A structured distributional model of sentence meaning and processing

Most compositional distributional semantic models represent sentence meaning with a single vector. Inthis paper, we propose a structured distributional model (SDM) that combines word embeddings withformal semantics and is based on the assumption that sentences represent events and situations. Thesemantic representation of a sentence is a formal structure derived from discourse representation theoryand containing distributional vectors. This structure is dynamically and incrementally built by integratingknowledge about events and their typical participants, as they are activated by lexical items. Event knowl-edge is modelled as a graph extracted from parsed corpora and encoding roles and relationships betweenparticipants that are represented as distributional vectors. SDM is grounded on extensive psycholinguisticresearch showing that generalized knowledge about events stored in semantic memory plays a key rolein sentence comprehension. We evaluate SDM on two recently introduced compositionality data sets, andour results show that combining a simple compositional model with event knowledge constantly improvesperformances, even with dif ferent types of word embeddings.