Abstract
The thematic role 'agent' is known to be semantically very flexible. English speakers often use agentive expressions not only to describe intentional events, but also accidental events (Fausey et al. 2010). Also, inherently agentive verbs like English eventive hit can take inanimate subjects (Fillmore 1967). The flexibility of the role agent is not without limit, however. Even in `flexible' languages like English, agentive language is context-sensitive and not appropriate to describe any kind of accidental events (Knobe, Joo, Yousif, Martin & Keil, in prep.). Furthermore, speakers of languages that have morphological markers of reduced agency (such as Japanese or Tagalog) tend to disprefer (unmarked) agentive language in accidental contexts.
In this talk, we explore the hypothesis that the predicate 'agent' shows the same context-sensitivity as gradable predicates (Kennedy 2007). Speakers judging the applicability of the predicate 'agent' introduced by agent Voice make reference to a `standard degree' of agency with respect to the relevant comparison class. But while research on gradability traditionally has focused on adjectives such as tall, defining a single unanalysed dimension, we observe that the predicate `agent' has greater affinity with gradable multidimensional predicates (Kamp 1975, Sassoon 2012, Sassoon and Fadlon 2017). Sassoon shows that some of these predicates, such as healthy or optimistic, have readings based on universal quantification over dimensions, as revealed for instance by exception phrases (see, e.g., Hamida is healthy except for blood pressure). We note that agentive predicates behave the same (Hamida did it except she wasn't aware of it/she didn't do it on purpose/she doesn't know how she managed to).
We identify grammatically relevant agentive dimensions on the basis of typological work on reduced-agency marking, and adapt Sassoon's semantic analysis of multidimensional predicates to introduce degrees and dimensions in the semantics of the predicate 'agent' introduced by agent Voice.