Income? AN IKDIVIDUAL DIFFERENCE VARIABLE .

Thesis for’the Degree of Phi" D
Michigan State University

TERRY DEAN TENBRINK

1968

 
      
  

  

. LIBRARY

Michigar State
Univcn .ity

This is to certify that the

thesis entitled
Recoding — An Individual Difference Variable
presented by

Terry Dean TenBrink

has been accepted towards fulfillment
of the requirements for

_Bh_LD_._degree in Education

0-169

 

 

 

 

‘...
_
—
A

 

ABC) 1".(AC P
decoding - in Individual Difference Variable

3y ferry Dean Fendrink

The major purpose of this study was to utilize the transformational
model of sentence structure to investigate individual differences in
prose learning. Five tests were constructed to measure a person's
ability to understand and recall sentences of varyinq levels of com-
plexity (recoding skills). One hundred fifty four high school 85
were administered these tests as part of a battery of tests which also
included measures of IQ, vocabulary, digit Span, readin: rate, and
reading comprehension.

Several statistical analyses were made upon the data in an effort
to discover the relationships which existed among the various measures
obtained from the test batterv. Intercorrelation matrices and factor
analyses indicated that high linear relationships existed awong all
the variables under study except reading rate and digit Span which had
low correlations with the other variables and tended to load on separate
factors.

Further analyses were made on the sentence skill tests alone in
an effort to determine whether or not they were all measurinz the same
general skill. The intercorrelations were high amonq three of the five
sentence skill tests. These three tests were proported to measure
the 3's sensitivitv to the level of grammaticality (grammaticality

test), his ability to understand deeply embedded sentences (self-

embeddedness test), and his ability to rely on syntactic meaninz in
the sentence, even when it conflicted with semantic meaninq (semantic-
syntactic conflict test). A fourth sentence skill test, designed to
measure the 3's sensitivity to ambiguity (ambiguity test), correlated
moderately with the first three tests discussed above. The fifth
sentence skill test, constructed to measure ability to recall sentences
exactly (fransformational complexity test) did not correlate well with
any of the other tests and in a factor analysis, this test loaded on
its own factor.

In an effort to discover the importance of these sentence skill
tests in prose learning, a multiple regression design was utilized.
Two different reading comprehension tests were constructed, one in-
volving memory and the other one not involving memory. These tests
were used in two different forms (form A - written in complex sentences
and form B - written in simple sentences) as the criterion variables
in the multiple regression analyses. IQ, vocabulary, reading rate,
digit span, and recoding (as specified by the five sentence skill tests)
were the predictor variables. A least squares ordered delete redression
program was used to determine the unique variance accounted for by each
of the independent variables. Vocabulary was demonstrated to contri-
bute more unique variance to reading comprehension on a whole than
any other single predictor. The next best predictor was the semantic-
syntactic conflict test which was the largest contributor to the
variance of the complex form of the reading test involving memory. The

other sentence skill tests contributed small but unique proportions of

variance. Neither reading rate nor digit Span contributed any unique
variance. IQ, a traditional predictor of reading comprehension, was
a poorer predictor than either vocabulary or recodinz.

The semantic-syntactic conflict test contributed more to the
variance accounted for in the complex versions of the reading tests
than in the simple versions but the other sentence skill tests did
not follow that pattern as was predicted.

There were no significant differences between the mean readint
comprehension scores of the two versions of the tests. 33 did as
well when tested on the reading passages containing complex sentences

as they did on the passages containing simple sentences.

"JCOUINJ - AN erIVIUU L DIFFeaffiCn VAUIADLL

5y

ferry Jean fendrink

A THQQIS

Jubmitted to
Lichiqan state University
in partial fulfillment of the requirements
for the dearee of

JCCTCK OF FHlLUdUtdY

Colleze of education
department of educational fsycholoqy

196R

.14 ar‘w 2
I ~ 2.

,7 ./ (j 0 s)
r

I
(’_ (I _, ét.

.42"

Copyright by
TERRY DEAN TENBRINK
1968

A CK N OWLE DGEMEN T

The author wishes to take this opportunity to express his
appreciation to Dr. Clessen J. Martin for his assistance and

guidance throughout this study.

Sincere appreciation is hereby also expressed to my wife, Betty,
for her encouragement and patience and to my three children, Susan,
Jill, and Douglas, for their understanding and patience during the

time this thesis was being prepared.

TABLE OF

Acknowledgement . . . . . . . . .

List of Tables . . . . . . . . .

List Of Figues 0 0 C O O O O O O 0

List of Appendices . . . . . . . .

I.

II.

III.

IV.

INTRODUCTION
General Problem.Area . . . .
Objectives and Hypotheses . .

REVIEW OF THE LITERATURE

CONTENTS

The Psychological Reality of Receding
Sentence Characteristics Which Hinder

Receding
Receding, an Individual Difference Variable . . . .

The Relationship Between Receding and Reading Compre-

Summary of Literature Review and Purpose of the Present
Study .

IETHOM O O O O O O O O O I 0
Subjects

Instrumentation and Procedures.

Objectives and Hypotheses . .

RESULTS . . . . . . . .
Means and Standard Deviations
Intercerrelatiens . . . . . .
Reliability Estimates . . . .
Factor Analyses . . . . . . .
Multiple Regression Analyses

DISCUSSION
Objective
Objective
Objective
Objective
Objective
Objective
Objective
Implications for further study

\onmirmmI—J

SWRY O O O O O O 0 O 0 I 0

iii

hension

Predictors of Reading Comprehension . . . . . . . . .

O O O O O O O O O

LIST OF TABLES

A list of the sample populations indicating the number
of Ss assigned to each group in the various samples . .

Number of words, number of sentences and number of dif-
ferent types of sentences and clauses in the two ver-

10

ll

12

13

14

15

16

siens of the RM and RNM reading passages . .

A list of abbreviations used in this thesis . . . . .

Means and Standard Deviations of all variables for

g roup A O I O O O O O I O O O O O O O O O O O I I O O 0

Means and Standard Deviations of all variables for
group B . . . . . . . . . .

Intercerrelation matrix of all variables for group A

Intercerrelation matrix of all variables for group B

Intercorrelation matrix of five sentence skill tests

for the

KudernRichardson r20 reliability coefficients obtained

combined groups . . . .

for groups A and B on the sentence skill tests, the

reading comprehension tests, and the vocabulary tests .

Varimax rotated factor loadings of variables: IQ, V0,
RR. RE. R1“ for group A O O O O O O O C O O O O O O O
Varimax rotated factor loadings of variables: IQ, V0,
RR, RE and RM for group B . . . . . . . . . . . . . .
Varimax rotated factor loadings of variables: IQ. V0,
RR, RE, and RNM for group A . . . . . . . . . . . . .
Varimax rotated factor loadings of variables: IQ, V0,
RR, RE, and RNM for group B . . . . . . . . . . . . .
Varimax rotated factor loadings of variables: IQ, V0,
RR, GR, SE, SS, AM, TR, and RM for group A . . . . .
Varimax rotated factor loadings of variables: Q, V0,
RR, GR, SE, SS, AM, TR and RM for group B . . . .
Varimax rotated factor loadings of variables: IQ,

RR, GR,

SE, SS, AM, TR, and RNM for group A . .

iv

52
59

61

62
64

65

67

68

68

69

70

71

72

17

18

19

20

21

22

23

2Q

25

Varimax rotated factor loadings of variables: IQ, V0,
RR, GR, SE, SS, AM, TR, and RNM for group B . . . . . .

Varimax rotated factor loadings for the sentence skill
tests for both groups . . . . . . . . . . . . . . . . .

R25 for full and restricted models using IQ, V0, and
RE as predictors of reading comprehension for groups A
and B O O O O O O O I O O I O O O O O O O I O O O O O 0

R23 for full and restricted models using IQ, V0, GR,
SE, SS, AM, and TR as predictors of reading comprehen-
sion for groups A and B . . . . . . . . . . . . . . . .

Predictability indices (R2 for FM-II minus R2 for R2)
of IQ, V0, GR, SE, SS, AM, and TR when used to predict
RM and RNM for groups A and B . . . . . . . . . . . . .

Predictability indices (difference between R; and R32)
of important combinations of the independent Variables
IQ, V0, GR, SE, SS, AM, and TR when used to predict RM
and RNM for groups A and B . . . . . . . . . . . . . . .

Predicted RM and RNM scores for four hypothetical Ss
differing on IQ and RE scores . . . . . . . . . . . . .

Predicted RNM scores (simple and complex versions) for
four hypothetical Ss differing on RE scores only. Pre-
dictions were made from FM I regression Equations .

The means and standard deviations of the RM and RNM
scores for the 33 who scored in the upper and lower 20%
on recording (RE) . . . . . . . . . . . . . . . . .

73

7a

79

81

82

83

98

100

103

LIST OF FIGURES

Diagram representing the relationships among the key
concepts in Chomsky's model of transformational
gramr O O O O O O 0 O O O C C C O O I O O O I O O O O 5

Graph indicating relations among eight types of

sentences formed by the negative, passive, and inter-
rogative transformations (Suggested by Mehler's data--

from Miller, 1962) . . . . . . . . . . . . . . . . . . 32

Graph indicating relations among eight types of san-
tences formed by negative, passive, and interrogative
transformations (suggested by Clifton and Odom's 1966
data) . . . . . . . . . . . . . . . . . . . . . . . . . 32

\O CDV O\U\ (two N

|-'
0

LIST OF APPENDICES

Mean item difficulty and mean item discrimination
indices for the sentence skill tests, the reading

comprehension tests, and the
groups A and B . . . . . . .

Vocabulary Test
Digit Span Test

Grammaticality Test

SelféEmbedded Sentence Test .

Semantic-Syntactic Conflict Test .

.Ambiguity Test . . . . . . . . .

Transformational Complexity Test .

Reading Comprehension Test (Memory): Forms A and B

Reading Comprehension Test (NonéMemory): Forms A and

B

vocabulary test

for

119
120
124
125
129
133
140
1244

152

165

‘31.}"1‘317; I - INTRODUCTIOE‘Q

General Prpblem Area

a”..-

 

The transformational model of the structure of language as pro-
posed by Chomsky (1957, 1965) has stimulated a great amount of research
dealing with the problem of sentence learning. Psycholinguists, now
having at hand a carefully specified model of the structure of.£ng1ish
sentences, have begun to investigate some of the variables in sentence
learning which have been suggested by transformational grammar. Many
variables have been found to be important in sentence learning. Among
them are, the level of grammaticality (Miller and Selfridge, 1951),
the number of transformations within a sentence (Mehler, 1963), and
the amount of agreement between the grammar and the semantics of a
sentence (Marks and Miller, 1964). For the most part the studies have
investigated the effects these manipulations have upon sentence recall
and comprehension. The process underlying the behavior of the 35 has
been postulated by Miller (1962) to be a process of utilizing trans-
formational rules to "break-down" each sentence into its kernel elements.
According to killer, the process he calls recoding involves the assign-
ment of a constituent structure to each sentence so that it can be
stored in kernel form. Miller (1962) cites several studies which
:indicate that the recoding process has psychological reality--it is not
rnerely a process in the mind of the grammarian but there is evidence
lior the fact that peeple do engage in some sort of recoding process in

Eittempting to comprehend and/or learn sentences.

Although there have been a number of studies indicating the
psychological reality of recoding, little or no research has been con-
ducted investigating individual differences in recoding ability. The
major purpose of this study is to utilize the transformational model
of sentence structure to investigate individual differences in the
skills of recoding and to examine the importance of these individual
difference variables in prose learning.

Although the studies in sentence learning have utilized the models
of grammar as proposed by the linguists, the prose learning studies
have more often been attempts to answer practical questions and if the
studies were based upon any theory or theoretical model, they were
based upon incomplete models or at least models which have not accounted
for the grammatical structure of the language (c.f. Rothkopf, 1965).
Chomsky {1957a} maintains that not only is a theory of performance (pro-
duction and/or perception) important, but this theory must incorporate
the theory of competence, i.e. structure of the language. Ehe position
taken in this thesis is that a generative, transformational Iodel cf
language offers a {cod 'escriztiVe model of the structure of the
anglish language, and may provide a framework for the investigation of
individual differences in prose learning.

Probably the most complete model of prose learning offered to date
is that propcsed by RothkOpf (1965). RothkOpf's model is a conceptual
model of prose learning which leans heavilv upon traditional learning
theory of the Rullian type and is very similar to Cook and Kendler's

model for paired-associate learning. Rothkopf asserts that "subject

\O

matter learning is assumed, in the main, to depend on temporal contiru-
ity." According to hothkopf, associations are formed anon; sentence
components and a necessary althourh not sufficient condition for the
formation of these associations is temporal continuity anon; articula-
tion responses. .e further asserts that a study of the relationships
between sentence structure and intonation pattern on the one hand and
learning on the other is both feasible and desirable. @othkopf has
concentrated his research efforts on the relationship between the
nature of the prose material and the mathemagenic behaviors of the
subject reading the material.

