ARE PHONOLOGICAL ENTITIES FICTITIOUS OR REAL?

1
ARE PHONOLOGICAL ENTITIES FICTITIOUS OR REAL?

• Prepared by Agnieszka Sowinska and Beata
  Szymczak
• Based on Self-organizing processes and the
  explanation of phonological universals by
  B.Lindblom

2

• Where do phonological universals such as segments
  and features come from?

3
...we will be using

• a self-organizing model of phonological
  structure
• implemented in a series of computational
  experiments organized to simulate the emergence
  of segments and features
• which was to select a subset of k phonetic
  signals from a larger inventory of n universally
  possible gestures.

4
holistic coding

phonemic coding

• every derived syllable remains a gestalt pattern
  that cannot be fractioned into smaller parts
  occurring also in other syllables
• one holistic phonetic signal per morpheme

• every selected signal can be reduced to subparts
  shared with other syllables of the subset
• mapping meaning onto sound by forming combinatory
  patterns of segments and features

5
THREE RESULTS LOOM LARGE

• 1) the occurrence of MINIMAL PAIRS, (instances of
  phonemic and segmental coding)
• 2) the possibility of analyzing the derived
  contrasts in terms of DISTINCTIVE FEATURES (eg.
  grave-acute)
• 3) the RULE governing the distribution of palatal
  and velar allophone of the /g/ phoneme

SEGMENT, FEATURE, RULE - are not explicit
constructs of the present theory - BUT are
IMPLICIT properties of the phonetic signals
6
The conditions under which structuration into
segments and features arise

• the mechanism favouring phonemic coding ( the
  repeated contrastive use of a syllable onset and
  offset) requires that k - the lexicon - be much
  greater than the number of available onsets and
  offsets
• IF
• 1 the performance constraints severely limit the
  phonetic variation of onsets and offsets
• 2 k becomes sufficiently large relative to the
  phonetic repertoire
• THEN
• speakers can find a way of making their
  inventory of phonetic signals grow ONLY by
  invoking gesture onsets and offsets repeatedly
  and in new combinations

7
It leads us further to...

• THE SEGMENTAL AND FEATURAL STRUCTURATION BUILT
  INTO THE PHYLOGENY AND ONTOGENY OF SPEECH AS A
  STATISTICAL BIAS, AND ARISING IMPLICITLY IN A
  SELF-ORGANIZING MANNER

8
And now its high time we got down to the
nitty-gritty
WHERE DO SEGMENTS AND FEATURES COME FROM?
9
Two general approaches

• MENTALISTIC
• based on information theory that treats speech as
  an error-correcting code ( eg. our early
  ancestors...)

• MECHANISTIC
• eg. children-appear to use words as unanalyzed
  wholes never aware of having acquired phonemic
  coding, which seems to emerge in an automatic
  and implicit manner

10
THE THEORY OF SELF-ORGANIZING SYSTEMS (a
mechanistic explanation)

• a scientific paradigm which has recently arisen
  at the intersection of physics, chemistry,
  biology, and sociology

• aims at formulating the general laws governing
  the spontaneous occurrence of order in nature and
  the evolutionary dynamics

11
PHONOLOGICAL STRUCTURE

• INTERACTION AMONG SUBSYSTEMS

• SELF-ORGANIZATION

STRUCTURATION
12
TERMITE NEST BUILDING (M. Turvey)

• The form of these nests appears to arise as a
  result of a simple local behavioral pattern
  followed by each individual insect in the
  presence of certain local stimulus conditions

13
SIMULATIONS OF EMERGING PHONETIC STRUCTURE

• Size of lexicon or signal inventory - k
• Universal phonetic signal space (specific at
  three levels articulation, acoustics,
  perception)

