THE ROLE OF consonm cousmcnvss AND ,
TRANSITIONS m THE PERCEPTION OFONGOING SPEECH

Thesis for the Degree of Ph. D,
MICHIGAN STATE UNIVERSITY
ALLAN KARL BIRD
19‘22

- smut-r" V-o-I'fl'f
‘uy ‘s

y ‘ - 1" ,=
LIBK'MI‘; 13." °

t-

Michigan L at l

] Universi : y f

1 ._4.’

I
I
This is to certify that the

thesis entitled

THE ROLE OF CONSONANT CONSTRICI IVES AND
TRANSITIONS IN THE PERCEPTION
OF ONGOING SPEECH

presented by

Allan Karl Bird

has been accepted towards fulfillment
of the requirements for

PhD degree inAUdiOlOfl and
Speech Sciences

Major professor

 

 

 

‘Ibate //’/0’7L/

0-7639

 

 

ABSTRACT
THE ROLE OF CONSONANT CONSTRICTIVES AND
TRANSITIONS IN THE PERCEPTION
0F ONGOING SPEECH
by
Allan Karl Bird

Considerable research has been undertaken in an
effort to determine the critical acoustical elements
involved in the perception of Speech. Almost all of the
studies to date have used either nonsense syllables or
single words as the Speech sample. The major research
has proposed two major sources for identification of
consonants. One of these is the consonant constriction
and the second is the transition or the shift from the
constriction to the vowel. Recently the question has been
as to whether the perception of ongoing Speech is similar
to the perception of syllables or-words.

This study was designed to answer the following
questions:

1. Does the use of context provide cues for the
perception of words which have been acoustically
altered by the removal of various phonetic segments?

2. How much intelligibility is maintained because of

co-articulation when only the vowel sound is heard
both in context and out of context?

Allan Karl Bird

3. Is there a significant difference in the perception
of contextual speech when transitions are removed
and when constrictive elements are removed?

#. If significant differences occur in ongoing
contextual Speech, do the same differences also
appear in non-contextual sentences?

Eight experimental tapes were made. Four of the
tapes used sentences of five words each. The other four
tapes had the same 30 words in random order. The tapes
within an order were altered as follows: one had all of
the transitions missing: one had all of the constrictive
portions missing; one had the constriction and transition
portions missing; and one had the transitions and vowels
missing.

Different groups of 10 listeners each heard each
condition and every person in the group was asked to write
down the words which they heard.

A statistical analysis of the obtained data was
done to determine whether differences in results were
significant. The analysis of variance was used and an
F-ratio score indicated a significant difference between
orders of presentation. differences in segmentation
effect and interaction at the 0.01 level. Graphs were
made to demonstrate what the interaction effect was.

A post hoc comparison was performed showing the
significant seven crosswise comparisons to be contributing

to the overall significance. Of Special importance was

the statistically significant comparison between the

Allan Karl Bird

meaningful sentence with transitions omitted and the
meaningful sentence with constrictions omitted.

A confusion matrix was constructed to examine
the type of perceptual errors which were made under each
of the orders of presentation. It was found that. in
classifying errors as errors of place. voicing and
affrication. differences in these three perceptual
categories did exist.

The vowels alone were found to contain little
information although the subjects did appear to apply
some linguistic constraints to the material as shown by
the high number of‘thg's that the subjects guessed. In
terms of five word sentences, one would eXpect to find a
the in each of the sentences.

All of the findings of the study suggest that
contextual perception is a different process than when
one hears the words only as single entities. Further.
it appeared that the segments of the Speech signal

removed seemed to make a difference.

Accepted by the faculty of the Department of
Audiology and Speech Sciences. College of Communication
Arts. Michigan State University. in partial fulfillment
of the requirements for the Doctor of PhilosOphy degree.

Thesis Committees CoéDirector
Leo V. Deal. Ph.D.

Co-Director
Oscar Tosi, Ph.D.

 

Herbert J. Oyer. Ph.D.

 

Gordon Aldridge. Ph.D.

THE ROLE OF CONSONANT CONSTRICTIVES AND
TRANSITIONS IN THE PERCEPTION
OF ONGOING SPEECH

By
Allan Karl Bird

A THESIS

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Audiology and
Speech Sciences

1972

TABLE OF CONTENTS

LIST OF TABLES . . . . . . . . . . . . . . . . . .

LIST OF FIGURES O O I O O 0 O O O O I O O O O O 0

Chapter
I.

INTRODUCTION . . . . . . . . . . . . . . .

Statement of the Problem and Purpose of
the Study 0 I O O O O O I O 0
Importance of the Study . . . . . . . .
Definitions of Terms Used . . . . . . .

II. REVIEW OF THE LITERATURE . . . . . . . . .

Theories of Speech Perception . . . . .
Distinctive Features Theory . . . .
Motor lheory of Speech Perception . .
Coarticulation Theory of Sound
Production . . . . . . . . . . . . .
Phrase Theory . . . . . . .
Contextual Probability Theory . . . .

Studies Related to the Perception of

Consonants . . . . -
Studies Related to the Import ace of
the Constriction in Speech
Perception . . . . . .
Studies Related to the Importance of
the Transition in Speech Perception .

Studies Relating to the Role of Context

in Perception . . . . . . . . . .

Studies Relating to Time Compressed and

Interrupted Speech . . . . . . . . . . .

III. EQUIPMENT, MATERIALS. SUBJECTS. AND

PROCEDURES . . . . . . . . . . . . . . . .

Equipment . . . . . . .
Materials . . . . .

Preparation of Materials
SUbjeCtS o o o o o o o o

O C I C
Q C O O
O O O O
O O O I
O I O O
O C O 0
O O O O
1‘ I I .

ii

Procedures . . . . . . . . . . . .
Analysis of the Data . . . . . . . . .

IV. RESULTS

AND DISCUSSION . . . . . . . . . .

Results of Comparing Conditions . . .

V. SUMMARY.

Effects of Context . . . . . . . .
Effects of Segmentation . . . . .
Amount of Information Contained in
Vowels . . . . . .

Perceptual Differences in Ongoing
Contextual and Non-contextual

SPBBChoooeoooooooooo

CONCLUSIONS. AND RECOMMENDATIONS

Summary-o.............

00113111310118 0 o o o o 0

Recommendations for Further
Research . . . . . . . . . . . . . . .

BIBLIOGRAPHY .

APPEND Ix I C C O O O O O O O O O O O O O O O l 0

iii

84
91

91
9A

96
98
106

LIST OF TABLES

Table Page
1. Contents of the Experimental Tapes . . . . . . 6h

2. Group Means and Standard Deviations of Ten
- Subjects for Each of Eight Test Conditions
Ranked in Descending Order . . . . . . . . . . 70

3. Analysis of Variance . . 3 . . . . . . . . . . 71

4. Differences in Averages Contrasting Each of
the Conditions with all other Conditions . . . 75

5. Number of Correct Identifications for Each
of the 30 Words under Each of the
Experimental Conditions . . . . . . . . . . . 76

6. Soderberg's Infermation Values of Words in
Clauses Classified by Length . . . . . . . . . 78

7. Total Number of Correct ReSponses for Each
of the Five Sentence Positions . . . . . . . . 79

8. Classification of the Consonants Used to
Construct the Confusion Matrices . . . . . . . 87

9. A Confusion Matrix for Meaningful Sentences
with Transitions Omitted or Constrictions
Omitted and First Order Approximations with
Transitions Omitted or Constrictions
Omitted . O O O I O O O O O O O 0 O O O O O I 88

iv

LIST OF FIGURES

Page

Schematic Diagram Showing the Place of an

Incoming Perceptual Pattern in a Phoneme

Class with the Aid of Statistical Information

Based on the Immediate Past of the Message

(Fry! 196“! p0 16) o o o o o o o o o o o s o o 17

An Expanded Model Showing the Place of an
Incoming Perceptual Pattern in Speech
Perception . . . . . . . . . . . . . . . . . . l9

Schematic Diagram of Identification
Procedure 0 O O O O D O O I I O O O O O O O O 27

Diagram showing Preparation of Tapes to Form
the Eight EXperimental Tapes used in this
Study0.006060000000000... 63

Standard Deviation and Mean Number of Words
Correctly Identified for each Condition Based
on 10 Judges in Each Group . . . . . . . . . . 73

Confusion Errors for Contextual and
Non-ContBXtual SpeeCh o o o o o s o o o o o o 89

CHAPTER I

INTRODUCTION

Of all the abilities which man possesses. the one
which gives him his most distinctive nature is the ability
to use language as a tool of communication. It is little
wonder that an active interest in Speech and language
has been present in man's search for Self understanding.
The question of language and Speech has been of great
interest to the audiologist. the linguist, the psycholo-
gist, and the speech pathologist. From the communication
standpoint. there has been an active interest in the
information-bearing elements of Speech. Each person's
interest in communication is concerned with knowing and
understanding what the elements of the oral language
code are which allow man to communicate with man. How
one knows what another person has said and what allows a
person to say it are questions of extreme interest.

Traditionally. the study of speech perception
has examined the phoneme as the basic unit of Speech;

and the inquiries on speech perception have been directed

at discovering what the elements were which allowed the
listener to recognize the differences in phonemes. A
myriad of studies have been conducted looking at the
phoneme as the unit of Speech perception.

Recently. however. the use of the phoneme as the
unit of perception has been questioned: and it has been
suggested that perception of speech actually occurs
along more meaningful lines with more meaningful units.
Ladefoged and Broadbent (1957) suggest that the study of
the isolated phoneme is probably of only limited value
in understanding the perception of Speech. Brady (1960)
and Brady. House and Stevens (1961) found an increasing
volume of evidence becoming available that the perception
of ongoing speech and language entails Operations that
differ markedly from the perception of other acoustical
events. Mattingly and Liberman (1970) have stated that
analysis has shown that the same acoustical cues carry
information for different phonemes and. furthermore, that
the same phoneme is often signaled by differing acoustical
information.

Another problem in the study of the isolated
phoneme has been pointed out by Hendrik (1965). He
states that it is only in the study of the isolated
signal that the listener has all possible cues available
to him. He says:

phonemic system only pertains to the exceptional and
rather abnormal situation of isolated words enunciated

by one particular speaker. Application of the

phoneme system to other situations. including the
normal. is an as yet unproven extrapolation. (Hendrik,
p. #26)

Harris (1953) conducted a study designed to look
at the effects of using the phoneme as building blocks
for ongoing Speech. Consonant phonemes and vowel phonemes
were joined together to form words. The resulting
”ongoing Speech” was found to be very unnatural sounding
and highly unintelligible.

Lisker and Abramson (1967) found that certain
cues. which on the basis of acoustical analysis allowed
for the distinction between sounds in single words. were
much less useful in ongoing Speech. They suggested that
in ongoing speech the cues which are available to the
listener in the isolated word either are blurred or not
present at all. They further hypothesized that correct
perception was still possible because of the linguistic
and contextual constraints which were available in
ongoing speech. Therefore. they argue that even though
the acoustical one was not clear. the slack created was
compensated for by context. Fry (196R) has stated that the
most important perceptual information that a listener
can have about an utterance is that it makes sense. The
role of context appears to be one of considerable

importance. Context can perhaps best be defined using

the concept of information theory:

If we define a message as a sequence of code symbols.
in the present case. phones. the accuracy of decoding
a message of a given length will be greater than the
accuracy of encoding the individual symbols if and
only if there are conditional probabilities that are
not all equal (Strauss. 1950. p. 709).
This idea can perhaps best be demonstrated by looking at
the graphic symbols of our language. If. for example.
a person is asked to spell a word and is given no
information as to what the first letter of the word is.
he must randomly guess what the first letter is.
Obviously even in this random guessing. the probability
of certain letters occuring is greater. Assume now that
the person has guessed that the first letter is ”q." If
the person has a knowledge of the English language. the
chances of the person correctly guessing the next letter
are excellent. Most people knowledgeable of the language
would guess the next letter to be "u.” Even when the
case is not as obvious as the previous example. the
speller is able to limit his alternatives by prior
information; There are. as a result of certain learned
or innate rules. constraints placed upon our selection.
The suggestion has then been made that the idea
of context might also apply to the perception of the
Spoken code. Three types of context have been suggested
which may aid the listener in the comprehension of
ongoing Speech (Ladefoged. 1967). The first type of

context which he suggests might be of assistance to the

listener is the ”context of the situation.” These are

events which occur at the time of the utterance but are
not an intrinsic part of them. For example. a person
watching a sporting event might assume that someone
addressing him may be discussing the Sporting event. The
second type of cues are the linguistic cues of the language
which the listener might apply. For example. the person
hearing "he say” might well be able to supply the "S” to
the utterance even though he had not heard it since he
knows some linguistic rules regarding singular and plural.
The third type of contextual cues are the acoustical

cues.

Lieberman (1967) and Fry (196%) have both
suggested that the listener relies heavily on the
linguistic cues for Speech perception.

Black (196#) has pointed out that since English
words are not nonsense syllables. they cannot be randomly
assembled. There are limitations which are placed on the
alternatives from which a listener might select. Reed and
Wang (1961) were able to demonstrate that contextual cues
override acoustical cues for perception.

While several people have alluded to the fact
that perception of ongoing Speech varies from the
perception of the isolated word. only a slight beginning
has been made to investigate this question. Warren (1970)
and Warren and Obsek (1971) found that a listener was able

to supply a phoneme even when it was not present by

Simply removing one sound from a sentence. They called

this ability phonetic restoration.

STATEMENT OF THE PROBLEM AND
PURPOSE OF THE STUDY

9 This study seeks to answer some basic questions
regarding the perceptual ability of the speaker.
Specifically. the study is designed to determine whether
those elements of Speech perception which have been
suggested as being important in the perception of the
isolated phoneme carry Significance in ongoing Speech.

The following questions were formulated to define
the research:

1. Does the use of context provide cues for the
perception of words which have been acoustically

, altered by the removal of various phonetic
segments?

2. How much intelligibility is maintained because of
co-articulation when only the vowel sound is heard
both in context and out of context?

3; Is there a Significant difference in the perception
of contextual speech when transitions are removed
and when constrictive elements are removed?

4. If significant differences occur in ongoing

contextual speech. do the same differences also
appear in non-contextual sentences?

IMPORTANCE OF THE STUDY

It has been hypothesized that context is
important in the perception of speech. Further it has

been suggested that various portions of the Speech signal

carry varying degrees of information (Dukelskiy. 1967).
This question has not been examined in ongoing Speech.
It is the purpose of this study to attempt to provide
some basic information relative to phonemic restoration
ability of the listener and to the present knowledge
about the information-carrying units of Speech.

The information obtained from this study is viewed
as having practical application in the area of sound
discrimination and information processing as well as
having possible significant impact upon future research in

Speech perception.
DEFINITION OF TERMS USED

Constriction;--The term constriction or
constrictive element as used in this paper refers to the
portion of the consonant Which represents the quasi ”steady"
part of that consonant. It is that portion of the
consonants which Potter. Kepp and Green (19h?) refer to as
Spikes and fills. The more generic term "constriction”

was selected for use in this paper.

Context.--For the purposes of this paper. context
has reference to the effect of linguistic restraints upon
the perception of Speech. It refers to the meaning the
listener might gain as a result of syntactical. semantic,

and morphological rules which the listener applies.

First order approximation.--A first order
approximation sentence for the purposes of this study
refers to a sentence which is constructed out of
randomly selected words. The 30 words from the meaningful
sentences were randomly assigned to the six first order
approximation sentences by use of random number tables;

These sentences have no contextual information.

'Phonetic restoration.--Phonetic restoration is a
term coined by Warren and Obusek (1971). It refers to
the ability of the listener to supply or restore
perception of the missing phonetic elements of a Speech

signal;

Segmentation.—-Segmentation has reference to the

process of temporally removing portions of the phoneme.

Steady state vowel.--The steady state vowel is
the portion in which the amplitude spectrum of the vowel

is quasi constant.

Transition.--In ongoing speech. there is a shift
which occurs as the flow of Speech moves from one phoneme
to the next; In the case of a consonant-vowel syllable
this would be a Shift from the constriction to the steady

state vowel.

