Capitalistic vs. Militaristic

Download Report

Transcript Capitalistic vs. Militaristic

Paradigm Uniformity and Analogy:
The Capitalistic versus Militaristic Debate
David Eddington
Brigham Young University
càpitalálistic
capi[ɾ]alistic
mìlitarístic
mili[th]aristic
Same prosodic
realizations of /t/.
structure,
but
different
càpitalálistic
capi[ɾ]alistic
mìlitarístic
mili[th]aristic
Same prosodic
realizations of /t/.
structure,
but
Why the difference?
different
Paradigm Uniformity!
If a base has a particular non-contrastive
phonetic feature, derivatives of that base will
tend to keep that feature (Steriade 2000)
capi[ɾ]al explains the flap in capi[ɾ]alistic
mili[th]ary explains the stop in mili[th]aristic
Steriade's Experiment
Subjects read list of 10 words including:
ro[ɾ]ary
volun[th]ary
Then they read neologisms based on those
word such as:
rotaristic
voluntaristic
Steriade's Experiment
11 of 12 subjects used same allophone in
base word as in neologism:
If ro[ɾ]ary
If ro[th]ary
then
then
ro[ɾ]aristic
ro[th]aristic
Riehl's Study (2003)
Reihl's subjects repeated four bases and
derived forms 12 times (negative/istic,
positive/istic, primitive/istic, relative/istic)
Riehl's Study (2003)
Reihl's subjects repeated four bases and
derived forms 12 times (negative/istic,
positive/istic, primitive/istic, relative/istic)
Some intersubject variability was found (e.g.
primi[th]ivistic vs. primi[ɾ]ivistic).
Reihl claims this invalidates the influence of
the base form.
Riehl's Study (2003)
Riehl expected all or nothing behavior.
Human behavior isn't all or nothing, but
stochastic.
Riehl's Study (2003)
Riehl expected all or nothing behavior.
Human behavior isn't all or nothing, but
stochastic.
A correlation between use of flap in base and
in derived is highly significant (r (14) = .748, p
< .0005, two-tailed)
So, analogical influence isn't nullified by the
slight variation Riehl found.
Rule and Analogy Interactions
Steriade (2000) and Davis (2005) claim analogy
(paradigm uniformity) interrupt the application
of the flap rule.
Rule and Analogy Interactions
Steriade (2000) and Davis (2005) claim analogy
(paradigm uniformity) interrupt the application
of the flap rule.
For Steriade, militaristic and capitalistic should
both be flapped. The [th] in militaristic is
analogical.
For Davis, both should contain [th] and the flap
in capitalistic is analogical.
The Point of this Presentation
There is no interplay between analogy and
rules.
The Point of this Presentation
There is no interplay between analogy and
rules.
It's all analogy
Problems with Traditional Analogy



Analogy only serves to patch up cases rules
can't account for.
No constraints are put on analogy.
No computationally specific method is used.
Linguistic Behavior is Analogical




Speakers store all past linguistic
experience.
Phonetic detail and redundant features
stored also.
Stored exemplars are consulted rather than
abstract rules or constraints.
Semantic, phonetic, and orthographic
similarity are used to find relevant analogs.
Linguistic Behavior is Analogical





In determining the pronunciation of /t/ in
capitalistic, many stored words are
consulted.
These words may influence it in varying
degrees to be either [ɾ] or [th].
Since capital shares so many
characteristics with capitalistic it is a major
analog.
Analogical pull is gradient: 90% [ɾ] and
10% [th]
This accounts for some of the variability
Traditional Allophonic Distribution



Generalizations are gleaned from input
during acquisition
Generalizations are stored as rules or
constraints.
Rules are used in subsequent linguistic
processing.
Analogical Modeling


A theory and computer algorithm (Skousen
1989)
Determines outcome (e.g. [ɾ] or [th]) based
on the similarity of the test form to a
database of stored instances.
The Database

3,719 instances of allophones of /t/ taken
from TIMIT



630 speakers read 10 sentences.
Utterances transcribed
644 [ɾ ], 234 [ʔ ], 284 [Ø], 760 [t], 860 [t˭],
and 969 [th], 48 [d].
The Database

Each instance of /t/ is encoded to include its
allophonic realization and the context it
appears in.
The Database


Each instance of /t/ is encoded to include its
allophonic realization and the context it
appears in.
The phones or boundaries three slots to the
left and right of /t/, and stress are encoded.


e.g. I know I didn't meet her
1) [ɾ], 2) word boundary, 3) [m], 4) [i], 5) word
boundary, 6) [ɚ], 7)pause, 8) primary stress, 9)
unstressed
How the Algorithm Works
How the Algorithm Works

The details would put you to sleep so I'll
spare you.
How the Algorithm Works


The details would put you to sleep so I'll
spare you.
To predict the pronunciation of /t/ in
capitalistic, look for similar words in the
database and see how they are
pronounced. Apply that pronunciation.
How the Algorithm Works


The details would put you to sleep so I'll
spare you.
To predict the pronunciation of /t/ in
capitalistic, look for similar words in the
database and see how they are
pronounced. Apply that pronunciation.


