The Source of Passive Sentence Difficulty: Task Effects and Predicate Semantics, Not Argument Order

Passive sentences are claimed to be more difficult cross-linguistically and -populations (children, adults, people with aphasia). In healthy adults, Syntactic Complexity accounts attribute this difficulty to the complexity of computing an additional syntactic dependency (i.e., movement), while Good Enough Processing attributes it to a contradiction between an agent-first heuristic and syntactic processing outputs. Offline accuracy data support a greater difficulty for passives, but online self-paced reading data conflicts with these accounts in showing faster reading times for passives than actives. Here, we use acceptability judgments to exclude the possibility that the self-paced reading data may not have been sensitive to a late complexity effect. Experiment 1–2 further show that passive sentences are no more complex to parse and interpret than active ones, as acceptability judgments do not differ between them. These 2 experiments also replicate the previously observed accuracy effect, whereby passives are more errorful in comprehension. The offline comprehension difficulty is considered to arise from a task bias on Working Memory demands rather than parsing or interpretation. Experiment 3 controls these demands through balancing the voice of the comprehension question and no difficulty of passivization is observed in accuracy. In line with prior acquisition work, we do observe an interaction between predicate type (state, event) and voice (active, passive) in accuracy judgments. Stative passives are more errorful than stative actives, but no (or less) difference is observed between eventive passives and actives. Acceptability judgments show a similar interaction in the collapsed data, although marginal. Passives with a stative predicate are temporarily ambiguous (verbal vs. adjectival passive) and require disambiguation to a less preferred interpretation (verbal). Stative passives have also been argued to require coercion of a state subsequent to an event, which could increase interpretation difficulty. These results argue that passivization difficulty arises from task artefacts and additional processing factors introduced by stative predicates, and run counter to mainstream theories that posit a role for noncanonical argument order.