Experimental research has uncovered language learners’ ability to frequency-match to statistical generalizations across the lexicon, while also acquiring the idiosyncratic behavior of individual attested words. How can we model the learning of a frequency-matching grammar together with lexical idiosyncrasy? A recent approach based in the single-level regression model Maximum Entropy Harmonic Grammar makes use of general constraints that putatively capture statistical generalizations across the lexicon, as well as lexical constraints governing the behavior of individual words. I argue on the basis of learning simulations that the approach fails to learn statistical generalizations across the lexicon, running into what I call the GRAMMAR-LEXICON BALANCING PROBLEM: lexical constraints are so powerful that the learner comes to acquire the behavior of each attested form using only these constraints, at which point the general constraint is rendered superfluous and ineffective. I argue that MaxEnt be replaced with the HIERARCHICAL REGRESSION MODEL: multiple layers of regression structure, corresponding to different levels of a hierarchy of generalizations. Hierarchical regression is shown to surmount the grammar-lexicon balancing problem—learning a frequency-matching grammar together with lexical idiosyncrasy—by encoding general constraints as fixed effects and lexical constraints as a random effect. The model is applied to variable Slovenian palatalization, with promising results.

No abstract

A speech production experiment with electroglottography investigated how voicing is affected by consonants of differing degrees of constriction. Measures of glottal contact [closed quotient (CQ)] and strength of voicing [strength of excitation (SoE)] were used in conditional inference tree analyses. Broadly, the results show that as the degree of constriction increases, both CQ and SoE values decrease, indicating breathier and weaker voicing. Similar changes in voicing quality are observed throughout the course of the production of a given segment. Implications of these results for a greater understanding of source-tract interactions and for the phonological notion of sonority are discussed.

This paper carries out a detailed investigation into new data from Maragoli displaying an interaction between reduplication and hiatus repair. The data give rise to paradoxical, opportunistic orderings of phonological processes: in one set of inputs, copying before repairing avoids a complex onset, while in another set, repairing before copying avoids an onsetless syllable and maximizes word-internal self-similarity. Based on attested words and nonce probe data elicited from a native speaker, I argue that a successful analysis of the interaction requires direct comparison between forms derived by opposite orders of phonological changes. The orderings receive a full analysis in Parallel Optimality Theory (Prince & Smolensky 1993/2004) but translate into constraint ranking paradoxes in Harmonic Serialism with Serial Template Satisfaction (McCarthy et al. 2012). The data thus constitute evidence for irreducible parallelism in the sense of McCarthy (2013).

To what extent can a learning bias be defied in language? This question bears directly on the theory of phonological learning, as it addresses the limits of learner capability.A growing family of findings suggests that learners tend to favor phonological constraints that are morphosyntactically general-that is, obeyed by at least several morphemes, or in multiple or all grammatical contexts. That phonological alternations are typically corroborated by the phonotactics of a given language was observed as early as Chomsky and Halle 1968 and Kenstowicz and Kisseberth 1977, but the generalizing tendency just mentioned has also been observed in a number of recent corpus studies. Martin (2007, 2011), Shih and Zuraw (2018, and Breiss and Hayes (2019) observe cases of grammatical "leaking," in which strong phonotactic restrictions tend to manifest across word boundaries or compound boundaries, or affect the choice between grammatical constructions. Chong (2017) found that certain alternations purported to be apparent derived-environment effects are just that-merely apparent. For example, though Korean t-palatalization triggered by high front vocoids was previously proposed to constitute a derived-environment effect because [ti]-sequences exist in some roots, Chong showed that such sequences are highly underattested in the Korean lexicon. Generalization effects have also been borne out in artificial-language-learning experiments: Myers and Padgett (2014) found that participants generalize a phrase-final devoicing pattern to the word-final domain without exposure to unambiguous evidence; Chong (2017) found that participants more readily learned a suffixal harmony alternation when they were exposed to higher rates of root harmony, corroborating proposals that phonotactic generalizations assist in acquiring alternations (e.g., Tesar and Prince 2003, Hayes 2004, Jarosz 2006.In light of these findings, Martin (2011) and Chong (2017) propose learning models whereby whenever the learner weights positively a structure-specific constraint (e.g., applying only across a suffix boundary), it gives positive weight to an analogous structure-insensitive constraint, leading to the generalizing tendency. If there were to exist an alternation that applies consistently in a constrained morphosyntactic context without even an analogous statistical tendency in phonotactics to accompany it, then that would complicate our understanding of learners' preference for morphosyntactically general pat-For invaluable input, thanks to