This dissertation is about ellipsis, a natural language construction in which a word or phrase is understood even though it is not pronounced. This project brings together two strands of research: formal theoretical research on the relationship between the interpretation of an ellipsis site and its surrounding linguistic context, and psycholinguistics research on the processes by which we build comprehensive meaning from the silence of an ellipsis input. I argue that only by considering these two research programs together can we fully understand ellipsis as a natural language phenomenon. I first present novel English sluicing data that challenge even the most successful existing theories of the relationship between antecedent and elided content in ellipsis constructions by showing that the elided content and antecedent content in a sluicing construction can mismatch to a greater degree than previously thought possible. I use this data to argue that an interpretation condition for ellipsis must be sensitive to pragmatic content, and motivate a proposal in which sluicing is treated as a pragmatics-sensitive phenomenon licensed by local contextual entailment. I then present the results of eight experiments on ellipsis comprehension that use offline comprehension measures and online measures of incremental processing. I show that anaphoric and cataphoric ellipsis comprehension is subject to a proximity bias, wherein comprehenders prefer to resolve an ellipsis site to the candidate antecedent in greatest proximity to the ellipsis site, regardless of whether the site precedes or follows the antecedent. I also show that cataphoric ellipsis comprehension is subject to an active, forward search strategy. I argue that ellipsis processing ultimately shares features both with anaphoric processing and with the processing of long-distance dependencies, such as filler-gap processes.

This project looks at loanword adaptation in Finnish, focusing on the role of perception in the adaptation of voiceless stops as geminates. In the Finnish loanword process, non-geminate donor voiceless stops are often adapted in geminated form. This particular adaptation strategy is interesting because it serves a purpose other than to correct a phonological ill-formedness in the borrower language, meaning that it cannot be accounted for by appealing to native Finnish structural constraints. This project will argue that, while there is evidence that word-structure constraints play a role in the adaptation process, gemination is also influenced by perceptual cues. Specifically, I will show using on-line adaptation data that geminates are adapted as such because of two perceptual factors: stress placement in the donor and loan word and the existence of subsequent stops in the donor word.