Eva Neu (University of Massachusetts)

Encoding knowledge about verb distribution: Lessons from direct causatives of unergatives 
A standard theory of argument structure holds that intransitive verbs are stored in the lexicon with one of two basic syntactic frames: unaccusatives (1b), which have an internal argument assigned a patient theta-role, and unergatives (1b), which have an external argument assigned an agent theta-role:

(1a) The glass broke.  
(1b) The girl laughed. 

Unaccusatives can form direct causatives via the addition of an agent argument (2a). Adding an agent argument to an unergative would result in an event description containing two agents, which is widely assumed to be blocked. This predicts direct causatives of unergatives to be impossible (2b):

(2a) I broke the glass. 
(2b) *I laughed the girl. 

In this talk, I look at data from three languages that violate this generalization and allow direct causatives of unergatives like (2b): Hindi-Urdu, Turkish and Sason Arabic. I show that the object argument of these constructions behaves syntactically like an internal argument and semantically like a patient. I go on to discuss what this finding tells us about how knowledge of the different syntactic uses of verbs is encoded in speakers’ grammars. The standard theory outlined above is too rigid and underpredicts the distributional flexibility of verbs. Instead, I advocate using word embeddings to develop an alternative theory of verb distribution that can deal with variation and gradience, and I present some of my ongoing work on using interpretable dimensions to predict the syntactic behaviour of intransitives.