CHAPTER II

REVIEW OF THE LITERATURE

This chapter contains a discussion of the theories
of speech perception. a review of literature relevant to
the perception of the consonants of the English language.
a review of literature on the role of context in the
perception of Speech. and a discussion of time compression

and speech interruption.
THEORIES OF SPEECH PERCEPTION

A variety of Opinions has been expressed
concerning the nature of the perception of speech. The
earliest idea concerning the speech code was that Speech
was a series of non-overlapping phonemes which were each
self contained. It was assumed that traditionally Speech
was analyzed as a series of concatenated phonemes (Pant.
1962). Many of the early studies related to Speech used
this model.

Recently new models have been preposed in an
attempt to explain the way in which man perceives Speech.
A brief discussion of these theories follows: Distinctive

Feature Theory. Motor Theory of Speech Perception.
9

10

Co-articulation Theory of Speech Perception. the Phrase
Theory of Perception. and the Contextual Theory.

Distinctive Features Theogy

It has always been considered that the phoneme
was the smallest unit of speech which could be
differentiated. 0n the basis of analysis of spectral
representations of sound. Jakobson. Belle. and Pant (1961)
were able to suggest that smaller perceptual and production
units existed. These smaller units they called
”distinctive features." They suggested that the
individual phonemes were composed of bundles of smaller
features. Twelve distinctive features were described by
them. They further postulated that each of the features
was perceived in a dichotomic manner. i.e. each feature
was perceived as either being present or not being present.
In the case of a phoneme. the listener decides not to
what degree a feature is present but whether it is present
or not. It was felt that in this manner the process of
perception was considerably simplified. Other systems
of distinctive features have been suggested by Chomsky
and Halls (1968). Wicklegren (1966). and Miller and Nicely

(1955).

Motor Theogy of Speech Pepception
Motor theory suggests that the perception of Speech

is based upon awareness of articulatory movement of the

11

listener. attempting to produce unconsciously the same
Speech he is receiving. It suggests that the articulatory
movements and their kinesthetic feedbacksof the listener
mediate between the incoming acoustical events and the
perception of Speech. This theory was outlined by

COOper. et a1 (1952). Mattingly and Lieberman (1970)
pointed out the same acoustical cues sometimes Signal the
presence of different phonemes while differing acoustical
cues are sometimes present to make the same phoneme. They
suggest that this makes clear the need for some mediational
factors in the perception of the speech other than the
acoustic information. They point out in the acoustical
analysis of Speech that different articulatory patterns
are often represented by similar acoustical events and yet
a listener is able to make the prOper perceptual decision.
Further. similar articulatory patterns are often signaled
by varying and diverse acoustical representation. Still
the listeners hear the correct sound. Lane (1965) points
out that further support is lent to this theory because
the perception of sounds more closely parallels the

articulatory changes than it does the acoustical changes.

ngpticulation Theory_of Sound Production

C. M. Harris (1953) conducted a study that
recorded consonant-vowel syllables. The consonants were
then spliced away from the original vowels with which

they were recorded and spliced with new vowels. It was

12

found that this yielded unnatural and unintelligible
Speech. Harris suggested on the basis of his study that
the use of individual phonemes as building blocks for
Speech was most likely an unrealistic assumption. He
suggested that there was some type of interaction between
sounds in combination.

The coarticulation theory of Speech assumes that
one or more of the sound features characterizing a sound
segment may extend over several temporal segments of Speech.
This idea was suggested by Peterson and Barney (1952) and
Hughes and Halle (1956); they concluded that it is not
usually possible to find segments of the acoustical signal
which can be uniquely identified as single and discrete
entities during the flow of normal speech production. They
speculated that this might mean that the unit of Speech
perception is larger than the phoneme. Stevens and House
(1963) found evidence that the formant frequencies of
vowels are modified in a systematic way by the consonantal
contexts in which they occur and noted that the effects of
coarticulation upon the vowel could often be found
extending over one phoneme in either direction.

Stevens. House and Paul (1966) used Consonant-
Vowel—Consonant utterances and found that the formant
structures of certain vowels were modified by the
consonant environments in which they were placed and

found evidence of co-articulation over two phonemes.

13

This view was further supported by Ohman (1966) using
Vowel-Consonant-Vowel syllables.

Two basic theories of coarticulation have
evolved. One stresses that the perceptual unit is
supra-phoneme in size. The other is explained on the
basis of phoneme-sized units.

Several advocates of the supra-phoneme unit have
proposed theories. Daniloff and Moll (1968) suggested
that the unit of perception was a unit the size of the
Consonant-Vowsl-Consonant. They felt that this perception
was not affected by word boundaries. Kozhevnikov and
Chistovich (1965) agree that the unit of perception is
larger than the phoneme but feel that the perceptual
unit is the ConsonanteConsonant—Vowel.

Hanks (1966) has suggested that the unit of
perception is still phoneme-Sized but that the high
level perceptual and production mechanism is responsible
for the coarticulation effect. The process is one of
higher level look-ahead and scanning.

Numerous studies have been conducted which have
shown that in the production aspect of speech.
coarticulation does occur (Fujimura. 1961: M011. 1960.

1962: Daniloff and Moll. 1968: Amerman. et al.. 1970).

Phrase Theory
It has been argued by several that the Spoken

language is decoded by the listener in word groups called

l4

phonemic clauses (Garrett. 1965: Garrett. Bever and

Fodor. 1966: Fodor and Bever. 1965: Dittmann and
Llewellyn. 1967: Fodor. Bever and Garrett. 1968: Bever.
Lackner and Kirk. 1969: Bever. lackner and Stolz. 1969).
The general method of investigation was to place a click
at various locations within a sentence. A point was
marked at the place of the phrase juncture of the sentence.
This position was called the zero point. Clicks were then
recorded at the zero point and at points one syllable.

one word. etc. away from the zero point in each direction.
The findings were that the click or noise was generally
mislocated and usually in the direction of the phrase
juncture or zero point. Based upon the idea that a
perceptual unit resists interruption. the general
conclusion of the investigators was that since the noise
or click migrated toward the phrase juncture. this indicated
that the unit for decoding was the phrase or phonemic

clause.

Contextual Probability Theogy
Lieberman (1967) prOposed a model for the

perception of ongoing speech. He suggests that the
entire context of the sentence plays an important part in
determining how essential any particular cue is for

the perception of the message. As Black (1962) pointed
out. words are not nonsense syllables. Words in any

language have sounds that are not randomly assembled.

15

Brown and Hildum (1956) ran a study in which they
constructed three sets of words. The first group of
words were highly unfamiliar words of the English
language. the second set were non-English words and the
third set were words whose phonetic context could not
occur in English. They found that the English:words were
remembered by native Speakers of English about 69 percent
of the time. the English nonawords were recalled with
46.8 percent accuracy. while the non-English.words were
only recalled 9.7 percent of the time. They concluded
that the perception of speech is directed by knowledge
of the sequential probabilities.

Reed and Wang (1961) also provided evidence of
this probability phenomenon. Three words were recorded
in their study. They were bat. pat. and gpgg. The /S/
was then removed from the word gpgg and placed before pg;
and pgi. Subjects perceived both.§_plus pg: and‘g plus
‘pgp as being gpgp. The non-English combination ofI§p
was not perceived even when actually present. They
concluded that if the analysis were acoustical or motor
rather than contextual probability, the listeners would
have heard bat.

Harris (1960) has suggested that there are
enough cues furnished by restriction placed on utterances
by syntax and context to give the listener a good idea

of the missing parts.

l6

Lisker and AbramSon (1967) conducted an
eXperiment using the analysis of speech in isolated
words and in ongoing Speech. It was found that the time
from the onset of the burst to the onset of periodicy was
sufficient to classify the consonants as being voiced or
voiceless. However. when the voice onset time was
evaluated in ongoing Speech. it became useless in
predicting voiced and voiceless. They offered two
possible explanations as to what happens. The first
possibility suggested was that there is a general
blurring of the distinctiveness of the sound. This
blurring is compensated for by the linguistic
constraints. thus taking up any slack in the intelligi-
bility. The second alternative possibility offered by
them was that there are certain cues in deliberate
speech of single words which are merely redundant.
Studies by Pickett and Pollack (1963) and Fry (1964)
suggest that the blurring and linguistic constraint is
the correct alternative.

Lieberman (1967) suggests that the perception
of ongoing speech is based on both contextual and
acoustical cues in combination. It is suggested that a
listener first performs a preliminary analysis of
the speech signal on those acoustical cues which are
available to him. When the acoustical information has

been processed. the listener postulates a hypothesis

17

regarding what the message might possibly be. This
hypothesis is then referred to a process where it is
tested using those semantic and syntactic constraints
which exist in the listener's language system. If the
hypothesis appears tenable in light of the constraints.
the listener interprets the message as having been heard.

Fry's model (l96h) shown in Figure 1 illustrates this

 

 

 

 

 

 

viewpoint.
Phonemes
...kAm
‘ PHONEME
STATISTIC

 

 

 

 

 

 

Figure l.--Schematic diagram showing the place of an
incoming perceptual pattern in a phoneme class with the
aid of statistical information based on the immediate
past of the message (Fry. 1964. p. 16).

In this model the incoming message is first analyzed

according to the pitch. quality. loudness, and duration

18

of the component portions of the incoming message. Once
as much information as possible has been gained from the
acoustical message. on the basis of those phonemes which
have been interpreted so far. a tentative hypothesis

is formed; The hypothesis is then referred to the area
called the phoneme statistic area. The probability of
our hypothesis being correct is assessed; and if the
assumption appears reasonable. the message is considered
heard. If the hypothesis does not seem reasonable.
further analysis of the message is made or the message

is assumed not to have been heard. A further exPlanation
using the example in the above model may make it

clearer. Assume that because of the blurring which
occurs in the acoustical prOperties during ongoing
speech. according to the acoustical signal. we are able
to determine that the sound which was heard was one of
the nasals. i.e; /m/. /n/. or /9/1 The message can then
be referred to the probability or statistical phoneme
area. Here it is found that the most likely correct
choice for that context is the /m/. This is then referred
back to see if the assumption can be supported by what is
further heard: The model as explained above would hold
true if speech perception were based on the phoneme;
However. it is very simple to expand this model to include
all possible perceptual units; The expanded model is

shown in Figure 2.

19

 

_ .. .. ._y Sentences

 

 

L. Sentence
— '- ""—' Words I Statistic

 

 

 

 

 

 

Word
" "‘ " Morpheme [ Statistic _
"" """ ‘4 Phonemes

 

 

 

 

Morpheme
Statistic

 

  

Error Correcting Mechanism

 

   

I
I

I

l

I

I
r.
I

I

I

I
I_.
I

I

I

l

I

I

I

I
I-

    
 

—

 

Phoneme
Statistic

 

 

 

 

Figure 2.-—The principle of statistically-based
identification of linguistic units extending
to all levels of Operation (Fry. 1964).

20

At the present time. none of the above suggested
models have been conclusively validated as being the
model whereby perception occurs. It is further possible
that under varying conditions. differing methods of

perception are employed.

STUDIES RELATED TO THE PERCEPTION
OF CONSONANTS

Three approaches have been employed in attempting
to discover’which elements of the speech signal allow
for its perception. The first method is analysis. In
this method. a recording of some Speech signal is made.
This sample is then analyzed by use of electrical
equipment such as the Spectrograph which allows the
Speech signal to be displayed in a spectrogram. The
display is then examined for any consistency which might
occur. The basic problem existing in this method is that
one cannot be sure that those elements which are selected
for the basis of analysis are the same elements which
are used by the human auditory system in analyzing the
same signal.

The second approach. and the one used in the
current study. is distortion. In distortion the speech
Signal is altered in some way such as compression.
clipping. or bandwidth limiting. When the distortion

has occurred. the signal is then presented to listeners

21

in such a way that the human ear serves as the analyzer
for the Speech signal.

The third approach employed is synthesis. In
synthesis the Speech signal is in some way artificially
created. and the signal is then presented to the human
auditory system to ascertain what it perceives. The
earliest method employed involved simplified handpainting
of Spectrograms. These are drawn and then are converted
into Speech sounds. This is the playback pattern system.
The main type of synthesizers presently used is the analog-
computer type. Each of the methods has some strong points
and each has some inherent weaknesses.

Various phases of the consonants have also been
explored. Liberman (1957) has suggested that cues for
the perception of the consonants can be divided into
‘three major classes: constriction. transition. and
resonation. Constriction cues are those cues which are
the noise of the consonants. In the case of the plosives.
the constriction cues are the stops and bursts of sound.
For the fricatives. the constriction cues are the
frictional portion of the sound. The affricate sounds
have both burst and friction constriction cues. The
transitional cues are those based on the movement of
the acoustic energy from the locus of the consonant to
the steady state portion of the vowel. Resonant cues are

used to classify the nasal sounds as nasal consonants

22

and all other consonants as being non-nasal. The
majority of the studies which have been performed looking
at these classes of cues have been done using either words

or monosyllabic nonsense syllables.

Studies related to the importance

of the constriction in Speech
perception

One of the earliest studies relating to the
importance of the constrictive in the perception of
sounds was conducted by Liberman. et al. (1952). Using
Speech synthesis. they found that the perception of the
/t/. /p/ and /k/'was based upon the presence of noise
at various frequency ranges. For they/p/ sound. the
presence of noise in the area of the first formant was
significant. For the /k/'and /t/} it was the presence
of noise in the area of the second and third formant
areas respectively.

0; m. Harris (1953) found that the /t/. AW.

4773 /§/} /1/n and /r/ sounds were recognized on the basis
of the constriction. regardless of the vowel context
away from.which the consonant was spliced.

Harris (l95#. 1958) synthesized the fricative
consonants /s/. 07. /9/. and /f/. Based on the
constrictive portion of the consonants. the listeners were
able to divide the sounds into two groups. the /s/ and /f/
group and the /o/ and /r/ group. Further. on the basis of

the constrictive component. the listenerS'were able to

23

differentiate between /s/ and m. The /s/ was perceived
when the lower cut off frequency was at 3600 Hz. For the
{yy'it was best at a lower cut off frequency of 2000 Hz.

COOper. Delattre. et al. (1952) speculated that
it was possible to distinguish the /p/'and /k/ sound from
the /t/'sound on the basis of the burst portion of the
Speech signal. These findings were based on work done
with speech synthesis.

In 195# Fischer-Jorgensen conducted a study of
the acoustical elements of the Danish stop consonants.
The results of this study were based upon spectrographic
analysis of the stops /p/. /t/. /k/. /b/. /d/ and /g/ in
the Danish. Several factors were found to be different
in the constrictive or burst portion of the /p/. /t/. and
/k/. Similar results were found for the cognates of each
of the phonemes;

He found that the intensity of the explosion
showed characteristic differences. It was found that /k/
was more intense than /p/. and /p/'was more intense than
/t/. It was pointed out. however. that the results were
somewhat compounded by the environment in which the sound
occurred. Differences in duration were also present. The
/g/ cognates were found to be longer than the /d/. which
was longer than the /b/L Duration of the aSpiration was
also found to be a factor in the Spectral analysis of the

step consonants. Several other variations were pointed

2#

out which were Spectrally present. In summarizing the
acoustical characteristics of the stOps. Fischer-Jorgensen
suggests that adequate recognition of the plosive sounds
is based on certain characteristics. The p/b cognates
were found to have a fairly neutral explosion with
relatively low resonance qualities. The t/d cognates were
recognized by the high eXplosion and the relatively high
resonance qualities. The k/g were recognized by the strong
concentration of explosion in the area of the frequency of
surrounding sounds and a resonance quality which is also
greatly influenced by the presence of surrounding sounds.
An interesting observation was made by Fischer-

Jorgensen regarding the influence of sounds which surround
the sound under investigation:

From a phonemic point of view we would say that only

differences between conditioned variants (or

”allOphones") must be phonetically constant. not

differences between phonemes: thus the difference

between /k/ and /t/ before /i/ must Show some

constancy if communication by means of speech is to

be possible. but even this constancy is of limited

kind. for there may be a bundle of differences which

are all present in Optimal cases. but which need not

all be there. In some cases the frequency of the

explosion might be the same but the duration mi ht

be different. or vice versa. Final /k/ and /p after

/u/'may have the same formant bendings but different

explosions or if they are unexploded. there must be

differences of formant bending. etc. But there has

to be some kind of consistency (p. 58).