Clearly within Rothkopf's model there is a need for a precise
description of the structure of sentences and a careful investigation
of the individual difference variables interacting with the sentence
structure. Perhaps due to his emphasis on the effect of manipulations
such as amount and Spacing of repetition and test like events within
the prose passaqe he has not incorporated within his model any careful
description of sentence structure. It has already been suggested that
transformational grammar offers an excellent description of the struc-
ture of language and therefore a brief summary of Chomsky's transforma-
tional grammar is given below and may provide the basis for investiga-
tion of some of the "unobservable" mathemaqenic behaviors referred to
by dothkopf.

Eghomsky's‘Theory_gf;Iransfgrmgtipn_l,Grammar
At the core of Chomsky's transformational grammar is the notion

tfliat language is generative, i.e. it makes infinite use of finite means.

Languaqe is a structure of forms and concepts interrelated by a system
of transformational rules. This generative grammar, accordinq to
Chomsky, is internalized in every normal human. It acts as a theory
of irammar for the person and provides him with a sound-meaning cor-
relation for an infinite number of sentences. At the base of this
fienerative grammar, the person has a set of "base" rules which provide
him with a lexicon (including phonological and semantic features) and
a categorical system (a kind of context-free phrase structural trimmer).
From these two sets of base rules, deep structure sentences are derived.
These deep structure sentences are in the form of phrase markers, and
transformational rules apelied to this deep structure yield surfice
sentences. There are two types of transformational rules--obliqatory
and optional. When only obligatory rules are performed on the deep
structure, the result is a phrase-marker which, when represented
phonetically, is a kernel sentence. When any optional transformation
is applied to the deep structure, phrase markers result and can be
represented phonetically or can be further changed by other optional
transformations. Any number of these optional transformations can be
mapped onto these phrase markers creating more and more complex sen-
tences which, when finally represented phonetically, result in complex
Surface structure sentences. For example, when only obligatory rules
are applied to the deep structure. "John aux. hit Pete," the resultinz
Surface structure would be, "John did hit Pete." However optional
‘transformations can be made upon that same deep structure to yield

snlrface structures like: "Pete was hit by John," ”Did John hit Pete?",

\Jl

"Fete was not hit by John, was he?", etc.

Just as the surface structure can be represented phonetically, so
the deep structure has a semantic representation. It is this deep
structure in its phrase marker form and its corresponding semantic
representation which, according to Killer and others, is stored in
memory. (For a diagrammatic representation of this system see Figure 1.)
This model of the structure of language will be used as a theoretical

linguistic basis for the discussion of the recodinz process which

 

 

 

follows.
Base Rules
context-free Cate~orical Lexicon phonoloaical and
phrase structural system semantic features
qrammar \\ \/ /
\ V. ~ ., . -_, '_
\ Jgqu .) EUUQ'JUJM
\
Semantic lepresentation \ Transformation nules
\
\ ,‘ . fl. "N .fi - y f" I-, I
\ bUzil‘ACfl. JI‘AUQTUJLL‘.
\
\ _ ,.
\ Phonetic nepresenta-
\ . 1 - V
\ tion (whicn in many
\ languages can be trans-
\
\ lated into written
‘\symbols)
\
\

 

Figure 1: Diagram representing the relationships amonq the key concepts
in Chomskyjs model of transformational grammar.

 

The central problem of this study is the determination of whether
<>r not an ability to recode sentences is an individual difference
\rariable that correlates positively with a person's ability to compre-

Puend prose material. However, prior to the development of measurement

tcchnieues dcsirned to measure recoding skills, an attempt was made
in this study to define the psycholotical processes involved in recoding.

The notion of recoding was first stated by Liller (1962) and grew
out of a lorical arqumcnt for the proposition that ”7e cannot under-
stand a sentence until we are able to assirn a constituent structure
to it." Killer argued that the proper function of this syntactical
skill necessary to assignins a constituent structure to a sentence is
not only an essential ingredient in the process of understanding a
sentence, but it hizhlizhts the fact that the problem of “caning
involves a urea deal more than the natter of reference alone. Killer
then extended this notion to the sumrestion that what people do in
order to remember sentences could be called recoding. He argued that
what people remember is a kernel sentence with a "footnote" about the
syntactical structure. This footnote according to Viller enables the
subject to make the necessary grammatical transformations when he is
called upon to recall the sentence verbatim.

Tiller cites several studies which he claims lend empirical evidence
to this notion that sentences are stored in kernel form. In 1951,
miller, Heise, and Lichtcn conducted a study in which they reported
that 35 heard words more accurately in sentences than in isolation.
'This stucy was followed by a series of studies in which tiller and
cathers compared the recall of sentences with the recall of random
iatrinzs of words. On measures of latency, recognition, and verbatim
Ivacall; sentences which were most wranmatical were recalled with least

~‘I

GPIVDr° Viller concluees, then, that the mechanism in undwrstfndinz

sentences is complex and that the context of the sentence allows us
to organize the flow of sounds into position units which are larger
than individual words. These "larger-than-individual~words" units
might be considered phrase structures, kernel sentences, or, in terms
of Chomsky's most recent theorizing, deep structures.

Transformational grammar postulates that transformational rules,
acting upon the deep structure, result in surface structure sentences.
Our general definition of recoding might be stated as a reverse of this
process, i.e. persons, upon perceiving a surface structure sentence,
somehow become aware of the deep structure(s) underlying the surface
structure; and they store this, perhaps with some transformational tag
indicating the transformational rules which were involved.

In order to gain some insights into the possible kinds of skills
necessary for recoding, it may be well to speculate concerning the recod-
ing process, noticing in particular what the language user perceives be-
fore the recoding process begins, and what he perceives when he is fin-
ished. Although this examination of what the language user perceives may
involve some conjecture as well as a certain amount of introspection, it
should nevertheless at least provide a starting point toward a better
understanding of recoding and sentence comprehension.

Before the recoding process begins, a person has manipulated the
linguistic input to the point where he has finally identified the pro-
per referents to each of the meaning units in that input. This means
that for a given sentence the person has before him a sequence of mean-
ingful linguistic units. He must now see the interrelationships between
'these meaning units so that he will "understand" the sentence. This

'understanding involves being aware of who or what is doing, being done

to, describing, being described, giving, or receiving; and where in

time or place these things have occurred. In other words he must be
able to associate the subject with its predicate, the predicate with

its object, the adjective with its noun, etc. In the deep structure

or kernel sentence only those elements which must be associated together
are included. These simple relationships which exist in deep structure
sentences can be clearly seen from an analysis of the phrase markers

which comprise the ten basic kernel sentences described by Roberts

(1962).
1. N -- v 6. Nlnv 312192
2. N -- V Adj 7. Nl--V NZN3
3. “1-- V N1 8. Nl--be Adv (time and place)
4, N1-- v NZ 9. Nlnbe Adj
50 Nl-- V N2 Adj 10. N1.--” N1

This analysis goes beyond "the matter of reference alone" and is
in agreement with Miller's (1962) statement that meaning is more than
mere reference. Reference, although necessary is not sufficient for
sentence understanding; the relationships among the grammatical parts of
the sentence must also be correctly perceived. Vocabulary and recoding
then are both likely to be important predictors of reading comprehension
for the S must conjure up appropriate referents first of all and then
he must be able to form the correct relationships among the referents as
dictated by the syntactic structure of the sentence.

It may be postulated that there are certain cues which a person
ggets from the sentence which help him make the appropriate associations

Eunong the various sentence parts. One type of one which is particularly

valuable in a language like English, where word order is so important,
might be termed position-cue. The position of a unit within an English
sentence or phrase gives a hint as to its function as well as to its
meaning in that particular sentence. For example, the function and
meaning of the word "fast" changes from sentence to sentence and the
changes are reflected in the position within the sentence that the word
occupies. Notice the shift in position and meaning of "fast" in the
following sentences:

1) John is a fast boy. (between a determiner and a noun - modifies

boy - an adjective)

2) John is running fast. (following an intransitive verb -

modifies is running - an adverb)

3) A fast is difficult for John. (preceding the verb, following

a determiner - subject - a noun)

4) John could not fast for long. (following subject, auxillary,

and negative adverb - a verb)

A second type of cue which is useful in understanding which ele-
ments of a sentence function in which capacity is word-form cue. The
form of a word or the endings it takes is an indication of how it is
functioning in a sentence. For example, only nouns (and pronouns) can
be pluralized: the -ly ending usually signals an adjective or adverb;
the verbs take the endings showing time ( -ed for past, for example).

Then there are, of course, semantic cues which point up the im-
‘portance of vocabulary in sentence understanding. If a person were not
érware of the various meanings that "fast" could have, he would have a

ctifficult time understanding the four sentences used above.

10

Another type of cue which relies to some extent upon the other cues
(especially position and word form) is termed transformation cue. The
correct position of a subject, for example, will differ, depending upon
what transformations have been made on the sentence. Subject-verb-
object is the correct word order for an active declarative sentence, but
that order reverses if the sentence is passive. The 3 must recognize
which transformations have been made in order to correctly identify the
various sentence parts on the basis of position. Word form also
changes with the various transformations and thus transformation cues
are important aids in sentence understanding.

The above four kinds of cues enable a reader to "understand" Lewis
Carroll's "Jaberwocky" (1961) which begins:

"Twas brillig and the slithy toves
did gyre and gamble in the webs

all mimsy were the borogoves
and . . ."

Note that some "understanding" is possible despite the fact that the
semantic cues are at a minimum in this poem.

The usefulness of these cues lies in the fact that they help a
person to reduce the sentence to some simple or kernel (deep structure)
form and then to make the correct associations among the major sentence
parts ("doer," "did," "done to"). But how does a person get from sur-
face structure to deep structure, from complex sentence to kernel sen-
‘tence? Perhaps as a first step, one must discriminate among the func-
‘tional units of a sentence. He must decide which element is the subject
sand which the verb: which the mofifier and which the modified. Occasion-

Eilly in making these decisions some of the cues found in the sentence

11
are in conflict. For example, semantic cues and word-form cues tell one
that baseball is a noun and usually functions as a subject or object.
However, occasionally a noun may be in a modifier position as in "bagg-
ball bat" and therefore one must suppress the semantic and word-form
cues and rely on position cue realizing that the "noun" is a modifier
and adjusting the sentence meaning accordingly.

Another step a person must take in getting to the "kernel" of a
sentence is to decide which functional units are to be associated to-
gether. All of the cues discussed are important in this decision but
word form cues are often most important in this step. In a sentence
like "fkfinacarried the glove and the ball and the bat was carried by
Pete," there are many possible interpretations of who carried what.
Pete, however, did not carry anything but the bat. The important one
in this sentence is the form of the verb "be." "Was" is singular and
therefore agrees with the subject bat which is also singular. If the
subject was "the ball and the bat? then the appropriate verb form would
be "were" to agree with the compound subject. It should be noted in the
above example that temporal contiguity is not a sufficient condition to
yield a correct understanding of the sentence. Reference alore is also
not sufficient. In this case, as in many other compound and/or complex
sentences, it is necessary to take note of word form cues, position cues,
and transformation cues (the first clause is active, the second passive
tin the sample sentence above) as well as semantic cues.

It follows from the above analysis of recoding that any character-
:istic of a sentence which distorts or minimizes the useful sentence-cues

‘Jill cause that sentence to be more difficult to understand. An

12

examination of the sentence learning studies discloses at least five
characteristics of sentences which have been studied and which have
provided the basis for his investigation of individual difference var-
iables in sentence comprehension. These characteristics are listed
below followed by a discussion in each case of how any difficulties in
the comprehension of prose material may be interpreted in view of indi-
vidual differences in the ability to utilize appropriate sentence cues.

1. Degree of grammaticality

2. Sentence depth
Q Semantic - syntactic conflict

4. Structural ambiguity

5. Number of transformations
Degree of Grammaticalitl:

Ungrammatical sentences are more difficult to learn or identify
than grammatical ones (of. Killer, 1952; harks and Miller, 1964). An
analysis of a number of ungrammatical sentences will serve to illustrate
how the structural cues are less clear than in similar sentences ex-
pressed grammatically.

1. we elected president of the senior class Al.

-Unnatural separation of the verb-object and the object-
objective complement - the position cues are misleading.

2. Nobody in his right mind are willing to do such a thing.

~Agreement between subject and verb is missing and causes
confusion in perceiving the subject-verb relationship-the

word form cue is yielding incorrect information.

