The nature of distinctive features

advertisement
Dept. for Speech, Music and Hearing
Quarterly Progress and
Status Report
The nature of distinctive
features
Fant, G.
journal:
volume:
number:
year:
pages:
STL-QPSR
7
4
1966
001-014
http://www.speech.kth.se/qpsr
I.
A.
THEORY O F DISTINCTIVE FEATURES
THE NATURE O F DISTINCTIVE FEATURES
*
G. F a n t
Introduction
The following e s s a y i s intended a s a review of m y own thinking on
distinctive f e a t u r e s . Of a l l that h a s been w r i t t e n o n t h i s topic I have
followed up only a p a r t .
I a m m o r e p r a c t i c a l l y oriented than linguists
and l e s s bound t o orthodox acceptance of working p r i n c i p l e s but I find
i t fascinating to s i t down and w o r k out a l t e r n a t i v e solutions, e. g. , t o
o r d e r the Swedish vowels which was one of my e a r l y i n t e r e s t s .
A substantial r e v i s i o n of o u r old " P r e l i m i n a r i e s " h a s long b e e n
overdue.
T h e m a j o r principles a r e s t i l l valid but I f e e l we need much
m o r e factual d a t a before a substantial r e v i s i o n c a n be undertaken.
T i l l t h e n the following m a t e r i a l m a y s e r v e a s a n e x p r e s s i o n of m y
views on the subject.
S p e c i a l i s t s i n language and s p e e c h have displayed r a t h e r diverging
reactions t o the theory of distinctive f e a t u r e s and misunderstandings
have b e e n frequent.
Does i t provide a condensed p r e s e n t a t i o n of the
m o s t useful f a c t s about speech o r is i t just a n intellectual g a m e , a
purpose of i t s e l f f o r the s t r u c t u r a l linguist?
High i n i t i a l expectations
of finding new and s i m p l e solutions t o c e n t r a l p r o b l e m s of s p e e c h
a n a l y s i s , such a s automatic s p e e c h recognition, have b e e n followed by
distrust.
Now, the t h e o r y of distinctive f e a t u r e s is not intended as a
working r e c i p e f o r technical application but i t c a n provide s o m e organizational principles and suggestions. The r a t h e r specific terminology
of the " ~ r e l i m i n a r i e s ' 19
(ddea not stand f o r e i t h e r radically new o r v e r y
special features.
The a r t i c u l a t o r y , acoustic, o r perceptive c o r r e l a t e s
of distinctive f e a t u r e s should c o m p r i s e condensed t r a n s f o r m s of the
m o s t relevant information f r o m any of t h e s e s t a g e s within the complete
*
T h i s i s a p r e p r i n t v e r s i o n of a p a p e r t o be published b y Mouton & Co.
i n F e s t s c h r i f t f o r Roman Jakobson on t h e o c c a s i o n of his 70th
birthday, October 11, 1966.
-
speech communication systems. A continued study of alternative solutions and descriptive f o r m s i s needed i n the development of the d i s tinctive feature theory a s i n speech analysis i n general.
What is then really the concept of distinctive f e a t u r e s ?
How a r e
they defined, f r o m the speech wave, f r o m articulation, o r f r o m p e r ception?
A r e they simply a p a r t of the linguistic code f o r decomposi-
tion of phonemes i n a bundle of s m a l l e r units?
Reading the ' ' P r e l i m -
inaries" one finds that the distinctive f e a t u r e s operate on a l l four
levels but f r o m which one do you s t a r t the analysis?
Here a s i n
phonemic analysis the s e t of distinctive f e a t u r e s constitutes a b s t r a c t
units of the message code.
one of two alternatives.
The distinctive f e a t u r e i s a choice between
In l l P r e l i m i n a r i e s " distinctive f e a t u r e s a r e
r e f e r r e d t o by t e r m s a s "discriminations",
s t r e s s i n g the linguistic level.
"choice",
and llselectionl'
A distinctive f e a t u r e generally r e c u r s
a s a choice situation i n s e v e r a l minimal distinction p a i r s within a
language ancl i t i s of c o u r s e required that the physical o r physiological
manifestations be consistent, i. e . , one and the s a m e feature shall have
qualitatively the s a m e articulatory, acoustic, and perceptive c o r r e l a t e s
independent of context of other f e a t u r e s within the bundle.
The modifica-
tion "qualitatively" h e r e implies that the relation between the two opposites i s the s a m e i n all contexts.