In general. Fischer-Jp’rgensen felt that his

findings were consistent with the findings of studies
which had been conducted earlier at Haskins Laboratory

using synthetic Speech.

 

pfintwi
the fri:
as fife,
the 3le
the has;

duratio:

noted ‘2:
Coupons:

Comte:—

a cones!
than t?“

/f/ and

25

'Gerstman (1956) found that duration of the noise
portion of the fricative was significant in its
identification. The syllable {ya/'was recorded and
successive portions of the fricative were removed. At a
point where approximately one-half ot the duration of
the friction had been removed. the syllable was perceived
as /tja/. If all of the fricative duration were removed.
the syllable was perceived as either /ka/ or /ta/. 0n
the basis of his study. Gerstman suggested that the
duration of the constriction portion of the consonant is
sufficient to allow the listener to identify the sound
as being a plosive. a fricative. or an affricate.

Hughes and Halle (1956) investigated the spectral
preperties of the constriction portion of the fricative
sounds of English. The method used was analysis. The
relative energy of the various frequency components of
the consonants was obtained. A strong energy component was
noted below 700 Hz for all of the voiced sounds. but this
component was never present in any of the voiceless
counterparts.

Further analysis showed that the /S/'cognates had
a concentration of energy which was consistently higher
than those present in the ([7 cognates. Results for the
/f/ and /v/'were somewhat confusing.

The authors devised a mathematical computation

in which three measures were made. First. the energy

 

bathe

subtrac
720 Hz

band f:
the ban
last me
region

energy

subtrac
the pea
procedu
procedu
each of
There, a
Demept
those 5

mere PE

analys;
they ‘va

consfir:

26

in the frequency range between #200 Hz and 10.000 Hz was
subtracted from the energy present in the range between
720 Hz to 10,000 Hz. Second. the energy present in the
band from 720 to 6500 Hz was computed and the energy in
the band from 720 to 2150 Hz was deducted from this. The
last measurement was made by locating the peak in the
region from 1500 cps to #000 Hz. .Once this was done. the
energy in dB in a band from 720 Hz to 1370 Hz was
subtracted from energy in a 500 Hz band centered around
the peak frequency. Figure 3 Shows the identification
procedure then preposed. Using this identification
procedure. Hughes and Halle were able to identify correctly
each of the sounds with no greater than a 10 percent error.
There appeared to be no consistency in the error. A
perceptual study was also conducted. and it was found that
those sounds which were within the criteria range were
more readily identified correctly by listeners.

sane, Hughes and Radley (1957) conducted an
analysis of the stop consonants. The first question which
they were interested in was whether or*not the burst. or
constrictive. portion of steps contained sufficient
information for the correct identification of the stop. In
order to answer this question. they recorded each of the
stops and then gated the monosyllables so that none of the
transition was heard. No information was available as to

how this gating was done. The portion of the burst heard

2?

Jensumooua Soapsoamwpsoea mo smswmwo owvssomomuaqn shaman

«HameHmo 20H940HmHazmQH

 

 

 

 

 

 

 

 

 

 

\m\ \m\ \u\
m N m 34
n .mme AmMmHIONb may I Axdmn Gd may AOMHN I 0N5 may I Aoomw OP CNN may N .mde
\m .. E- E4. 3.
as «A mm «o

 

 

hwx OH on com: mav W ADM 0H 0P can mav H ammo:

rm 3 E

 

 

 

 

 

 

g
o.
\o
0\
(5

scores w
eXperien
that lea
w8.3 repc

the basi

28

was 20 msec. long. The taped stimuli containing the

20 msec. bursts were then presented to subjects. The
percentage of correct identification for the five best
judges was 65 percent. 70 percent. 75 Percent. 80 percent.
and 96 percent. They pointed out that the three best
scores were obtained by people having had considerable
experience in listening to isolated bursts. It was felt
that learning could be a factor in these cases. Since it
was reported that correct identification could be made on
the basis of the burst. further analysis of the burst was
undertaken. A frequency analysis of the energy in the
various frequency bands showed that for the /p/ and /b/'the
concentration of energy was in the frequency range between
500-1500 Hz. For the alveolar plosives there was a
predominance of energy in the frequencies above 4000 Hz.
The velar plosives had concentration in the middle
frequencies. These results were in good agreement with
the findings of Fischer-Jdrgensen.

Malecot (1958) found that suppression of the
release in identification of final stOps drastically
altered the perception of the voiceless stops. It was
also felt that the release played a major role in
conveying the manner of articulation.

Hoffman (1958) used a Speech synthesizer in order
to compare the perception of the voiced plosives /b/. /g/.
and /d/. The results of his study yielded findings similar

 

nasal cc

identifi

nume rous
that the
to allow
stOps,

analysi
that 0t

the one

29

to the results of the voiceless counterparts conducted by
COOper et al. (1952). He further found that with the
exception of the /b/. the constrictive portion of the
speech signal provided adequate information for the
perception of the voiced plosives.

Nakate (1959) found that the duration of the
nasal consonants was significant in making the correct
identification.

Liberman (1957) in summarizing the results of
numerous studies conducted at Haskins Laboratory stated
that the position of the frequency burst was sufficient
to allow the listener to distinguish within the voiceless
steps. This seemed to agree with the findings of other
analysis and synthesis studies. Liberman also speculated
that other factors such as the duration and the nature of
the onset of noise or the intensity of the noise might
serve as cues for the perception of the consonant sounds.
While no research data were presented. these speculations
appear to agree with the analytical findings of
Fischer-Jorgensen.

Ingeman (1960) suggested that low hiss sound.
duration of friction and the intensity of the sound were
all factors_which were helpful in differentiating
between /s/ and /z/.

Lotz. et a1. (1960) compared the voiced and

voiceless stops. It appeared that the lack of aspiration

.,La

 

follOWlfié
be the :'

a.sound

interval
to inse:
(Bastiar
Bastian
silence
results.
rather

no cons
reason

it appe
basis C

a meani

fOund .1
/f/, /}
that t]

in the

30

following the initial constrictive release seemed to
be the major cue in causing the listener to evaluate
a sound as being voiced rather than voiceless.

It has been found that the introduction of an
interval of silence is enough to cause the listener
to insert a plosive sound even when none was present
(Bastian. et al.. 1961). Using synthesized speech.
Bastian created the words pppg and Elii- A period of
Silence was then introduced following the /S/. This
resulted in the hearer perceiving the words as ppppg
rather than p232. and pplip rather than plifi. While
no consideration was made in the article regarding the
reason for the selection of the stop which was inserted.
it appears that the selection could have been made on the
basis of meaning. since the stop inserted still maintained
a meaningful word.

Heinz and Stevens (1958) using synthetic speech
found that the center frequencies for the five consonants
/f/. /s/. /_f/. /0/, and /;/ were different. He suggested
that there is a consistent pattern which would be useful
in the identification. It was found that for the ([7.
the center frequency was around 2000 Hz. The center
frequencies were around 3500 and 5000 for the /§/'and
/S/ respectively. No consistent difference was found
between the /1‘/ and /e/. This would appear to agree
with the findings of Harris (1958). findings which

31

suggest that it is not possible to distinguish between
the /f/ and /0/'on the basis of the constrictive portion
of the consonant.

Strevens (1961). using Spectrographic analysis.
studied nine fricatives. Line spectra were made from
the spectrograms. It was found that the relative
intensity and the frequency components of the nine
fricatives used in the study varied. He reported the
following findings:

1. [5V' Lowest frequency at which energy is visible
on the Spectrogram is between 1600 and 1650 cps.
Low peaks of energy tend to occur around 1800-
2000 ops.. h000-h500 ops.. and 5500 cps. Energy
rarely above 6500 cps. Intensity ranking:
lowest of 9.

2. /f/ Lowest frequency is around 1500-1700 cps.
Low peaks of energy tend to occur around 1900 ops..
4000 cps.. and occasionally 5000 ops. Upper
limit of frequency is rarely below 7000 ops..
usually around 7500 cps. In general. a higher
upper limit than No. l. Intensity ranking:
3rd in ascending order.

3. /8/ Lowest frequency varies. but lies between
1400 and 2000 ops. Low peaks of energy tend to
occur. the lowest being close to 2000 cps.. the
upper peaks varying somewhat. but tending to lie
about 1000 cycles apart. Upper limit of frequency
rarely below 7200 ops.: some speakers reach
8000 cps. In general. a somewhat higher upper
limit than No. 2. Intensity ranking: 2nd in
ascending order.

h. ./s/ .Lowest frequency almost always above
3500 cps. Peaks of energy tend to occur with
no apparent pattern. except that they do not lie
closer to one another than 1000 cycles. Upper
limit of frequency exceeds 8000 ops. in most
cases. Intensity ranking: 5th in ascending
order.

 

7.

5.

7.

9.

32

lflf? Lowest frequency varies between 1600 and
2500 cps. Peaks of energy tend to occur not less
than 1000 cycles apart and the aspect of
amplitude cross-sections shows a sharp cut-off
around 7000 ops. Intensity ranking: 7th in
ascending order.

/§/ Lower limit of frequency varies generally
between 2800 and 3600 ops. Peaks of energy tend
to appear at roughly 1000 cycle intervals: these
peaks are sharper than those in No. 5. Upper
frequency limit very variable. but usually
between 6000 and 7200 cps.. i.e. lower than for
either No. 4 or No. 5. The general Shape of the
Spectrum is like that of No. a /S/. but with
all values tranSposed 1000 ops. down. Intensity
ranking: greatest of all 9 items.

‘/s/ Lower limit of frequency usually between
1200 ops. and 1500 ops. There is always a strong
peak of energy below 2000 cps.. with others above
about 3500 ops. The aspect of amplitude cross-
sections gives a hint of formant-like structure:
the low peak is steeper than the upper peaks.
which are often double peaks. some 500-600 cycles
apart. Upper frequency limit is very variable.
usually between 5000 ops. and 7500 ops. A
considerable variety of different sound qualities
was obtained from the subjects. as was to be
expected. Versions judged in phonetic terms to
have a more back place of articulation tended to
approach more closely to a formant—like structure.
Intensity rating: 6th in ascending order.

‘AKV 'Lower limit of frequency varies between

700 ops. and 1200 cps. All spectra bear a marked
resemblance to vowels. with a "formant” of one or
two high peaks between 1000 ops. and 2h00 cps.
Sometimes there are 3 or even # "formants”
altogether. with l.or 2 of them having rather
high peaks of intensity. in the region from

3000 cps. to 6000 ops. At first glance the
Spectrographic pattern is that of a vowel rather
than a voiceless fricative. Upper limit of
frequency variable between 6000 ops. and 7000 cps.
Intensity rating: nth in ascending order.

/h/ Lower limit of frequency usually varies
between 400 cps. and 700 ops. The peaks of
intensity which occur are so marked as to suggest

‘a multi-formant vowel. One major peak occurs

around 1000 cps.. one around 1700 ops. At least

 

a
q
1

H)

signi
signal1
perceil
burst J
consis‘
cited.
with 0'
steps,

enVil‘C:

33

5 majer peaks occur in each pattern: spectra for
women subjects exhibit more Of these peaks than
for men. Upper limit Of frequency is usually
around 6500 cps. Intensity ranking: 8th in
ascending order. but for technical reasons the
data (and thus the ranking) for this item are

suspect (pp. 211-213).

In the most recent study conducted on the
significance Of the constrictive portion of the Speech
signal. Ainswerth (1968) found that the phoneme /t/’was
perceived on the basis of the frequency of the noise
burst rather than any other factor. Again. this remains
consistent with the findings Of many Of the other studies
cited. The results on the /p/ and /k/'were also consistent
with other findings cited. However. for the voiced
steps. the results were not as clear. as the effect Of
environment seemed to Show up more strongly. For the /b/
sound. the presence Of the interval Of voicelessness.
the spread of the transitions. and the noise probabilities
were all important in order for this sound to be correctly
identified. This shows some agreements with the findings
Of Hoffman (1958). who found that the burst was not
sufficient to identify the /b/. The recognition of the
/g/ appeared to be based on the frequency of the noise
burst near the second formant. This view appears to be
in disagreement with the findings of Delattre (1955)
and Harris (1958). One of the most important findings
of Ainsworth's study was the implication regarding the

effects of the environment in determining factors

3#

important for perception. It was found that the vowel
following the voiced consonants had an effect upon which
factors were important for perception and in what way
the important factors were affected. A case in point
can be made from looking at the /d/. In the situation
where the /d/ precedes one of the back vowels. it
appeared that the formant transitions were the most
important cues for perception. When the /d/'was presented
in conjunction with one of the front vowels. the
constrictive element of the sound was the cue which
provided for perception. The above generalizations did
not hold true for two vowels. however. The two vowels
which showed deviation from the rule were A27 and /£7;
Ainsworth explained this digression for A97. citing the
fact that with this particular phoneme almost no second
formant transition is present and. therefore. the listener
was taking cues from the third formant transition. With
.657: the frequency of the burst is located very near the
locus Of the transition such that it is possible to
perceive /d/’On the basis of either the burst or the
transition. The hypothesis formulated regarding these two
exceptions was not investigated.

While several peeple cited earlier have suggested
that high frequency components of fricatives are sources

' Of identification. it has been shown (Lawrence and Byers.

35

1969) that persons with high frequency hearing losses are
able to identify fricatives correctly.

§tudies related to the importance
of the transition in Speech

perceptien

The suggestion that Speech perception might be
dependent in part upon the consonant—vowel transition
was first made by Potter. Kepp and Green (1947) based
upon the extensive Spectrographical studies undertaken
by these authors.

Early exploration of the question was undertaken
by Coopen.et a1. (1952) and Liberman. et al. (195%).
They looked at the role Of the transition in the
perception of the English consonants /b/.{/p/. /d/. /t/.
/x/, and /g/. They found that the direction of the
second formant allowed the listener to make certain
judgments. 0n the basis of a rising second formant. the
listener heard the sound as a /b/ or as its voiceless
cognate. A falling transition caused the sound to be
heard as either Of the other four sounds. Further results
showed that the difference between the voiced and
voiceless sounds could be made on the basis Of the first
formant transition along with the presence or absence
Of the voice bar. These findings. along with those
previously mentioned regarding the constrictive portions
of the same sounds. yield the following criteria for

selecting the plosive sound: The labial plosive is

 

betwee
of the

the co

identi
et al‘
Simpli
envirc
examp]

by /i,

36

signalled by the presence of a low burst of sound and a
rising transition: the alveolar plosives have a high
energy burst and a falling transition: the velar plosives
have a characteristic low burst and falling transition;

Harris (1954. 1958) found that differentiation
between the /f/ and /9/ could be made on the basis
of the second formant transition but not on the basis of
the constrictive portion of the sound.

Fischer-Jdrgensen (1954) felt that the
identification criteria for the stops suggested by Cooper,
et alL (1952) and Liberman. et a1. (l95h) were overly
simplified because they didn’t take into account the
environment in which the sound occurred. They found, for
example that the /t/ had a rising formant when followed
by /i/v /e/9 and /00

Gerstman. et al. (195#) undertook to determine
which factors were important in the transition. They
found that the duration of the transition was more
important than the rate at which the transition occurred.

Schatz (195h) found that the transition had an
important role in the perception of the /k/. Actual tape
recordings were made of a speaker saying the words kee .
co . coo . 113359. hgp. and £293. Each of the consonants
was then spliced away from the context in which it was

spoken. Initial findings indicated that as long as the

3?

transition was present. the isolated consonant was
correctly identified but that if the transition was
removed. the perception changed.