Outcome is stochastic (e.g. 90% [ɾ], 10% [th]).
Highest probability considered the “winner”.
Results of Previous Simulations


Allophone of all 3,719 database items
predicted by analogy.
Most predictions were either correct or the
“error” was a possible alternative
pronunciation.

e.g. amoun[th] of or amoun[Ø] of
Results of Previous Simulations


Allophone of all 3,719 database items
predicted by analogy.
Most predictions were either correct or the
“error” was a possible alternative
pronunciation.



e.g. amoun[th] of or amoun[Ø] of
Little change in predictive power when only
fraction of database used.
Little change when “critical” variables such
as stress are eliminated.
Present Simulations


Test words: capitalistic, negativistic,
positivistic, primitivistic, relativistic,
habitability, irritability, immutability,
dissatisfaction.
Two simulations:


Base words of test words contain [ɾ] in
database.
Base words of test words contain [th] in
database.
Test Word
Simulation Type
tʰ
ɾ
t
∅
ʔ
t=
d
capitalistic
Flapping Simulation
12
78
0
3
3
4
0
Aspiration Simulation
90
0
0
3
3
4
0
Flapping Simulation
1
98
0
0
0
1
0
Aspiration Simulation
93
6
0
0
0
1
0
Flapping Simulation
0
99
0
0
1
0
0
Aspiration Simulation
96
3
0
0
1
0
0
Flapping Simulation
0
96
1
1
0
1
0
Aspiration Simulation
94
2
1
1
0
1
0
Flapping Simulation
10
90
0
0
0
0
0
Aspiration Simulation
86
14
0
0
0
0
0
Flapping Simulation
8
80
0
0
0
14
0
Aspiration Simulation
80
0
0
0
0
20
0
Flapping Simulation
3
95
0
0
0
2
0
Aspiration Simulation
96
3
0
0
0
1
0
Flapping Simulation
4
93
0
1
0
3
0
Aspiration Simulation
83
12
0
1
0
4
0
Flapping Simulation
0
100
0
0
0
0
0
100
0
0
0
0
0
0
negativistic
positivistic
primitivistic
relativistic
habitability
irritability
immutability
dissatisfaction
Aspiration Simulation
Results of the Simulation

The pronunciation of the base form
influences that of the derived for per
analogy (paradigm uniformity).
Results of the Simulation


The pronunciation of the base form
influences that of the derived for per
analogy (paradigm uniformity).
The variability seen in Riehl's experiment is
seen in predicted probabilities less than
100%.
Results of the Simulation



The pronunciation of the base form
influences that of the derived for per
analogy (paradigm uniformity).
The variability seen in Riehl's experiment is
seen in predicted probabilities les than
100%.
The base form is not the only word
influencing the derived form.

capi[th]alistic predicted at 90%, yet capital only
accounts for 30% of this. Words such as
appetite, hepatitis, and particular also influence
the outcome.
What about Morphologically
Simple Words?



Mediterranean and Navratilova have same
stress as capitalistic and militaristic, yet
have no base form.
Steriade's rule incorrectly predicts
Medi[ɾ]erranean and Navra[ɾ]ilova.
Davis' rule correctly predicts
Medi[th]erranean and Navra[th]ilova.
What about Morphologically
Simple Words?




Mediterranean and Navratilova have same
stress as capitalistic and militaristic, yet
have no base form.
Steriade's rule incorrectly predicts
Medi[ɾ]erranean and Navra[ɾ]ilova.
Davis' rule correctly predicts
Medi[th]erranean and Navra[th]ilova.
Analogy correctly predicts Medi[th]erranean
and Navra[th]ilova.
So, is Davis' or Steriade's
Rule Correct?



Analogy corresponds with the outcome of
Davis' rule for these two words.
But, analogy works on a case-by-case
basis unlike rules which are global
generalizations.
Analogy can't be used to verify the
“correctness” of a rule.
Conclusions


Previous simulations show analogy alone
can account for the realizations of /t/.
No need to posit a rule plus analogy
dichotomy.
Conclusions





Previous simulations show analogy alone
can account for the realizations of /t/.
No need to posit a rule plus analogy
system.
Analogy accounts for influence of base
forms on derived forms.
Analogy account for monomorphemic forms
also.
Analogy accounts for variability in
pronunciation.
My Rant: Why Rules Suck





They can't be proven or disproved so they
fall outside of the realm of scientific
investigation.
L2 speakers may know the rules but can't
apply them to speak.
Some rules are extremely complex for
linguists to arrive at, yet are assumed to be
“subconsciously” learned by preschoolers.
Rules can't account for variability.
Same goes for constraints.