Current typologies of voice assimilation between obstruents include languages that exhibit either assimilation to voicelessness (a type of emergence of the unmarked effect) or onset-controlled patterns, where the value controlling the change is in the onset obstruent. In either case, this type of local assimilation is considered to result in (contextually) unmarked structures. This article presents data that highlights a previously unrecognised pattern: assimilation resulting in voicing (an ‘emergence of the marked’ effect). This pattern has implications for how markedness is expressed in grammar. It is argued here that voicing is a privative feature, and that faithfulness constraints regulating the feature [voice] yield a rich typology that includes emergence of both marked and unmarked patterns. In addition, this typology yields benefits that are lost if voicing is considered a binary feature. This is illustrated by extending the dynamics of this voicing typology to other laryngeal features, such as [spread glottis].

According to many works on English phonology, word-final alveolar consonants – and only alveolar consonants – assimilate to following word-initial consonants, e.g. ran quickly → ra[ŋ] quickly. Some phonologists explain the readiness of alveolar consonants to assimilate (vs. the resistance of velar and labial articulations) by proposing that they have underspecified place of articulation (e.g. Avery & Rice 1989). Labial or dorsal nasals do not undergo assimilation because their place nodes are specified. There are reports that velar and labial consonants sometimes assimilate in English, but these are anecdotal observations, with no available audio and no statistics on their occurrence. We find evidence of assimilation of labial and velar nasals in the Audio British National Corpus, motivating a new, quantitative phonological framework: a statistical model of underspecification and variation which captures typical as well as less common but systematic patterns seen in non-coronal assimilation.

In this paper, I provide a unified account of three frequency effects in phonology. First, typologically marked elements are underrepresented. Second, phonological changes are underrepresented. Third, morphologically conditioned phonological changes are overrepresented. These effects are demonstrated with corpus data from English and Welsh. I show how all three effects follow from a simple conception of phonological complexity. Further, I demonstrate how this notion of complexity makes predictions about other phenomena in these languages, and that these predictions are borne out. I model this with traditional Optimality Theory, but the proposal is consistent with any constraint-based formalism that weights constraints in some way.

OT error-driven learning admits guarantees of efficiency, stochastic tolerance and noise robustness which hold independently of any substantive assumptions on the constraints. This paper shows that the HG learner used in the current literature does not admit such constraint-independent guarantees. The HG theory of error-driven learning thus needs to be substantially restricted to specific constraint sets.