Delattre. Cooper, and Liberman (1955) used
synthetic speech to investigate the locus of the
transition for the voiced stops. They found that for
these plosives, loci could be determined. The second
formant locus for the /d/'was at 1800 Hz, for the /b/
at 720 Hz and for the /g/ at 3000 Hz. For the /g/, this
occurred only when the adjoining vowel has its second
formant above 1200 Hz. It also was shown that a silent
period prior to the transition resulted in better
perception of the sound than when the transition began
right at the locus. There needed to be a silent period
before the onset of voicing. The steady state level of
the first formant appeared irrelevant except when there
was a Straight second formant transition occurring
about halfway between the locus for the /g/ and the /d/.
In that situation. a rising first formant caused one to
perceive the /d/. whereas a falling one was perceived
as /g/.

Malecot (1956) used distortion in order to
determine the role of the transition in the perception of
the nasal sounds. Two methods of preparing the stimulus
were attempted. In the first procedure. the tape was

edited using the human ear to determine the points where

 

 

 

I
14-

trans..i
method 6

purposes

the sour
monosyli
Ralecot

for the

StOPS. :
importa
these c
Speech
duratic
Q0 age<
the tr;
it Was

in the

for 1:57
VOWelS

 

38

cuts were to be made. The human judgment was aided by
the use of the oscillosc0pe. This method of determining
transitions was found to be unreliable. The second

, method employed used the spectrograms for editing
purposes. This method was found to be suitable. Since
this method involved the use of the peak amplitude of
the sound wave. it would have application only for the
monosyllables said in isolation. Using this technique.
Malecot determined that the transition was important
for the perception of nasal sounds.

Liberman. et al. (1956) again using the voiced
stops, found that the duration of the transition was an
important factor in allowing for the distinction between
these consonants and certain semivowels. Using the
speech synthesizer. it was found that by increasing the
duration of the transition for the /b/ sound beyond
40 msecl. the listener perceived the step as a /w/. When
the transition for the /g/”was increased beyond 60 msec.,
it was perceived as a /j/£ Further increase resulted
in the listener perceiving a /u/ for the /b/’and an /i/
for the /g/: This was found to be true in front of the
vowels /(/. /i/. /e/. /a/. /7/, and /o/.

Stevens and House (1958) found that the
perception of the‘/f/ from the /b/'was dependent
upon the second formant transition. The transition did

not appear important in recognizing the difference between

the /s/ and _/f/.

39

Harris, et al. (1958) examined the importance
of the third formant transition in the speech perception
process. It was not possible to find third formant
criteria for the perception of sounds, and it was found
that the presence of the third formant did help to make
for more accurate and consistent perception of phonemes.

O'Conner, et al. (1957) suggested that the
extent and direction of the second formant transition
was an important cue for distinguishing among the /w/,
/j/. /r/. and /l/. This was again based on synthesis
studies.

Green (1958) applied Speech analysis to the study
of the importance of the transition in consonant
recognition. The long vowels (/i/, /a1’/. /o/, /.9/, and
ARV) were combined with 22 English consonants to form
monosyllables. The monosyllables were then presented in
the fOIIOWing fashion: /bisbi;bi/. Through spectro-
graphic analysis of the consonants. Green was able to
establish loci for all but four of the consonants. The

loci were determined as follows:

600-

;p;_ #00- #60- average
/h/ 300. 100, 700. 650, 500-

/W/ 450. 550. 450- = 480- n
/r/ 1300—

/v/ 800. 350. 150. 700- = 660- v
/e/ 1650. 1500, 1hoo-

A27 1250, 1200, 1250. 1400— e 1380- n
/t/ 1600. 1600. 1700. 1850, 1850-

/d/ 1850, 2000- = 1780- n
/n/ 1850-

/l/ 1700, 1650, 1700. 1800- = 1710- "

#0

/s/ 1800. 1850. 1750-
/z/ 1750. 1700, luoo- = 1620- average
/r 1100-
/ / 2000— - zooo- "

/ 2000-

j/ 2300. 2500. zuoo- = 21.00.. ..
/k/
5g? = "
fl/ = ..

(p. 59)

Information on the transitions further showed that the
/p/. /b/. and /m/ phonemes had negative second and third
formants: the /k/. /g/, and /h/ had positive second and
negative third formants; and /t/. /d/. and /h/

had both positive and negative second formants but
positive third formants. It appears. then. that these
sounds can be classed according to place of articulation
on the basis of the transition. Time lag between the
locus and the onset of the transition also appeared to be
a useful one in assessing place of articulation. It also
appeared that it might be helpful in distinguishing
between cognate pairs.

Nakata (1959) found that the duration of the
transition was important in identification of the nasals.
The transition alone was adequate to provide consistent
identification between the /v/ and the A37 (Delattre.
Liberman. and Cooper. 1960). Kelly. Anthony and Uldall
(1963) found that the locus of the transition was the
important feature for perception. Lindblom and

Studdert-Kennedy (1967) found that the recognition of

#1

monosyllables could be altered significantly by
altering the rate and direction of the adjacent formant
transition.

Grimm (1966) created 42 CV syllables. These were
then altered by removing an increasing number of 10 msec.
segments from the syllables. They found that perception
was random until the segment containing the noise-non-
noise portion was present.

Delattre (1969) suggested that the first formant
transition may be important for the perception and
recognition of sounds with regard to the manner of
articulation. whereas the second formant may be
important for the determination of place. This shows
agreement with the analytic study conducted by Green (1958).

In a study looking at initial and final consonants,
Ahmen and Fatechand (1959) used 16 consonants and
8 vowels in order to make up 80 monosyllabic consonant-
vowel-consonant meaningless words and 80 vowel-consonant
meaningless words. A large-amplitude pulse was placed
as near the start of the word as possible for the
purpose of determining the beginning of the word.

Various portions of these words were then gated out

and a perceptual task was undertaken. The main interest
of this study was the effect of duration. The duration
of the sounds was determined by the aural method.

The authors played the tape. and the gate was operated

42

manually. The duration of the gate was adjusted

until the listeners agreed that only the consonant

was heard without a trace of the vowel. This was
called the ”pure consonant.” The ”vowel contaminated
consonant" was determined by adjusting the gate to a
point where only the vowel-consonant was heard or to
that point where the consonant disappeared completely.
Similar procedures were employed to determine the length
of the "pure vowel" and the ”contaminated vowel." The
duration of the ”pure vowel” was found to be from

100 msec. to #60 msec. with an average duration of

300 msec. The "contaminated vowel" was found to average
4&0 msec. The average initial consonant length was

70 msec. They found that it was possible to remove the
first 30 msec. of the fricative and plosive with no loss
in intelligibility. whereas #0 msec. could be removed
from the semivowels. It was further found that the
portion of the consonant nearest the vowel provided more
information than the initial portions. All of this
tends to point to the importance of the transition in
Speech perception.

The results of the studies on speech perception
are often confusing. and it has been Shown that possibly
any one of Several cues may provide adequate information
for the perception of certain consonants. Grimm (1966)

has stated:

“3

(1) The perception of the initial consonants was
dependent upon the different elements of the syllable--
i.e.. the plosive bursts. the noise between the
plosive burst and the juncture of the noise and
nonnoise portions. and the transition area itself.
Exce tions to this were with the syllables /di/'and
/b4 . These syllables were perceived with a high
probability even when all of_these elements were
excised. (2) No rule could be established to Show
that any one of the elements of the different plosive
syllables contributed uniformly to the perception of
all syllables. (p. 1h60)

Similar statements could be made for the fricatives.

STUDIES RELATING TO THE ROLE OF
CONTEXT IN PERCEPTION

Fisbher-Jorgensen (195h) found that the vowel
which followed a consonant tended to affect whether the
‘transition of the vowel was rising or falling.

Brown and Hildum (1956) looked at the effect
of a listener's expectation in his ability to recall
words. Three word lists were constructed. The first
word list contained English words which were infrequent
in occurrence and usage; The second list was composed
of nonsense syllables whose form would be possible in the
English language. The third group were nonsense syllables
which would not be possible in English. They found that
the mean recall for the words was 69.2 percent: for the
English nondwords. #6.8 percent: and for the non-English
syllables. 9.7 percent. The authors concluded that the
perception of speech is directed by the knowledge of the

sequential probabilities.

#4

It has further been shown that the recognition
of the final consonant is improved by the knowledge of
the initial consonant of a nonsense monosyllable (Ahmen
and Fatehchand. 1959).

Reed and Wang (1961) found that when listeners
heard non-English syllables. they rejected them in
favor of the English alternative.

Shearme and Holmes (1962) studied the first
and second formants in ongoing Speech and isolated
words. They found that in ongoing Speech the formants
of the vowels were located in a much more neutral
position than the formants in isolated carrier words.

Swaffield. Shearme and Holmes (1961) suggested
that the differences in formant transitions for conversa-
tional Speech and monosyllables based on analysis of the
acoustics are striking. They note two major differences.
The first was that the set of vowel transitions in
ongoing Speech tended to be much more neutral. and the
second difference was that many times the position of the
transition in ongoing Speech didn't even correspond to
the region for the monosyllable.

Pickett and Pollock (1963) found that the
intelligibility of ongoing speech depended on the
duration of the excerpt listened to. They recorded
conversational Speech and then gated out all of the

message except for a single word. They found the Single

45

word to be unintelligible. They then allowed the gate
width to increase until the initial word was perceived.
For example. if the word to be heard was‘gf. the listener
first heard‘gf. then of the. then of the world. It was
found that one word was usually inappropriately
recognized and the correct perception occurred only
after a certain length of signal was allowed to pass.

In 196# they replicated the study. However. the
interest this time was to determine whether the results
obtained in the previous study were a result of
contextual cues or the increasing acoustical information
available. Subjects read the sentence which they were
going to hear. One word was omitted from the
sentence. This word was then presented in the method
previously described for the 1963 study. It was
determined that even if contextual cues were available.
correct perception of isolated words gated out of
ongoing speech was difficult and that correct identifica-
tion continued to improve to the point where three words
were contained within the gated segment.

Lisker and Abramson (1967) attempted to study
the time between the burst and the onset of periodicy
(voice onset time. or VOT) in order to determine its
effectiveness in classifying sounds as voiced and

unvoiced. They used both isolated words and ongoing

46

Speech and found that in isolated words the VOT time was
useful in making the classification. In the ongoing
Speech it was found not to provide the differentiation.
They hypothesized that this was due to the fact that the
linguistic constraints take up any change in the
information lost because of the "blurring” of ongoing
Speech.

Mattingly. Liberman. Syrdal. and Halives (1969)
found that the perception of stOp consonants depends on
whether or not the cue is presented in a Speech-like
context or in isolation.

It has also been Shown that a sound can be
removed from ongoing Speech and still be perceived
as being there. This has been termed Phonemic Restoration
(PhR). Warren and Obusek (1970. 1971) found that if the
/S/ sound were removed from a word. the listeners still
replaced it. They removed the /S/ out of the word
legislatures in a sentence. The /%/'was replaced by
either a cough. a loud buzz. a soft buzz. a period of
silence. or a loud 1.000 Hz tone presented at 8 dB
above peak intensity of the Sentence and one presented
at the peak intensity of the sentence. The listeners were
asked to locate where the noise or Silence occurred in
the sentence. In all conditions except for the silence.
PhR occurred and the inserted sound was difficult to

locate. In the case of the silence. the PhR was made

 

.1“ rm-

-—_—H-‘ .n—— em _—. ’—

._1
(D
U)
U)

 

47

less often and location was more often correctly
identified.

In a preliminary study by Sherman (1971) it was
found that the listener was able to maintain acoustical
information long enough to wait for subsequent context
in order to perceive a message correctly. The study
presented a sentence beginning. ”There was time to‘_ave
. . . . .” The missing fragment could have been ghave.
gave. wave. etc. It was found that the subsequent
information allowed the listener to make the correct
selection.

Ali. et al. (1971) recorded consonant-vowel
syllables. Each of these syllables was then combined
with either a plosive, a fricative. an affricate. or a
nasal. The tapes were then processed so that the final
consonant including the transition was removed using
spectrcs-.ns which Malecot (1956) had found to be most
reliable. The data were then analyzed to determine
whether it was possible for the person still to perceive
the nasal sound. It was found that five or more of the
nine asals were perceived correctly by all persons who
listened. The statistical analysis Showed that the
perception of nasals was better than chance. Plosives
tended to be perceived more often as nasals. The vowel
of the preceding syllable also appeared to influence

the listener's perception of a nasal.

48

Two tentative conclusions appear warranted based
upon the studies reviewed. The first conclusion is that
the presentation of phonemes in a contextual environment
appears to blur some of the acoustic elements which are
present when the phoneme is produced either in isolation
or in syllables. The second factor appears to be that
the listener to a language with which he is acquainted
applies constraints to what he hears based upon his
knowledge of the language system employed. This constraint
causes him to select certain choices even though

acoustically the choice may be unclear.

STUDIES RELATING TO TIME COMPRESSED
AND INTERRUPTED SPEECH

The question of how much speech could be removed
and still not cause a significant loss in correct
perception has been the subject of considerable interest.
Perhaps one of the earliest studies dealing with
interrupted speech was carried out by Miller and
Licklider (1950). They considered three factors in
looking at the intelligibility of ongoing speech. Factors
considered were the number of interruptions per second.
the proportion of time the Speech was on. and the
regularity of the interruptions. The PB word lists of .
Egan were employed. They found that it was possible to

remove up to 50 percent of the signal and still have

“9

better than 90 percent intelligibility. It was further
found that as long as the interruption occurred more
than 10 times per second. intelligibility was not
impaired. The regularity of the interruption was not a
factor in the study.

As a result of this study by Miller and
Licklider. the interest was aroused as to what might
happen if the speech segments were removed rather than
just having periods of silence. For this reason. studies
in time compression began.

Two methods of Speech compression have been used.
The earliest method was that of Speeding up the speech.
The message was recorded at a certain rate and then played
back at a much more rapid rate. This method has been
employed by several researchers (Fletcher. 1929; Foulke.
1966: Garvey. 1953: Klumpp and Webster. 1961: Kurtzrock,
1957, and McLain. 1962). The second method is the
sampling method. This method entails removing portions
of speech from the signal. The first method used was
that of miller and Licklider (1950): they used a switching
arrangement which permitted a recorded speech signal to
be turned off periodically during the playback of the
recording. Garvey (1953) manually removed portions of the
tape and then spliced the tape together. thus resulting‘
in time compression. Fairbanks, et al. (1954) devised an

electro-mechanical machine for removing Speech. The machine

50

reproduced periodic segments of the speech signal while
discarding other segments. Computers also have recently
been used to sample the speech signal (Scott. 1965).

Several factors have emerged as being important
in the perception of time compressed Speech. One of the
factors is the method of compression used.

Klumpp and Webster (1962) used the speed changing
method of compression and found that when this method
was employed, the intelligibility dropped by 40 percent
or more with only a 33 percent savings in time. This
agrees with findings by Garvey (1953). Kurtzrock (1957),
and Foulke (1966).

Using the manual sampling method, words can
withstand considerably more compression and still be
intelligible. With the speech compressed by 60 percent.
the intelligibility still remained approximately 90 percent
(Garvey, 1953). However. the intelligibility drOpped to
50 percent when the compression was increased to 75 percent.

Using the electromechanical method appears to give
results similar to the sampling-by-hand method. Fairbanks.
Guttman and Miron (1957) performed a study in which
the message was compressed at five different compression
levels. The material was factual information about
which it was haped foreknowledge would be small. At
50 percent compression, the reduction of message

effectiveness was small. The comprehension was still

 

51

about 90 percent of that for subjects who heard the
original version of 1&1 words per minute. At 60 percent
compression. however. the perception of the speech was
only at 50 percent of maximum. A sharp decline in
accurate perception occurred between 50 percent and
60 percent compression. Again. these results showed
good agreement with the findings of Kurtzrock (1967),
Fairbanks and Kodman (1957). and Dukelskiy (1967).

