Constraint-based Phonology

Will Styler - LIGN 111

Today’s Plan

• Conspiracies

• Optimality Theory

• Cats, Dogs, and Dishes

• Strengths and Weaknesses of OT

So far, we’ve been using SPE-style Generative Phonology

• Underlying forms are transformed into surface forms by a series of ordered rules.

• Variation is either conditioned (changes by rules) or speakers screwing up

• We’ve expanded it a bit with a nice feature theory

• … and I’ve added dashes of laboratory phonology

We define rules…

• Voicing Assimilation: [-son] -> [-voice] / [-voice]__#

• Vowel Epenthesis: ∅ -> [-front,-back] / [+coronal,+strident]__[+coronal,+strident]#

Then we apply them…

This approach has a lot of benefits

• It’s relatively simple

  • You find a UR, you make a rule, and you get a surface form

• It’s been around for 50 years

  • It’s been well studied and expanded upon
  • (Nearly) everybody still knows it and can use it
  • It’s the Phonological Lingua Franca

• Folks in other fields are likely still using SPE-style phonology

• Other theories are often based on its shortcomings

  • Even the haters have to respect

• It’s quick to deploy

  • Just write down a couple of rules

It’s the best way to start out in Phonology

• There’s not so much ‘technology’ to get to your first analysis

• It often feels intuitive

  • “Huh, that sound changes there”

• It exposes the elements of phonology that are most relevant for fieldwork/clinical work

• Nearly everybody started off learning this way!

… but it’s not the only approach

We’ve been seeing a lot of coda-hatred

• Some languages epenthesize vowels to create more syllables. (Yokuts!)

• Many languages try very hard to remove codas using a variety of deletion rules. (Russian)

• Some languages transform codas into other sounds.

• Some languages force codas towards sonorance (see Dark vs. Light /l/, nasal-coda-only languages)

• … but all these processes are doing the same damned thing

It’s not just codas…

• By and large, languages don’t like hiatus (two vowels next to one-another)

• Most languages don’t care for huge consonant clusters

• We don’t like violating the sonority hierarchy

• Two adjacent stresses stress us out

Over and over, a language will use multiple different rules to accomplish the same damned thing

“Why don’t we just talk about the things we’re trying to avoid, instead of creating crazy conspiracies of rules to deal with them?!”

Optimality Theory (OT)

Optimality Theory

A theory of phonology which relies on ranked constraints to solve phonological problems

Optimality Theory

• Developed by Prince and Smolensky in 1993. “Optimality Theory: Constraint Ranking in Generative Grammar”

• “Don’t talk about the individual rules, just talk about the preferences!”

We don’t need a conspiracy of rules to solve problems

• Don’t worry about V Epenthesis, C Deletion and Forced Sonority

• Just say “We don’t like codas. Don’t allow those.”

• “Let’s find the form that’s least worst”

Three basic steps

• GEN - GENerate a list of possible forms from the underlying form

• CON - Rank the CONstraints which help you decide on the best ones

• EVAL - EVALuate the possible forms, and choose the optimal candidate

Let’s indulge in a fantasy

• Let’s pretend I could afford to buy a house in La Jolla!

We use OT when buying a house

GEN in home-buying

• “Generate a list of possible candidate houses”

• Go online, or get a realtor

• Find all the houses for sale in the area

• Gather information about their characteristics

CON in home-buying

• “Let’s find the constraints on our purchasing and rank them”

• Everybody has the same set of possible things they could care about

• You decide what you want

• … what you don’t want

• … and what’s a deal-breaker

• Then rank them according to your desires

Possible Constraints

• “We need two bedrooms”

• “Houses with turrets on the side look cool!”

• “We want to have a public transit commute”

• “We don’t want a lawn”

• “We don’t want to pay more than $350,000”

  • 🤣😭

• “We don’t want an area with bad schools”

* in OT means “Don’t allow this”

• *CC would mean “No consonant clusters”

• *VV would mean “No Hiatus”

• *$350K+ would mean “No houses over $350k”

So, we have constraints

TwoBedrooms

HasTurret

NoCommute

NoLawn

*$350k+

GoodSchools

For us…

• *$350k+

• TwoBedroom

• NoCommute

• NoLawn

• HasTurret

• GoodSchools

For somebody else…

TwoBedroom

GoodSchools

*$350k+

NoLawn

HasTurret

NoCommute

EVAL for Real Estate

• Look at all the listings, and figure out which one will be best for us!

We have a house!

• … at least, in this fantasy

So, Optimality theory lets us pick the least-worst candidates from a massive set of possibilities

So, uh, Phonology?

GEN

• Generate EVERY POSSIBLE SURFACE FORM from the URs given

CON

• Every language shares the same set of constraints!

  • … but rank them differently

• There are markedness constraints

  • “This thing sucks, don’t do that”

• … and faithfulness constraints

  • “At least try to be similar to the UR

EVAL

• Count the number of times each form violates the constraints

• Violating higher-ranked constraints is more important

• When you violate a constraint which is higher ranked than another candidate’s highest violation, it’s a fatal violation.

• The ‘winner’ is the candidate that has the lowest-ranked first violation

(There’s more to it, but that’s the basics!)

How does this work in practice?

Back to Cats, Dogs, and Dishes

/kæt+z/

GEN - Let’s create a list of every possible surface form

• EVERY ONE OF THEM

CON - Generate and rank the constraints

• *SS: “Don’t put two strident sounds next to each other”

• Agree: “Voicing should agree in adjacent consonants”

• Max: “The segments in the UR should be there on the surface”

• Dep: “Don’t Epenthesize segments that weren’t there”

• Ident: “Don’t change segments in the underlying form”

• Ranked in that order

EVAL - Evaluate the Candidates

EVAL - Evaluate the Candidates

Dogs

GEN - Come up with every possible form

• … again

CON - Generate and rank the constraints

• You’ll use the same constraints and rankings in all forms in the problem

• In fact, you’ll use the same rankings for every form in the language!

EVAL - Evaluate the Candidates

Dishes

GEN and CON work the same

• New surface candidates, same constraints

EVAL

OT’s Strengths and Weaknesses

Optimality Theory Strengths

• Magnificent for handling “conspiracies” of rules in languages

• Amazing for phonotactics

• Some types of analysis get very simple, very quickly

• It handles cross-linguistic patterns, and “rules” that show up over and over again

• Sometimes, it’s just more graceful to do it with OT

Optimality Theory Weaknesses

• GEN makes us examine EVERY POSSIBLE CHOICE

  • “Penguin” is a valid candidate for /kæt+z/

  • This may be cognitively nutty

• Constraints (but not their rankings) are generally considered innate

  • Acquisition gets dicey without innate constraints

• OT is rekt by counter-bleeding phenomena

• Out of the box, it’s just as bad at handling variation as SPE-style phonology

  • … but there are some interesting hacks

Optimality Theory has been expanded and improved

• ‘Harmonic Grammar’ is a broader approach

• Quantitative and computational approaches to OT

• Doing ‘Old School OT’ like I’ve taught you is SO 1993

… but constraint-based approaches are everywhere, and with good reason!

Wrapping up

• (Some) Conspiracies are real!

• Optimality Theory is amazing at handling them

• Cats, Dogs, and Dishes can be handled with constraints too

• Constraint-based phonology is a powerful tool!

For Next Time

We’ll work with an OT analysis!

Thanks!