13

3. To whom you give the map that shows where the money is buried?
-Improper or incomplete interrogative transformation (the do
and verb is missing) makes it difficult to determine which
clause takes the interrogative, "to whom. . ." or "where . . ."
position cue as well as word form cues are misleading.

Sentggce erth:

This phenomenon (of. Yngve, 1960) can best be illustrated with
self-embedded sentences. An analysis of some fairly complex self-
embedded sentences will quickly illustrate why this type of sentence is
difficult to recode.

l. The girl that baked the cake won a blue ribbon.

~The subject (girl) is separated from its verb (won) by an em-
bedded clause which ends in an object (cake) which semantically
could serve as the subject of the verb, won, and which is much
closer to that verb than its real subject, girl. Semantic cues
are of no value as an aid in interpreting position cues; trans-
formation cues, although subtle, are the important cues in
understanding this sentence.

2. The dog that the cat that the bird fought scolded approached
the colt.

-Which subject goes with which verb is clearly an enigma in this

extremely self-embedded sentence. Again, transformation cues

need to be relied upon; semantic cues are of little or no value.
ngntic—Smtggug antlict:

Syntactic and semantic rules can act independently (of. Marks and

litller, 196“), but they also interact and can sometimes dictate

14

different interpretations causing confusion and interpretation errors
(cf. Gough, 1965; Slobin, 1965). Illustrations of this phenomenon can
be easily constructed:

1. The canary swallowed the cat.

-Semantics would indicate that the cat is the deer and the
canary the "done to," but syntactical rules tell us that the
canary is indeed the subject - position cues take precedence
over semantic cues in this sentence.

2. Colorless green ideas sleep furiously.

-here the syntactical rules dictate very clearly the underlying
phrase markers (Adj Adj N V Adv) but the semantic rules make
the sentence difficult to interpret. In recalling a sentence
like this a subject may bring the semantics in line with the
syntax by substituting other syntactically correct words which
may follow semantic rules somewhat better (e.g. Odorless green
frogs sleep soundly) - the important cues here are word form
(colorless) and position (modifier-modified; actor-acted upon).
Structural Ambiguity:

Persons respond more slowly to sentences which are ambiguous even
though they may not be consciously aware of the ambiguity (cf. HacKay,
1966). It is the nature of structural ambiguities that they allow for
more than one interpretation, i.e. it is unclear which element in a
sentence is related to which of the other elements. When this occurs,
Iwacoding is more difficult because the deep structures which reflect

'these relationships cannot be accurately ascertained:

15

l. The king's picture fascinated me.

-Is this a portrait of the king (i.e. kernel: picture is
king “N1 be N1) or is this a picture belonging to the king
(i.e. kernel: picture is king's -N1 be Adj)? - there is no
resolve here; all cues are open to more than one interpreta-
tion.

2. The girl in the car that moved.

-Which is the correct subject-verb relationship: girl-
moved or car-moved? - Again, the structural cues are of no
help: the result: structural ambiguity.

Number of Transformations:

The more transformations which are necessary in explaining the
relationship between the deep and surface structures of a sentence, the
more difficult the surface structure sentence will be to recall (of.
hehler, 1963; Clifton and Odom, 1966). A careful examination of some
of the more complicated surface structures such as the passive or the
passive-negative-query will serve to show the importance of structural
cues (especially transformation cues) in interpreting these more com-
plex sentences.

1. The girl had been distracted by her boyfriend.

-3irl is in the normal subject position and is often in this
position as the doer of the sentence, but here the boyfriend
(at the end where the direct and indirect objects often occur)
is the deer and the girl is being done to. A similar con-
struction: The girl had been distracting. . . would be

interpreted as girl = doer.

16

2. flhy did the voters not come to the polls?
JJith the subject separated from its verb by the negative
and the verb Split (did - come) and an object of the preposi-
tion but no object of the action in the sentence, the reader
has many things to consider before he can know for sure who
is doing what, to whom (if to anyone at all). Again, the
subtle transformation cues are necessary to help make some
sense out of the position cues.

There are aspects of these five characteristics of English sen-
tences which have been the object of a number of sentence learning
studies. Each characteristic may make the task of recoding more dif-
ficult by minimizing or distorting the useful semantic and structural
cues to the sentence meaning. Because each characteristic poses a
slightly different problem for the person who is trying to recode the
sentence, tests constructed to measure a person's ability to recode
sentences of each of the five types should be useful subtests which
together could be thought of as measuring the skill of recoding. If
recoding is an important process involved in sentence comprehension and
recall, and if sentence comprehension and recall are important in prose
learning, then tests designed to measure these individual differences
should facilitate the prediction of reading comprehension scores.

Another psychological factor that needs to be considered in any
study which considers the variables affecting learning from prose
inaterial is the factor of memory. Although memory per se is not a
Imajor variable of interest in this study, it is a variable which nust

lDe considered. Accordingly, two types of reading comprehension tests

17

were developed. One of the tests was constructed so that the Js could
not look back at the story to answer the questions. In this test
memory was considered to be an important factor. in the other test
the is were allowed to look back at the story and little or no memory
was needed to answer the questions. fhis test was designated a non-
memory test. Remory was get considered to be a primary factor in the
test. lf individual differences in recoding correlated reliably with
both of these tests, then it would seem reasonable to conclude that
individual differences in readinj do exist (i.e. they are not just a
function of the type of test).

A further important factor that needs to be accounted for in a
study of prose comprehension is the nature of the material itself. Je-
cause the recodin: skills being measured by the five subtests in this
study are postulated to measure ability to comprehend sentences of
various kinds and various levels of difficulty, an important kind of
manipulation of the prose material used should be a manipulafion of the
level of difficulty of the sentences within the prose passage. Accord-
ingly, two forms of each type of test were constructed. in one form
the prose passages were written in highly complex sentences and in the
other form the passages were written in simple active declarative sen-
tences. It was hypothesized that the complex form of the test would be
more difficult and that recoding would predict the scores of the com-
plew passage better than the simple passage. It was postulated that
(complev sentences require more recoding skill to understand than simple

Ekéntences, and therefore recoding would be a more important variable

111 the comprehension of a passage containing complev sentences.

The first major objective of this study was the desijn and develop-
ment of sentence skill tests which would reliably measure individual,
differences in sentence skills that have been postulated to be important
in recoding. A second major objective was to determine the importance
of these tests in the prediction of reading comprehension. it was
hypothesized that these tests would predict reading comprehension better
if the passage used in the comprehension test contained complex, dif-
ficult-to-recode sentences than if the passage contained mostly simple,
active declarative sentences. it was further hypothesized that a com-
prehension test which did contain complex sentences would be more dif-
ficult than a similar test containing simple sentences. Another objec-
tive of this study was to determine whether or not these sentence skill
tests would add any precision to the prediction of reading comprehension
over and above that already available throuzh the use of IQ and vocabu-
lary as predictor variables. It was also an objective of this study
to determine whether the recoding subtcsts were measuring the same
general factor or several unique factors. Finally, it was an objective
of this study to determine the relationships which exist among the sen-
tence skill tests and the following variables: Ia, vocabulary, reading

rate, d' it span, and readinz comprehension.

CdAFTsP Il - LLVIQJ OF Tdd LIFelATUi;

In the first chapter the general notion of recoding as an indi-
vidual difference variable was discussed. The argument was forwarded
that in order to recode a compler sentence the 3 must be able to
utilize the information contained in various structural and semantic
cues found within the sentence. It was stated that at least five
characteristics of Lnglish sentences have been studied which make
sentences more difficult to learn. fhe studies which have investigated
granmaticalness, self-embeddedness, semantic-syntactic conflict,
ambiguity, and transformational complexity will now be reviewed in
order to validate the use of these characteristics as bases for the
construction of sentence skills test used in this study. However,
before the studies which investiratei the various sentence character-
istics are reviewed, a number of studies will be examined which tend
to confirm the psychological reality of recoding in general--i.e. that
process of reducing surface structure sentences to some simpler, deep
structure form in readiness for information storage.

The Psycholozical iealitv_of Vecoding
gehler and Carey (1947) have conducted a study which strongly sup-

ports the notion that 35 do respond to the deep structure of a sentence.

.In this study college 35 listened to groups of ten syntactical homoren-
(sous sentences mived with white noise. inomalous test sentences fol-
Ilomed each group of ten. lerception scores for the test sentences

ifliowed that changes in both surface structure ani deep structure can

19

20

disrupt perception. These authors were able to successfully develop
within the 33 a set to respond to deep structure, therefore demonstra-
ting the psychological reality of this phenomenon.

The results of a large number of studies have indicated that 35

do in fact utilize transformational rules in producing and recalling

L‘J

-nglish sentences (of. filler, 1962: hehler, 1963; Farx and miller,
196Q; Savin and rerchonock, 1965, etc.). These studies which most
directly deal with the notion that sentences are coded in deep struc-
ture form are the studies discussed by miller (1962) and the well-
known study by Savin and Perchonock (1965).

Miller (1942) discusses several experiments dealing with the recall
of sentences of varying degrees of grammaticalness and differing in trans-
formational type. He cites Eehler's 1963 study which demonstrated the
high frequency of syntactic errors made by 85 when recalling sentences of
different types and sujzests the hypothesis that people remember the
kernel sentence plus some footnote about syntactic structures. Finally,
he constructed an equation for estimating the probability of recall with
various numbers of syntactic errors. Using the equation with Iehler's
data, miller obtained fairly good agreement between the estimated prob-
abilities and the obtained errors. He concluded that the hypothesis of
recoding could not be rejected on the basis of Lehler's preliminary
<iata.

An extremely clever and useful study by Savin and Perch nook (1933)
Jfisnds evidence to the fact that in storing sentences the subject first
Iwneodes these sentences into a kernel or deep structure form. In this

stikdy Savin and Perchonock, using colleqe SS, asked the 33 to memorize

21

a list of sentences varying in transformational complexity. \fter

".;«:rg ‘le sentences the 55 were requested to read a number of un-
related words. The recall task was one of asking the 33 to recall,
not only the sentences they memorized, but also the words that fol-
lowed those sentences. The results of this study showed that the 35
were able to recall more words when the words followed a kernel or
simple surface structure sentence than when they followed more com-
,

elex trnasformational sentences. For exam le an averaqe o; a prox-
. D . _ ‘

inately ‘ ° words were recalled when the words followed basin active

x" .1

declarative sentences and only an average of approxirately 3.9

words
when the words followed a passive-nerative-query. The interpretation
that Savin and Perchonock placel on this study, was that sentences With
more transformation, i.e. fireater transformational complexity,

occupied more memory space than sentences of less transformational
complexity. In fact, in their experinental material, shorter sen-
tences with more transformational conplexity occupied more "space"

in memory than longer sentences that involved less transformational
complexity. In a follow-up study reported by Chomsky (1967b), Savin
considered paired sentences with approximately the same deep structure,
but one of the pair being more complex in surface structure. fihen he
presented these sentences followed by a seluence of unrelated words
and then determined the number of the unrelated words recalled by the
he discovered that the paired sentences were indistinguishable.
Iiowever, when the 35 were presented a sequence of unrelated words pre-
<3edinq the sentences being tested, then the more complex the sentence

“was in surface structure the fewer the words the 3 was able to recall.

22

Savin interpreted this to mean that the sentences were stored in memory
in terms of the deep structure. When the unrelated words preceded the
sentences, these words used up a certain amount of short-term memory,
and the sentence that was more complex in surface structure could not
be analyzed with the amount of memory remaining. However, if the sen-
tence precedes the unrelated words, it is analyzed and stored in deep
structure, and then about the same amount of memory Space is taken up
for either of the sentences in a given pair.

It has long been known that economy is an important determinant
in man's use of language (cf. Zipf, 1935). Certainly it is more
economical to store information in a "kernel" form than it is to store
it in the form of a more complex sentence which is typical of Spoken
and written discourse. Furthermore, it seems logical to suspect that
if stored in some common form, the deep structures representing various
ideas retained from written or spoken discourse can be better associ-
ated, or subsumed in the cognitive structure. If we store information
in deep structure form, this may allow us to make generalizations more
easily because we have, as it were, stripped the surface structure of
the Specific aSpects of the message which were important only at that
time. For example, it may be in a given situation that a person may
want to emphasize the object of some action more than the subject of
that action. A sentence like "John was hit by fete," may therefore
be more appropriate in communicating that message than a sentence
like "Fete hit John." The passive transformation is more complicated
and.would probably require more "memory Space" than the simple active

declarative, "Pete hit John." However, the central idea of the message

9')
,/

can be stored in the simplest form and be used at a later time.

fhe sam. appeal to economy cited above as a reason for storing
information in a "kernel" form can be cited as a reason for commwnica-
ting in a more complicated form (such as is often found in surface
structure sentences involving several transformations). It is more
economical in terms of Speaking time or in terms of amount of written
material to put more than one idea in one sentence. Although this is
useful for the person transmittin: information, it is not necessarily
useful for the person receiving it. The person receivin: a number of
ideas may wish to store those ideas separately, perhaps independently
forming associations with previously stored knowledge. A kernel sen-
tence or a sentence in deep structure form contains only one main idea
and is therefore a useful form for memory storage.

meirical evidence can also be found to support the fact the is
engaging in recall of sentences take into account not only the semantic
elements in a sentence, but also the syntactic elements. harm and
Miller (1964) conducted a study in which 35 were asked to repeat, upon
hearing; grammatical sentences, anomalous sentences and anagrams. an
analysis was made of the semantic and syntactic errors. Semantic
errors occurred most often when semantic rules were violated. Fhe
authors concluded that the 35 were performing both syntactic and
semantic manipulations on the sentences.

