This dissertation investigates imitation in speech, which is the general tendency shown by a speaker to become more similar to another speaker in the way they speak. Many of us have experienced this while talking to someone who is speaking the same language but with a different accent. Conversing with such a person can affect some characteristics of our speech, so that we come to sound more like them. Imitation in speech has been very extensively studied, especially over recent years. To contribute to this line of research we provide an account of imitation in speech at the allophonic level, that is at the level of the possible phonetic realisations of a phoneme. We are interested in whether imitation of the sound of a given phoneme in a particular word position can influence the other possible realisations of that phoneme in the same word position. We are also interested in determining whether imitation of a speech sound in a particular word position for a given phoneme can affect the realisations of that phoneme in a different word position.

New Zealand English provides of wealth of allophonic variation across word positions for the phoneme /t/. Therefore it is an ideal language to investigate imitation and allophony. Before presenting our experimental designs and our results on imitation however, we verified and further extended the work that has been conducted on the dialect. We analysed large corpora of spoken NZE and found new allophones of /t/. We discuss a fricative realisation in particular.

The fricative realisation in NZE was further examined by means of a palatographic pilot experiment, as well as a perception experiment. Building on our findings from the perception experiment, we investigated imitation in speech towards an artificially created novel fricative allophone in medial position using acoustic and EPG data. For some speakers, the mere exposure to the novel allophone affected the realisation of other allophones in the same word position.

A series of acoustic experiments were then conducted to examine allophonic imitation across word positions. We found that repeated exposure to a given allophone can drive allophonic selection across word positions. We also found that positional transfer can occur, such that exposure to an acoustically manipulated allophone can affect the same allophones in a different word position. Interestingly it can also affect other allophones in a different word position.

Our results are discussed according to hybrid exemplar models and the Direct Realist view. We discuss which theoretical framework best accounts for the results we obtained.