencer (annoy) and experiencer-stimulus (admire) verbs. Thus, causality prompts (because) display a stimulus bias, whereas consequentiality prompts (so (that)) display an experiencer bias (Crinean & Garnham 2006, Hartshorne et al. 2015). Consequently, Crinean & Garnham (2006), Pickering & Majid (2006), and Hartshorne et al. (2015) argued for a Contingency Mirror Hypothesis: What triggers the IC bias, should also be responsible for ICons, causes and consequences being part of one relation. Although Crinean & Garnham (2006) as well as Hartshorne et al. (2015) are ‘verb-based accounts’ and Pickering & Majid (2006) is based on world knowledge, they both predict causes and consequences to be equally available.

In my talk I will argue for the Causal Asymmetry Hypothesis instead, and propose that different mechanisms apply to IC and ICons. For IC, verb semantics triggers causal specifications, involving a contingency relation with an underspecified cause (i.e., what is it about John that annoys Mary?; Hartshorne & Snedeker 2013, Bott & Solstad, 2014, Bott & Solstad, to appear). We assume no parallel mechanism for consequences, as the effect is already specified in the verb (i.e., annoying causes being annoyed).

Evidence for the Causal Asymmetry Hypothesis will be presented from two lines of written production studies. The first set of experiments provides direct evidence for our lexical theory of IC. The second set of experiments directly contrasts IC and ICons showing that the two discourse biases are in fact very different from each other once we look beyond coreference into discourse structure..