Chomsky (1967a) suagests that both the form and meaning of a sen-
tence are determined by syntactic structures that are not represented
directly in the surface structure. This notion suggests that the syn-

tactic structure of a sentence feverns to a large degree the central

21+

meaning which the 3 stores in memory. A recent study supporting the
idea that syntactic form influences meaning was conducted by Johnson
(l967). In this study nonsense syllables were used in the subject and
object positions of simple active and passive sentences. Phe refer-
ential meaning of these sentences was controlled, and the 35 rated the
nonsense syllables on a Semantic Differential scale. The authors con-
cluded that the animate-inanimate component of meaning may be struc-
turally determined, i.e. related to the position that words occupy in
sentences. This study might also be interpreted as supporting
Chomsky's suggestion that the base structure involves both a categor-
ical component and a lexicon. Both of hese elements of the base

structure ar

3 subject to transformational rules.

 

Sentence Charact
Although only a few studies such as those described above can be

legitimately used to support the psychological reality of the recoding

process, there have been a number of studies which have examined the

F

effect of grammatical complexity on learning. Fhese studies are re-
viewed in order to establish the validity of the recoding subtests sug-
gested by the five sentence characteristics discussed previously.
Degree of 3rammaticalitv

fhc earliest attempt to study the effect of the degree of gramma-
ticality upon recall was conducted by Killer and Selfridge (1950).
These authors constructed several strings of words of different lengths
and varying in statistical approximation to English from O-order to 7th-

order and normal textual material. They found that regardless of the

length of the word strings (from 10-50 words) the percentage of words

25
recalled correctly v15 jreater for each successively higher-order ap-
proximation to English up to 5th order apprOUimation (5th, 7th, and
text yielded resentially the same results). As the word strings
approached grammatical correctness, they became easier to learn.

Another early study which investigated the psychological signi-
ficance of grammaticalness was conlucted by miller, flaise, and
Leichten (1951, reported by Liller, 1952). This study demo stratel
that words can be phrCEived nor; accurately when they are heard in the
context of a sentence, than when they are pronounced separately as
indiviiial items. In 194? .iller inollower‘l up this ifivastifation with
a more carefully controlled experiment in which 25 words which had been
previously overlearned by the 35 were arranged in four lists. One of
the lists was arranged as a grammatical sentence, another as a pseudo
sentence, and the last two lists were random strings-~one containing a
52 word vocabulary and the other a 25 word vocabulary. These sentences
were read in the presence of white noise. hhen sentences were read
rapidly, the pseudo sentences were perceived no more accurately than
the random 25 word lists. The grammatical sentences, however, were
most intelligible under conditions of slow and fast reading.

A series of studies by Epstein (1961, 1962, 1967) demonstrated
basically the same findings as those reported by Miller et al. In all
of Epstein's studies it was shown that grammatical material is easier
to learn than ungrammatical. In this series of studies the 35 were
asked to recall the sentences rather than simply to recognize them in
the presence of some masking noise.

Another study which supported the finding that 55 immediately re-

call grammatical sentences best was carried out by Killer and lsard

26

(1963). Three sets of stimulus material were created. The first set
included grammatical sentences like "Gadgets simplify work around the
house." The second set included anomalous sentences like "Gadgets kill
passengers from the eyes." The third set included ungrammatical sen-
tences like "Between gadgets highway passengers the steal." Sentences
were recorded on tape, and the 85 were asked to repeat aloud immediate-
ly afterward what was heard. The experiments were performed such that
the Ss heard the sentences in the first experiment without a masking
noise and in the second in the presence of a masking noise. Both ex-
periments yielded virtually the same results: grammatical sentences
were repeated with least errors and ungrammatical with most errors.

In 196# Marks and Miller conducted a study similar to that of
Miller and Isard‘s 1963 study. In this study the intelligibility of
grammatical, anomalous, and ungrammatical sentences was again com-
pared. However, Marks and Miller added a fourth type of sentence:
anagram strings. These four kinds of sentences or strings of words
were so constructed that semantic and syntactic rules could be
violated independently. The 85' performances in a free-recall learn-
ing situation were compared for various types of strings. Again
grammatical sentences were recalled with least errors and ungram-
matical with most. In this study anomalous sentences yielded slightly
.fewer errors when scored for complete recall, but when scored for
Isecall regardless of position, then anagrams were superior to anoma-
ILous sentences. Error scores were divided into semantic errors and
Styntactic errors. Syntactic errors occurred most frequently in anagram
srbrings (word lists where syntactical rules were violated). Semantic

earrors, however, occurred most frequently in anomalous sentences (word

27

lists where semantic rules were violated). This finding suggests
that syntactic and semantic rules have psychological reality.

Foder and Bever (1965) conducted an interesting study in which
they were trying to show that Speakers of English do respond to
grammatical structure; and in fact anticipate that structure. In
this experiment the Ss heard recorded sentences which were interrupted
occasionally by small clicking noises. The subject had before him a
manuscript of the sentences which he had heard and his task was to
mark the positions where he thought he heard the clicks. The results
of this study showed that the Ss placed the clicks closer to the
phrase boundaries than to the position where they actually occurred.
The 85 seemed to be aware of the major grammatical elements and placed
the clicks in such a way that they did notinterrupt those intact
elements.

Another study which supports the psychological reality of the
phrase structure of grammar was done by Johnson (1965). This experi-
ment was designed to determine the extent to which 35 would use their
knowledge of grammar to break a sentence into functional units as
they attempted to learn it. The task was a PA task, and the Ss
learned eight sentences as responses to the digits one to eight.

Their responses were scored for the conditional probability that the
words in the sentences were wrong given that the immediately preceding
‘word was right. It was argued that if 83 break the sentences into
subunits using their knowledge of grammar, the conditional probability

:of an error should be high between subunits and law within subunits.

28

The results, in fact, supported this argument.

A final study which investigated the effect of the level of gram-
maticality on learning was reported by Coleman (1965a). He took the
position that the level of grammaticality should be classifiable by
college 85 and that they would learn the grammatical sentences most
easily. He conducted three studies using sentences which represented
four levels of grammaticality. In the first experiment the subject
successfully ranked the sentences according to the level of grammatial-
ity. In the second experiment sentences were presented on a memory
drum, and the most grammatical sentences were recalled best. A cloze
technique was used in the third experiment and again the 85 performed
best on the grammatical sentences.

The kinds of measures that were obtained by the experimenters in
the above group of studies could be classified into three categories:

1. A measure of the 5's ability to recognize and/or recall un-

grammatical or pseudo grammatical strings (e.g. Miller, 1962;
Epstein, 1961; Foder and Bever, 1965, etc.)
2. A measure of the 3's sensitivity to grammatical structure
(e.g. Johnson, 1965: Coleman, 1965a)
3. A measure of the 8's ability to judge the level of grammatical-
ity (Coleman, 1965a)
Dgpth 91 §gntence

Yngve (1960) presented a theoretical model of language structure
IJhich was based on the notion of sentence depth. He hypothesized that
‘the greater the depth of a sentence the greater the load it placed on

rmemory.

29

Martin and Roberts (1966) attempted to show that sentence depth
as measured by Yngve's theory accounts for the structural complexity
of sentences which cause problems in intelligibility in recall. The
measure of sentence depth which they used entails a phrase-structural
analysis of the sentence and an accounting of the grammatical commit-
ments incurred by each word of the sentence. A word is said to be
structurally embedded in a sentence to the extent that it determines
the structure of those parts of the sentence that follows. Sentence
complexity (measured by this depth concept) and sentence kind (desig-
nated by the number of transformations) were manipulated independently
with sentence length held constant. Both sentence complexity and
sentence kind were found to affect recall, and support was given for
Yngve's theory.

In a study by Blumenthal (1966) the 83 were asked to rewrite
self-embedded sentences in order to render them more acceptable or
understandable. Eight college 35 responded to ten sentences, each
with two multiple-embedded clauses. Thus there were 160 observations
and in only 41 of these were the grammatical relations correctly
assigned according to the embedded structure.

A final experiment in this group was conducted by Stolz (1967).
This experiment was designed to test the notion that English speaker-
hearers can successfully decode sentences which are grammatically
novel to them. Stolz chose self-embedded relative clause sentences
as the kind of novel sentences to present the 33. The 85 were pre-

sented the sentences in spoken and written form and were given

 

. 2'. p!.iflpillli “a

a. i.

 

30

unlimited time to decode them. The results of this study indicated
that only one half of the college 85 used in this study could success-
fully decode these sentences without the aid of semantic constraints
or feedback.

In the above studies which manipulated sentence depth, the
researchers were able to get measures of the 85' ability to untangle
self-embedded structures as well as measures of recall of self-
embedded sentences.

N ans 3

One of the earliest studies which dealt with the psychological
reality of transformation rules was carried out by Miller (1962).

In this study a series of sentencermatching tests were used. Each
test was constructed of two lists of sentences such that one list
included transformations of one type (for example, passives) and

the other list included transformations from the same deep structure,
but of a different type (for example, passive negatives). The task
of the subject in each case was to match the sentences from one list
with a sentence from the other list so that the matched sentences were
surface structures representing the same deep structure. The total
time the subject took to match the sentences was recorded. The authors
found that whenever transformational theory required more than one
transformation from the sentences in the first list to the sentences
in the second list, the total time to match these sentences was
greater. In general, the data from these studies supported the trans-

formational theory.

. ,,
‘ L
A 1.115;;- 1'5. w «

 

A study by Mehler (1963) supported the evidence found by Miller.
In this study the 35 were asked to learn sentences of eight basic types
varying in transformational complexity. These types are as follows:
kernels, negatives, passives, queries, passive-negatives, passive
queries, negative queries, and passive-negative queries. The more
complex sentences (e.g. passive-negative queries) were more difficult
to learn. In fact, the data suggests that the number of errors a
subject made in recalling a sentence seemed to be, at least in part,

a function of the number of transformations in the sentence. The
relations among the eight sentence types, as suggested by these data,
are shown in Figure 1.

Clifton and Odom (1966) reported a number of studies conducted
by Clifton and others in which the difficulty of recall of the various
sentence types was further explored. Their data suggests that the
"psychological distance" between the various sentence types is not as
uniform as suggested by the graph of Mahler's data in Figure 1. Accord-
ing to their study, a graphic representation of the psychological dis-
tance between sentence types is more accurately depicted as shown in
Figure 2. These relations not only held up when recall was measured,
but also when a measure of confusion between sentence types was
obtained. Although the fine details concerning the relative dif-
ficulty of learning the various sentence types are not agreed upon, as
can be seen in the differences between Figures 1 and 2, there is a

large amount of agreement with respect to the fact that, generally

3?

 

 

 

 

 

N} . I-‘.
1' 1 \21‘

 

 

 

Figure 2: Graph indicating relations among eight types of sentences
formed by the negative (N), passive (P), and interrOgative
(Q) transformations (suggested by Nehler's data - from

Killer. 196214

 

 

 

\
1t tyoes of sentences

Jioxs among e13
passive (1“). ".nd interrojativc ()

Gated by Clifton and Odom's lQ‘fi

we
‘1

3: Graph indicatinj r
formed by negative
transformations ( u;
46"”)

\gh‘I‘LI.

