Phonetically Driven Phonology The Role of Optimality Theory and by rogerholland


									Phonetically Driven Phonology:
The Role of Optimality Theory
   and Inductive Grounding
      by Bruce Hayes (1996)

     Presented by Anne Rimrott
       LING 800 Phonology
         February 24, 2003
     What This Paper Is About
• Connecting phonetics and phonology
  – phonological constraints are often phonetic in
    character, but they are not phonetics itself
• Bridging gap between functional and formal
• Defining grounded constraints
• Presenting an inductive grounding algorithm
  which “generates” grounded constraints
• Presenting a Phonetically-Driven Optimality-
  Theoretic Approach to Phonology
• Deep divide between formalist and functionalist
  approaches in phonology
• Functionalist approach (traditionally)
   – Phonetic in character: sound patterns of languages
     arranged to facilitate ease of articulation and
     distinctness of contrasting forms in perception; “good
• Formalist approach (contemporary)
   – Phonology modeled as formal system set up to mirror
     characteristic phonological behavior of languages
• Optimality Theory: the solution?
    Optimality Theory - Outline
• Prince and Smolensky (1993)
• Formal theory
• The phonology selects an output form from the set
  of logical possibilities using constraint rankings
• Phonotactic (markedness) constraints: “The output
  should have phonological property X”
• Faithfulness constraints: “The output should
  resemble the input for property Y”
          OT – Advantages
General advantages of formal theories:
   falsifiability  improvement/replacement
  Increased pattern recognition capacity
Solutions to problems that were not
 treatable in earlier theories (e.g. ordering
 paradoxes involving phonology and
           OT – Advantages
Allows incorporation of general principles
 of markedness into language-specific
  previously: arbitrary-looking rules
  OT: principles of markedness form sole
    ingredient of the language-specific analysis
  o Rankings are the only arbitrary element in the
   Principled character of language-specific
    analysis greatly increased
          Principled Constraints
• OT thrives on principled constraints
• Definition: constraints that “do work” in many
  languages, and do it in different ways
• Justified on functional grounds: render speech
  easier to articulate or render contrasting forms
  easier to distinguish perceptually
• Functionally motivated phonetic constraints are
  inherently principled
 “Phonetically-Driven Optimality-theoretic
  Phonology” (bridge gap between phonetics and phonology)
       How to Bridge the Gap
• A research result in phonetics is not the
  same as a phonological constraint
• Question of how to incorporate the phonetic
  research that explains a phonological
  pattern into a formal theory of phonology
• How to attain phonetically-driven
  phonological description/phonetically-
  motivated constraints
Why Phonetic Patterns Cannot Serve
 As A Direct Basis For Phonology
1. Phonetics involves gradient and variable
   phenomena, whereas phonology is
   characteristically categorial and far less
   variable (ex.: postnasal obstruent voicing
   in English and Ecuadorian Quechua)
2. Phonetics is asymmetrical, whereas
   phonology is usually symmetrical (ex.:
   production of stops  next slides)
    Relating Phonetics and Phonology:
           Production of Stops
•    Stop voicing influenced        •   Reflected in
     by phonetic factors:               phonological patterns:
     a. Place of articulation           a. Gaps in stop inventories
        (front: voicing
                                           typically *[p] or *[g]
        easier, back:
        voicelessness easier)
     b. Closure duration                b. Obstruent + Voiced
        (voicelessness easier in           obstruent - clusters
        post-obstruent position)           avoided
     c. Postnasal position              c. Frequently no voiceless
        (voicing easier)                   stops after nasals
     d. Phrasal position
                                        d. Voicing favored in
        (voicing harder in
        utterance-initial and              medial position
        utterance-final position)
     Phonetics and Phonology