14
UNIVERSAL PHONETIC SIGNAL SPACE

• We specify the lg-independent universal class of
  possible articulations confined to vowels and
  voiced stops
• these articulations involve transitions from a
  closed (stoplike) to and open (vowellike) state
• possible CV event (a straight line coursing
  between a possible locus - assigned to each
  closure location - and a possible vowel )
• possible CV syllable - any trajectory running
  between an arbitrarty but possible locus and an
  arbitrary but possible vowel in the space defined
  by the first four formants
• the CV space- a continous one made up of
  infinitely many holistic signals (quantally
  structured)

15
Figure 1. The main result of the study
demonstrating that in the presence of certain
constraints a continuous space can become
quantally structured.
16
Phonetic constraints

• Talker-based conditions
• sensory discriminability
• preference for less extreme articulation

• Listener-based conditions
• perceptual distance
• perceptual salience

17
Where the open-close feature comes from in vowels?

• THE AIM the assigment of phonetic shape to a
  minilexicon consisting of k lexical elements
  with distinct meanings
• THE RESULT systems of CV syllables tend to be
  selected in such a way that they achieve
  sufficient perceptual differences at acceptable
  articulatory costs
• every syllable was used once as the initial item

18

• SYLLABLE END POINTS

Figure 2. Occurence of derived CV
combinations
19
OUR CONCLUSIONS ARE...

• The computations were insensitive to how the
  recursive search was initiated
• the most favored vowel turned out to be ?
• Figure 2 differs from a holistic coding in that
  individual CV transitions can indeed be
  fractioned into smaller parts also occurring in
  other CV sequences

20
Figure 3. Tendency towards complementary
distribution of /g/ allophones
21
It will take us to two extreme outcomes...

• We need a minilexicon containing 12 words, in
  the form of CV syllables

• We search for sets of 12 CV sequences

22
The holistic coding -no minimal pairs Every
CV transition is a gestalt that cannot be
fractioned into smaller parts also occurring in
other CV sequences
23
END POINTS 1 2 3 4
STARTING 1 2 POINTS
3
ALL POSSIBLE COMBINATIONS OF POINTS USED
24
running out of onsets/offsets as a numerical
artefact

• The first item ba - the gradual emergence of a
  near-optimal sequence of syllables in a schematic
  way

• We ask the computer to generate no more than 7
  syllables

25
(No Transcript)
26
THE RESULTS ARE

• minimal pairs the simulated phonemic
  constraints are not artefacts, and they could by
  a result of running out of onsets/offsets
• the size of that inventory is determined by the
  severity of the performance constraints
• the rank order is determined by the perceptual
  distance and salience criterion

27
The origin of segments and features an
explanation based on the concept of
self-organization

• - Random sampling of the possibilities offered by
  the universal phonetic space should make all such
  possibilities equally possible. However, in the
  presence of certain constraints, nonuniform
  preferences for certain syllables over others
  arise (quantal structuration).
• - The notion of system implies certain
  paradigmatic relations among the elements of the
  system must hold. When those relations are
  present, structuration occurs gtimplicit form
  emerges and the causes of such pattern formation
  are indirect.
• Let us describe our ancestors joint effort to
  define the phonetic shapes of a growing set of
  concepts as A RANDOM SAMPLING OF THE UNIVERSAL
  PHONETIC SPACE IN THE PRESENCE OF PERFORMANCE
  CONSTRAINTS. ?

28

• ? Our preceding reasoning that quantal
  structuration is built into the phylogeny of
  speech as a statistical bizs should apply. When
  several individuals find that their random
  samplings sometimes converge and similar signals
  are favoured, a situation that might be conducive
  to socially conventionalised naming appesrs to
  be at hand.
• Results
• 1) evidence of phonemelike or segmental
  coding
• 2) the predicted syllables bear some, if not
  strong, resemblance to natural sets of syllables
• 3) the rather realistic allophonic variation
  of the /g/ phoneme
• FEATURE, PHONEME, and ALLOPHONIC RULE are present
  only IMPLICIT properties of the behaviour. They
  are derived rather than axiomatically postulated
  as substantive universals.

29
It was worth wading through...because now, its
time for...
THE END