Speaking, the greater the number of transformations from deep structure
to surface structure, the more difficult the sentence will be to learn.
One exception to this generalization occurs frequently in the
literature. The negative is often more difficult than the more com-

plex transformations, such as the passive. A good example of this

can be found in Slobin's study (1965). In this study a number of

sentences of the different transformational types was read to the

35. In each case the S was shown a picture related to the sentence,

and was asked to respond with "yes" or "no" depending upon whether

or not the picture correctly demonstrated the statement contained

in the sentence. Again, the data supported the notion that the

sentences involving a greater number of transformations caused the

most errors. This study, as well as another carried out by dough

(1965), did not measure simple recall, but instead, understanding

of the various sentence types. Slobin's study also renders support

flirthe idea that acquisition of transformational rules occurs at an

early age, for the results were generalizeable from age five to adult.
One further study which measures more than simple recall was

conducted by Savin and Perchonock (1965). They asked 85 to learn

sentences and a number of unrelated words which followed them, and

then they obtained a measure of the number of words that the 85

were able to recall. These authors demonstrated that the sentences

having the greater number of transformations took longer to learn,

or took up more immediate memory space, and therefore left less space

for recall of the unrelated words.

 

.19 blulfl‘JldJl’ .
...: o A. . n
_.

I Utrnl'rl. II}.

 

The most important generalizations that can be made from the
above studies are:
l. The influence of the number of transformations in a
sentence upon learning caibe measured, and
2. A number of different measures of this phenomenon all
seem to yield reasonably the same results.
The kinds of measures that were obtained in the above studies
can be classified into at least three categories:
1. Ability to recognize sentences of different sur-
face structures having the same deep structure (e.g. Tiller,
1962), and Clifton and Odom, 1966).
2. Ability to recall a sentence verbatim (e.g. Nehler, 1963
and Savin and Perchonock, 1965).
3. Ability to understand or comprehend the meaning of
a sentence (e.g. Slobin, 1965; and dough, 1966).
Semantic - Syntactic Conflict
The study by Mahler (1963) discussed above was one of the early
studies in which an assessment was made of the relative effect of
syntax and semantics upon recall errors. The hypothesis was advanced
but not directly tested in this study, that 85 analyzed sentences
into a semantic component plus syntactic corrections, and that this
separation of semantic content from syntactic form is one reason
Why the general meaning of the message is usually so much easier
to recall than its exact wording.

Harks and Miller (1964) (also discussed previously) arranged

35

their materials so that semantic and syntactic rules could be indepen-
dently violated. Their results showed that when semantic rules were
violated, more semantic errors occurred: and when syntactic rules were
violated, more syntactic errors were made.

Cough (1965),in his study which was described previously, found
that the true-false variable was an important variable, at least when
measuring verification time, sentences were more difficult when the
semantics and the syntax were in'bonflict."

Although the exact nature of the interaction between semantics
and syntax is at best unclear at this point, the above studies do
demonstrate that 33 respond to both semantic and syntactic cues.
Furthermore, they demonstrate that these two types of ones can be
independently manipulated, and errors affected by this manipulation
can be measured.

Ambiggity

This final characteristic of English sentences which we have
proposed as affecting the accuracy of recoding has been studied
very little. Only one study has been located which attempted to
measure a person's reactions to ambiguity. In this study, MacKay
(1966) presented the $5 with a long list of sentences containing
normal sentences and sentences which included lexical, surface and
underlying ambiguities. These sentences were incomplete, and the
35' task was merely to complete them by adding a number of words or

phrases. The time to complete each sentence was recorded, and it

36

was found that although the 85 were unaware of the ambiguities, they
did take longer to complete the ambiguous sentences than they did the
normal ones.

Although this is the only study located which measured ambiguity,
it seems reasonable that some measure of the 35' ability to recognize
and resolve amibguities could be obtained.

Re 0 n A I f en 9 Va 'ab

Support for the proposition that recoding is an individual dif-
ference variable arises from both a theoretical and empirical analysis
of prose learning. Although Chomsky (1965) maintains that theoretically
all mature native speakers have the same competence in their native
language, it is not inconsistent with Chomsky's notion to postulate
that all persons do not utilize that competence with the same degree of
skill. Chomsky (1967b) discusses the fact that the rules which deter-
mine surface and deep structure, and their relationship, are abstract
and remote from consciousness. It is not unlikely that these complex
rules are used more freely, more swiftly and more accurately by some
language users than by others. In a discussion about a theory of
language-users, Chomsky and Miller (1963) argue that differences in
memory limitations, in coding skills, etc. do probably cause differences
in performance among individuals and differences between an actual user
of the language and what has been referred to by grammarians as the
"ideal-speaker-hearer."

It should be noted that there are individual differences in a

person's abilities to apply transformational rules to the morpho-

?7

phonemic level of language. It seems reasonable that there may
also be individual differences in persons' abilities to apply trans-
formational rules to the phrase structure level of the language.
Furthermore, individual differences exist in the use of the lexical
component of the base of the language, and it is reasonable to
assume that they also exist in the use of the categorical component.
In the various studies cited throughout this paper, there were
obviously many errors made by the 85 as they were called upon to
produce and/or perceive sentences of varying grammatical complex-
ities. However, these errors were treated as error variance, and
no analyses of individual differences were made. The strongest
suggestion of individual differences in this ability to recode comes
from Stolz's (1967) study where he found that approximately 50% of
college 35 couldn't recode embedded sentences without the aid of
semantic constraints or feedback. ESpecially when one considers
the homogeneity of this population of college 85 (with regards to
language ability) and the ceiling effect which may have occurred due
to the use of 35 who are undoubtedly more sophisticated in language
usage than the general population, it becomes evident that individ-

ual differences must occur in this skill.

 

B tween Recodin And headinr C morehension

 

Having supplied some evidence for the fact that recoding skill
may be an individual difference variable, it now becomes important
to establish a rationale for the relationship of this variable to

reading comprehension. If this relationship can be established and

if tests of recoding are highly correlated with ability to learn
from connective discourse and subsequently correlated with success
in school, then it is reasonable to begin a research program which
will seek to discover the effect of the recoding skill on learning

and to seek to discover the predictive value of that skill.

Although it is not difficult to see how reading comprehension
tests are measures of a person's ability to learn from connective dis-
course, it may not be quite so easy to see how these tests are re-
lated to recoding. A careful survey of reading comprehension tests
show that they obtain no direct measure of recoding; however, it seems
Ilogical that if recoding is a process which is important in under-
standing the ideas contained in a single sentence, then in order
:for a S to correctly perceive the relationships which exist
eunong the major ideas in a prose passage, each of the ideas must be
cuerrectly recoded before they can be stored and/or compared to each
(ather or to other ideas already in the 3's repertoire of knowledge.
:Dt must be noted that the argument is not being made for equating re-
cxmding and prose comprehension, but rather for the fact that recod-
iqu is a process the S engages in, in his attempt to prepare
liJuguistic material for storage. Certainly if this recoding is
irucorrectly performed, then the comprehension of a sentence and
IXEFhaps subsequently of a paragraph will be in error. It cannot
‘be'-argued, however, that if errors in comprehension are made, they
aITB necessarily due to the recoding error. Errors in comprehension

mag; also arise from errors in discrimination, semantic coding and

39

other individual difference variables such as attention and moti-
vational factors. (Perhaps some of these are individual differences
being measured by I. Q., vocabulary, and reading rate--the tradition-
al predictors of reading comprehension.)

Two studies have been located which tend to support the notion
that recoding as defined in this thesis is a skill which is important
to reading comprehension. The first is a study by Bormuth (1964)
which tested Yngve's method of measuring grammatical complevity.

The purpose of the study was to see if the method could be used to
predict the difficulty children have in comprehending written verbal
materials. Nine prose passages were used which covered three sub-
ject matter areas (literature, science, social studies) at three
.Dale-Chall readability levels (4.5, 5.5, 6.5). A fifty item cloze
<3omprehension test was administered for each passage to 150 4th,
_5th, and 6th grade children. The mean word depth was obtained for
eeach passage by using Yngve's procedure. A high correlation was
(ibtained between the mean word depth of the passages and both the
13ale-Chall readability and the cloze test estimates of passage dif-
:Piculty. The authors concluded that mean word depth may be one of
tflie most powerful predictors of the comprehension difficulties of
{Nissages and that this measure accounted for some relevant factors
tflmat the Dale-Chall readability measure did not account for. It
Seems reasonable to hypothesize that if mean word depth accurately
Predicts the difficulty level of a passage, then a measure of an

irkiividual's ability to recode sentences with large mean word depth

40

indices should assist in predicting Ss' comprehension of similar
passages. This is the skill being measured in the present study by
subtask 2, “depth of sentence.“

A second study indicating a relationship between recoding and
reading comprehension is a doctoral dissertation by Rinne III (1967).
The study was designed to test the hypothesis that training remedial
reading 83 in pattern awareness (recognizing sentence patterns in
written communications) would improve their reading comprehension as
measured by a sentence comprehension test. Although this hypothesis
was not confirmed, a significant positive correlation was found
between pattern awareness (measured by a test designed for that pur-
‘pose) and literal comprehension in reading (r = .41, n = 364, P <
.001). Awareness of sentence patterns is akin to awareness of kernel
rflnrase markers and thus some support for the relationship between
rwacoding (reducing surface sentences to kernel phrase markers) and
stading comprehension is obtained from this study.

Thredictors of Reading Comprehension

As the above reported studies indicate, there seems to be some

 

rmilationship between reading comprehension and the ability to handle
tflie complexities of English sentences. Perhaps such sentence skill
113 only another way of measuring the same thing the traditional
Prwsdictors of reading comprehension have been measuring. In order
tI) answer this question these "traditional" predictors must be Speci-
fixed so that measures on them can be obtained and compared to the
Serrtence skill measures. Althohgh virtually all textbooks which deal

‘Witfli the problems of reading cite IQ, vocabulary, and reading rate as

bl

the major factors in reading comprehension, there is not an overabun-
dunce of studies which have carefully examined the relationship
between these variables and reading.

Strang (1940), in her book dealing with the problems of reading in
high school and college, reports several studies in which correlations
between various standardized tests and reading comprehension were
obtained. She suggests that one of the most complete studies was con-
ducted by G. H. Hillgard in the early 1920's. Hillgard consistently
obtained the highest correlations with IQ and vocabulary. The next
highest correlations were between reading rate and reading comprehen-
sion but they were always considerably lower than the correlations
<dbtained between IS and comprehension and vocabulary and comprehension.

The studies which have been conducted, since Hillgard's work, all
seemIto replicate his findings fairly consistently. Strang (1943) in
:1n.article appearing a short time after her book, reviewed a number
cxf studies which investigated the relationship between IQ and reading
(namprehension. These studies reported correlations between IQ and
Iweading comprehension which ranged from the low .50's to the high
-70's.

Wheeler and wheeler (1949) confirmed this relationship between
112 and reading ability for university freshmen. They obtained a
Cerrelation of .70 between the ACE psychological test and reading
Seores.

Robeck (1964) conducted a study in which he attempted to examine
tine strengths and weaknesses of reading clinic 55 on the WISC. Eighty

elfflnentary and junior high reading clinic 85 were given the WISC, and

42

their subtest scores analyzed. As a group the is were strong on com-
prehension, similarities, vocabulary, picture completion, and block
design. This finding is especially surprising in view of the high cor-
relation between vocabulary and reading comprehension. These 33 were
weak in Information, Arithmetic, Digit Span, and Coding. This finding
was interpreted to mean that the reading clinic Ss lacked ability to
recall specific verbal material. The results of this study seem to
indicate that there is something besides vocabulary and ability to
comprehend that is accounting for weaknesses in reading because the
clinic 35 were strong in comprehension and vocabulary. The ability
to recall Specific verbal material is suggested as a possibility. But
‘why can't these as recall Specific material? Note that they were also
1weak in coding. Perhaps these is are not storing correct information.
Perhaps they have recoded the material incorrectly and have therefore
stored faulty information and in turn recalled faulty information.
Although the studies relating I; and vocabulary with comprehension
all yield approximately the same results, the results of the studies
correlating reading rate and reading comprehension are not consistent.
{Some investigators maintain that low correlations are found consistent-
?ly'between reading comprehension and rate (e.g., Enrich, 1930), while
(Ithers argue that valid measures of reading rate correlate strongly
Vtith comprehension (of. Tinker, 1932). Carlson (19A9) suggests that
'the strength of the relationship between reading rate and comprehension
<iepends on the f; level of the 85, the purpose of reading, and the
(tifficulty of the material. de does conclude, however, in all kinds

'Of situations he obtained low correlations between rate and comprehension.

”3

These low correlations were not useful in the prediction of reading
ability.

Almost all of the studies which have examined the relationship
among various "predictor" variables and reading comprehension were
simply correlation studies and not predictive studies. One of the few
studies which did utilize a multiple regression design was conducted
by Artley (lQNB). Of the variables investigated, he found vocabulary
to be the most important predictor of reading comprehension. He ex-
amined the relationships between general and specific reading abilities
within a specific subject matter area (social studies). His 35 were
200 eleventh graders from an eastern high school. The beta weights
.for the various predictor variables are given below and illustrate the
ixnportance of vocabulary in his study: social studies vocabulary
(.432), general reading vocabulary (.396), ability to interpret (.277).
eibility to obtain facts ( 277). ability to organize (.274), ability to
rxzrceive logical relations (.228), ability to apply generalizations

(.156), and ability to evaluate arguments (.127).