 Phonetics can serve as source of
  phonological explanation
• BUT: if we try to do the matching between
  phonetic mechanisms and phonological
  typology in a naïve, direct way we run into
  difficulties  next slides
  Problems With Matching Directly
Phonetic Difficulty Map:       • The greater the value the
                               more difficult to pronounce a
                  b d      g   voiced stop in the specified
[-son] ___        43 50 52     •Trading relationships: [g] not
                               always harder than [b]
# ___             23 27 35     •Contour line for phonetic
                               difficulty: 25
[+son, -nas] ___ 10 20 30
                               •Phonetically rational: ban all
                               configurations that exceed
[+nas] ___        0   0    0   contour line  next slide
  Phonetically Rational Vs. Formally
       Symmetric Constraints
• Phonetically rational but     • Phonetically sensible
   unattested:                     constraints which possess
• Based on phonetic map            formal symmetry (attested):
   from previous slide          * Voiced obstruent word-finally
• ban any voiced stop that         (Polish)
   requires more than 25        * Voiced obstruent after another
   units of effort, i.e.:          obstruent (Latin)
* post-obstruent voiced stops   * Voiced obstruent geminate
* [d,g] in initial position        (Japanese)
* [g] after oral sonorants      * Voiced velar obstruents
 Phonological constraints tend to ban phonetic
 difficulty in simple, formally symmetrical ways
the influence of phonetics in phonology is not
 direct, but is mediated by structural constraints
 that are under some pressure toward formal
gap between phonetic difficulty patterns and
      Background: Phonological
          Grammar Design
•    The language learner as a designer of
•    Compromise between formal symmetry
     and reflection of phonetic difficulty
•    Phonological grammar design under OT:
    1) Gain access to constraints by inventing them
        inductive grounding (the process by
       which constraints are invented)
    2) Forming a grammar by ranking the
        Inductive Grounding
• Assumptions:
  – Child has access to degree of phonetic
    difficulty (through own production/perception)
  – Child can generalize across tokens and thus
    create a phonetic map
  – This map is used to construct phonetically
    grounded constraints  inductive grounding
 Inductive Grounding Algorithm
• Assumptions:
  – Constraints are constructed profusely, as
    arbitrary well-formed combinations of the
    primitive elements of phonological theory
  – Candidate constraints are assessed for their
    degree of grounding by calculating constraint
      Grounded Constraints And
       Constraint Effectiveness
• Grounded constraints are phonetically sensible:
  ban things that are phonetically hard, allow things
  that are phonetically easy
• Constraint effectiveness:
Effectiveness = Correct predictions / (Correct pred. + errors)
• Perfect constraints (ban hard things): value 1
• useless constraints (arbitrary): 0.5
• utterly perverse constraints (ban easy things): 0
       Selecting the Grounded
• The inductive grounding algorithm should
  generate a very rich and principled
  constraint set
• The constraints should not correspond
  exactly to phonetic patterns but deviate to
  achieve structural simplicity (cp. above)
 Local maxima of constraint effectiveness
  instead of just one global maximum
Redefining “Grounded Constraint”
• Given a phonological constraint C and a
  phonetic map M, C is said to be grounded
  with respect to M if the phonetic
  effectiveness of C is greater than that of all
  neighbors of C of equal or lesser complexity
• Two constraints are neighbors if the structural description
  of one may be obtained from that of the other by a single
  substitution (e.g. * [+nas] [-voi] and * [+nas] [+voi])
• Constraint C1 is less complex than C2 iff the structural
  description of C1 is properly included in the structural
  description of C2 (e.g. *[LAB] is less complex than
  *[LAB, -voice])
How Does It All Work? An Example
• Phonetic difficulty map for six stops:
                     p      t      k      b      d     g