Absolute values of descriptive
p a r a m e t e r s , however, generally v a r y with context.
F a i l u r e to r e -
cognize the role of contextual bias i s a frequent s o u r c e of misunderstanding of the nature of distinctive features.
by definition the s a m e i n a l l contexts.
A distinctive feature is
The associated physical phe-
nomena, on the other hand, r e f e r r e d to a s "correlates", "cues1t, o r
" p a r a m e t e r s " need exhibit only relational invariance.
Distinctive features a r e really distinctive categories o r c l a s s e s
within a linguistic s y s t e m but just like i n accepted phonemic analysis
it i s required that they a r e consistent with the phonetic facts and these
phonetic facts on various levels have lent t h e i r name to the features.
It i s not within my competence to d i s c u s s the generality of distinctive
f e a t u r e s but i t i s apparent that they c o m p r i s e , a s they should, the
essentials of the framework of c l a s s i c a l phonetics and i n addition s o m e
categorizations of a m o r e novel appearance.
I s h a l l now proceed to some specific points concerning distinctive
f e a t u r e s some of which a r e often brought up i n discussions.
How important i s the binary principle?
The binary principle obviously has i t s b a s i s i n the presence v e r s u s
absence of an articulatory o r phonatory event, e. g . , presence v e r s u s
absence of voicing, nasality, occlusion, etc. o r i n a selection of one
of two polar alternatives along a continuous p a r a m e t e r scale, e. g
more open a s opposed to l e s s open.
.,
In the analysis of vowels i t can
be motivated to recognize m o r e than two significant levels of one and
the same p a r a m e t e r , cif., the discussion on compactness i n " p r e l i m As indicated i n the analysis of the Swedish
inaries", paragraph 2.414.
vowel s y s t e m l a t e r i n this paper one could conceive of instances, where
up to 4 distinct levels of one and the same feature might be considered.
In these instances a decompoqition i n t e r m s of two binary categories
i s generally undertaken i n o r d e r t o allow a consistent use of the binary
principle within the whole system.
The distinctive feature represents the linguist' s condensed view of
the minimal units for composing speech messages.
If properly applied,
categorization according to binary principles need not come i n conflict
with the physical reality.
It i s a m a t t e r of coding convehience only.
Economy, but at what p r i c e ?
If alternative solutions of distinctive features a r e possible i t i s
sstablished policy to adopt the one providing the best economy i n t e r m s
of the smallest number of features o r r a t h e r the least redundancy
minimizing the number of alternatives that can be generated by the
specific s e t of categories.
However, this requirement can come i n
conflict with the principle of consistency, i. e . , t h e r e i s the r i s k that
one o r more minimal p a i r s i n which a feature i s supposed to operate
do not conform sufficiently well with the r e s t of the system.
Seemingly
elegant solutions may thus have to be rejected by failures to apply i n
specific contexts.
It i s also apparent that the economy gained i n t r e a t -
ing consonants and vowels with the s a m e features at times leads to
somewhat remote analogies.
Are distinctive features always orthogonal?
Distinctive features a r e handled a s independent units on thc linguistic level but t h e i r phonetic manifestations often lack orthogonality.
The
phonetic quality of a vowel ma;r to a f i r s t approximation be specified
by F 1 and F2. In the F
1'
F plane, however, gravity, compactness,
2
f l a t n e s s , and t e n s e n e s s a l l occupy specific v e c t o r s and interdependency
i s thus unavoidable.
E v e n when s u c h p a r a m e t e r s a s F3,Fo, o v e r a l l
intensity and duration a r e added i n o r d e r to provide a b e t t e r approximation i t i s not phonetically r e a l i s t i c to choose a consistently orthogonal
s e t of f e a t u r e s .
How a r e distirictive f e a t u r e s distributed i n t i m e ?
P e o p l e who l a c k training i n experimental phonetics a r e g e n e r a l l y
r a t h e r s u r p r i s e d when they l e a r n that the acoustic s p e e c h wave does
not stand up v e r y well to the i d e a l concept imposed by o u r intuitive
phonemic view of s p e e c h a s a sequence of d i s c r e t e units with distinct
boundaries.
