I’m broadly interested in instances of neutralization: when two phonemes are pronounced the same, e.g. in certain environments, /t/ and /d/ are neutralized to flaps in most varieties of American English. Some of the time, learners can tell which phoneme the flap derived from based on morphophonemic alternations: waiter, pronounced with a flap, is derived from wait and therefore the flap is obviously an allophone of /t/. Similarly, wader is pronounced with a flap (and exactly the same as waiter), but is morphologically derived from wade: learners can know that the flap is an allophone of /d/, not /t/. However, there are also instances of surface flaps which appear in words that do not alternate and therefore lack evidence for being allophones of /t/ or /d/. For example, spider, better, water, and Adam all contain ambiguous flaps: presumably, these could be either /t/ or /d/ underlyingly, and the learner has no access to alternations on which to base the decision.
My senior thesis looked at this issue in depth using Canadian Raising from elicited production data. Later, I expanded the project to also consider vowel durations and corpus data. Currently, I’m developing further experiments of different types including an artificial grammar learning task and a priming experiment.