3

   

e P esr;t

 

S f Li e Revi w and Pur ose of

     

The above reported studies indicate that althcugh the relation-
stnips of IQ and vocabulary with reading comprehension are fairly strong,
tinere seems to be a numtar of other factors accounting for some of the
Vaffiance in reading comprehension. It has been suggested in this
tfuasis that perhaps one important skill which way be related to reading
cC>I7Iprehension is a recoding or sentence comprehension skill. A review
0f.‘the literature in sentence learning disclosed the fact that sentence

cuflutilization can be measured and some of the studies suggested that

an

this may be an important variable in reading comprehension.

The purpose of this study is to attempt to construct reliable
tests based upon the 3's ‘esponse to the sentence characteristics
which have already been studied (grammaticality, sentence depth,
semantic-syntactic conflict, ambiguity, and transformational complexity)
and discover the relationships which exist among these tests and reading
comprehension. Furthermore, an attempt will be made to discover the
strength of these tests as predictors of reading comprehension com-

pared to the traditional predictors (1Q and vocabulary).

CdAPTEd III - hETHOUS

,"‘

subjects

The 85 used in this study were high school students who were in-
volved in various summer programs in hichigan during the summer of
1968. The number of students who volunteered from each of the follow-

ing programs and were randomly assigned to groups A and B are indicated

in Table 1.

TABLE 1

A List of the Sample Populations Indicating the Number of

Ss Assigned to Each Group in the Various Samples

 

 

 

Sample Group A Group B
1. Youth for Christ International 32 Ss 28 Ss
Conference, Winnona Lake,
Indiana
.2. Seventh-Day Adventist Summer 4 35 i 5 55
Youth, Grand Ledge, hichigan i
:3. Educational Park Summer Program, 32 Ss 3 36 35
Grand Rapids, Michigan 3
3
C
1*. Walker's Youth Camp, 6 Ss : 11 SS
Eckerman, Michigan 1
TOTAL 7:. as E 80 Ss

Ir} all, 160 33 were tested, but 6 35 failed to complete the entire test

battery and so were eliminated from the study.

 

 

 

 

There were 39 females

afki 35 males in group A and #9 females and 31 males in group B.

afinarage age of the Ss in group A was 17 years (S.D. = h.l) and the

aVerag-e L; was 115.1 (5.1). = 17.7)-
45

For group B the average age was

46

16.6 (S.D. = 1.0) and the average IQ, 114 (S.D. = 12.1). Within each
sample the 35 were assigned randomly to one of two forms of the Read-
ing Comprehension tests. All 55 were administered identical tests
measuring the various independent variables.

Instrumentation and Procedures

A battery of tests was used in this study and a complete copy of
all the tests except for the 33 test (the Otis Quick Scoring - form EM,

Gamma series) appears in the appendix. Each of the tests is listed

 

"A- '.

below along with a brief description and some sample test items. Each
test was pilot tested several times and each time revisions were made
on the basis of information from a complete item analysis. A summary
of'the item analysis data based on the final data collection appears in
appendix 1. Mean item difficulty and mean item discrimination indices
.are presented for the sentence skill tests. the reading comprehension
tweet, and the vocabulary test for groups A and B.

l. IQ test - The IQ test used was the Otis Quick Scoring, form
End, Gamma series (1956 revision). The only reported reliability co-
exfficient for this test is an alternate-form correlation of .97. This r
i4; probably spuriously high because the 1918 sample of 4th through 8th
graders is a much more heterogeneous sample than the restricted age
range of the 83 used in this study (13 to 19 years of age).

2. Vocabulary test - The vocabulary test used was constructed by
the author of this study and consisted of 32 items. For each item the
S ‘was to select from among a number of alternatives the one which was
clxasest in meaning to the first word of that item. The words were

Selected at random (not randomly) from texts, novels, etc. which are

14,7

widely read by high school students.

Sample items:

1. boat a. animal b. ordinary c. ship
d. sea 9. fruit
2. elongate a. stretch b. shrink c. magnify

d. control e. deserve

3. labyrinth a. laboratory b. bassinet
c. maze d. cork e. experiment

This test appears in appendix 2.

3. Digit Span Test - This test consisted of 12 different digit
span sequences. Two different sequences from each of the following
digit span lengths were constructed: 5, 6, 7, 8, 9, 10. After the 3
read all the digits in a specific sequence, the 85 recorded this
sequence on an answer sheet provided for them. This answer sheet con-
tained, for each sentence read, the same number of blanks as the actual
number of digits in that sequence. This procedure allowed the 85 to
leave a blank if they could not recall a digit and in so doing keep the
rest of the digits in the proper place in their sequence. This test
appears in appendix 3.

4. Grammaticality - This test consisted of 28 "sentences", each
sentence having little or no meaning because most of the words were
nonsense words. Some of the "sentences" were grammatically correct
having the sound and rhythm of a complete English sentence. The other
"sentences" were grammatically inconect and did not have the sound or
rhythm of complete English sentences. The 85' task was to read each
"sentence" and decide whether or not it sounded like a complete English

sentence.

 

rem-

Sample items:

1. When the klappy, the snordle blumped along.
2. Zurpy gazzle tamps were sloopy.
3. Two zweep which were galumphing away.

This test appears in appendix 4.

5. Self-embedded sentences - This test consisted of seven self-
embedded sentences. Following each sentence was a number of phrases.
Some of the phrases were correct according to the sentence preceding
them: others were not. The 85' task was to read each phrase and decide
whether it was true or false according to the self-embedded sentence
preceding it.

Sample items:

The birds that the fence that the men put up held sang a song.

1. the birds sang a song

2. the men held the fence

3. the men sang a song.
This test appears in appendix 5.

6. Semantic-syntactic Conflict - This test consisted of a number
of sentences which were syntactically correct but which had unusual
meanings. Below these sentences were a number of multiple choice items
consisting of phrases which were either correct or incorrect according
to the sentences above them. The Ss' task was to choose from among the
alternative choices the correct or incorrect phrase as specified by the
direction.

Sample items:

Spilled chaos saves baseball love

 

49

The bottled young rhythm drank grape wonders.

1. Choose the correct phrase
1. Chaos saves baseball
2. Chaos was saved
3. Love was saved

2. Choose the igggrgegt phrase
l. The rhythm drank wonders
2. The rhythm drank grape
3. The rhythm was young

This test appears in appendix 6.

7. Ambiguity - This test consists of 28 items some of which
were structurally ambiguous and the rest of which were unambiguous.
The 35' task was to read each item and decide whether or not it had
more than one possible meaning.

Sample items:

1. I am going away today

2. The book in the car that is red

3. he took the stone from the cloth and washed it.
This test appears in appendix 7.

8. Transformational Complexity - This test consisted of 28 sen-
tences which the 85 were asked to memorize so that upon seeing the
sentence at a later time they could recognize it when it was surrounded
by a number of other sentences which were transformations of it (i.e.
the original sentence and the alternatives all had the same deep struc-
ture). The Ss memorized, and were tested on the sentences, in groups

of four and five in the following manner: Upon turning a certain page

 

C
)0

he 35 were given 20 seconds to study the four or five sentences on
that page. Next they were signaled to turn the page again where they
were faced with four or five multiple choice items. For each item
they were to select from among the possible alternatives the exact
sentence which they had studied on the previous page. This test
appears in appendix 8.

9. Reading date - This variable was simply the number of words
read per minute by each of the 55. The scores were obtained by having
all Ss begin reading the first reading passage at the same time (300
words of this passage were exactly the same for both forms). After
30 seconds the 83 were told to stop reading and to indicate on an
answer sheet the number of the line and the word they were reading
when they were stopped. After the testing session the experimenter
computed the number of words read.

10. Prose Comprehension - Two types of reading tests were used.
The first type of test required the 85 to remember the important facts
and relationships in the story. In answering the questions about the
story the 83 were not allowed to look back for the answers. For this
test an adaptation of a story used by Martin and Alonso (1967) was
utilized. The second type of reading test did not require the 35 to
remember all the facts and inferences in the story for they were
allowed to look back at the story when answering the questions. This

test was adapted from two stories found in the Iowa test of basic

skills, reading comprehension section (Lindquist and Hieronymus, 1955).

The first type of reading test contained a story about two coun-

tries who were at war with each other. Although the subject matter

 

51

was new to all 85, the vocabulary used throughout the story was rela-
tively simple and few uncommon words were used. The Dale Chall read-
ability formula indicated a 7th grade reading level. Following the
story, 33 multiple choice items tested the 53' recognition of major
facts, inferences, and relationships that existed within the story. F
The 35 were instructed to read the story and then, without looking
back at the text, to answer the multiple choice questions. Becausw

the 85 were not allowed to look took at the story for the answers,

 

this test is referred to 's a memory test and will be designated by g
the abbreviation An (leading - aemory). This test is presented in
appendix 9.
The sceeni tvn a of readinz test utilized two stories from the

Iowa Test of basic skills desiwned for eighth grade level. The first
story described the history and characteristic of the qinqko tree.
'Ten multiple c:7oiC7 ouestions followed this story. The second story

escribed the process involved int the treatnent of bronze disease.
’fhis story was 1150 followe lby ten multiple choice questions. Phe

(

(questions which followed these stories again measured the 55' under-

standini of the facts and inferences found in the stories. For this
‘test the 53 were instructed to read the stories and then to answer the
:7uestions, but in this case as were told that they could look back at
‘the stories for the answers. This was called a non-memory test and
tIill be des i*nated by the abbreviation Aha (headinq-don-hemory). This
'test appears in appendix 10.

Two forms of each of the two types of readinq tests were con-

svtruct3:i. Form A was constructed by using as many complex SOYWances

 

52

in the body of the text as was practical. hany subordinatinq clauses

and compound sentences were used. Furthermore, passives, negatives, and
paSSiVe‘nevative constructions were used throughout the story. Form 5
included exactly the same story as Form A, but sentences were of the

simple active declarative type. Thus the difference between Form A and
Form B was in the type of sentence structure used. The number of words
used was almost identical for the two forms and the vocabulary was un-
chanmed from.Form A to Form 5. The number of words and number of sen-
tence types for the two forms is presented in Table 2. Both forms were

followed by exactly the same multiple choice test questions. Both forms

of each list are presented in appendices 9 and 10.

PH BLs‘J 2

w-‘

{umber of ”ords, lumber of Sentences, and Lumber of different
Types of dentences and Clauses in the Two Versions of the
db and gun neadin: Passages

 

 

 

,—‘ ' RNI-l
7 :E4 Passage On Passage Two
A .3 I A 13 A U
r% of’words t 96* 913* I 445 460 635 635
5% of sentences 3 26 73 I 16 £12 20 53
r; (If Simple senten- E 37 96 I 23 4Q 17 54
cos and clauses
%7<1f negatives 7 3 4 2 1 1
j of passives 3?. l 6 l 28 6
7,? of neg. passives 2 O 2 0 1L O
of adjectivaliza- 0/3. 0 19 O 8 l
tions

 

 

 

*The first lfl9 words were exactly the same for both versions of AL
and therefore the first part of the Rb passage was eliminated
from this analysis.

 

53

Pilot Studies

In order to establish the reliability of the various tests used in
this study, several pilot studies were conducted. Various junior hiah
school, high school, and summer "teen" camps throuahout mid-iichigan
were used to obtain 33 for the pilot study. When the rec ding subtests
were used with junior high students, a large number of students scored
below chance level. Because of the apparent difficulty of these tests,
it was decided to use high school students throughout the study. Item
analyses were conducted for each test and several revisions were made
in order to increase the reliability of the tests and to insure an
apprOpriate level of test difficulty.
Final Data Collection

All of the tests were bound in a spiral booklet and two versions of
the booklet were constructed. The tests were the same for both versions
except for the reading comprehension tests (form A being used in one
version of the booklet and form 8 in the other). At each testing session
the test booklets were distributed randomly forming two groups (desig-
nated as groups a and B). The total testing time for the complete bat-
tery was approximately two hours. The 35 were given a small break mid-
tway through the test session. The directions for each of the various
“tests are recorded in the appendix along with the tests just as they
aappeared in the test booklets.
(Dbjectives and Hypotheses

The major objectives of this study were to:

1. design sentence skill tests which would reliably measure in-

ciividual differences in sentence recoding and recall.

 

54

2. determine the extent to which the sentence skill tests comprise
a general skill factor or specific skill factors.

