How is the social meaning of linguistic variation stored in memory?

This thesis uses experimental methods to investigate the storage of the social meaning of accent variation in memory, focussing on two main aspects: 1. The connection between linguistic variation and its social meaning in memory storage 2. The size and detail of the linguistic representations that are used to access social meaning The first aspect is explored by testing whether social and linguistic information are joined together within the same representations in memory or not. This would be the case if the episodic memory traces that exemplar theory postulates incorporate social meaning (Docherty & Foulkes 2014). Alternatively third-wave variationist research posits that there is a fluid link between variation and its social meaning (Eckert 2008), implying separate storage in memory. Innovations to exemplar theory allow for both exemplars and more abstract generalisations across exemplars to be used in speech processing. This introduces new types of representations to its theory of speech processing, which may store social and linguistic information separately, although this has not been explored extensively yet. In this PhD I present three highly automatic ‘socio-contextual priming’ experiments which suggest that it is possible for listeners to process linguistic variation without its social meaning becoming activated as well, suggesting that the two types of information are stored separately. The second aspect is explored by means of two accent recognition experiments, which tested whether listeners recognise accents more accurately depending on whether the stimulus word is frequent, not frequent or a non-word. These experiments found that listeners are able to recognise accents even in non-words, implying that lexical information is not required for the retrieval of social (accent) information, and that sub-lexical information is used to access social meaning. Accuracy was higher in the experiment which used real words, but there was no difference between accuracy in high-frequency and low-frequency words within the experiment. This suggests that the number or strength of lexical representations that are used to access social meaning does not facilitate access to social meaning, but that the presence of a lexical phonological form is helpful for the contextualisation of sub-lexical information used to access social meaning. Furthermore, the accent recognition experiments found that listeners perceived fewer voices in the experiments than were used in reality, suggesting that the representations listeners used for recognition were only detailed enough for them to distinguish between more general character types rather than individual speakers.