Another factor which appears to be important in
the perception of the compressed Speech is the duration
of the discarded portion. For example, it is possible
to get 50 percent time compression by discarding 10 msec.
of a 20 msec. portion of Speech or by discarding a
30 msec. portion of a 60 msec. segment. As reported
earlier, Killer and Licklider (1950) found that as long
as ten samplings per second occurred. there was not any
loss of inzelligibility. .

'Sarvey (1953) used discard intervals of 40, 60.
80. and 130 msec. intervals. The resultant intelligi-
bility scores were 95, 96, 95. and 86 percent. Fairbanks
and Kodman (1957) also investigated this question. Their
results indicated that if intervals as large as .16 or
.24 seconds were discarded, intelligibility would never
reach 100 percent. They found a considerable loss of
intelligibility when the compression sample removed was

greater than 80 msec.

52

The frequency and duration of sampling and
discarding also appears to make a difference. Fairbanks
and Kodman (1957) found that if the portion removed was
less than 10 msec.. intelligibility was also decreased.

Foulke (1966) has suggested that masking may
occur if sampling occurs with sufficient frequency. He
illustrates this by presenting the following example.

He found that if sampling occurred every 2.5 msec.. a
#00 Hz tone was generated. This then could serve to mask
the signal.

It was further found that the length of the
utterance made a difference in the perception. Henry
(1966) found that the more the number of phonemes. the
greater the intelligibility. He used words with from
3 to 9 phonemes and found a continued improvement as the
number increased. Klumpp and Webster (1961) found that
short phrases were more intelligible than single words.

In comparing the results of studies on time
compression with the study by Miller and Licklider (1950)
on interrupted speech. Garvey (1953) found that with up
‘ to 50 percent time compression or interruption. the
results were similar. However, when 62 percent of the
signal was discarded, the interrupted technique yielded
#0 percent better results than the time compression.

Ahmend and Fatechand (1959) in their study using

clipped speech made the following observations. They found

53

that where the interruption was made had an effect on
the perception. If the initial portion of the consonant
were allowed to pass unaltered, the next segment of the
tape then gated out or suppressed. and then the remainder
of the consonant allowed to pass. the recognition for
the plosives, fricatives, and semivowels was 52, 60. and
93 percent respectively. If. however, the initial
segment was suppressed, the perception for the consonant-
vowel-consonant nonsense words was 36, 60 and 80 percent.
The results of time compression study have shown
that up to 60 percent of the signal can be removed with
only slight deterioration in the listener's ability to
perceive the message. Frequency of removal and the
duration of removal also appear to be a factor. It was
also shown that the more contextual cues that were
available, the more one was able to compress Speech
without loss of information. A further important
suggesti:: which has been made is that the place of the

interruption was significant.

CHAPTER III

EQUIPMENT, MATERIALS, SUBJECTS, AND PROCEDURES

The following sections contain a detailed
description of the subjects. eXperimental materials.

and procedures used in this study.
EQUIPMENT

The following equipment was used during the
various phases of the study:

Tape Recorder I (Amp ex AG #40 - recording)

Tape Recorder II (Ampex AG 600 - dubbing)b

Tape Recorder III (Ampex AG 500 - playback)

MicrOphone (635A Electrovoice)

Level Recorder (Bruel and Kjaer 2305)

Commercial Test Room (Industrial Acoustic Company,
Inc.. Single-walled booth, series 402A)

Magnetic Recordin Tapes (type 201. Scotch Brand)

Permanent Magnet %Alnico V3 /16" diameter x 1” long)

Sound Spectrograph Model PVlOA (Voiceprint
Laboratory)

Sound Level Meter (B & K 2204s)

Earphones (TDH 39 300 Ohms)

Filter (B & K 1613)

Measurement Amplifier (Hewlett-Packard #50 A. 20 dB gain)

MATERIALS

Six sentences were constructed to meet the

following criteria:

5#

 

“.5. it;

II

 

55

a. At least seventy-five percent of the
consonants contained in each of the sentences
were either stOps or fricatives. The
remaining consonants were nasals. affricates,
glides. and semi—vowels. This criterion
was established to allow for some comparisons
between the current study and the previous
literature written dealing with perception
of consonants in words and syllables.

b. All of the sentences used in the study were
five words in length. Miller (1956) suggests
that the human brain is able to store up to
seven items for recall with little difficulty.
Because this study was not to be a test of
short term memory but rather one of perception
the use of five words was felt to be well
within the recall limits of the subjects.

2. All of the sentences were of the simple.
declarative type in order to not introduce
linguistic complexity as a variable.

i. All words in the sentences were mono-syllables
selected from the mono-syllable word lists
of Moser (1969).

Using these criteria. the following sentences

were constructed for use in this study:

56

1. Big dogs ate the food.

2. The boys bought five fish.

3. Cups should go up there.

4. She gave us six pigs.

5. His wife baked this cake.

6. You pushed the door shut.

These 30 words then served as the basis for
constructing 6 word lists of 5 words each. The word lists
were randOmly constructed and are as follows:

1. cups six go the boys

2. five cake their fish you've

3. gave baked the door bought

h. dogs this she the food

5. pushed ate should pigs big

6. Shut his up wife us

PREPARATION OF MATERIALS

The master tape was recorded in the Speech Science
Laboratory at Michigan State University by one male
speaker who Spoke with a General American Dialect. The
Speaker is a member of the faculty in the Department of
Audiology and Speech Sciences at Michigan State University
and has had extensive phonetic training as well as voice
and diction training. The stimulus material was
recorded using the Ampex AG #00 tape recorder employing
an Electrovoice 635A microphone in a single-walled

sound-treated booth. Each of the sentences was

57

recorded as naturally as possible to assure that stress
and intonation patterns remained normal. The sentences
were read at the rate of about two and one-half words
per second. This rate is at approximately the lower
limits for normal reading. Care was taken in reading
the sentences to assure that each of the phonemes was
articulated. Following the taping of the sentences. the
sound spectrograph PVlOA (Voiceprint Laboratory) was
employed to obtain spectrograms of each of the sentences
as a check to make certain that each of the consonants
was visually present. In one of the sentences all of
the phonemes were not present and so the sentence was
re-recorded and again visually displayed. The second
recording had all of the phonemes present and was,
therefore. used. The master tape was then played to
three listeners who had some training in Speech pathology.
and these listeners were asked whether they detected
anything unnatural about the sentences. One stated that
the Speech was slightly exaggerated. All of them felt
that rate, stress. and intonation were normal.

This master tape provided the stimulus material
for the two additional tapes. Each of the additional
tapes. herein referred to as sub-master tapes 1 and 2.
was dubbed using the Ampex AG 440 in conjunction with
the Ampex AG 600. Each of the sub-masters was then

placed on the sound spectrograph. Use of the spectrograph

58

follows the suggestion of Malecot (1956). Spectrograms
were made of each of the sentences. Each Spectrogram

was then removed from the recording drum of the
Voiceprinter and the beginning of the consonants was
marked on the spectrogram. Criteria used for establishing
this point are as follows:

Voiceless stone. The V0 boundary was taken to be

the point where the last trace of the formant transi-
tion of the initial vowel was visible in the sound
Spectrograms. The CV boundary was located at the
first transient spike of the release.

Voiced stogg. The V0 boundary was identified with
the point where the F1 first reached the low
frequency of the so-called voicing bar of the closure
Segment. The boundary was put just before the voice
pulse in which the higher formants reappeared in

the segment after the closure segment.

Voiceless Fricatives. The friction noise of all
stimulus words containing these consonants was
preceded and followed by brief aspirative segment,
the middle of which was taken as the VC and CV
boundaries.

Nasals. The segments representing the nasal murmur
of the nasal consonants were easily delimited from
the adjacent nasalized vowel segments by observing
the very different Spectrum characteristics of these
two segment types. The boundaries were put where the
Spectrum change was most abrupt. '

Sonorants. The segmentation of the intervocalic /1/

presented no problems since the lateralized closure

of this consonant is associated with a very rapid

change of s ectrum character at the VC and CV

boundaries Ohman. 1966. p. 984).

These-provided a general guide for determining

the point to be marked. Potter. Kopp and Green's (19h?)
information was used to aid in determining the steady

~state portion of the vowels as well as the spike and

59

fill (constrictive) portions of the consonants. Areas
between the steady state portion of the vowel and the
constrictive element of the consonant were considered to
be the transition. Spectrograms were made of the
sentences and steady state vowel, transitions and
constrictives were marked by the experimenter on the
Spectrograms. A second person. an expert and recognized
authority in spectrography, was then asked to examine
the spectrograms and agreement was made regarding the
segmentation of the Spectrograms. Only in two cases was
there a discrepancy. and in these cases an agreement was
reached as to where the removal should be made. In

each case the discrepancy was a minor difference.

Once the segmentation of the sentences had been
determined. the tapes were again placed on the voice-
printer and spectrograms made. For locating the onset
of the cznsonant constrictions the spectrograms were
left in place on the recording drum and the recording
drum of the voiceprinter was rotated until the printing
stylus was in direct alignment with the mark on the
spectrogram indicating the onset of the consonant.

Because the rotation of the recording drum and
the playback head of the scanning drum are synchronized.
it was possible to find the onset of the constriction
portion of the consonant on the tape using this

synchronization.

60

By locating the revolving playback head beneath
the portion of the tape which was loOped about the
scanning drum of the voiceprinter. the location of the
onset of the consonant on the magnetic tape was located.
The same procedure was used for determining the onset
of the transition. the onset of the steady state portion
of the signal. and the onset of the vowel-consonant
transition. These points were marked on the back of the
tape with a felt tip pen.

Removal of the segments of the tape was
accomplished by passing a permanent magnet (Alnico V
which had a 3/16” diameter) between the lines marked on
the magnetic tape. thus erasing the portions of the tape
contained between those lines.

To check that the correct portion of the speech
signal had been removed. a Second Spectrogram was made upon
the completion of the deletion. The two spectrograms were
then placed upon each other and the portion which was
removed was compared to the portion which Should have
been removed to determine that the correct portion of the
sounds had been removed.

Sub-master l was altered initially so that all
of the consonant constrictions contained in each of the
six sentences were removed. This left only the
transitional section of the consonants and the steady

state portion of the vowels remaining. Sub-master 1 was

61

then dubbed twice yielding experimental tape 1 and 2.
EXperimental tape 1 was left unaltered at this point.
Experimental tape 2 was spliced so that each word was
contained in a separate segment of Spliced tape. Words
were then randomly assigned to Six word lists of five
words each and the tape was resplioed. EXperimental tape 1
was a meaningful sentence with consonant constrictives
removed. Experimental tape 2 was first order approxima-
tion sentences with consonant constrictives removed.

Sub-master l was then further altered by removing
the transitional element. leaving only the steady state
portion of the vowel remaining. Two eXperimental tapes
were then dubbed from sub-master l, yielding experimental
tapes 3 and h. Experimental tape 3 was left and was a
meaningful sentence with the consonant constriction and
transition removed. leaving only the steady state vowel.
Experimental tape b was Spliced and the words reordered
as in experimental tape 2. resulting in a first order
approximation sentence with consonant constriction and
transition removed.

Sub-master 2 was altered by initially removing
the transitional segment of all of the words. leaving
the consonant constrictives and steady state portions
of the vowel. Two experimental tapes were then dubbed.
Experimental tape 5 was a meaningful sentence with

transition removed. Experimental tape 6 was spliced

62

auui the words reordered. resulting in a first order
approximation with transition removed.

Sub-master 2 was then further altered by removing
the steady state portion of the vowel. leaving only the
consonant constriction. Experimental tape 7 and 8 were
then dubbed from sub-master 2. EXperimental tape 7 was
a meaningful sentence containing only consonant
constrictions with the transition and steady state portion
of the vowel having been deleted. Experimental tape 8
had the same segmentation effect as experimental tape 7
but was rearranged in first order approximation. Figure 4
shows the procedure involved in deve10ping the eight
experimental tapes. Table 1 lists the tapes and explains
which portions were removed. the order of presentation and

the portions which the subjects actually heard.
SUBJECTS

The 80 subjects used in this study were all
adults ranging in age between 19 and 30 years old. The
mean age of the subjects was 23.8 years and the median
age was 2h. There were 20 male and 60 female subjects.
All of the subjects were residents of East Lansing and
Lansing. Michigan. and all of the subjects had completed
high school and had some college education. All were
native English Speaking and Spoke general American English.

None of the subjects had clinically significant hearing

63

heave was» as poms moms»

Hmpcmsflnmmxe Psmfio map Show op mommy mo nowvmnmaosm wswsosm smnmmHQIIJS oSSMwm
F

commences cam pooaamm

m ease HMPSmSSSmaxmw_

as made ampsmsflmmaxov
peed

 

Am mace HmpSmsHSmmxmv
commences cam emowaam

An wasp Hmpcmsfianva
Pena

1
Vowels

 

 

A: meme Hmpcesfimmaxmv
eosmpuoms use emowaam

mCOHPHmfldHP
®>OEQ N mm9m<slmbm
to
O
or
we.
MA<B mmsm<2
e
1w
0
“w
e
WCOHPOthmgoo WI.

 

Am mama HmPSoSSSmonv
pMmH

 

AN made HmPSoEHSonmV
commences was emoHHAw

 

 

AH can» Hmpcmsanmaxev

 

PHmH

 

 

 

 

pamcoacoo m>osom#mlmmem<slmpm

TABLE l.--Contents of the

6#

experimental tapes

 

 

 

Experi- (Order of 'wPortions
mental Presenta- Portions Subjects
_Tape tion Removed Heard
1 Meaningful Consonant Transitions
Sentence Constrictions and Steady
State Vowels
2 First Order Consonant Transitions
Approximation Constrictions and Steady
State Vowels
3 Meaningful Consonant Steady State
Sentence Constrictions Vowels
and Transit-
ions
b First Order Consonant Steady State
Approximation Constrictions Vowels
and Transit-
ions
5 Meaningful Transitions Consonant
Sentence Constrictions
and Steady
State Vowels
6 First Order Transitions Consonant
Approximation Constrictions
and Steady
State Vowels
7 Meaningful Steady State Consonant
Sentence Vowel and Constrictions
Transitions
8 First Order Steady State Consonant
Approximation Vowel and Constrictions

TranSitions

65

losses as determined by a pure tone screening test
administered at 20 dB ISO. The subjects were tested just
prior to their participation in the present study. The
three listeners used to validate the master tape were
subjected to the same criteria as the experimental
subjects. Each of the subjects used in the study heard
only one of the eight tapes and was not aware at the time
of testing specifically how the tape had been altered.

Assignment to groups was done randomly.
PROCEDURES

The tapes were played to each of the subjects.
Each group of subjects heard one of the four taped
conditions. Each of the subjects was seated in a room.
The tape was then presented at 70 dB SPL to the right
ear of each of the listeners. The calibration of the
tape for tresentation was accomplished by averaging the
peak rue value of the tape from the graphic level
recording. A 1000 Hz puretone calibration tone was then
placed at the beginning of the tape and was used for
adjusting the volume of presentation. The tape was
played on the Ampex AG500 with output to a Dynaco 60-watt
amplifier and was then fed to 12 sets of TDH 39 300 Ohms
telephonic headphones.

The following instructions were given to the

subjects who heard the meaningful sentences:

66

You are going to hear six sentences of five words
each. Each sentence will be preceded by the phrase.
"Sentence number . . . . ." Portions of the sentence
have been removed from the tape recording. Your
task is to write the sentence you hear in the
apprOpriate space on your answer Sheet. For example.
sentence number one will be written in Space number
one. After the completion of the sentence you will
have fifteen seconds. If you are uncertain of what
you have heard, you are urged to make the best
response possible. that is. leave as few blanks as
possible. Record your response only after hearing the
entire sentence. Are there any questions?

