Discourse connectives vs. paraphrases: what influences likelihood and form of reference in the absence of verb semantics?

The subject of like has the semantic role of Experiencer, while the object is the Stimulus. Like and similar verbs belong to the class of Experiencer-Stimulus verbs (ES). In continuation experiments, where participants are asked to write continuations of sentences/clauses after certain verbs followed by a connective and (often) an ambiguous pronoun, as in (1), participants show a strong and robust tendency to continue with a reference to the Stimulus argument: in (1-a), he = Bill. But connectives also influence continuation likelihood. Stimulus continuations are preferred after because but Experiencer continuations are preferred after so, see (1-b). ES verbs and Stimulus-Experiencer verbs (SE) like amaze, that have Experiencer objects and Stimulus subjects are all State verbs. Stevenson et al. (1994; hereafter Steve&co) and Fukumura & von Gompel (2010; hereafter F&vG) found semantic role continuation preferences hold for State verbs with consistent effects for connectives, regardless of the grammatical role of the Experiencer or Stimulus. However, not all verb groups that show consistent continuations biases find an interaction with connectives. Steve&co found connective choice did not influence Transfer verbs(e.g. give) or Action verbs (e.g. hit). Transfer verbs are consistently biased to Goal/object continuations, and Action verbs to Patient/object continuations. An explanation may be that connective-verb semantic interactions only occur when connective meaning relates to verb meaning. For State verbs the Stimulus is the cause, so it is not surprising that after because most continuations refer to this semantic role. The opposite is the case for so (used with Cause-so-Effect sequences). The connection between Agent/Patient roles or Source/Goal roles and causal connectives may be more indirect with Action and Transfer verbs. Nonsense verbs can be used to study the contribution of connectives separate from preferences stemming from verb biases. Given this background, our first research question is: