How exactly do learners generalise in the face of ambiguous data? While there has been a substantial amount of research studying the biases that learners employ, there has been very little work on what sorts of biases are employed in the face of data that is ambiguous between phonological generalisations with different degrees of complexity. In this article, we present the results from three artificial language learning experiments that suggest that, at least for phonotactic sequence patterns, learners are able to keep track of multiple generalisations related to the same segmental co-occurrences; however, the generalisations they learn are only the simplest ones consistent with the data.

We present an empirical challenge to Jardine's (2016) assertion that only tonal spreading patterns can be unbounded circumambient, meaning that the determination of a phonological value may depend on information that is an unbounded distance away on both sides. We focus on a demonstration that the ATR harmony pattern found in Tutrugbu is unbounded circumambient, and we also cite several other segmental spreading processes with the same general character. We discuss implications for the complexity of phonology and for the relationship between the explanation of typology and the evaluation of phonological theories.

This paper employs a computational framework to demonstrate that two competing feature-geometric models of tonal representation are notationally equivalent. A model-theoretic analysis of these structures using a low-complexity logic yields two main results. First, the current study demonstrates that the models do not differ in their empirical coverage of assimilatory tone-sandhi processes in Chinese dialects, contrary to previous claims. Second, the models are shown to be bi-interpretable (using a more restrictive definition of bi-interpretability than earlier studies), thus providing a formally rigorous demonstration that the differences between the structures of the models are superficial, rather than substantive. The computational characterisation pursued here is well suited to questions of notational equivalence, because it allows for a principled comparison of the empirical coverage and structural content of two models using a single formalism.

Based on primary data from Tundra Nenets, this paper explores phonological patterns which seem to require restrictions on the input, and thus present a particular challenge to Optimality Theory. In these patterns, a contrastive segment appears only in the environments where it is also derived by active alternations in the language. I illustrate this with the behaviour of Tundra Nenets /k/, and argue that these patterns can be analysed as distributional generalisations that hold only at early derivational levels. A Stratal OT analysis is proposed. Tundra Nenets also presents a pattern which appears to involve unnatural classes, but is reanalysed with only natural class alternations in my account.