One m a j o r shortcoming of " P r e l i m i n a r i e s " is the l a c k of
a r e a l i s t i c d i s c u s s i o n of the t i m e - v a r y i n g a s p e c t s of s p e e c h p a t t e r n s
and the t e m p o r a l distribution of t h e acoustic, articulatory, and perceptual
c h a r a c t e r i s t i c s underlying the distinctive f e a t u r e s which might have
s a v e d phonetically inexperienced people f r o m developing a n o v e r s i m plified, often naive, view of the segment s t r u c t u r e of speech.
It i s s a i d i n p a r a g r a p h 2.14 of " P r e l i m i n a r i e s " that " F o r p r a c t i c a l
p u r p o s e s e a c h phoneme c a n be r e p r e s e n t e d by a q u a s i - s t a t i o n a r y s p e c t r u m i n which t h e t r a n s f e r function i s invariable with r e s p e c t to t i m e ,
except i n the m a n n e r stated f o r t r a n s i e n t effects".
Phoneme boundaries
a r e s a i d to b e r e l a t e d to rapid changes e i t h e r i n the s o u r c e function o r
i n the vocal t r a n s f e r function.
It is a l s o s a i d t h a t inherent f e a t u r e s i n
c o n t r a s t t o prosodic f e a t u r e s a r e definable without a r e f e r e n c e to the
sequence.
T h e s e r u l e s a r e oversimplified and need t o b e reformulated
and expanded.
T h e s p e e c h s p e c t r o g r a m displays a m i x t u r e of continuous and d i s continuous e l e m e n t s .
A s u c c e s s i v i t y of "segments" i s to be s e e n but
what f r o m the s p e c t r o g r a m s a p p e a r s to b e a n a t u r a l unit m a y constitute
only a f r a c t i o n of a phoneme, e. g. , the a s p i r a t i o n segment o r the occlusion segment of a noninitial unvoiced stop.
In o t h e r i n s t a n c e s a
piece of s p e e c h t h a t stands out f r o m the r e s t of the sequence a s a s e p a r a t e unit, e. g . , i n v i r t u e of a continuity of voicing, m a y be a s s o c i a t e d
with s e v e r a l s u c c e s s i v e phonemes.
No unique and s i m p l e rule e x i s t s f o r segmentation of s p e e c h on the
b a s i s of nonphonemic c r i t e r i a .
I would c l a i m , however, that a non-
phonemic scgmentation (1)(2) could be of value a s a rationale f o r a r t i c ulatory o r s p e c t r o g r a p h i c s y s t e m a t i z a t i o n s p e r s e and a s a n i n t r o d u c t o r y
,
I
f o r m of transcription before imposing the linguistic message concepts
on the signal data.
The outcome of such confrontations of preconceived
linguistic s t r u c t u r e with the observed acoustic-phonetic s t r u c t u r e i s
that the number of physical sound segments comes out t o be l a r g e r than
the number of phonemes.
Because of coarticulation effects one sound
segment generally c a r r i e s information on two o r m o r e successive
phonemes.
Conversely,
a single phoneme e x e r t s an influence on sev-
e r a l successive sound segments of the signal structure.
When i t comes to discussing the distribution of distinctive features
i n time one must make c l e a r if it i s the abstract message s t r u c t u r e o r
the physical manifestation of speech s i g n a l s that i s intended.
In the
f o r m e r c a s e distinctive features a r e bounded a s the phonemes although
t h e r e can be continuity of a feature f r o m one phoneme to the next.
It
i s thus said in paragraph 1.1 of "Preliminaries" that "The difference
between the distinctive f e a t u r e s of continuous bundles p e r m i t s the
division of a sequence into phonemes.
This difference may be either
complete, a s between the l a s t two phonemes /i/ and
/r~/ i n the word
wing (which have no distinctive features in common) o r partial a s between the l a s t two phonemes of the word apt".
This statement i s c o r r e c t on the message level.
Indeed, the
phoneme /i/ i s categorized a s nonconsonantal, vocalic, nong rave,
acute, and noncompact whereas th6 phoneme /Q/ is labelled consonantal,
nonvocalic, nasal, compact.
One would accordingly expect a maximum
of acoustic contrast between the /i/ and the / r ~ / of wing.
In a multidimensional articulatory o r acoustic space, however, the
zontrast between the two corresponding segments i s minimal only.
The
place of tongue articulation i s identical o r almost identical i n American
English, and both segments a r e produced with a lowered velum, the
anticipatory nasalization being a normal feature generally affecting the
entire segment assigned to the /i/.
The raising of the tongue against
the palate closing off the mouth cavity does not affect the sound much
since a substantial part i s directed through the nose already i n the /i/
segment.
