formant frequency values for control of a resonance synthesizer.
Target values of articulatory parameters stored for each phone are
distinguished as 'marked' or 'unmarked', depending on whether they
are characteristic of the articulation of the phone; the distinction
is much the same as Coker's important v. unimportant. A further
distinction is made between position and closure parameters. The
transition of a parameter from the midpoint of one phone to the
midpoint of the next is made up of linear segments, and varies
depending on the type of each phone (vowel, consonant, or pause),
the marking of the two target values, the characteristic rate of
each parameter, and the parameter type (position or closure). The
rather complex transition rules insure that marked target values for
consonant closure parameters will be attained and held, and that
progress towards other marked target values will occur over a longer
time and at a more rapid rate than toward unmarked values. Thus
coarticulation and centralization are provided for. In general a phone
can influence only the adjacent phones, but nasalization is provided
for by allowing the velar closure parameter to influence several
preceding phones.
Henke's (1967) model attempts to handle the same coarticulatory
phenomena as Haggard's and Coker's while avoiding a commitment to
a parametrization in favor of a naturalistic representation of
articulation. Each articulator is represented by a family of
'fleshpoints' on the midsagittal plane. During the motion of an
articulator, each point moves along a vector determined by a target
location and a target articulator shape. During the early part of its motion, a point first accelerates as the inertia of the articulator is overcome, then attains an appropriate steady velocity, and finally slows as it approaches the target point. The motion of the articulators is determined by a set of attributes stored for each phone. A configurative attribute corresponds to a target location and shape; a strength attribute, to the force which moves an articulator. At any moment, motion may be controlled by attributes associated with one or several successive phones. However, different attributes referring to the same articulatory region cannot both apply at once. A change of attribute will occur at a time dependent on the attributes of the current phone and of the following phone, and upon the progress of articulatory movements determined by other attributes. For example, when articulation of a stop consonant begins, the relevant stop attributes, specifying the shape and location of the articulator and the force of the closure, assume control of the articulator. When
closure is attained, the attributes of a following vowel, except
those which conflict with the stop attributes, are applied, and
attributes of earlier phones are dropped. After the stop is released,
all the attributes of the following vowel apply for enough time to
allow the articulators to approach the vowel target.
Systems such as those of Coker, Haggard and Henke are more
theoretically adequate than shape systems; we are clearly closer
to the level of phonetic features. The translation from discrete
to continuous domains is more natural because a target is defined
for each articulator. We might compare the kind of description
|