Learning a Frequency-Matching Grammar together with Lexical Idiosyncrasy: MaxEnt versus Hierarchical Regression

Abstract

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.

Cophonologies by Ph(r)ase

“…In this weighted constraint version of CBP, the R specification consists of constraint weight adjustments, rather than subrankings. Constraint resolution reduces to summing the weights of the morpheme-specific R specifications within a phase 8 As Zymet (2018Zymet ( , 2019 shows, lexically-specific weights, or multiple weights for a single constraint in the same language, may be best modeled in a hierarchical regression model, rather than the single-level logistic regression model of traditional MaxEnt. We expect that future work might show that a hierarchical regression model would be a better fit for the data presented here than the output of the MaxEnt Grammar Tool.…”

Identity Avoidance in Turkish Partial Reduplication: Feature Specificity and Locality

“…Methodologically we demonstrate that the precise nature of the identity avoidance effect can be revealed using hierarchical regression and statistical model comparisons (Graff and Jaeger, 2009;Zymet, 2019).…”

Exceptionality and derived environment effects: a comparison of Korean and Turkish

“…This is a subject of ongoing research (e.g. Moore-Cantwell & Pater 2016, Zymet 2018), and is beyond the scope of the current paper.…”

“…Two further conclusions can be drawn from an examination of deletion rates in these contexts. Firstly, although deletion is by far the most common pattern, each of the elicited suffixes seems to have its own rate of deletion, in what Zymet (2018) terms ‘lexical propensities’: 78% for the predicative, 91% for the accusative and 93% for the possessive. Secondly, a given root does not always behave in the same way across morphological contexts.…”