    [-son] __       a: 7   b: 0   c: 0   d: 43 e: 50 f: 52

    # __           g: 10   h: 0   i: 0   j: 23 k: 27 l: 35

    [+son, -nas]
                   m: 45 n: 28    o: 15 p: 10 q: 20 r: 30

    [+nas] __      s: 155 t: 135 u: 107 v: 0   w: 0   x: 0

  Derive phonological constraints by means of
 inductive grounding (e.g. * [+nas] [+voice] or * [LAB,
 -voice]  next slides)
     How To Compute Effectiveness
•   Pairwise comparisons between banned cells and unbanned
    cells of the phonetic map  banned cells should be more
•   E.g. * [LAB, -voice]:
•   Correct predictions: 50         •   Incorrect Predictions: 29
•   a > b, c, h, i, v-x             •   a < d-f, j-l, n-r, t, u
•   g > b, c, h, i, v-x             •   g < d-f, j-l, n, o, q, r, t, u
•   m > b, c, d, h-l, n-r, v-x      •   m < e, f, t, u
•   s > b-f, h-l, n-r, t-x          •   s < (none)

     Effectiveness Value: 50 / (50 + 29) = 0.633
     If this value exceeds the effectiveness value of all the
    neighbors of the constraint, the constraint is grounded
                Does It Work?
• The algorithm produced grounded constraints (table
  18) and constraints that are not grounded (table 19)
• All the constraints from table 18 are in fact attested in
  languages except for *[-nas][+voice]. But: nasal could
  be privative
• The constraints in table 19 are rare or unattested
 The inductive grounding algorithm indeed singles out
  phonologically-stated constraints that match typology
 Some of the grounded constraints do not have high
  effectiveness scores, but are well attested  formal
  symmetry and not just phonetic naturalness important
  in constraint creation
  The Remainder of the Task of
    Phonological Acquisition
• So far: inductive grounding algorithm 
  large number of grounded constraints
• Now: Tesar-Smolensky algorithm  rank
  the constraints
•  Together: construction of a large chunk
  of the phonology
• Still missing: constraints with no phonetic
  origin (e.g. faithfulness constraints)
Evaluation: Acquisition Evidence
• Is there any evidence that children can generate
  formally symmetrical constraints like the ones we
  just generated from their own phonetic
• Huge research body: children have capacity to
  obtain constraints that are phonetically
  grounded, formally simple, and not available from
  the ambient language data
• They do have a means of fabricating constraints
  that reflect phonetic naturalness
   Innate Knowledge: Different
       Assumptions in OT
• Tesar and Smolensky (1993) and others:
  constraints are innate. Child’s task: ranking
• Hayes (this paper): constraints not
  necessarily innate, but accessible through
  inductive grounding. Child’s task: finding
  and ranking constraints
       Ungrounded Constraints
• Phonological structure not always sensible (e.g.
  Northern Italian: intervocalic but not postnasal /s/
• Why? Result of a conspiracy of historical
  circumstances (e.g. peculiar sequence of changes,
  each one natural)
• Important points:
   – If grammars permit ungrounded constraints they must
     have some source  induction over the input data
   – If the distinction between inductively grounded
     constraints (learned from internal experience) and
     learned constraints (learned from gaps in the input data)
     is true, then it should be detectable. E.g.: children
     should never innovate ungrounded constraints.
     Consequences of Inductive
    Grounding for Feature Theory
•   Hayes makes suggestions regarding how feature
    theory might work under an inductive grounding
•   He thinks inductive grounding can help solve
    two problems of current feature theory
    1) Phonetics is very complicated (many different
       interactions) but phonological representations are not
       rich enough to characterize all the things that can
    2) Phonological assimilations that have more than one
•    Phonological constraints are often phonetic in
     character (but: no direct mapping)
•    Hypotheses:
    a) Learners extract phonological constraints from their
       own experience
    b) In constructing constraints, learners execute a trade-
       off between phonetic accuracy and formal simplicity
    c) Learners construct a phonetic map and then compare
       candidate constraints through the inductive grounding
       algorithm to find grounded constraints
•    Child data support the view that children can
     and do create constraints by inductive grounding
            More Conclusions
• Basic idea: functional factors are represented
  indirectly: they enter in at the level of language
  design, leading to the construction of formal
  grammars that are functionally good, with a bias
  toward formal symmetry
• Functional factors make themselves apparent in
  “maps” compiled from experiences of the
  language learner
        Even More Conclusions
• Inductive grounding creates constraints that reflect
  the functional principles in an indirect way
• Constraint ranking molds the raw set of
  constraints into a full and explicit grammar
• Innate: only procedure for inductive grounding
  and the mechanisms of OT
   – Arguments for innate principles can only be made
     stronger when inductive alternatives are addressed and
            The Final Slide
• Questions?
• Comments?
• View the complete presentation and other
  very interesting presentations on ejective
  stops at

To top