Logical metonymies (e.g. the student began the book) have often been treated as a case of type-clash (Pustejovsky 1995): an event-subcategorizing verb (begin) is combined with an entity-denoting object (the book), leading to (1) extra processing costs, ascribed to coercion, a compositional operation needed to construct an event sense for the object (Frisson and McElree 2008) and (2) the recovery of a covert event from complex lexical entries (object → reading, writing). Nevertheless, Lascarides and Copestake (1998) have observed that lexical information is not enough to account for the range of interpretations in logical metonymy (for example, to retrieve the interpretation for My goats eat anything. He really enjoyed your book). They claim that pragmatic inference is often needed to retrieve discourse-relevant interpretations, and thus advocate for an interaction between lexicon and pragmatics. While focusing on the role of type clash, previous experimental studies on logical metonymy have only marginally considered effects of discourse in logical metonymy interpretation (but see de Almeida and Dwivedi 2008) and their interaction with type. Selectional preferences are shaped by generalized event knowledge (Matsuki et al. 2011), which can be described in terms of thematic fit, that is, the typicality of a filler for an argument slot (e.g., the fact that eat requires a [+edible] object or that thief is a more fitting object for arrest than policeman). Thematic fit can be influenced by inter-sentential context but also by a wider discourse context; psycholinguistic studies have shown how people make extensive use of knowledge of typical scenarios to exploit contextual cues and build expectations about upcoming input in language (that is, input with the highest thematic fit with previous context, McRae et al. 1998; Matsuki et al. 2011). We propose a first step towards an expectation-based account of logical metonymy interpretation, where processing costs for logical metonymies are modulated by discourse-driven expectations about upcoming input. We suggest that thematic fit plays an important role in logical metonymy interpretation, by (1) distinguishing metonymic contexts (begin has a low thematic fit for entity-denoting objects such as book, making book a less expected object for begin than an event-denoting object), and (2) determining the most expected interpretation (the interpretation with the highest thematic fit with the context: the author began the book → writing, the student began the book → reading, Zarcone and Padó 2011).