Inferring Covert Events in Logical Metonymies: a Probe Recognition Experiment

Inferring Covert Events in Logical Metonymies: a Probe Recognition Experiment Alessandra Zarcone (alessandra.zarcone@ims.uni-stuttgart.de) Institut fur Maschinelle Sprachverarbeitung, Azenbergstr. 12 70174 Stuttgart, Germany Sebastian Pado (pado@cl.uni-heidelberg.de) Institut fur Computerlinguistik, Universitat Heidelberg, Im Neuenheimer Feld 325 69120 Heidelberg, Germany Alessandro Lenci (alessandro.lenci@ling.unipi.it) Dipartimento di Linguistica “T. Bolelli”, Via Santa Maria 36 56126 Pisa, Italy Abstract It has been widely acknowledged that the interpretation of log- ical metonymies involves the interpretation of covert events (begin the book → reading / writing). Whether this implicit content is part of our lexicon or rather derives from general prag- matic inference, it is currently subject of debate. We present results from a probe recognition experiment, providing novel evidence in support of early metonymy processing, consistent with the hypothesis that covert events are retrieved from knowl- edge of typical events activated by lexical items. Keywords: Logical metonymy; generalized event knowledge; qualia structure; covert events; probe recognition. Logical metonymy: lexicon, world knowledge, typical events In logical metonymies, an event-subcategorizing verb is com- bined with an entity-denoting patient. Contrast the German non-metonymic “long variant” (Ex. 1) with the metonymic “short variant” (Ex. 2), which is understood to have the same meaning: 1. Peter begann das Bier zu trinken. (“long variant”) Peter began the beer to drink. Peter began drinking the beer. 2. Peter begann das Bier. (“short variant”) Peter began the beer. → drinking the beer Covert events from lexical knowledge In (2), the event-subcategorizing verb beginnen (begin) com- bines with an entity, das Bier (the beer), but the clash is re- solved and the interpretation constructed by the recovery of a covert event which mediates between matrix verb and ob- ject (e.g. begin → drinking the beer). 1 The reconstruction of the covert event has well-known behavioral correlates – pro- cessing metonymic sentences is more costly than processing non-metonymic ones (Pylkkanen & McElree, 2006; Baggio, Chroma, Lambalgen, & Hagoort, 2010). On the level of theory, logical metonymies pose a challenge to compositionality (Partee, ter Meulen, & Wall, 1993; Bag- gio, Lambalgen, & Hagoort, 2012) and therefore touch on a foundational principle to language research. One of the points 1 This presumed process, namely the coercion of the object into an event associated with it, explains the use of the term “metonymy”. of debate is where covert events are retrieved from. The two main accounts of logical metonymy have suggested that covert events are part of our lexical knowledge (“lexical hypothesis”) or that they are retrieved through post-lexical inferences trig- gered by our general world knowledge and communication principles (“pragmatic hypothesis”). We present experimental evidence in support of a third hy- pothesis, namely that covert events are retrieved from knowl- edge of typical events stored in our long-term memory. A previous self-paced reading study (Zarcone & Pado, 2011) had presented evidence for generalized event knowledge inte- gration in logical metonymy, but its results were relying on a strong methodological assumption, namely that the same cog- nitive processes are used to interpret “long variants” (Ex. 1) and “short variants” (Ex. 2). The present study uses a probe recognition paradigm in order to avoid this assumption. Addi- tionally, the new paradigm allows us compare reaction times for low and high inter-stimulus intervals to assess the time course of metonymy interpretation in more detail. We find ev- idence for early, expectation-driven processing for metonymy as opposed to later, strategy expectancy generation, which points towards a central role of generalized event knowledge in logical metonymy interpretation. Pustejovsky (1995) and Jackendoff (1997) provide an account of logical metonymy which we call the “lexical hypothesis”: Logical metonymy is a type mismatch between the (semantic) subcategorization of a metonymic verb for an event and the en- tity denoted by the object on the other side, which requires the integration of an event to be resolved. The event is retrieved from complex lexical entries (qualia structures) associated with the object in the mental lexicon. In particular, the “agen- tive quale” (the event that brings about the object) and the “telic quale” (the main purpose of the object) are the relevant components of the qualia structure which can be retrieved as covert events in metonymic contexts. Being part of the mental lexicon, qualia model linguistic knowledge – in opposition to world knowledge and pragmatic inferences. Psycholinguistic work has identified experimental correlates for the lexical hypothesis (see Pylkkanen and McEl-