Phonological constraints as filters in SLA

1
Phonological constraints as filters in SLA

• Raung-fu Chung
• rfchung_at_mail.nsysu.edu.tw

2
1. Introduction

• The main components of this article are
• The framework of Optimality Theory
• Acquisition and learnability in OT
• Our model
• Concluding remarks

3
2. The framework of Optimality Theory

• (1) The model of OT

4

• For instance, the English morpheme for plurals
  /s/ can be realized either as s, z, depending
  on the preceding sound of the stem

5

• (2)
• cat k?? cats k??s
• dog d?g dogs d?gz
• hen hen hens henz
• The input form is taken to be s. Then we
  propose the following constraint
• Voiced Obstruent Prohibition)?

6

• (3) Voiced Obstruent Prohibition, VOP
• No obstruents can be voiced.

7

• Another constraint called for is
• (4) Obstruent Voicing Harmony, OVH
• The adjacent obstruents should share the same
  value for voice
• The third constraint is a universal constraint,
  which is one component of IO, here referred as
  Ident-IO(voice) Ident identical, IO Input and
  Output)
• (5) Ident-IO(voice)
• The value for voice feature of the Output
  should be identical with tha of the Input.

8

• As for the ranking, it is obvious, as shown
  below.
• (6) OVH gtgt VOP (?gtgt? be preceded or be prior to)
• Adding to IO, we have the following raking for
  all the three constraints we just proposed.
• (54) OVHgtgt Id-IO(voice)gtgt VOP

9

• (55)

/d?g-z/ OVH Id-IO(voice) VOP
? a. d?g-z
b. d?g-s !
c. d?k-s !
10
3. Acquisition and learnability in OT
11
3.1 The notion of learnability

• a. The formal learnability in the sense of Tesar
  Smolensky, 1993 assumed that all constraints
  started out being unranked. In later empirical
  studies (e.g. Gnanadesikan, 1996 Levelt, 1995),
  it is pointed out that outputs are initially
  governed by markedness constraints, rather than
  by faithfulness constraints. This leads to the
  proposal that in the initial state of the
  grammar, all markedness constraints outrank all
  faithfulness constraints, or M gtgt F for short
  (Kager, Pater Zonneveld, 2004 Hayes, 2004
  Prince Tesar, 2004).

12

• b. There are two algorithms accounting for
  learnability of constraint rankings Constraint
  Demotion Algorithm (CDA) and Gradual Learning
  Algorithm (GLM). Constraint Demotion Algorithm
  (CDA), proposed by Tesar Smolensky (1993, 1998,
  2000), ranks a set of constraints based on the
  positive input. For example, L1 acquisition can
  be interpreted as constraint demotion (Tesar
  Smolensky, 1996)

13

• c. Gradual Learning Algorithm (GLA), developed in
  Boersma (1997, 1998) and Boersma Hayes (2001),
  handles variation in the input and explains
  gradual well-formedness. GLA is helpful in
  accounting for categorization errors a learner
  makes in both production and perception. L2
  learners with restricted constraint sets have to
  gradually learn to rerank the constraints by
  raising or lowering the existing ones.

14
4. Our model

• L1 Filter Hypothesis OT

L1 filter
Interlanguage- ranking
15

• Empirical arguemtns
• An VOT-baased analysis of VOT production by
  Taiwanese EFL learners
• An diphthong construction of Mandarin and English
  for Taiwanese learners
• Errors of production of yi and wu for
  Mandarin EFL leasrners

16
An OT-based Analysis of VOT Productionby
Taiwanese EFL Learners

• Acoustic values of VOT (Liou, 2005)

• Note NSE native speakers of English HEFL high
  proficient EFL learners LEFL low proficient EFL
  learners MAN Mandarin SM Southern Min

17
Constraints for VOT

• 1. the CATEG(ORIZE) family, which punishes
  productive categories with certain acoustic
  values. For example, CATEG(VOT /91.5ms/) is
  against producing /91.5ms/ as a particular
  category.
• 2. WARP family, which demands every segment be
  produced as a member of the most similar
  available category. For instance, WARP(VOT
  9.3ms) requires that an acoustic segment with a
  VOT of 91.5ms should not be produced as any VOT
  category that is 9.3ms off (or more), i.e. as
  /82.2ms/ or /100.8ms/ or anything even farther
  away.

18
Constraint-ranking for English ph by NSE

• Tableau 1 English ph of NSE

19
Constraint-ranking for ? by Taiwanese EFL
learners

• Tableau 2 Mandarin ? by Taiwanese EFL learners

75.4 ms Intended(?) CATEG (/82.2/) CATEG (/91.5/) WARP (16.1) WARP (6.8) CATEG (/75.4/)
a. 91.5ms !
b. 82.2ms !
c. ? 75.4ms
20
Constraint-ranking for Interlanguage ph by
Taiwanese EFL learners

• Tableau 3 Interlanguage ph by HEFL

91.5 ms Intendedph WARP (16.1) CATEG (/91.5/) CATEG (/75.4/) CATEG (/82.2/) WARP (9.3)
a. ? 91.5ms !?
b. ? 82.2ms ? ?
c. 75.4ms !? !?
21
Tableau 4 Interlanguage ph by LEFL
91.5 ms Intendedph WARP (16.1) CATEG (/91.5/) CATEG (/75.4/) CATEG (/78.7/) WARP (12.8)
a. ? 91.5ms !?
b. ? 78.7ms ? ?
c. 75.4ms !? !?
22
An OT-based Analysis of Mandarin and English
diphthongs for Taiwanese EFL (MSL) learners
23

• (1)

Mandarin vowels Mandarin vowels Mandarin vowels SM vowels SM vowels SM vowels
i u_ u i u
e o e o
a a
24

• (2)

Mandarin diphthong construcion principle Mandarin diphthong construcion principle Mandarin diphthong construcion principle Mandarin diphthong construcion principle
Front back Front back Front back
i u
e o

-back vowels for

• (3)

ie ie (?)?t?ie (?)
ei pei (?)?kei (?)
back vowels for

• (4)

uo uo (?)?kuo (?)
ou ou(?)?kou (?)
25

• (4)
• N (N??,????)
• ? ? -? ?

26

• (5)

SM vowels SM vowels SM vowels
front back front back front back
i u
e o
a
Different back featues for

• (6)

iu iu (?)?kiu (?)
ui ui (?)?kui (?)

• (7)

io ?? kio (?)
io (?)

• (8)

ue ?? kue (?)
hue (?)
27

• (9)
• N (Nnucleus,?same,ungramm
  atical)
• ? ? ? ?

28
Input violation adjustment results samples
a. /ie/ Same front features ideleted e ?????en
a. /ie/ Same front features ideleted e ????? e
b. /ei/ Same front features i deleted e ????? pe
b. /ei/ Same front features i deleted e ????? ke
c. /uo/ Same back features u deleted o ????? o
c. /uo/ Same back features u deleted o ????? to
d. /ou/ Same back features u deleted o ????? o
d. /ou/ Same back features u deleted o ????? to
29
5. Concluding remarks

1. theoretical implications
2. empirical supports

30
The end