3. determine the relationships which exist among the various
sentence skill tests and the following variables: 14, vocabulary,
reading rate, digit span, and reading comprehension.

4. determine the proportion of variance in reading comprehension
scores uniquely accounted for by each of the following variables: IQ
vocabulary, reading rate, and the sentence skill measures: grammatical-
ity, self-embeddedness, semantic-syntactic conflict, ambiguity, and
transformational complexity.

5. increase the predictability of reading comprehension by the
addition of the sentence skill (recoding) tests along with the tra-
ditional predictors (11 and vocabulary).

6 determine if sentence skill measures were better predictors
of reading comprehension when the reading passage contained complex
sentences than when the passage contained simple sentences.

7. determine if comprehension of a nassage including complex
sentences is more difficult than comprehension of a passage containing
simple sentences.

The following hypotheses were tested in an effort to meet objectives
5, h, and 7.

1. When recoding is included in a multiple regression equation
along with IQ and vocabulary, a greater proportion of variance in read-
ing comprehension (measured by both memory and non-memory comprehension

'tests) will be accounted for than if recoding is deleted from the re-

gression equation.

 

55

symbolically:

RE? - lgf where 32 = variance accounted for by prediction
equation. F} = Full model - i.e., lg,
70, and Sm used as predictors of Tead-
in: Comprehension. {L = Yestricted
model - i.e., only la and V0 used as
predictors of Aeading Comprehension.

2. ‘Ihen recoding is deleted from a regression equation which

also included IQ and Vocabulary as predictor variables, the drop in n2
will be greater for group 3 (those receiving the reading test written

in simple sentences).

symbolically:
. ,_ ‘ 9

(i2 - m2>~) (<%.n - 2..) where A“ = variance accounted
F r19. “‘4 A 1‘ 1“ D C 31."

for by prediction
equation,
Ffll = Full model for
group A,
P73 = Full model for
group 8,
35A = aestricted model
for group A, and
RZB = hes ricted model
for group B.
7. Group A (reading the passages which include complex sentences)
xaill obtain lower mean reading comprehension scores than group 8 (read-

i;nz the passaqes which contain mostly simple sentences). 3ymbolically,

56

/ EA<X3 where 2A = the mean reading comprehension score for group A
and is = the mean readingr comprehension scores for group B.
In order to accomplish the objectives outlined above and to test
the three hypotheses, a number of statistical analyses were performed.
Means and standard deviations were computed for all variables and the
intercorrelation matrix of all variables was obtained. In an attempt
to better understand the nature of the skills being measured by the
various tests, factor analyses were performed on the major predictor
measures and the two reading comprehension tests. Factor analyses
were also performed on the five sentence skill tests in order to deter-
mine the factor structure of these tests.

In order to determine which of the several independent variables
served as the best predictors of reading comprehension, all the inde-
pendent variables were entered into least squares delete and least

squares add multiple regression programs. The general procedure used
in the least squares delete program is to begin with a multiple regres-
sion equation which utilizes all the independent variables and then
deletes the variables one at a time. The order of deletion is made on
the basis of the amount of variance accounted for by the specific indi-
vidual variables. Those variables that account for the least amount
of variance are deleted first and those variables accounting for the
most variance remain in the equation. The least squares add program is
essentially the reverse process of the least squares delete. Beginning
with the single most important variable (i.e. the independent variable
accounting for the most variance) the program continues to add variables

in the order of the magnitude of the variance accounted for. Each

57

variable added must contribute a significant amount of unique variance,
and for this study the significance level was set at the .05 level.

The program continues until all variables contributing unique variance
(at the .05 level) have been added to the equation. The program then
terminates and those variables not added to the equation are assumed to I
contribute no unique variance. Although the results of these programs

were difficult to interpret, the results obtained.were useful in identi-

 

fying the major variables accounting for the variance.

P—

Once the major predictor variables were identified by using the
least squares add and delete programs, a more precise method of deter-
mining the amount of variance accounted for by a given independent
variable or group of independent variables was used. This is a method
suggested by Bottenberg and Ward (1963) and is referred to as the
least squares ordered delete program. The research strategy utilized
in this procedure involves determining the multiple regression equation
which contains the major predictor variables for a given criterion
measure. This equation is referred to as the full model (FM). Next
a number of restricted models (RZs) were constructed. Each restricted
equation was constructed by deleting a given variable or group of
variables. This was accomplished by setting the beta weights for the
variable(s) of interest at zero. Various hypotheses were then tested
by comparing the R2 (multiple R squared) between the full model and
any given restricted model. Two basic full models were constructed.
The first contained the independent variables of IQ, vocabulary, and a
total recoding score. The second contained the independent variables

of'IQ, vocabulary, grammaticality, self-embeddedness, semantic-syntactic

58

conflict, ambiguity and transformational complexity. Each of these
full models were used for groups A and B, when reading comprehension
involved a memory factor and when reading comprehension did not involve
memory. Several restricted models were constructed and were compared
to these full models.

In order to aid the reader, Table 3 was constructed. This Table
contains a list of all the abbreviations used in the results and dis-

cussion chapters which follow.

F-I v ‘Lam ‘7 ‘ einq-

59

TA BLE 3

Abbreviations Used in this Thesis

 

 

I

IQ
V0
RR
DS
GR
SE
83
AM
TR
RE

V ab

intelligence (Otis Quick Scoring Test)
vocabulary

reading rate (words per minute

digit span

grammaticality

self-embedded sentences

semantic-syntactic conflict

ambiguity

transformational complexity

recoding (sum of GR, SE, 83, AM, and TR scores)

Depgndegt Variables

RM = reading comprehension--memory involved

RNM = reading comprehension--non-memory (meaning minimized)
Form A = passages of RM and/or RNM written in complex sentences
Form B = passages of RM and/or RNM written in simple sentences

Sgbjegts

Group A

33 administered form A of the criterion tests

Group B = 83 administered form B of the criterion tests

Stgtistiga; Terms

R2

FM

HZ

Multiple correlation squared--interpreted as amount of variance
accounted for

Full model regression equation--all independent variables of
interest included

Restricted model--certain variables deleted from FM by setting
their beta weights at zero

 

 

FEW “- tr. .4

CHAPHfith -EUSULTS

The basic design of this study was a multiple regression design,
and the major hypotheses relate to the amount of variance accounted
for in reading comprehension by various predictor variables. Because
two forms of the reading comprehension measures were used (Form A
written in complex sentences and Form B written in simple sentences),
separate analyses were made for the two groups. This procedure not
only allowed for a comparison of the variance accounted for by the
various predictor variables but also permitted a replication. The
means and standard deviations for all variables are presented in
Tables 4 and 5. Examination of these tables reveals little difference
between the two groups in the means and standard deviations with the
exception of reading rate, reading time and age. The average age of
group A was 17 years with a standard deviation of 2.1 years. Group B
on the other hand, had an average of 16.6 years with a standard devia-
tion of 1.1.

Means and Stgndagd Deviations

The means and standard deviations for all variables are presented
in Tables 4 and 5. Reading rate (BB) is reported in words per minute,
and total reading time (RT) is the total number of seconds to read the
first story (RM). This measure was obtained so that it could be deter-
mined whether or not there was any difference in reading time between
the different forms of the story. No difference occurred. A z-ratio
of 1.08 was obtained and was not significant. All other measures are
reported in terms of the mean number of correct responses.

60

 

61

TABLE 4

Means and Standard Deviations of all Variables for Group A

 

 

 

 

 

Standard

Variables Means Deviations
Age 17.03 4.07
IQ 115.09 17.67
V0 i 19.81 6.84
an j 197.30 66.92
HT § 289.64 E 64.89
08 E 64.08 ; 12.59
on § 18.91 § 4.09
SE 5 20.17 5.06
SS f 20.27 5.84
AN E 17.28 » 3.90
Ta : 7.59 3.04
RN ‘ 17.18 4.22
RNM ‘ 14.07 u,3®
BB N 34.23 ‘ 16.19

 

 

 

O\
N

TABLE 5

Means and Standard Deviations of all Variables for Group 8

 

 

 

 

 

Standard

Variables Means Deviations
Age 16.67 1.13
14 113.99 12.10
v0 17.84 6.33
RR 209.78 114.41
RT 293.75 72.91
LS 35.99 11.51
33 19.04 4.29
3 19.00 4.60
53 19.66 6.05
A. 16.71 3.82
‘31 8.10 2.92
.1. 13.33 4.37
h. . 14.43 4.19
L3 82.i1 15.63

 

..o
._.

 

It had been hypothesized that the mean reading comprehension scores
both when memory was a factor (RF) and when it was not a factor (REE),
would be significantly higher for group B than for group A. In order
to test this hypotheses z-tests were made on the mean RM and RNM read-
ing comprehension scores between groups A and B.

The mean reading score for group A on the RM test 17.2 (S.D. =
4.22) and the mean for the same test for group B was 18.3 (S.D. = 4.37).
The zrratio between these two means was 1.5 and.was not significant.

The mean reading score for group A on the RNM test was 14.1 (S.D. =
4.38) and the mean for this test for group B was 14.5 (S.D. = 4.19).

The z-ratio obtained for these two means (z = .61) was also non-signifi-
cant. The null-hypotheses that the reading comprehension test scores
for group A and group B would be the same could not be rejected when RH
was the reading comprehension measure used nor when RNM was the measure
of comprehension.

Intercgrrelations

The intercorrelation matrices of all variables for both groups are
presented in Tables 6 and 7. Correlations of .19 or greater are signifi-
cant at the .05 level and .26 or greater, at the .01 level. First of
all, it should be noted that the correlation between IQ and reading
comprehension is about the same for group B as the correlations reported
between these variables in most studies found in the literature (r
between IQ and RM = .49 and between IQ and RNM = .62). However, when
complex sentences are used in the reading passages (group A), the cor-
relations are much lower (r between IQ and RM = .22 and between IQ and

RN = .04). Vocabulary (V0) and total recoding (RE) on the other hand,

64

do not exhibit different correlations with reading comprehension between
groups A and B. For group A the correlation between V0 and RH is .50
and between V0 and RNM is .62. For group B similar correlations exist:
the correlation between RE and reading comprehension is also much the
same for both groups. For group A, r between RE and RM is .49 and
between RE and RNM, r = .52.

Reading rate and digit span are the only variables which consis-
tently exhibit correlations which are very low. The correlations
between these two variables and the other variables are also more un-
stable than the other correlations in the matrix. Note that the cor-
relation between D.S. and IQ is -.32 for group A and .36 for group B.
Furthermore the correlation between 0.3. and R.R is .33 for group A
and only .03 for group B.

TABLE 6

Intercorrelation Matrix of all Variables for Group A

 

 

IQ VO RR 93 GT SE SS AN TR RM RN RE
I: --
yo .43 -_

a1 -.03 .05 ~-
DS .24 -.22 .
an .31 .42 .12 .00 --

SM .55 .48 .15 -.11 .55 --

SS .41 .59 .11 .05 .50 .65 --

he: .21 .52 .11 .04 .34 .39 .52 --

TR .67 .33 -.07 -.22 .24 .41 .21 .03 --

in» .33 .59 .10 -.11 .29 .39 .55 .40 .20 --

rams .13 .70 .17 -.02 .37 .41 .64 .61 .07 .67 --

as .57 .65 .13 -.06 .73 .36 .85 .64 .46 .57 .61 ~-

 

 

If” 4“! e-i‘

65

TABLE 7

Intercorrelation Hatrix for group B

 

IQ V0 RR DS GR Se 85 Am Th RM RNN RE
10 -—
v0 .71 --
an .02 -.08 --
03 .31 .22 .09 --
s; .55 .52 -.08 .17 --

se .57 .56 -.17 .12 .55 --

ss .68 .58 -.01 .14 .54 .56 --

AM .35 .36 -.09 .05 .22 .23 .34 --

TR .36 .28 -.09 .01 .32 .29 .28 .31 --

an .51 .57 -.16 .02 .29 .50 .47 .37 .45 -—

RNM .61 .65 -.01 .02 .31 .42 .55 .29 .31 .64 --

RE .73 .67 -.16 .15 .75 .77 .84 .56 .54 .59 .55 --

 

Table 8 presents an intercorrelation matrix for the five recoding sub-
tests with groups A and B combined (15455). An examination of this
matrix reveals that the intercorrelations among GR, SE, and 88 are
moderately high while the correlations among Ah, TR, and the other vari-
ables are relatively low. The highest correlation was between SE and
SS (r = .60) and the lowest was between AN and T3 (r = .16).