If there were any questions. they were answered.
The only question raised was regarding guessing. To
answer this question. the portion of the original
statement about uncertainty was reread.
For the subjects who heard the first order
approximations. the following directions were given:
You are going to hear six lists of five words each.
Each list will be preceded by the phrase, "List
number . . . .” Portions of the list have been
removed from the tape recording. Your task is to
write the list you hear in the appropriate Space
on your answer sheet. For example. list number one
will be written in Space number one. After the
completion of the list you will have fifteen seconds.
If you are uncertain of what you have heard. you are
urged to make the best response possible. that is.
leave as few blanks as possible. Record your
reSponse only after hearing the entire list. Are
there any questions?
The tape was then played and the listeners recorded their

responses.
ANALYSIS OF THE DATA

The total number of words correctly identified

by each listener was tabulated and an initial preview

67

of the data was conducted by using a two-way analysis

of variance. It was determined that the F Distribution or
F- ratio would be used to determine whether the null
hypotheses could be rejected. This test was accepted
because it analyzes the significance of the difference
between variances. A critical value at the .01 level of
confidence was accepted as the level that must be reached
in order to reject the null hypothesis. If significance
was reached, a Sheffe post hoc comparison would then be
employed to determine where the Significant difference

occurred.

CHAPTER IV

RESULTS AND DISCUSSION

The purpose of this study was to determine the

effects which context had in the recognition of meaningful

and non-meaningful sentences. A further factor considered

in the study was the importance of the transition and the

constriction in the recognition of words Spoken in

context but heard both in context and out of it. To aid

in guiding the study, the following questions were

formulated:

l.

2.

3.

adults.

Sees the use of context provide cues for the
reiception of words which have been altered by
the removal of various acoustic segments?

itw much intelligibility is maintained because of
gs-articulation when only a vowel sound segmented
from a word is heard both presented in context and
out of context?

is there a significant difference in the perception
cf contextual Speech when transitions are removed
and when constrictive elements are removed?

If significant differences occur in ongoing
contextual speech, do the same differences also
appear in non-contextual sentences?

The study was conducted using 80 normal hearing

The subjects were divided into eight groups of

ten, and each group heard one of the taped conditions.

68

69

Scores for each person were computed based upon the
number of words which he was able to perceive correctly.
The results of the study appear to provide.
within the confines of the experimental procedure.
significant information relative to the perception of

language and the phoneme.
RESULTS OF COMPARING CONDITIONS

Initially. means and standard deviations for each
of the taped conditions were computed. The means were
based on the average number of words correctly identified
by the ten people who heard each of the conditions.

Table 2 on the following page presents the means and
standard deviations for each of the eight conditions.

Table 2 Shows that the range of standard
deviations for the eight conditions is 0.2< s.d (3.7.
being the lower end for experimental tape a and the
upper end for experimental tape 2. The means for the
eight conditions ranged 0.6 <mean<26.l experimental
tape A to eXperimental tape 5 reSpectively. The F
distribution, or F ratio. was used to further analyze the
significance of the differences in variance of the scores
obtained. Table 3 summarizes the results of the analysis.
The differences for the columns (order effect). the
differences for rows (segmentation effect) and the

interaction effect were all significant at the 0.01 level

70

TABLE 2.--Group means and standard deviations of ten
subjects for each of eight test
conditions ranked in descending

 

 

 

order

Wtal Mean (Tn number 5%
Tape of words). Deviation

5 26.1 2.6

1 19.2 3.1

6 11.2 3.9

2 9.0 3.7

7 7.2 2.3

3 3.6 2.2

8 2.5 1.5

a 0.6 0.2

 

71

TABLE 3.--Analysis of variance

 

 

 

Source SS df MS F
Rows (order of

presentation) 1,371.2 l 1,371.2 133.1*
Columns

(segmentation

condition) 3,637.2 3 1,212.4 117.7%
Interaction 414.5 3 138.2 13.4*
Error (w £2111.

cell) 735 72 10.3

Totals 6.157.9 79

 

* = significance at 0.01 level

72

of confidence. An F of 2.09 was needed for significance
at this level. The magnitude of the segmentation and the
order effect allow us to conclude with considerable
certainty that both of these factors are significant in
speech perception. The F for segmentation was 117.7 and
for order, 113.3. The F for interaction was only 13.“.

In order to examine the interaction effect which
was also found to be significant, Figure 5 presents a
visual display of the results obtained in the study
relative to the interaction effect. As can be seen in
this figure. the interaction which occurs between the
segmentation condition and the order effect is that
context seems to become less significant as more and
more of the signal is removed. It might be hypothesized
that as more and more information is removed from the
contextual sentences through the segmentation process,
the contextual cues which had aided in understanding the
meaningful sentences are lost and therefore, that the
performance which occurs parallels more closely the
performance of the non-contextual or first order
approximation sentences.

As it was a concern of this study to know which
of the individual comparisons contributed to the overall
significance. a post hoc comparison according to Scheffe
(Hays, 1963) was employed. The Scheffe comparison is

suitable for this analysis because it allows one to

30

23

26
2A
22
20
18
16
14
12

Words Correct

10

OJ

U\

u

73

x = sentence

 

O .. first order
approximation

 

 

$2 $1
8 o o H
OH OH ’H G.)
+9 +J +>3
0H o 0". o
(D ..-g U) :>
ii 3 5% s
3.. on $4
E-* s: E-4

O

O

Portion Removed

Transition

& Constric-

1

tion

FIGURE 5.--Standard deviations and mean number of words

correctly identified based on ten
listeners under each condition

7t:

look at all possible pairwise comparisons. The Scheffe

analysis resulted in the following confidence interval:
Te - 6.8 5Ys5§s+ 6.8

To test for significance, Table A was constructed to

compare each of the eight conditions with every other

condition. In order for a comparison to be significant,

the difference as shown on the table must be outside the

limits of the confidence interval.

Effects of Context

The importance of context in perception of the
spoken word can be seen by looking at the number of times
each word was correctly identified when placed in a
meaningful context compared to the number of times the
same word was correctly identified in a first order
approximation sentence. Table 5 shows the number of times
each word was identified correctly under each of the
eight presentation conditions.

As can be seen by examining Table 5, the context
has a general effect across all words. Particular
attention will be paid to eXperimental tape 1 (meaningful
sentence minus the consonant constrictions), experimental
tape 2 (first order approximation minus the consonant
constrictions), eXperimental tape 5 (meaningful sentence
minus the transitions), and experimental tape 6 (first
order approximation minus the transitions). Contrasting

experimental tape 1 with experimental tape 2, the words

75

Hm>mH no.0 was an pesoaeacmnm.

 

 

5.: ~.a a
m.H m.w: p. :
H.H w.m o.n e.m m
*m.m o.# *o.oa $0.5 N.Ha w
*o.mm em.ma .m.mm .m.mm am.sH H.6m m
“.0 s.d *s.m e.m «.m- .H.afl- o.m N
$5.6H *o.mH *o.mH so.mfl *o.m *m.m- *N.OH «.ma a
mama
HmPSmsHSmme
m.~ m.s e. 6.m N.HH H.6m o.m cams
m al. a m o m, m news

 

mama HaemoSSSmaxm

 

mcowpwnsOo heave
Ham sens acowpaesoo one we some msfipmwnpcoo mmwmno>m cw mmocmnommwouu.: mqm<e

TABLE 5.--Number of correct identifications for each of

76

the 30 words under each of the

experimental conditions

 

 

 

Experimental Tape

 

Word ‘13 2 g3* u 5* 6 7* 8
big 6 8 1 -- 8 -- 1 --
dogs 1 -- —— —— 9 6 1 -_
ate 10 8 7 2 10 8 l —-
the 30 22 16 4 30 23 22 12
f°°d 7 8 2 -- 8 6 2
boys 8 10 -- -- 1o 10 6 3
bought 10 9 -- —- 9 6 2 _-
five 10 10 l -- 10 7 1 --
fish 3 10 1 -- 1o 4 1 —-
cups 1 —- -- -— 6 _- -- _-
should 3 ~» —— _- 8 2 -- __
go 10 7 1 -- 9 -- -- ~-
UD 10 8 4 -- 10 8 —— __
there 10 9 —- -- 8 -- -- -_
she -- -- -- -- 10 2 5 2
gave 6 -- -- -- 9 -- 1 _-
us 6 2 —- -- 8 -- -_
six 6 -- -- ~— 10 -- -- _-
pigs 6 -— l —- 6 1 1 --
his 9 —- ~- -- 7 2 -- --
wife 9 —- —— -- 7 2 -- 1-
baked 7 2 -~ —- 10 5 1 _-
this -- -- -- —— 7 -- -_ 1
cake 8 -- l -- 10 -- -— --
you’ve 2 —— -- -- h -_ -- --
pushed 2 -- —- —- 8 —- 8 --
door 6 3 1 -- 10 10 10 4
shut 6 ~- -- -- 10 7 9 1

 

*These are the contextual conditions

77

in context were consistently recognized more often. In
the case of 20 of the 30 words, the words in the first
order approximation sentence were recognized only half

as often as the same words in the sentential presentation.
Only in one case did experimental tape 2 exceed experi-
mental tape 1 and that was for the word 392$, In this
case. the word was recognized seven times in context but
eight times in the first order approximation sentence. In
one further case the word was not correctly recognized

in either condition. This was the word gag. In the
sentential order this was perceived as pg. Here the
linguistic constraints known to an English speaker of
that sentence would have allowed for the use of either
word.

In comparing eXperimental tape 5 and experimental
tape 6, the same results were obtained as for eXperimental
tape 1 an: experimental tape 2. Words in eXperimental
tape 5 were correctly identified more frequently than
the same words in the first order approximation sentence
used in experimental tape 6 for all but two cases. In
these two cases the words pgyg and £23; were both correctly
identified by all of the subjects who heard these
conditions. In 21 of 30 words presented. experimental
tape 5 words were recognized at least twice as frequently
as experimental tape 6 words.

Further support can be found for the influence

of linguistic constraint by looking at the influence of

78

word position. Soderberg (1967) looked at the information
values of words in clauses of varying length. The
information value was determined by the extent to which
college students could predict what that word was to be.
Table 6 presents Soderberg's results. Of particular
interest to this study are the findings for five-word
clauses.

TABLE 6.--Soderberg's information values of words in
clauses classified by length

 

 

No. of Words
in clause 1 2 3 E 5 6* 2 8 9 10 Totals

 

2 70 65 135
3 75 56 51 182
a 71 79 no 23 213
5 59 69 67 6A #6 305
6 76 56 52 77 52 41 359
7 76 an 71 77 35 43 61 407
8 82 77 74 us 77 53 66 39 519
9 61 56 62 75 50 #3 62 35 64 508
10 97 5a 6817 62 15 50 85 59 10 517

Totals 667 556 485 379 322 195 239 159 123 10 3,135

 

In this current study, using sentence position as
a variable, the results in Table 7 were obtained.

It can.be seen that the score for experimental
tape 5 and experimental tape 1 both follow the general
distribution of Soderberg's findings with positions one
and five being the poorest and the medial positions being
better. The first order approximations. however, do not
follow this order. In neither case is position one the

poorest. and in both cases the last position is the

79

TABLE 7.--Tota1 number of correct responses for each of
the five sentence positions

 

 

 

 

Experimental tg WordIPOSItion
Tape 1 2 31 4 .5
5 “9 51 55 57 52
1 28 29 53 42 4O
6 20 15 20 32 22
2 10 10 31 17 23

 

second best. It would appear that in the case of the
sentential ordering, the recognition of words is in some
way related to or parallels the information value of
5-word clauses as found by Soderberg. It would appear
that subjects were using some of the predictive value of
linguistic constraint just as Soderberg's college students
did. In the first order approximation sentences, however,
subjects had no linguistic information so that correct
responses were based upon the acoustic cues alone and
word position was no longer a factor in whether a word
was correctly perceived. An example of linguistic

context affecting perception is shown to demonstrate this.
In sentence one, Big dogs ate the food, fpgg.was perceived
as 9.9.1325 The /b/ for /f/ and /n/ for /d/ were both rare
confusions in Miller and Nicely's (1955) study. All of
this would again support the fact that much of the

information of ongoing speech is determined by linguistic

80

constraints as Opposed to the acoustic constraints which

function in the first order approximations.

Effects of segmentation

A comparison of the effects of temporal segmentation
was undertaken using Table 9 (page 79). First of all.
if the effects of segmentation upon the contextual
sentences are compared. it is found that those contextual
sentences in which only the transition had been removed
were superior to all of the other segmentation conditions.
This significant difference between the condition in which
only the transition had been removed and in which only
the consonant constriction had been removed will warrant
further consideration. It can also be noted that the
condition in which only the consonant constriction had
been removed. while being inferior to the experimental
tape 5 conditions (without transitions), was superior
to the other two conditions, experimental tape 3 (without
constriction and transition) and experimental tape 7
(without transition and vowel). Between the experimental
conditions on experimental tape 7 and experimental tape 3
there were no significant differences. Among first order
approximation sentences no significant differences were
noted between experimental tape 6 (without transitions)
and experimental tape 2 (without consonant constrictions),

nor between experimental tape 8 (without vowels and

81

transitions) and experimental tape 9 (without constric-
tions and transitions). However, there were significant
differences between experimental tape 6 (first order
approximation without transitions) and experimental

tape 8 (first order approximation without vowels and
transitions), between experimental tape 6 and experimental
tape 9 (first order approximation without consonant
constrictions and transitions). between experimental
tape 2 (first order approximation without consonant
constrictions) and experimental tape 8, and also between
experimental tape 2 and experimental tape 4.

Before a more indepth comparison is attempted
between experimental conditions in tape 5 and tape 1,
which appear of particular importance in this study, some
general comments are deemed necessary.

The first question which will be raised at this
point is the question of what factor might account for
the differences which were found in the segmentation
effect. In terms of the literature dealing with time
compression, the simplest explanation which could be
made would deal with the amount of the signal which was
removed. In accounting for the differences between
conditions in which only one portion of the Speech
signal was removed compared to those in which two portions
were removed, it is clear that in these comparisons

approximately twice as much of the signal was removed

82

when two of the speech elements (either transition.
constriction, or vowel) was removed as when only one was
deleted. The Significant differences between experimental
tape 1 as compared to 3 or 7 and experimental tape 5 as
compared to 3 or 7 are most easily eXplained on the basis
of this time removal factor. The fact that no differences
occur between conditions on tapes 3 and 7 can be SXplained
in the same way. Since approximately the same amount of
information was removed. the results would be expected

to be the same. However, when one compares the meaningful
sentences with the transitions removed with those in
which the constrictions were removed, one finds again
approximately the same amount of Speech Signal was removed.
This was determined by measuring the amount of the
spectrogram removed under each condition. In comparing
the amounts deleted. one finds that slightly more was
removed when all of the transitions were removed and. yet.
this coniition was superior to the condition in which

the constriction was removed. The findings here that
supression of the initial part of the word is more
detrimental to the speech signal is in agreement with

the findings of Ahmand and Fatechand (1959) who found _
that with the transitions removed perception was at 527

93 percent and with constrictions removed it was at 36-

80 percent. One possible explanation of why this is so
might be that the removal of the initial portion of

words makes it difficult for the listener to segment the

acoustical signal into individual words. While no
evidence of this was seen in the contextual situation.
several examples of this were seen in the non-contextual
or first order approximation sentences where subjects

tended to run words together. hearing she‘pig as sheep Egg.

Amount of information contained in vowels

In the present study it was found that in ongoing
speech as well as in word lists, almost no information
is carried in the vowel itself. If one looks at the
findings in Table 5, it is seen that in the meaningful
sentence condition where only the vowel was remaining
(eXperimental tape 3) subjects were only able to give
36 correct responses.' Even in these figures some doubt
exists as to the correct reSponses given. Sixteen of the
36 correct responses were for the word :22, However.
during this particular tape. subjects reported hearing
#2 333's so that it is possible that subjects, knowing
they were to hear sentences simply guessed that many
of the sentences would have the in them and therefore
placed them in. The other word which was heard frequently
was'gtg which was correct seven times. Here again. while
at; only occurred one time in the six sentences, the
subjects guessed this an average of 2.4 times each or a
total of 2h times. The fact that gtg occurs so commonly
in English may have led the subjects to guess 332 any
time they heard /e/ plus an apparent plosive. The fact

8#

remains, however. that even under the most favorable
interpretation of these data the subjects who heard only
the vowels in meaningful sentences were only able to
identify an average of 1.2 words or a recognition of

4 percent.

