Functional pressures and linguistic typology
The explanation of linguistic variation and change is one of the central questions in the language sciences. Functional explanations focus on how the needs and abilities of language users shape the distribution of linguistic structures that typically conventionalize -- e.g. structures that are harder to perceive or learn accurately are less likely to conventionalize accurately. Perceptibility effects are common sound patterns that seem closely related to the relative confusability of different speech sound sequences. One class of explanations -- purely phonological accounts --have assumed speakers (implicitly) know how confusability varies as a function of immediately adjacent sounds, and that this is a rich enough description of confusability to explain perceptibility effects. Chapter 2 shows that the perceptibility of tokens of any given sound in American English systematically varies based on a listener's incrementally-adjusted expectations about what the speaker intends to say, and shows that this variation is significantly greater than variation due to immediately adjacent sounds. To derive this result, I present a computational psycholinguistic model of word recognition and apply it to experimental confusability data and a transcribed lexicon of 10^4 words. I conclude that purely phonological accounts of perceptibility effects need to be more complicated and less modular than currently appreciated. Chapter 3 applies the same word recognition model and novel information-theoretic measures of confusability to two conversational corpora and shows that words that are more contextually confusable are lengthened in contexts where they are more confusable, and shortened where they are less so. This is a crucial step towards a linking hypothesis between the realtime perceptibility of different speech sound sequences and conventionalized perceptibility effects. Chapter 4 considers morphology. Prior research has observed an inverse relation between morphological complexity and demographic variables like speech community size and proportion of adult learners. Recent work has hypothesized that higher complexity may be helpful to child learners, and that populations with differing demographics constitute environments with different 'selection pressures' for language variants to 'evolve' in. I argue that mathematical formulations of Darwinian evolution suggest a more likely explanation: 'neutral' change caused by random fluctuations in variant frequency ('drift') is much more powerful in small populations and can easily overwhelm selection relative to large populations.