In the c a s e of a m o r e reduced articulation the tongue never
reaches the stages of full contact with the palate and the phoneme / r ~ /
i s signalled merely a s the nasalization of the sound segment.
The p e r -
ceptual importance of nasalization of a vowel a s a cue f o r identification
of an adjacent nasal phoneme i s considerable ( 3
.
The theory of segmentation of speech on various levels i s a well
worth object f o r f u r t h e r r e s e a r c h and descriptive studies. The e s s e n t i a l point to consider i s that we can m e a s u r e the duration of physical
events such a s sound segments i n a s p e c t r o g r a m but t h e r e e x i s t s no
unique method o r convention of measuring the duration of a phoneme o r
of a distinctive feature.
Specific features.
Vocalic and consonantal
One of the weaker p a r t s of the distinctive feature theory i s that of
defining consonants and vowels. It i s i n my opinion quite motivated to
categorize liquids a s being both vocalic and consonantal but the c l a s s i fication of the consonant h (and glides) a s being nonvocalic and nonconsonantal i s a m o r e a r b i t r a r y construction although arguments c a n be
raised i n favor of such a classification.
The physical c r i t e r i a f o r the
vocalic and consonantal f e a t u r e s have not been v e r y rigid.
A small
damping of vowel formants has been one of the requirements i n all
versions of the system.
The f i r s t edition of " P r e l i m i n a r i e s " s t r e s s e d
the voiced s o u r c e of vocalic sounds but the Addenda and Corrigenda
chapters of the l a t e r edition turned the emphasis on the formant pattern
and formulated the consonantal feature a s almost the negative of the
vocalic f e a t u r e i n t e r m s of formant reduction.
In "Fundamentals of Language" Jakobson and Halle (4) limit the
consonantal feature to a low intensity alone.
In a study of the classifica-
tion of Swedish phonemes ( 5 ) I introduced a new formulation retaining
the concept of formant reduction i n defining the consonant feature but
with intensity associated with the vocalic feature.
formulation i s that the phoneme /h/
a s being l e s s intense.
One gain of this
accordingly c o n t r a s t s with vowels
Also /h/ differs f r o m other consonants i n the
l a c k of p a t t e r n contrast with adjacent vowels, motivating the minus
consonantal feature.
The study of Fant ( 1 9 6 0 ) ~s e e i e f . (5)y
the theory of the syllable.
also includes comments on
The syllable nucleus must p o s s e s s the voc-
alic feature and i t displays a temporal contrast with respect t o adjacent
sounds i n t e r m s of either higher intensity o r a m o r e vowellike structure.
Syllabicity should not be ascribed to intensity alone.
~ e n s e / l a xand voiced/voiceles s
The subject of tense and lax vowels and consonants has been given
a thorough treatment by Jikobson and Halle (6).
T h e i r view on the
subject does not depart substantially f r o m that e x p r e s s e d i n o u r e a r l i e r
joint work.
The tense v e r s u s lax opposition i s intended to operate i n
Tenseness i s phonetically described
vowels a s well a s i n consonants.
by an articulation with g r e a t e r o v e r p r e s s u r e behind the place of the
active source; i n the c a s e of vowels a higher subglottal p r e s s u r e and
i n the c a s e of stops and constrictives a higher p r e s s u r e behind the
place of articulation.
F u r t h e r m o r e tenseness i s associated with a m o r e
e x t r e m e articulation and with a g r e a t e r t i m e spent i n a n e x t r e m e a r t i -
,,.- .
culatory position.
The two l a s t mentioned c h a r a c t e r i s t i c s w e r e mentioned i n o u r ear,
l i e r work but I a m somewhat sceptic about the higher o v e r p r e s s u r e ,
This factor when p r e s e n t indeed adds emphasis to tense consonants
but i n my opinion i t has not been sufficiently well documented i n experimental work.
6'
Recent studies of Mal6c t suggest that the combined ef-
fect of p r e s s u r e and duration expressed a s a pulse integral could have
a role i n proprioceptive feedback.
In my experience Swedish voiced
and unvoiced stops a r e produced with the s a m e subarticulatory p r e s s u r e
at the instance before release.
This does not prove anything f o r Eng-
l i s h but i t s e e m s probable that the p r e s s u r e factor, i f present, a s a
constituent i n the opposition between American English unvoiced and
voiced stops, i s of relatively s m a l l significance and a secondary effect
of glottal articulation.
