Predicting plurality: an examination of the effects of morphological predictability on the learning and realization of bound morphemes.
Darcy Elizabeth Rose · UC Research Repository (University of Canterbury) · 2017
This thesis examines the learning and production of bound morphemes and how this linguistic behavior is influenced by the contextual predictability of the message those morphemes signal. Using the grammatical category of plurality (e.g. cup ~ cups) as a case study, and treating language as a system of message transmission, it demonstrates that the contextual predictability of a grammatical morpheme is correlated with variation in the learning and realization of that morpheme. Building on previous work examining the influence of contextual predictability on linguistic behavior at other levels of linguistic representation, this thesis suggests that a language user’s knowledge of morphemes includes some representation of morphological predictability. This informs the larger question of what constitutes a language user’s knowledge of language, and how linguistic behavior varies as a function of that knowledge. The influence of the contextual predictability of bound morphemes and the messages they signal is evaluated via three studies. These studies use the Rescorla-Wagner model and Message-Oriented Phonology, two frameworks which quantify the amount of information carried by a linguistic unit, to examine how the learning and production of bound morphemes is influenced by two biases that shape communication systems. These communicative biases are: a pressure to accurately transmit messages and a pressure to minimize resource costs. Study 1 explores the effects of morphological predictability on the learning of plural morphemes in an online artificial language learning experiment. Given multiple cues to the morphological category of plurality, this study shows that the second cue is learned less well when the message of plurality is more predictable, given first cue. Study 2 investigates gradient realizations of plural marking in a spoken corpus of New Zealand English. Using a measure of contextual predictability based on how often the preceding word occurs before a plural noun, this study shows that plural morphemes with higher predictability in context tend to have more reduced realizations. After demonstrating in Study 2 that contextual predictability plays a role in morphological reduction, Study 3 uses an online rating task to explore how large the relevant context is over which morphological predictability is calculated, and whether this predictability is accessible through subjective ratings. Using extracted contexts of one or five words from the corpus used in Study 2, judgments about the likelihood of a plural occurring in the given context were solicited from native speakers of English. While moderately correlated with the corpus measure of predictability used in Study 2, neither of the subjective measures of plural predictability is found to be predictive of plural duration. This finding suggests that while the contextual predictability of morphemes does influence production, either language users may not be able to access fine-grained morphological predictability in an overt task, or this task may not have been able to capture such fine-grained intuitions. Further work is required to determine whether an alternative experimental design might elicit subjective ratings which are predictive of plural durations, enabling exploration of the size of the relevant context. The above studies demonstrate that at the level of the morpheme, in both learning and production, language users are sensitive to the pressures present in any system of communication, and suggest that communicative biases shape human language at the level of the morpheme. These findings invite further research into the interaction of influences of contextual predictability from multiple levels of linguistic structure, as well as exploration of how morphological systems are shaped over time, and how, cross-linguistically, the biases of accurate message transmission and conserving resource cost are balanced to create effective morphological systems.