This thesis explores the production of loanwords. When a word is borrowed from one language to another language, the word may contain a sound or structure that is not permitted in the grammar of the borrowing language. The non-native structure may be adapted to make it well-formed in the borrowing language, or it may be imported without modification. For example, a word borrowed from te reo Māori to New Zealand English (NZE) may contain a non-native rhotic sound [ɾ], and the non-native sound is sometimes adapted as a native rhotic sound (e.g. ko[ɹ]u and ma[ɹ]ae), and sometimes imported (e.g. ko[ɾ]u and ma[ɾ]ae). By exploring the variation in /r/ realizations in Māori loanwords, this thesis addresses research questions regarding (a) the likelihood of choosing a variant and (b) the degree of phonetic redundancy in the pronunciation.

The first part of this thesis (Chapters 3-5) explores the effect of sociolinguistic factors on the likelihood of choosing adapted structure [ɹ] vs. imported structure [ɾ], by running three experiments. It is demonstrated that the choice is affected by some sociolinguistic factors, i.e., imported structure is more likely to be produced in speech associated with Māori, by speakers associated with Māori, and in loanwords associated with Māori. Addressing these questions increases our understanding of the relationship between loanword phonology and sociolinguistic factors, which has been less studied in previous literature. These findings suggest that the variation in loanword phonology can be socially meaningful, because variation in loanword phonology is a type of inter-speaker and intra-speaker variation (Bell 2014). Namely, adapted structure and imported structure carry different social messages. In particular, it is argued that imported structure has a social message associated with Māori.

On the basis of the argument for the social messages, the second part of this thesis (Chapters 6-7) explores the relationship between the phonetic redundancy in a variant in loanword phonology (i.e., adapted structure and imported structure) and the social message predictability, by analysing the data collected from the three experiments again. It is shown that imported structure [ɾ] is produced with less acoustic redundancy (i.e., shorter formant cessation) when it is more predictable from context; adapted structure [ɹ] is produced with more acoustic redundancy (i.e., a lower F3 value) when it is more predictable from context. This increases our understanding of the relationship between the realization of a linguistic unit and predictability in general. More specifically, this indicates that the phonetic redundancy in a signal is not only influenced by lexical message predictability as has been demonstrated in previous literature, but it is also affected by social message predictability.

At a theoretical level, investigating the two research questions provides insights to the cognitive process of loanword phonology. More specifically, the research on the likelihood of choosing a variant in loanword phonology develops our understanding of how adapted structure and imported structure are represented in the mind of a borrower. The result could be captured by positing that exemplars with adapted structure and imported structure are represented, the exemplar space is developed through the usage of loanwords, and the representation of imported structure is stored in relation to a social category “Māori” (Pierrehumbert 2001; Foulkes and Docherty 2006). The research on redundancy in a variant increases our understanding of how adapted structure and imported structure are processed in production. The result could be encapsulated by positing that the phonetic realization of imported structure [ɾ] is highly influenced by articulatory biases which balance accuracy of message transmission and resource cost (Hall et al. 2016; 2018), and the realization of adapted structure [ɹ] is highly influenced by averaging exemplars with adapted structure and those with imported structure. The reason why the two structures are influenced by different mechanisms is due to the saliency of the social messages. It is speculated that the realization of a linguistic variant is more likely to be affected by message-oriented production biases, when the social message is salient; otherwise, it is more likely to be affected by exemplar-averaging processes.