Several recent studies support the view of L i s k e r
and Abramson ( 8 ) that i t i s the glottal articulation that i s the basic
factor.
The longer duration and higher intensity of the noise interval follow-
.
.
ing at the r e l e a s e of an unvoiced stop compared with voiced stops a r e
physiologically due to a l a t e r closing of the vocal c o r d s a f t e r the r e lease.
The pulmonary p r e s s u r e appears to be the same.
subglottal p r e s s u r e i n Swedish speech do
Studies of
not reveal any difference i n
the pulmonary activity comparing voiced and unvoiced consonants o r
s h o r t o r long vowels.
I really doubt that subglottal p r e s s u r e would
have anything to do with the tense-lax opposition among English and
F r e n c h vowels.
On these grounds I hesitate to accept the use of the tense-lax opposition among American English consonants a s well a s vowels suggested
by Halle (9). It should be used i n the vowel s y s t e m but withbut reference to subglottal p r e s s u r e .
Within the consonants i t would be just
a s motivated to use the voiced-voiceless distinction a s the tense-lax
distinction.
The economy gained by one and the s a m e feature operat-
ing i n vowels a s well a s i n consonants i s of course desirable but I find
e
-
--
between phonetic
- a. close correspondence
- important..-..
t o- retain
..
i t more
... . .
facts "-.
and
. fcature c r i t e r i a .
7
,-
We need m o r e experimental data to illu-
minate this very interesting problem.
Distinctive features and perception
I do not hold the view that the decoding of speech in the brain up to
the level of phonemic identification has to follow a functional scheme
s t r i c t l y conforming with a distinctive feature system of language analysis.
This does not imply that I consider distinctive features unimpor-
tant i n speech perception.
Distinctive features a s phonetic c l a s s e s a r e
a psychological reality a s judged f r o m confusion t e s t s under varying
types of distortion o r mental disturbance.
Even i n rapid mimicking ( 10)
the decoding proceeds along phonetic c l a s s e s s o that, e. g. , place of
articulation may be confused whilst the category of stop sound is c o r rectly recognized.
When constructing models of speech perception we should not limit
our choice to a representation in t e r m s of either allophones o r features.
On the contrary i t s e e m s reasonable that a decoding i n t e r m s of phonetic c l a s s e s (distinctive features) i s paralleled by a direct attempt of
all ophone decoding.
I a m thus more i n favor of a parallel analysis of
features than of a s e r i a l analysis with a succession of decisions.
The question whether a translation to equivalent motor instructions
precedes phonemic identification i s not important.
Of g r e a t e r impor-
tance i s to study what aspects of known features a r e of p r i m a r y importance to perception (I1)
.
It i s observed that i n some instances i t i s the
temporal contrast of two successive sound segments (e.g.,
stops,
l a t e r a l s , nasals) r a t h e r than the inherent quality of each of the segments
that evokes the particular auditory sensation associated with the particu l a r c l a s s (feature). The pertinent problem i s to find what t r a n s f o r m s
we should apply to the speech wave data i n o r d e r to extract the information bearing elements that operate i n speech perception.
Attempts to
avoid the s e a r c h for auditorily relevant sound characteristics by an
uncritical acceptance of the view that perception is m e r e l y a reconstruction of the production does not appear v e r y fruitful.
Given a distinction, let u s say between the phonemes /k/ and /t/ o r
between the phonemes /g/ and /d/,
the procedure would f i r s t of a l l be
to study a number of minimally contrasting p a i r s i n different vowel
contexts.. As a precaution for securing that we have not lost any significant information i n the acoustic specification of a feature, i t i s advisable to consider the underlying production events and apply known
transformation rules.
Second formant loci a r e known t o be effective
c u e s for discriminating place of articulation but they a r e not sufficient.
Indeed, a vowel context can be found such that the F2 transition is
almost the s a m e f o r /k/ and /t/ o r f o r /g/
1 .
and /d/.
This is the vowel
This ambiguity i n F2, observed by Liberman 1957 ('')(see
Fig.
3 in his paper) led him to conclude that "articulation goes one way and
the speech wave another way" and since we c a n discriminate
/ d a / perception follows articulation, not the speech wave.
lga/
from
The a m -
biguity is resolved i f we add the information of the third formant and
the b u r s t portion.
