Solstad & Bott (2022): On the Nature of Implicit Causality and Consequentiality #

This file formalizes the Two-Mechanism Account of Solstad and Bott. Implicit causality (I-Caus), the coreference bias of a continuation after because, is verb-semantic: on the authors' Empty Slot Theory, a stimulus argument causes the psychological state the verb names without the verb saying how, so the argument carries an underspecified proposition that an explanation fills. Implicit consequentiality (I-Cons), the bias after and so, is discourse-structural: on the Contiguity Principle of Kehler, a consequence continues from the final state of the prompt's eventuality, held by the experiencer of the state or by the participant the action affects. Both mechanisms read off a verb class's proto-role grid (Dowty): CarriesSlot is the stimulus label, HoldsEndState the experiencer label or causal affectedness, ICaus and ICons the resulting coreference predictions, and Bias routes them through the direction of the cause–effect relation a connective signals.

The rival One-Mechanism Account of Crinean and Garnham reads both biases off one causal decomposition, causes with agents and stimuli alike, consequences with experiencers and patients, so a continuation about the end-state argument should be a consequence. The Two-Mechanism Account lets slot filling take precedence over contiguity, so a slot predicate is continued by an explanation whichever argument the continuation is about. oneMechanism and twoMechanism state the two coherence predictions; they agree on the slot argument and diverge exactly on the end-state argument of a slot predicate, the paper's Asymmetry Hypothesis. The accounts also part on agent-patient verbs, whose agent is a cause on the One-Mechanism reading but carries no slot, since it causes by the very action the verb names.

Four sentence-continuation experiments on German stimulus-experiencer (ärgern) and experiencer-stimulus (bewundern) verbs test the hypothesis. Experiment 1 confirms the mirror coreference biases: after weil the stimulus argument, after sodass the experiencer, with the two per-verb biases almost perfectly anticorrelated. Experiment 2 (full stop) finds explanations the most frequent relation for both classes, about three times as frequent as consequences. Experiment 3 (forced reference) finds explanations at least as frequent as consequences even when the continuation is forced onto the I-Cons argument, against the One-Mechanism symmetry. Experiment 4 finds explanatory specifications, continuations giving the direct cause of the psychological state, almost only in I-Caus-congruent explanations, and consequence specifications, restatements of the experiencer's state, essentially never: consequences introduce an eventuality subsequent to the prompt's, as contiguity predicts.

Implementation notes #

The experimental results are reported in prose above; the library's format for experimental data is pending. The psych grids are the substrate's psychCausal and psychState, which are each other's flip, so the mirror theorem is one instance of bias_flip. The agent-evocator grid judgment follows the paper's description of Crinean and Garnham: the evocator is causally affected by the agent's act and, through the verb's presupposition of a preceding eventuality it took part in, the non-agentive cause of the agent's intention, so it is both the slot argument and the end-state argument. Agent-patient verbs take the manner-contact grid mannerContact.

Solstad & Bott (2022): On the Nature of Implicit Causality and Consequentiality #

Implementation notes #

References #

The two mechanisms #

The psych doublets #

The two accounts and the Asymmetry Hypothesis #

The four verb classes #