The Neural Representation of Polysemy: The Case of Dot-objects

Abstract and concrete concepts are generally considered fundamentally distinct categories. However, many concepts have both concrete and abstract senses, for instance "book" can refer to both a physical object (as in "a torn book") and the abstract content (as in "an interesting book"). How is the meaning of such concepts represented in the brain? In this thesis, we address this question in the light of Pustejovsky’s dot-object hypothesis (Pustejovsky, 1995, 2011). According to the hypothesis, words such as "book" and "lunch" are dot-objects: a class of logically polysemous words which have multiple senses that are closely bound together. As a result of the close binding, both senses can be accessed simultaneously (as in "read the book"), however sometimes only a single aspect is emphasised by the context (e.g. "the book burned", "he summarised the book"). We argue that the complex meanings of the dot-objects are represented and manipulated in semantic hubs in the brain, where all aspects of conceptual knowledge converge and are represented in a modality-independent manner (thus accommodating diverse aspects of knowledge such as that cakes are made of pastry and are related to both birthdays and diabetes). We present three experiments investigating the neural representation of three dot-object categories with clear concrete and abstract senses: print matter such as "book" (OBJECT • INFORMATION), meal concepts such as "lunch" (FOOD • EVENT), and institution such as "church" (BUILDING • ORGANISATION). In all the experiments, participants read the dot-objects in a minimal context which elicited either the concrete or the abstract interpretation (e.g. "open the book" / "consult the book", "cook the lunch" / "organise the lunch"). We found the neural distinction between the concrete and abstract interpretations of the dot-objects differed from the concrete-abstract distinction observed for mono-sense nouns; instead the differential effect was most evident in the anterior temporal lobe (ATL), an area argue to be the semantic hub. The result suggests that 1) the distinct senses of a dot-object are associated with a single, unspecified structure in the mental lexicon, thus aligning with the dot-object theory; 2) when in context, the semantic representation is specified by instantiation to a particular sense. In addition, we also observed variations between the book-like and the lunch-like dot-objects, suggesting a graded representation mechanism within the ATL. Finally the third experiment showed that the MEG gamma-band frequency power could distinguish the neural correlates of the concrete and abstract interpretations; notably the divergence occurred 400ms and later post-stimulus onset. Given the established role in the literature of the gamma-band in integration processes, we conclude that the meaning instantiation only occurred at the later integration stage.