In a recent experiment ( s e e Section 111. A) by Fant, Lindblom and
de s e r p a - ~ e i t g owe low-pass filtered Swedish and English stops a t
2000 c / s .
The result was that o u r bilingual speaker, now serving
a s a listening subject, classified a l l his filtered Swedish /t/ a s /k/ and
his /d/ and /g/
and likewise almost a l l his B r i t i s h English /t/ a s /k/.
Since the filtering removed the upper p a r t of the b u r s t s p e c t r u m of the
dentals t h e r e remained v e r y little difference between the unfiltered
/k/ and the filtered /t/.
This experiment supports the view that the effective acoustic feature
i s the combination of the second and the t h i r d formant and the b u r s t
concentrating the energy i n the palatals and v e l a r s i n a frequency region
at o r above the effective upper formant of the vowel, s e e Fant (13)
pp. 217-218.
My formulation above of the perceptually relevant c h a r -
a c t e r i s t i c s of compactness i s not claimed to r e p r e s e n t the final answer.
Also I do not claim that a formulation found effective f o r contrasting
/k/ and /t/ must be exactly the s a m e a s that contrasting
/n/
and /n/.
The human b r a i n s u r e l y has a sufficiently l a r g e capacity to identify / r ~ /
a s different f r o m /n/ i n a space s e p a r a t e f r o m that discriminating /k/
and /t/.
It s e e m s quite likely that feature analysis i s an important
aspect of speech recognition, adding to the economy of the system.
On the other hand, i t s e e m s likely that perception works according to
m o r e redundant principles than those guiding linguistic analysis. Here
I believe Roman Jakobson a g r e e s with me that speech perception i s
not m e r e l y the neural decoding of his twelve-features.
Recent studies of Chistovich et a1 i n Stockholm have revealed interesting results concerning vowel perception. These studies (14)
support the view of a categorical perception of isolated vowels and the
existence of a n FIFZquantization i n accordance with our working principles of acoustic phonetic analysis.
Swedish vowels
The Swedish vowel system i s generally presented i n t e r m s of 9
long and 9 short phonemes. Special p r e - r variants of the open unrounded and rounded front vowels i n Swedish orthography a and o stand
out a s well recognized allophones of an especially "open" quality.
The following phonemic symbols r e f e r r e d to a s the STA alphabet
will be used for the 9 vowels.
Approximate IPA symbols a r e included
a s examples of phonetic values i n contexts other than before [ r ]
.
Index 1 stands for long vowel, 2 f o r short vowel, and no index implies
either long o r short, i. e, length distinction omitted.
Group
I Back vowels
I1 Unrounded
front vowels
I11
front vowels
STA
O1
i
1
e
1
al
Y1
'1
IPA
STA
u:
O2
i:
i
e:
ez
e - E
E:
Z2
E
Y:
u:
Y2
2
'2
IPA
I
Y
0
In my f i r s t attempt of ordering ( I 5 ) I chose to oppose group I to
group I1 and I11 i n t e r m s of the grave/acute distinction and group I11
was naturally opposed to group I1 i n t e r m s of the f l a t / ~ l a i nfeaturc.
So f a r my views have not changed.
However, a division within the
t h r e e major groups i n t e r m s of articulatory opening, i. e. , the compactness feature classifying not only / b l / and / e l / but also by a n
intermediate
+- degree of
compactness did not conform well with phonetic
facts, the F1 of /u1/ being the same a s the Fl of /Y1/.
Also within
the short vowels the F I of /uh/ was generally not significantly different
from the F1 of /e2/. I therefore chose in l a t e r works (5)(16)(17) to
use three degrees of flatness assigning an intermediate value of flatness
to the phonemes /y/ and /ti/
and to the
honeme /u/ the maximal degree
Pl8)
This conforms well with
of flatness, a s suggested by Malmberg
the flatness c r i t e r i a of low F1
.
+ F2 + F3 and the
extreme degree of
In group I either flatness
lip-rounding i n the /u1/ and /ol/ phonemes.
o r compactness i n three levels can be used for the further division.
Feature
1
1
grave
I
o
+
3
a
+
+
!
I
i
e
1
y
u
o
1-
-
-
-
-
-
-
-
+-
+
+
+
-
-
+
I
I
I
+
flat
1
-
t
t
compact
i
-
-
t
f
An alternative solution which perhaps comes closer to the vicws of
Roman Jakobson would be the following.