In the non-contextual sentences similar results
were obtained; ‘223 and gig accounted for all of the
correct responses; but again, both words were guessed a
disproportionate number of times. The results of this
study in the area of vowels appears clears Vowels carry
very little meaning when said alone.

Perceptual differences in ongoinggcontextua;
and non-contextual_speech

Since it appears that the major differences which
were significant occurred in the case of the meaningful
sentences in which the transition had been removed and the
meaningful sentences in which the constrictions had been
removed. a confusion matrices of errors for these conditions
was constructed; The first order approximations of these
two conditions were also included so that some further
comparisons of perceptual confusion might be undertaken.
Because of the limited number of times each phoneme was
used. it was felt that an error matrix which would allow
for grouping the phonemes would be more helpful. Therefore.
the phonemes were grouped according to three major categories.
In order to allow for some comparisons with Miller and

Nicely's (1955) study. the same taxonomy system was

85

employed although two of the smaller classifications were

deleted.

The phonemes were grouped according to the

following categories:

18

2.

3.

Voicing.--Phonemes in which the vocal cords are
in vibration were contrasted against those in
which they do not vibrate; The voiced consonants
in this study were the /b/, /d/, /g/, /v/, /z/,
/m/, and /n/. The voiceless ones against which
they were contrasted were the /p/, /t/. /k/, /f/,
/3/v and /h/$

Affrioation.--If the flow of air is halted
completely so that the consonant is a step or
nasal or if a turbulence is only forced,
resulting in a friction noise, we get the
distinctions between /p/. /t/, /k/, /b/, /d/,
.x’s/. /m/. /n/ and /f/. /S/. /V/. and /2/.

Tiace of articulation.--The sounds were also
:cmpared for errors based upon the place in the
mouth in which either the plosion or the friction
took place. For this category, three places of
articulation were considered: Those consonants
which are produced in the front of the mouth,
/p/, /b/, /f/, /v/, and /m/: those which are
produced in the middle of the mouth /t/, /d/,
/s/, /z/, and /n/: and those which are produced
at the back of the oral cavity /k/, /g/, and /h/.

86

Table 8 shows the classification of the consonants used
to construct the confusion matrices.

As no errors were noted in duration which Miller
and Nicely used to distinguish the /0/ from the /s/ and
the /J/ from the /z/, this classification was deleted from
the study; The number of errors of each type was then
tabulated; Each time a consonant was used in place of
another consonant, the error'was analyzed and counted.

For example, if a./p/‘was heard as a /t/, this was
considered an error in place of articulation. If a./p/
was heard as an /f/, this was considered as an error

in affrication. If a./p/'was heard as a /b/} this was
considered an error in voicing. If a /t/”was heard as a
/v/; this was considered as an error in place, affrication
and voicing. The number of errors made in each category
are shown in Table 9. Several omissions of words
occurred, and these could not be taken into account in

the error matrices. Figure 6 shows a graphic presentation
of this material. Miller and Nicely (1955) found that
place was most often incorrectly perceived, with
affrication being next and voicing most often correctly
perceived; When one looks at the data for first order
approximations in this current study, one finds that the
same relationship holds true. However, if one looks at
the information that is available on the contextual

sentences which were altered, it appears that there is a

TABLE 8.--Classification of the consonants used to

87

construct the confusion matrices

 

 

 

nasal

Consonant Voicing Affrication Place
p O O 0
t 0 0 l
k 0 O 2
f 0 l O
O 0 1 1
s O l l
f o 1 2
b l O 0
d l' O 1
g l 0 2
v 1 l 0
at l l 1
z 1 l l
h 0 l 2
m l 2 0
n l 2 l

Voicing: O = unvoiced Place: 0 = front

1 = voiced l = middle
Affrication: 0 = plosive 2 = back
% : fricative

88

TABLE 9.--A confusion matrix for'meaningful sentences with
transitions omitted or constrictions omitted
and first order approximations with
transitions omitted or constrictions omitted

 

 

 

 

Meaningful First Order
Sentences Approximations
Transi; Constric- Transi- Constric-
tions tions tions tions
Omitted Omitted Omitted Omitted
Voicing 3 12 1 7
Affrioation 10 37 22 23

Place 10 36 23 42

 

89

voicing
affrication
place

is
(I)

D<3><

"I!"

Number of Confusions
to
N

 

 

 

s c s c

013 o 013 0
odd) "4'0 «4m u-I'o
we #0 pp Pm
we (no ~Hp op
{Doc-I «4+! can-1 "4+:
55 f!“ $8 4‘3"
‘4 mg $4 :08

5+ s E4 s

o o

c) c)

Meaningful First Order
Sentence Approximation

CONDITION

FIGURE 6.--Confusion errors for contextual and
non-contextual Speech

9O

much higher percentage of affrication errors than in the
first order approximation sentences and in Miller and
Nicely's study. It must be remembered that the types of
alterations done in Miller and Nicely's study were not
the same as for the present study: therefore, direct
comparisons with the current study are not possible;
However, that the same results occurred in the first
order approximations in the current study but not in

the contextual sentences appears to be significant and
casts further doubt upon the hypothesis that Speech in
single words is perceived in the same way as speech in
context; It would appear that Fry's (l96u) suggestion
that auditory information coming in is compared to the
person's linguistic model has a valid basis. These two
types of information then result in perception. Further.
results on perceptual studies using single words should

not be extended to ongoing speech and vice versa;

CHAPTER V

SUMMARY, CONCLUSIONS AND RECOMMENDATIONS

Eight experimental tapes were produced for the
purposes of this study; Each of the tapes presented
thirty stimulus words under differing conditions; Four
of the tapes presented words in a meaningful sentence
and four of the tapes presented the words in random
order; Under each of these conditions various portions
of the speech signal were removed. The portions removed
were the constrictions, the constrictions and the
transitions, the transitions, and the transitions and the
vowels; Eighty listeners were then randomly divided
into eight groups and one of the groups heard each of the
experimental tapes; Each person was asked to write down
what he heard:

The mean number of words correctly identified for
each of the eighty experimental conditions was then
computed and the standard deviations were determined; An
analysis of variance showed that a significant difference
existed both in terms of the portion of Speech removed
and in terms of the order of presentationl Interaction

effect was also significant.

91

92

A post hoc comparison was performed which showed

that the following conditions contributed to the

significance of the portion removed:

The comparison between sentences with the
constriction removed vs. the sentences with
the transition removed

The comparison between sentences with the
constriction removed vs. the sentences with
the transition and vowel removed

The comparison between sentences with the
transition removed vs; the sentences with
the constriction and transition removed

The com arison between sentences with the
transit on removed vs. the sentences with
the transition and vowel removed

The comparison between first order approximation
sentences with the constriction removed vs.
first order approximation sentences with the
constriction and transition removed

The comparison between first order approximation
sentences with the constriction and transition
removed vs. first order approximation sentences
with the transition removed

The comparison between first order approximation
sentences with the transition removed vs; the
first order approximation sentences with the
transition and vowel removed;

All comparisons contributed to the significance

in contextual and non-contextual word order except the

following:
1.

2.

Sentences with transition and vowel removed
vs. first order approximation sentences with
constriction and transitions removed

Sentences with the transition and vowel removed
vs. first order approximation sentences with
the transition removed

3;

75

93

Sentences with the transition and vowel removed
vs. first order approximation sentences with
the constriction removed

Sentences with the transition and vowel removed
vs; first order approximation sentences with
the transition and vowel removed

Sentences with the constriction and transition
removed vs; first order approximation sentences
'with the constriction and transition removed

Sentences with the constriction and transition
removed vs. first order approximation sentences
with the constriction and transition removed

Sentences with the constriction and transition
removed vs. first order approximation sentences
with the transition and vowel removed.

Each individual word was examined and it appeared

that the differences were across all of the words;

A confusion matrix.was constructed to examine

the type of perceptual errors which.were made under each

of the orders of presentation; It was found that, in

classifying errors as errors of place, voicing and

affrication. differences in these three perceptual

categories did existl

The vowels alone were found to contain little

information although the subjects did appear to apply

some linguistic constraints to the material as shown by

the high number of the's that the subjects guessed. In

terms of five word sentences. one would expect to find

a the in each of the sentences;

All of the findings of the study suggest that

contextual perception is a different process than when

9;»

one hears the words only as single entities; Further,

it appeared that the segments of the speech signal

removed seemed to make a difference;

CONCLUSIONS

Within the limitations of the instrumentation

employed and the design of this study, the following

conclusions are warranted:

l;

3.

The use of a magnet to remove selected portions
of a Speech signal in research on Speech
perception appears to offer potential for

future research;

The context in which a word is embedded has a
positive influence upon perception when either
the transitions or constrictions have been
removed;

then two of the three portions, either the
transitions, the constrictions. or the vowel.

are removed, contextual listening situations

are not significantly more intelligible than
non-contextual Speech; It is possible that

with so much information removed context is lost;
Linguistic rules appear to override auditory cuesl
gig being heard as the at the beginning of a
sentence indicates that linguistic rules are

being appliedl

7;

95

The total amount of the speech signal removed

has an effect upon perception. The more of the
signal removed, the less listeners are able to
understand correctly; This applies to both
contextual and non-contextual situations;

Despite the fact that approximately the same
amount of the speech signal is removed. the
removal of the constriction is more detrimental
to speech than the removal of the transition when
the speech signals are presented in a meaningful
linguistic context. but no significant difference
occurs in non-contextual situations.

Apparently in a situation where a forced choice
is not required, the vowels carry little or no
information: That is, if subjects hear only a
string of vowels and are not forced to decide

:n the vowel as a word. they appear to have no
information;

The perceptual errors in contextual speech differ
from the perceptual errors in isolated words.
Errors of perception based on affrication appear
much higher in contextual Speech than would be
expected on the basis of previous studies using
isolated words or than were found in the non-

contextual lists in the current study. This is

96

particularly true when the constrictive portion
is removed.
9. Perception of contextual material is a different

task than perception of isolated words.

RECOMMENDATIONS FOR FURTHER
RESEARCH

In view of the findings of this research. the
following recommendations for additional research are
presented:

1; This study should be replicated to see whether the
method of removal is adequate to obtain similar
results;

2. Sentences of various linguistic transformations
should be used to allow for further comparisons
*3 context;

3. Tarious orders of approximation could be employed
:3 allow for an analysis of varying degrees of

context.

{1"
.
:1»

large sampling of selected consonants should be
employed in contextual and non-contextual
situations to allow for more definitive results
in terms of perceptual confusions.

5. In looking at the amount of information available
in vowels, comparisons should be made between

forced choice and non—forced choice situations.

-4 As! WWI.”

 

 

7.

9;

97

Various speakers should be employed to determine
‘whether speaker variation results in any differences
in terms of effect of context or perceptual

removal.

An attempt should be made to compare perception

of the same listener on contextual and
non-contextual information to see whether one can
predict performance on one task based on their
performance on the other;

The study should be replicated on subjects of

 

various ages to see whether the influence of
context is affected by age. This would be useful
in understanding language acquisition;

Studies using similar materials could be employed
with clinical cases exhibiting various pathological
conditions: e;g.. hearing impairment: articulation

disorders: language delayed children.

 

BIBLIOGRAPHY

BIBLIOGRAPHY

Ahmend. R. and Fatehchand. R.. Effects of sample duration
on the articulation of sounds in normal and
clipped speech. J. Acoust. Soc. Ameg‘. 31. 1022-
1029 (1959).

Ainsworth. W. A.. Perception of stop consonants in
synthetic CV syllables. Language and Speech. ll.
139r156 (1968).

Ali. Latif. Gallagher. Goldstein J.. and Daniloff. R..
Perception of coarticulated nasality. J. Acoust.
Soc. Amer,. 49. 538-540 (1971).

Amerman. J.. Daniloff. R.. and Moll. K.. Lip and jaw
coarticulation for the phoneme AXV. J. Speech
Hearing Res,. 13. lh7e161 (1970).

Bastian. J.. Delattre. P.C.. and Liberman. A. M., Silent
intervals as a cue for the distinction between
steps and semivowels in the medial position.

J; Acoust.¥Soc. Amegi. 31. 1568 (1959).

Bever. T. G.. Lackner. J. R.. and Kirk. R.. The underlying
structures of sentences are the primary units of
immediate speech processin . Perception &

PS cho h sics. 5. 225-234 %l9695.

Bever. T. G.. Lackner. J. R.. and Stolz. W.. Transitional
. probability is not a general mechanism for the

segmentation of speech. J. exp, Psychol.. 79.
387-39“ (1969).

Black. J. W.. Predicting the intelligibility of words. In
A. Sovyarir and P. Aalto (Eds.). Proceedin s of
the #th International Con ress of Phonetic Science
(1952).

Brady. P. T.. Perception of s eech-like sounds. J.AAcoust.
Soc. Amer.. 32. 1501 $1960).

Brady. P. T.. House. A. S. and Stevens. K. S.. Perception
of sounds characterized by a rapidly changing
resonant frequency. J. Acoust. Soc. Amg§.. 33.
1357-1362 (1961).

98

 

 

99

Brown. R. w. and Hildrum. D. C.. EXpectancy and the
percgption of syllables. Language. 32. hll-hl9
195 .

Chomsky. N.. and Halle. M.. The Sound Patterns of En lish.
New York: John Wiley and Sons (1963).

COOper. P. S.. Delattre. P. C.. Liberman. A. M.. Borst.
J. M.. and Gerstman. L. J.. Some experiments on

the perception of synthetic speech sounds. ‘g;
Acoust. Soc. Amer.. 24. 597-606 (1952).

Daniloff. R;. and Moll. K.. Coarticulation of lip
rougging. J; Speech Hearinnges.. 11. 707-721
19 .

Delattre. P; C.. First formant transition as a cue to
place of articulation. J.gAcoust. Soc. AmegL.
46. 110 (1969).

Delattre. Pi C.. Liberman. A. M.. and Cooper. F. 8..
Second and third formant transitions in English
frigatives. JggAcoust. Soc. Ameghn 32. 1501
\19 0 .

Delattre. P. C.. Liberman. A. M.. and Cooper. F. 3..
Acoustic loci and transitional cues for
censonants. J. Acoust.gSoc. Amegi, 27. 769-773
.1955 .

Dukelsxiv, N. I. Cited in J. W. Falter and K. w. Ottan.
Tubernetics and Speech communication: A survey
*? Russian literature. IEEE Trans on Audio-
Electrpacoustics. AVlS. 27-3 1907 .

Fairbanxs. G. and Kodman. F.. Jr.. Word intelligibility as
a function of time compression. J, Acoust._Soc.
Amer.. 29. 636-6#1 (195?).

Fairbanks. C.. Gutman. N.. and Miran. m. 8.. Effects of
time compression upon the comprehension of
connected 3 eech. J. Speech HearinggDis4. 22.
10-19 (1957?.

Pant, C. G. M.. Descriptive analysis of the acoustic
aspects of Speech. Logos. 5. 3-17 (1962).

Fischer—Jdrgensen. E.. Acoustic analysis of stag
consonants. Miscellanea Phonetica. 2. 2-59

(1954);

 

100

FletcherE R..)Speech and Hearing. New York: Van Nostrand
1929 .

Fodor. J. Al. and Bever. T. C.. The psychological reality
of linguistic segments. Journal of Verbal
Learning and Verbal Behaviqg. 5. 5153520 (1965).

Foulke. E.. Comparison of comprehension of two forms of
compressed s eech. Exceptional Children. 33.
169-173 (1962).

Fry. D. B.. The correction of errors in the reception of
speech. Phonetica. 11. 16b-17h (196h).

Fujimura. 0.. Effects of vowel context on the articulation

of stopconsonants. J. Acoust. Soc. Amer.. 33.
842 (1961).

Garrett. Ml. Bever. T.. and Fodor. J.. The active use of
grammar in Speech perception. Perception and
PS cho h sics. 1. 30-32 (1966).

Garvey. w. D.. The intelli ibility of speeded speech.
J 9 . PS Ch°1!. 5' 102‘108 (1953).

