(With Sebastian Sauppe, Åshild Næss, Martin Meyer, Ina Bornkessel-Schleswesky, and Balthasar Bickel. You can find the published paper under my publications.)
In the field of language processing, it’s a known effect that if a sentence starts with a nominal (NP/DP), the listener will just make the default assumption that this NP is the agent in the sentence to come. If the rest of the sentence then reveals this first NP was not the agent (for example because the verb is passive, or the patient/theme has been scrambled linearly to the first position, etc.), then speakers get surprised, and one can measure this physiologically (with EEG measurements).
It’s then reasonable to ask whether this agent-first bias is built in or hardwired in our species, or whether it’s just learned from experience. After all, in most languages, the subject is by default the first constituent of the sentence, and subjects overwhelmingly tend to be agents. Thus, one could imagine that the agent-first bias is simply the result of having heard a lot of sentences that start with an agent.
To test this, we ran the same kind of experiment on a language that by default puts patients first. Together with Sebastian Sauppe, Åshild Næss, and other collaborators, we’ve run a study like this on Äiwoo, an Oceanic (Austronesian) language where the default transitive construction has OVS order, that is, the patient comes first. An experiment of this kind had never been done on a language like this before. The preliminary results seem to indicate that:
- Indeed, Äiwoo speakers make the reverse assumption of what has been found in other languages: they assume the first NP to be the patient of the sentence, and they show surprisal when it’s not
- …unless this first NP is a human, in which case they still assume it’ll be the agent!