In group I flatness i s used to
u) and /3/ (IPA o) f r o m /a/,
separate phonemes /o/ (IPA -.
opposed to /o/ by the g r e a t e r compactness.
and /&/
is
Now in group I1 and 111
the opposition ~ h a r ~ / ~ l ias i introduced
n
to differentiate /i/ and /e/
from /1/ and /y/ and /6/ from /u/.
*
The criterion of sharpness i s
higher F2 and F j everything else being equal.
In this sense sharpness
i s given a function similar to that of diffuseness in Halle' s vowel
analyses
(9). By this arrangement one avoids introducing three degrees
of flatness o r compactness and four binary features specify the entire
system, a s shown on next page.
*
Suggested by S. ohman, personal communication.
P
a
i
e
a
y
u
a
t
+
t
-
-
-
-
-
-
flat
t
t
-
-
-
-
t
t
t
compact
t
t
-
f
+
-
-
t
sharp
0
0
t
t
-
t
-
t
Feature
o
1
grave
I
I
0
In t h i s solution /u/ d i f f e r s f r o m
/a/
!
Within the
by two f e a t u r e s .
s y s t e m of s h o r t vowels /ti2/ i s opposed t o /uZ/ p r i m a r i l y by t h e s h a r p w h e r e a s within t h e long vowels
ening
/u
/
1
/al/
differs from
p r i m a r i l y i n t e r m s of g r e a t e r c o m p a c t n e s s .
A s m a l l variation of the s y s t e m allowing a c l o s e r relational c o r -
respondence of /e/ t o /u/ is t h e following:
--
Feature
jgrave
o
I t
a
+
+
I
flat
I
compact
e
2
Y
u
a
I
i-
-
-
-
-
-
I
-
i
t
t
t
-
t
1 -
-
t
-
t
t
t
-
- 1 -
t
+ I i
I
I
sharp
0
0
0
-
i
i
+
1
-
Concluding r e m a r k s
T h e concept of distinctive f e a t u r e s r e q u i r e s a n i n t e g r a t e d view of
speech p a t t e r n s a t e a c h stage of the s p e e c h communication chain and
should not b e confused with the concept of cue which is l i m i t e d t o a
d e t a i l of the pattern.
S e v e r a l c u e s make up a feature.
The p r i m a r y
a i m of t h e phonologically oriented a n a l y s i s s u c h a s the one underlying
m y w o r k together with Jakobson and Halle was t o m a k e a n integrated
formulation of t h o s e c u e s which a r e i n common i n a l l contexts w h e r e
the f e a t u r e o p e r a t e s .
T h e r e is a potential d a n g e r t h a t t h i s demand f o r
g e n e r a l i t y dilutes the d i s c r i m i n a t o r y power of the formulation i n any
specific context.
In s p i t e of frequent r e f e r e n c e to contextual b i a s i n o u r
w o r k misunderstandings have o c c u r r e d , especially among t h o s e entering
the field of phonetics.
T h e generalized formulations of o u r f e a t u r e s p r i m a r i l y s e r v e a
function i n s u m m a r y views of a phonological s y s t e m . If we a r e interest&n
s p e e c h communication m e c h a n i s m s we m u s t t a k e into con-
s i d e r a t i o n all perceptually important information i n e a c h specific
context w h e r e t h e f e a t u r e o p e r a t e s . Contextual b i a s r e p r e s e n t s " r e dundant f e a t u r e s " f r o m a r e s t r i c t e d s t r u c t u r a l point of view.
In m y view t h i s terminology is conceptually misleading when d i s cussing the perception of speech.
What is redundant f r o m a n a r r o w
metalinguistic point of view m a y c o m p r i s e e s s e n t i a l s of the actual code.
T h i s amounts t o saying that the m a j o r allophones of a language would
have a perceptual r e a l i t y a s individuals w h e r e a s we need not anticipate
t h a t the b r a i n p o s s e s s e s a catalogue of a l l possible contextual v a r i a tions of a f e a t u r e o r of a phoneme.
T h e concept of the distinctive
f e a t u r e a p p e a r s t o be a r e a l i t y i n perception.
Most of t h e m a n n e r of production f e a t u r e s a r e c l e a r cut and the
s a m e i n a l l contexts.
It is outside m y competence t o d i s c u s s the univer-
s a l i t y of f e a t u r e s except i n a v e r y g e n e r a l s e n s e .
