The effect of animacy on the agent preference: Self-paced reading evidence from Basque.

Language processing shows a tendency to prefer agents over other roles. For instance, when initial unmarked noun phrases (NPs) are ambiguous between agent and patient roles, there is a preference to interpret them as agents, with ensuing reanalysis effects if the NP later turns out to be a patient. Intriguingly, this preference also applies in languages where initial, unmarked (caseless) NPs would tend to be patients because agents are often dropped or marked by a distinct case, the ergative. However, an unresolved question is to what extent the agent preference can be modulated by animacy in a language with agent-dropping and ergative case. To address this, we performed a self-paced reading study exploiting a case-marking syncretism in Basque, which makes some NPs ambiguous between agent and patient readings despite otherwise consistent ergative marking of agents. We looked at the role of an animate vs. inanimate initial NP in transitive sentences, modeling self-paced reading times in a hierarchical Bayesian regression framework. When the role of the initial NP was disambiguated by the verb, we found no reanalysis effect. By contrast, when the role of the initial NP was disambiguated by a second, unambiguous NP, we found a slowdown after human patients compared to human agents, but not after inanimate patients, in the words following the disambiguating region. This suggests that the agent preference can be attenuated when initial NPs are inanimate.