-
Contrasting
Feature Systems:
- Jakobson-Fant-Halle vs. Chomsky-Halle
- September 27, 2005
Some Common Traits
- Each phonological segment type in any language is a static, serially-ordered
unit. No temporal properties are inherent to any of the features.
- Each segment is a vector (that
is, an ordered list) specifying the
value of each relevant feature.
- Combinations of the features are used to pick out particular subgroups of segments. Thus,
voiced stop segments are all the segments that are [+voice,
-continuant].
- The features (and segments) are invariant
across speaking rates, speaker's voice, position in a
sequence, voice quality, pitch range, etc.
- The lexical specification of words in human memory employs these features.
That is, memory does not employ any concrete, speaker- or
rate-dependent features.
J-F-K:
- Each feature is defined in both
articulatory and acoustic terms. The rationale
is that the features are communicative signals used by both speakers
and hearers.
Thus, they must have both an articulatory definition and an acoustic
definition.
The authors discuss the acoustic theory of speech production,
display spectrograms of various segment types, etc.
- A single universal list of features plays a phonological role for
surface contrasts
only. The features are assumed to be always audible and
visible.
Thus only a small set of features
is needed - enough to cover all the surface contrasts within any
language -- only about 12 features seem to be sufficient. Thus,
some phonetic variation
within a feature is inevitable. Thus, the feature [+flat] can be
``implemented'' in different languages or different speakers either
with lip rounding or larynx lowering or pharyngeal constriction.
- The features permit only binary
values: +/-, plus or minus.
Chomsky-Halle
- Each feature is defined in exclusively
articulatory terms. The features
are viewed as units in a speech production system. Perception is
assumed
to be via analysis-by-synthesis - that is, perception leans primarily
on production or generation mechanisms.
- There is large universal set
of phonetic features, 40 or so, ``the phonetic
capabilities of man''. All possible differences in speech control
that
languages `have available to them' are included -- that is, the
phonetic
features are universal and equally available to all humans at
birth. Of course, 40 or so features still permit many thousands
of consonant and vowel types.
- Individual languages select a small subset of these (often 10-15) for
the phonological contrasts
within each language - the surface or near-surface phonological
contrasts. Thus the features play two roles - as (relatively)
concrete specifiers of
universal phonetic control units, and as abstract contrast units for
particular
languages.
- When used phonologically,
the features permit only binary values,
+/-.
But when considered universally and phonetically, they are also
normally binary, but may have, in principle, many possible values
between 0 and 1.