I consider the dis-
tinctive f e a t u r e s t o be a v e r y powerful concept b a s e d on the n a t u r a l
c o n s t r a i n t s of o u r speaking mechanism.
been operating with is not unique.
T h e s e t of f e a t u r e s we have
Alternative solutions, affecting
s o m e of the f e a t u r e s , a r e possible within s o m e languages a s well a s
when dealing with s i m i l a r phenomena i n two different languages.
The universal acceptance of the b a s i c concept of f e a t u r e a n a l y s i s
i s much due t o the inspiring w o r k of Roman Jakobson.
References
( i ) F a n t , G. : "Descriptive a n a l y s i s of the acoustic a s p e c t s of speech1!
Logos 5 (1962), pp. 3-17.
-
(2) F a n t , G. and Lindblom, B. : "Studies of m i n i m a l s p e e c h sound
units", STL-CPSR 2/1961, pp. 1 - 1 1.
(3) MArtony, J.: "The r o l e of formant amplitudes i n s y n t h e s i s of
n a s a l consonants", STL-GPSR 3/1964, pp. 28 - 3 1.
(4) Jakobson, R. and Halle, M. : Fundamentals of Language
( ' s-Gravenhage 1956).
(5) F a n t , G. : f i S t r u c t u r a lclassification of Swedish phonemes",
STL-CPSR 2/1960, pp. 10-15.
(6) Jakobson, R. and Halle, M. : " T e n s e n e s s and l a x n e s s u , Solccted
Writin
I by K. Jakobson (' s-Gravenhage 1962)' pp.
&
cant. next page
( 7 ) Malecot, A. : "Mechanical p r e s s u r e a s an index of ' F o r c e of
Articulation' It, Phonetica 14 (1966), pp, 1 59- 180.
-
(8) L i s k e r , L. and Abramson, A. S. : "A cross-language study of
voicing i n initial stops: acoustical m e a s u r e m e n t s f t , Word
20 (1964), pp. 384-422.
-
(9) :Halle, M. : llPhonology i n generative g r a m m a r " , Word 18 (1962),
pp. 54-72.
(10) Kozhevnikov, V. A. and C histovich, L. A, : Speech: Articulation
US Dept. of C o m m e r c e ,
1965).
(1 1) F a n t , G. : "Auditory patterns of speech", t o be publ. i n the P r o c .
of the Symp. on Models f o r the P e r c e p t i o n of Speech and
Visual F o r m , Boston, Mass., Nov. 11- 14, 1964.
(12) Liberman, A. M. : "Some r e s u l t s of r e s e a r c h i n speech perception",
J. Acoust. Soc. Am. 29 (1 957), pp. 358 -368.
(13) Fant, G. : Acoustic Theory of Speech Production (' s-Gravenhage
1960).
(14) Chistovich, L. et a l . : ''Mimicking of synthetic vowelstf,
STL-GPSR 2/1966, pp. 1-18;' and Chistovich et a1. :
"Mimicking and perception of synthetic vowels, P a r t II",
STL-GPSR 3/1966, pp. 1-3.
(15) F a n t , G.: "Phonetic and phonemic b a s i s f o r the transcription of
Swedish word material", Xcta Oto-Lar . , Suppl. 116,
(1954), pp. 24-29.
(16) Fant, G. : "Acoustic analysis and synthesis of speech with applications t o Swedish", E r i c s s o n Technics 15 (1959), pp. 1-106.
-
(17) Fant, G. : "Modern i n s t r u m e n t s and methods f o r acoustic studies
of speech" , Acta Polytechnics Scandinavica 246/1958, 8 4 p.
(18) Malmberg, B.: "Distinctive f e a t u r e s of Swedish vowels. Some
instrumental and s t r u c t u r a l d a t a u , F o r Roman Jakobson
(' s-Gravenhage 1956), pp. 3 16-321.
(1 9) Jakobson, R. , F a n t , G. , and Halle, M. : " P r e l i m i n a r i e s t o speech
analysis. The distinctive features and t h e i r c o r r e l a t e s i t ,
Acoustics Laboratory, Massachusetts Institute of Technology,
Technical Report No. 1 3 (1 952), 58 pp. ; 4th printing publ.
by The M. I. T. P r e s s , Cambridge, Mass. 1963.
(20) F a n t , G., Lindblom, B . , and de s e r p a - ~ e i t c o ,A.: "Consonant
confusions in English and Swedish a pilot study", section
111. A i n this i s s u e of STL-C-PSR,
-
Download