Gerstman. L. J.. Noise duration as a cue for distinguishing
among fricative. affricate. and stop consonants.
J..Acoust. Soc. Amer.. 28. 160 (1956).

Gerstman. L. J.. Liberman. A. M.. Delattre. P. C.. and
Cooper. F. 5.. Rate and duration of change in
formant frequency as cues for the identification

of s eech sounds. J. Acoust. Soc. Amer.. 26.

952 €195fi).

Green. P. S..{Consonant-vowel transitions: A s ectrographic
study. Studia Linguistica. 12: 57-105 1958).

Grimm. W. A.. Perception of segments of English-spoken
consonant-vowel syllables. J._Acoust. Soc._Amer..
no, 1h5h-1h61 (1966).

Halle. M.. Hughes. G. W.. and Radly. J. P. 8.. Acoustical
properties of sto consonants. J. Acoust. Soc;
Amer.. 29. 107-11 (1957).

Harris. 0. M.. A study of the building blocks in speech.
girAcoust._Soc. Amggi. 25. 962-969 (1953).

Harris. K. S.. Cues for the discrimination of American
English fricatives in sgoken syllables. Laflggéga
and s eech. 1. 1—7 (195 ).

101

Harris. K. S.. Cues for the identification of the
fricatives of American English. ‘J. Acoust. Sog.
éflfiEi; 26: 952 (1954).

Harris. K. S.. Hoffman. H. S.. Liberman. A. m.. Delattre.
P. C.. and CoOper. F. 8.. Effects of 3rd formant
transitions on the perception of the voiced stop
consonants. J. Acoust. Soc. Amer.. 30. 122-126

(1958).

Hays. W. L.. Statistics. New York: Holt. Rinehart and
Winston. Inc. (1963).

Heinz. J. M.. and Stevens. K. N.. 0n the properties of
voiceless fricative consonants. J. Acoust. Soc.
ell-2.9.2.... 33. 589-596 (1958).

Hendrick. M.. Are phonemes really realized? In E.
Zwirnen and W. Bethge (Eds.). Proceedings of the

Eth International Congress of Phonetic Science.
2 30 1905 .

Henke. W. R.. Dynamic articulatory model of Speech
production. 1267 Conference on Speech Communica-
tion and Processing. Cambridge: The MIT Press
{1957).

 

 

. dun-11.x“ ‘— '

Hoffm.n. H. 8.. Study of some cues in the perception of
voiced stop consonants. J. Acoust. Soc. Amer..
j), 1035-10h1 (1958).

Hughes, 3. w.. and Halle. E.. Spectral preperties of
fricative consonants. J.AAcoust. Soc. Amer,.

;3, 303-310 (1956).

Ingemann, E.. Formants as a voicing one in the perception
of /z/S. J._Acoust. Soc. Amer.. 32. 1501 (1960).

Jakobson, R.. Fant. C.. and Halle. M. Preliminaries to
Speech Analysis. Cambridge: MIT Press (1963).

Kelly. 5.. Anthony. J. K.. and Uldall. E.. Tempo and
tragsition. J. Acoust. Soc. Amer.. 35. 1113
19 3 .

Klumpp. R. C.. and Webster. J. C.. Intelligibility of
time-compressed s eech. J. Acoust. Soc. Ame;..
31. 265-267 (1961).

102

Kozkevnikov. V. A.. and Chistovich. L. A.. Rech
Artikuliatsieu i Vos ruatie. Moscow-Leningrad.
(Translation Speech: Articulation and
Perception) Washington. D. 0.: Joint
Publications Research Service (1965).

Kurtzrock. G. R.. The effects of time and frequency
distortion upon word intelligibility. Speech
Monographs. 2n. 9% (1957).

Ladefoged. P.. Three Areas of E erimental Phonetics.
London: Oxford University Press (1967).
Ladefoged. P. and Broadbent. D. E.. Information conveyed

by vowels. J..Acoust.gSoc. Ame£&. 29. 98-10“
(1957).

Lane. H.. The motor theory of speech perception: A
irigigal review. Psychol. Review. 74:6. 431-461
19 7 .

Lawrence. D. L. and Byers. V. W.. Identification of
voiceless fricatives by high frequency hearing
in aired listeners. J Speech Hearin Res . 12.
42 -h34 (1969).

 

Liberman. A. M.. Some results of research on Speech
erception. J. Acoust. Soc._Amer;. 29, 117-123
1957 .

Lieberman. P.. Intonation. Perception and Language.
Cambridge: The MIT Press (1967).

Liberman. A. M.. Delattre. P. C.. and COOper. F. S.. The
role of selected stimulus-variables in the
perception of the unvoiced stOp consonants.
Amer J of Ps chol . 65. “97-516 (1952).

Liberman. A. E.. Delattre. P. C.. COOper. F. S.. and
Gerstman. L. J.. The role of consonant vowel
transitions in the perception of the step and
nasai)consonants. PsycholI Monog.. 68:379. 1-13

195 .

Liberman. A. M.. Delattre. P. C.. Gerstman. L. J.. and
Cooper. F. 8.. Tempo of frequency change as a cue
for distinguishing classes of speech sounds.

J. exp. PSXCthJ 529 127-13? (1956).

Lindblom, B. E. P.. and Striddert-Kennedy. M.. 0n the
role of formant transition in vowel reco ition.
J._Acoust. Soc. Amer.. 42. 830~893 (1967 .

103

Lisker. L. and Abramson. A. 8.. Some effects on context
on voice onset time in English stops. Lang.
Speech. 10. 1-28 (1967).

Lotz. J.. Abramson. A. A.. Gerstman. L. J.. Ingeman. P..
and Nemser. W. J.. The perception of English
steps by speakers of English. Spanish. Hungarian.
and Thai: A tape-cutting experiment. Lang.

S each. 3. 71-77 (1960).

Malécot. A.. Acoustic clues for nasal consonants: An
experimental study involving a tape Splicing
technique. Langggge. 32. 274-284 (1956).

Malécot. A.. The role of releases in the identification
of released final steps. Langggge. 32. 274-289

'(1955).

McLain. J.. A comparison of two methods of producing
rapid Speech. Internatienal Journal for the
Education of the Blind. 12. O- 3 l9 2 .

Mattingly. I. G. and Liberman. A. E.. The speech code and
the physiology of language. In K. N. Leibovic
(Bd.). Information Processing in the Nervous
Svstem. Springer Verlag (1970). _*

Prlattingly. I. C.. Liberman. A. 14.. Syrdal. A. K.. and
fialives. T.. Discrimination of F2 transition in
speech context and in isolation. JL Acoust. SopL
gear.. #5. 314-315 (1969).

Miller. :.. The magical number seven. plus or minus two:
some limits to our capacity for processing
information. PS chol' Rev . 63. 81-97 (1955).

Miller. 3. and Licklider. J.. The intelligibility of
interrupted s eech. J. Acoust. Soc. Amer.. 22.
167-173 (1950?.

Miller. 3. A. and Nicely. P. E.. An analysis of perceptual
confusion among some English consonants. jg.
Acoust. Soc. Amegi, 27. 338-352 (1955).

M011. K. L.. Cineflurographic techniques in Speech
fesgagch. J. Speech Reaping Res.. 3. 227-2hl
l9 0 .

Moll. K. L.. Velopharyngeal closure on vowels. J. Speech
Hearing Res,. 5. 30-37 (1962).

104

Moser. H. N.. One-syllable Words. Columbus. Ohio: C. E.
Merrill Publishing Company (1969).

Nakata. K.. Synthesis and perception of nasal consonants.
JI Acous . Soc. Amer.. 31. 661-666 (1959).

O'Connor. J. D.. Gerstman. L. J.. Liberman. A. N.. Delattre.
P. C.. and Cooper. F. 8.. Acoustic cues for the
perception of initial /w. j. r. l/‘in English.

.EQEQ: 13. 2fi-43 (195?).

ahman. S. E. C.. Perception of segments of VCV utterances.
. Acoust. Soc. Amer.. 40. 979-988 (1966).

Peterson. G. E.. and Barney. H. L.. Control methods used
in a study of the vowels. J.gAcoust. Soc. Amepi.
24: 175'184 (1952).

Pickett. J. M. and Pollack. I.. Adjacent context and the
intelligibility of Speech excised from fluent
speech. J. Acoust. Soc. Amer,. 35. 807- (1963).

Pollack. I. and Pickett. J. N.. Intelligibility of
excerpts from fluent speech: Auditory vs. struc-
tural context. Journal of Verbal Learning_and
Verbal Behaviog. 3. 79- 190 .

Potter. R. K.. Kepp. G. A.. and Green. H. C.. Visible
Speech. New York: Van Nostrand (1947 .

Reed. J. A. and Wang. W. S-Y.. The perception of steps
after 8. Phonetica. 6. 78- 1 (1961).

Schatz. C. D.. The role of context in the perception of
steps. Langi_Speech. 30. n7-56 (195h).

Shearne. J. N. and Holmes. J. N.. An experimental study
of the classification of sounds in continuous
speech according to their distribution in the
formant l and formant 2 plane. In A. Sovyarvi

and P. Aalto (Eds.). Proceedin s of the Fourth
International Congress of Phonetic Science (1962).
Sherman. G. Cited in R. M. Warren and C. J. Obusek.

Speech perception and phonemic restoration.
Perception and Psycho h sics. 9. 358-362 (1971).

Soderberg. G. A.. Linguistic factors in stutterin .
J. Speech Hearing Res,. 10. 801-810 (1967 .

105

Stevens. K. N. and House. A. S.. Perturbations of vowel
articulation by consonantal context: An acoustical
study. J;_Speech Hearing;Res.. 6. 111-128 (1963).

Stevens. K. N.. House. A. S.. and Paul. A. P.. Acoustic
description of syllable nuclei: An interpretation
in terms of a dynamic model of articulation. J;
Acoust. Soc. Amer.. #0. 123-132 (1966).

Straus. O. E.. Phonetics and communication. J. Acoust.
Soc; Amer.. 22. 709-711 (1950).

Strevens. P.. Spectra of fricative noise in human Speech.
Lang SpeeCh: 3v 33'14'9 (1960).

Swaffield. J.. Shearne. J. N.. and Holmes. J. N.. Some
measurements of vowel sounds of conversational
speech. J. Acoust. Soc._Amer.. 33: 1683 (1961).

Warren. R. m.. Perceptual restoration of missing Speech
sounds. Science. 167. 392-393 (1970).

Warren. R. M. and Obusek. C. J.. Speech perception and
phonemic restoration. Perception and Psycho-
LFM‘fSlCS, 99 358'362 (1971).

Wickelgren. A.. Distinctive features and errors in short~
term memory for English consonants. JLgAcoust.

m. 39. 388-398 (1966).

. u“ “‘2 "1' 3;-
f

 

APPENDIX
RAW SCORES FOR 10 SUBJECTS
IN THE EIGHT EXPERIMENTAL
CONDITIONS

106

EXPERIMENTAL CONDITION 1

(Meaningful Sentence with the Consonant
Constrictions Removed)

 

 

 

Subject 1 2 Sgntenci 5 6 Total
1 C 4 3 h 2 3 20
2 a L» 3 1: L: 1 20
3 3 4 3 O h 0 14
4 h h u a 4 l 21
5 2 5 h u a 4 23
6 3 5 3 h 4 3 22
7 3 4 3 O 4 l 15
8 l: 3 3 o - 5 5 20
9 3 4 4 3 # 3 21
1° .1 .3. .2 .9. J .2 16

Total 33 #1 33 23 38 24 192

 

107

EXPERIMENTAL CONDITION 2

(First Order Approximation with the
Consonant Constrictions Removed)

 

 

 

Sentence

Subject 1 2 3. h _5 6 Total
1 3 2 l 1 1 1 9
2 2 I 4 2 1 2 12
3 2 2 2 2 2 1 11
4 2 2 l 2 1 1 9
5 1 2 1 0 1 1 6
6 l 2 0 2 1 1 7
7 3 2 0 2 l l 9
8 2 2 1 2 1 0 8
9 2 2 2 2 1 1 10
10 .i .2 ...1. ._2. .2. ._1. ._2

Total 19 19 13 17 12 10 9O

 

108

EXPERIMENTAL CONDITION 3

(Meaningful Sentence with the Consonant
Constrictions and Transitions

 

 

 

Removed)
Sentence

Subject 1 2 3 4 5 6 Total
1 2 O O O O O 2

2 3 1 1 O 1 O 6

3 1 l 1 o 0 0 3

4 O 1 O O O O l

5 O 2 O O O O 2

6 O 1 O O O O 1

7 2 3 l O O 0 6

8 2 1 O O O O 3

9 2 1 o o o 2 5
10 .2 .1 .2 .9. .9. .1 .1
Total 15 12 5 O 1 3 36

109

EXPERIMENTAL CONDITION 4'

(First Order Approximation with the Consonant
Constrictions and Transitions
Removed)

 

 

Sentence

§ubject 1 2 3 4 5 6 Total
1 o o ' o o o o o

2 0 O O O O 0 O

3 0 0 1 O 1 O 2

4 O O 0 O O O O

5 O 1 O O O O l

6 O O 1 O O O l

7 0 0 O O O O 0

8 1 0 0 0 l O 2

9 O O O O O 0 O
19 .9 .9 .9. .9 .9. .9 .9.
Total 1 1 2 1 1 0 6

 

110

EXPERIMENTAL CONDITION 5

(Meaningful Sentence with Transitions Removed)

 

 

 

Subject 1 2 Sgntence 5 6 Total

1 5 5 5 5 5 4 29

2 4 5 4 3 5 4 25

3 5 4 3 5 3 4 21:

4 5 5 1+ 2 3 3 22

5 3 5 3 5 4 4 21+

6 5 5 5 5 2 1+ 26

7 5 5 5 4 5 5 29

8 5 5 5 5 5 5 30

9 5 5 3 4 S 5 27
10 .1 .5 ._{t ._‘i .i .3. .352
Total 45 49 41 43 42 41 261

 

111

EXPERIMENTAL CONDITION 6

(First Order Approximation with
Transitions Removed)

 

 

 

 

 

 

Sentence

 

Subject 1 2 3 4 5 6 Total
1 3 3 3 3 0 3 15
2 3 2 4 4 1 3 l7
3 2 o 2 0 1 0 5
4 2 1 4 3 l 3 14
5 2 0 3 3 2 2 12
6 2 2 4 2 l 2 l3
7 2 O 2 O 0 2 6
8 2 2 4 2 1 2 13
9 2 2 2 3 1 0 10
10 ..2. .9 .9 .9 .1 .9 _1

Total 22 12 30 20 9 19 112

 

112

EXPERIMENTAL CONDITION 7

(Meaningful Sentence with the Transitions
and Steady State Vowel Removed)

 

 

 

Subject 1 2 Sgntence 5_ 6 Total
1 O 2 O 1 0 4 7
2 0 2 0 1 0 4 7
3 2 o o 3 o 2 7
4 o 1 o o o 3 9
5 0 2 O l 0 4 7
5 5 3 0 o o 4 12
7 1 4 o 1 0 4 10
8 1 1 O O O 4 6
9 o 2 o o o 4 6
10 ._Q. ._2. .9. ._0. .9. .9. _‘.5.

Total 9 19 0 7 O 37 72

 

113

EXPERIMENTAL CONDITION 8

(First Order Approximation with the
Transitions and Steady State
Vowel Removed)

 

p

 

 

RIF—2 Sentence

Subject <;__ 2 3 4 5 6 Total
1 1 O l l O 0 3
2 O O 2 0 O O 2
3 O O O O 0 0 O
4 O O 2 O O l 3
5 0 0 3 o 0 o 3
6 1 O 0 O O O 1
7 l O 2 O O O 3
8 l O 2 O O O 3
9 2 O O O 0 O 2
lo .22. .9 .9 .1 .9 .9 .5

Total 8 O 12 4 O l 25

   

”'TITJiTIfliIT/LfljfllLufirLiIliujffl'fliyimm117111111“

8 0249