Frames, Polarity and Causation

Josef Ruppenhofer

We present a study of polarity-sensitive items (PSIs) such as _lift a finger_ based on corpus data and against the background of semantic frames. Whereas most prior research did not seek an explanation for why particular expressions become PSIs of particular types but instead focused on the syntactic licensing conditions, the work of Israel (Israel 1997; Israel 2011) ties different types of polarity sensitive items to lexical semantics. Based on Israel's model, we pursue several questions. First, we try to find evidence in corpora that it is lexical classes--operationalized as FrameNet frames (Baker et al. 1998)–that correlate with scalar rhetoric and polarity sensitivity. Second, Israel's analysis establishes a crucial link to scalar rhetoric in general: speakers employ scalar models widely and polarity items just happen to be lexical items that are restricted to occurring when such models are in use. We evaluate the assumption that, given knowledge about predicates with cause/agent and patient roles, we should be able to identify productively formed textual instances of scalar rhetoric that do not involve lexicalized PSIs. Finally, we explore the idea of using sub-classes of PSIs in order to mine lexical items with certain semantic properties from corpora.