Some notes on variable phonological processes

Table of contents

• Some notes on variable phonological processes
  • Factors that affect whether a phonological takes place or not?
  • Choices in modeling variation/the loci of variation
    • Locus of uncertainty
    • Variable rules vs. variable inputs: "Generative phonology" and "usage-based" models

Resources:

• Course notes on variation in phonology: Hayes & Zuraw Fall 2013
• Course notes on production planning: Zuraw Fall 2017
  • Idea: conditioning of phonological processes on syntax/prosody is actually epiphenomenal, it's all about production planning
• (And of course, we have local experts on variation in phonology!)
• Some more junior-ish scholars who work on variation
  • Claire Moore-Cantwell
  • Jennifer Nycz
  • Stephanie Shih
  • Brian Smith

Factors that affect whether a phonological takes place or not?

• "Free" (token) variation?

  • "Non-phonological" stuff: sociolinguistic variables like gender, class, socioindexical (identity-related)...
    

  • In sociolinguistics, modeled with (multionomial) logistic regression (like having a MaxEnt grammar for each rule to determine probability of application)

  • Classic Labov example from variation course notes, and example of role of gender

    

    

• Lexical (type) variation

  • Variation is across items, not within items

  • Example from Kie Zuraw, variation course notes

    

• Prosodically-conditioned variation

  • We saw an example: recall kipper example in Korean

  • A classic example from Nespor and Vogel's classic work Prosodic Phonology: Raddoppiamento Sintattico

    

    • Claim: the phonological phrase is the domain of application of this process (central and southern varieties of Italian)

    

    • And where does the phonological phrase determination come from? The syntax.

    - Here's an example of the syntax-prosody mapping:

Choices in modeling variation/the loci of variation

• Where is uncertainty in the grammar, and how to different factors enter the grammar (or not)?

Locus of uncertainty

• Constraint ranking (work by Antilla, stochastic OT, noisy harmonic grammar)
  • Weights of constraints selected from (independent) normal distributions at evaluation time
  • Stochastic OT: define probability distribution over constraint rankings
  • Noisy harmonic grammar: define probability distribution over constraint weights
    
• Uncertainty in output (MaxEnt)
  
  • Each output candidate is associated with a probability that is proportional to the exponential of its harmony
  • There is a single weighting of constraints
• Review: Gaja's handbook article on learning with violable constraints, especially Section 3.2

Variable rules vs. variable inputs: "Generative phonology" and "usage-based" models

From Jennifer Nycz paper, slides

• Three potential loci of variation

  • Choice of directly indexical variant (representational)
  • Variable evaluation (grammar, e.g., optional rules, variable ranking/weights)
    • (How does MaxEnt fit in here?)
  • Choice of realization based on other contextual effects (representational)
    

• See slides...