TABLE 8

Intercorrelation Matrix of Five Sentence Skill
Tests for the Combined Groups

 

 

GR SE 33 AM TR
an --
SB .54 --
ss .52 .60 --
AM .28 .32 .43 --
TR .29 .34 .24 .16 --

 

—— 0-.-

66

Reliability Estimate§_

Except for the IQ test (Otis Quick Scoring, Gamma Series, form 3H ,
the battery of tests used in this study were constructed by the author
and were not standardized. It was therefore important to get some
estimation of the reliability of these measures. The digit Span test P“
was given in alternate forms to a group of 40 high school students

and the obtained alternate - form reliability coefficient was .71. A:

an estimation of the relialfility of the five subtests of recoding, the

 

vocabulary test, and the two types of reading comprehension tests, Kuder- E
Richardson #20 reliability coefficients were obtained for each group.
These coefficients are presented in Table 9 and give an indication of
the internal consistency of these subtests. A comparison of the
reliability coefficients for each of the tests between the two groups
discloses little difference between the two groups and is another
indication of the stability of the measures.
TABLE 9
finder-dichardson 720 Reliability CoefficientSObtained for

Groups A and B on the Sentence Skill Tests, the leading
Comprehension Tests, and the Vocabulary Tests

 

 

 

Group G? SE 35 5 AM TR an RN V0
A .72 .83 .33 .64 .51 .71 .33 .90
3 .74 77 .88 .60 6% .74 .80 .86

 

.Factor Analyses

The decision was made to factor analyze the major variables in

order to determine if all the predictor variables were loading on the

67

same factor as the criterion measure(s). It was decided that the vari-
max rotation which mavimized the differences within a factor, across
variables, was the solution most appropriate in this situation.

The first analyses were made using the total recoding scores (la)
along with the independent variables 1%, V0, Rd and the criterion E?
variables HM and RNM. A separate factor analysis was made for each
criterion variable with the various predictor variables so that it

could be determined if the predictor variables loaded in approximately

 

the same way regardless of the type of reading comprehension test.
The varimax rotated factor loadings from these analyses for groups A
and B are presented in Tables 10-13. All four analyses yielded two
factors, the first accounting for 50-h01 of the variance and the second
for approximately 207. The first factor might be called a verbal or
language factor and the second a speed factor. Intelligence, vocabu-
lary, recoding, and reading comprehension all loaded heavily on the
verbal factor. The loadings on this factor were virtually the same
whether RH or AME was the criterion measure used.

TABLé lO

Varimax Rotated Factor Loadings of Variables:
IQ, V0, hi, hi, &M for Group A

 

 

 

 

 

Variables Factors
1 2
IQ .73 -. 20
V0 .9 .04
fin’ . 05 . QR
=13 .R‘? ‘ .14
(Ii-I . 7 3 . 10

 

 

 

68

Proportion of variance accounted for by each factor:

II
\J‘
to

factor 1

factor 2 = .21

TABLE 11

Varimax Rotated Factor Loadings of Variables:
RE, RE, and AM for Group B

IQ, v0,

 

 

 

 

Variables Factors
1 2
IQ .88 -.12
V0 .87 .02
RR -.04 .93
.ua .33 .0?
REM .75 .23

 

 

 

Proportion of variance accounted for by each factor:

factor 1 .53

factor 2 .21

TABLL 12

Varimax Rotated Factor Loadings of Variables:
IQ, VO, RR, 13, and iNH for Group A

 

 

 

 

 

 

Variables Factors
1 2
14 .67 -.h3
V0 .88 .03
Lin . 09 . 86
AB .90 i .02
:u .75 ' .39

 

Funaportion of variance accounted for by each factor:

factor 1 .52

factor 2 .22

 

69

TAbLb 13

Varimax metated Factor Loadings of Variables:
IQ, V0, an, lb, duh for Group u

 

 

 

 

Variables Factors
1 2
- It: . 99 . 06
V0 .89 -.07
Fill - . 0?. . 99
H3 . Q6 . 05
9%“ .Q1 -.13

 

 

 

Proportion of variance accounted for by each factor:

.59

Factor 1

.20

Factor 2

The differences in the loadings across groups were very slight, but IQ

did load less heavily on the first factor for group A. The speed factor

in all cases was accounted for almost entirely by reading rate which

'was the single variable with high loadings on this factor.

A second set of factor analyses was made using the same combination

(Jf variables as those presented in Tables lO-l3, with the exception of

time recoding variable. In this set of analyses the subtests of the

rwecoding measure were used instead of the total score. These analyses

aixied in determining whether or not all of the subtests loaded on the

sarme factor and if not, which ones loaded on the same factor as the

renrding comprehension measures. The varimax rotated factor loadings for

tfllijs set of analyses are presented in Tables 14-17. All four factor

arutlyses yielded three fairly stable factors after two rotations. The

tfliizwd rotation created a fourth facton but its structure did not appear

70

TABLE 14

Varimax Rotated Factor Loadings of Variables:
IQ, V0, an, GR, 35, 83, Ah, TR, and RH for jroup A

 

 

 

 

 

 

 

 

 

 

 

Rotations
First Rotation Second Rotation I Third Rotation
.Eagtors IFactors % I Factors1
Egriables l l 2 1 5 2, 3 ' l i ,2, 3. U
IQ .36 [ -.80 .26 I -.95 .03 f .25 .-.84 .02 .20
v0 .75 g -.25 .77 g -.30 5 .03 .76 -.29 .005;.20
a: E .32 .34 ! .ou f .07 , -.au .06 .05 §-.09 5.07
an i .67 ’ -.22 no 5 -.36 -.32 .20 -.13 5-.03 ;.a9
“ t 6‘ -.u; ‘ .53 l -.57 —.90 .35 -.03 1-.13 I.6R
as f .05 -.16 .72 = ~.2< -.13 .72 -.17 -.06 5.45
x f .70 .13 ‘ .73 .03. at»: .120 .14 . fiOOQ .34
r I 14 -.ea .02 ; .00 .c? .03 -.91 ; .05 '.09
an i .74 -.09 . .77 : -.1u .02 .s5 -.20 i-.09 -.009

.Prorortions of variance accounted for bv each factor:
. J
Ifiirst Rotation: Factor 1 = .39; Factor 2 = .21

:Seccmd Rotation: Factor 1 = .BQ; Factor 2 = .2“; Factor 3 = .12

Tlxird Yntation: Factor 1 .28; Factor 2 = .21: Factor 3 = .11:

Factor 4 .18

71

TABLL 15

Varimav Rotated Factor Loadinqs of Variables:

L4, v0, a2, Ga, 3%, 53, AP, rt, and a.
for Group B

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

notations
First Rotation Second Rotation Third notation
Factors Factors Factors
Variables l 2 1 2 3, 1 2 .3 4
12 3 .37 -.O? .80 .14 .33 Q0 '12! .29 .16
v0 % .31 -.19 .77 -.01 .31 .77 -.04; .35 .05
RR ; .17 .92 -.02 .97 -.09 -.023 .9“: -.05 -.03
SR é .72 ! -.C9 .77 -.O‘ 09 .7q; -.04: -.1O .22
SE ; .71 g -.27 .7‘?I -.23 .lhi .73; -.2°' .0t% .16
.3 .32 ‘ -.07 E .79; OD .2ug .79: .0a .20: .06
t ? n ' .
a: .30 l -.03 E .luf “a .77; .1Sf —.03 .901 .13
.- 1 1 I ‘
r1 .3Q - 50 g .10 - 07 , 735 19 -.03 .15‘ .95
7a sa - 51 1 .ué -.17 A9‘ a<~ -.1o .uo' .41
i
lawaportions of variance accounted for by each factor:
lfiirst flotation: Factor 1 = .42; Factor 2 = .17
seuoond iotation:" Factor 1 = .37; Factor 2 = .12; Factor 3 = .20

1%rird flotation: Factor 1 = .37: Factor 2 = .12; Factor 3

Factor 4

.13

= .15

 

“r
I
J‘v

TABLE 16

Varimax Rotated Factor Loadinzs of Variables:
Ia, V0, Ra, 3d, SE, 38, AM, TR, and RNM for Jroup A

 

 

 

 

 

 

 

 

 

 

l Rotations
—[ First Rotation Second Rotation Third Rotation
. Factors Factors , Factors
Variables : 1 2 l 2 I 3 1, l2 3_, 4
IQ P 15 -.97 i .17! -.93 g .0“ : .16 {-.97 ' .02 ‘-.22
V0 .72 -.37 E .77 5 -.33 E -05 i .79 f'-36 J““7_‘ 15
11%). : 3.3 .20 E .05 .04 -.95 : .C-‘? I .011- -.99 ;'-.07
SR } .55 - 33 .h? -.u0 - 27 g .?1 §-.10 - 02 g- 39
SE S .54 j -.60 .50 , -.62 -.P’ I .33 %-.U# '-.12 §-.63
: - . ;
so i .77 i -.34 .77 ‘ 33 ' *.13 .45 §-.23 §-.04 f-.52
AM 2 .73 -.03 ; .91 , .01, - 02 .90 % .02 :-.03 3*.13
Ti 3 -.04 -.83 ‘ -.0062 -.39 .10 . .0? l-.Ql , .05 ;-.09
RNM ; .97 -.02 .3“ .02 .33 i .02 “.10 “.13

-.07'

 

1

PrOportions of variance accounted for by eaca factor:

= o 2’?-

First Rotation: Factor 1 = .36; Factor 2

Second rotation: Factor 1 = .35: Factor 2

Third Factor 1

lotation:

Factor 4 .18

.30: Factor 2 = .223330tor 3

.11:

.26; Factor 3 = .12

TABLE 17

Varimax Kotated Factor Loadinqs of Variables:
12, V0, ii, 33, 33, 33, At, T: and 41$ for Eroup :

 

 

 

 

 

rotations
First Rotation Second “otation Third (otation
Factors Factors Factors
Variables l 2 l 2 3 1 2 3 4
I} .93 .02 .92 .12 .23 .33 .09 .22 .19
JO .34 -.07 .31 .02 .25 33 -.OS .27 .04
ii .06 .92 -.C3 .95 -.00 -.Ol .93 -.07 -.03
J\ .69 -.20 .73 -.17 .09 .68 -.10 -.21 .47
3 g 71 -028 a 7'3 -01-2'? 007 o 75 -025 -OO() .25
b) 00’2 -001 olTn .06 02]. 0:20 .03 015 015
an .45 -.93 .17 -.001 .32 .20 -.09 .33 .22
T: .45 - 36 .7? -.12 .72 .15 -.005 .2!Q 3”
3". .70 07 .67 .13 .32 .71 .09 .41 -.04

 

.Froportions of variance accounted for by each factor:

.First Aotation: Factor 1 = .46; Factor 2

lxecond iotation: Factor 1 = .40; Factor 2

IWuird notation: Factor 1

Factor 4

.13

.12; Factor 3 = .16

.40; Factor 2 = .12; Factor 3 = .13

.13

74

[‘A BLT!- 19.

Varimay Aotated Factor Loadinqs for the
Sentence Skill Tests for Both Groups

 

 

 

 

Factor Factor
Variable 1 2
Sm .62 .46
)3 .37 .49
s3 .91 .25
m; . 77 -.13
TH .05 .90

 

Proportion of variance accounted for by each factor:

Factor 1 = .42; Factor 2 = .27

to be as stable across the different conditions, and it is probably an
artifact of the criterion used to terminate the rotations (the computer
terminated the rotations when only one variable loaded heavily on a
single factor). Interpretations of these analyses were made upon the
basis of the factor loadings created by the second varimav rotation.
The 33 subtest consistently loaded heavily on the first factor. Also
loading heavily on this factor were the variables: V0, reading compre-
hension (1F and AME) and to some degree 1., 31 and 33. A second factor
which occurred consistently (although not always in the same order) is
the speed factor. For group A, T3 loaded along with I; on this third
factor but for group B it loaded with AM. 3% and 33 tended to load

along with 38 in the four analyses.

75

In order to gain further insicht into the relationship amonq the
five recoding subtests, a final factor analysis was performed. The
data for qroup A and B were combined and the five recoding subtests
were factor analyzed. The results of this analysis appears in Table 19.
Two factors eme" ad from this analysis. All of the subtests loaded
heavily on the first factor eycept Pa, which loaded by itself on factor
two. The first factor accounted for 42b of the variance while the
second factor accounted for 27$.

M ti Re ressi Ana 565

One of the important objectives of this study was to determine
which of several independent variables would serve as the best predictors
of reading comprehension. As an initial exploratory analysis all the
independent variables were used in least squares-delete and least
squares-add multiple regression programs. Analyses were made for both
groups A and B using the total score for recoding rather than the sub-
scores.

The least squares-delete analysis yielded the following results.
LJhen the RM reading comprehension test was used as the criterion measure
:and the total recoding tests were used rather than the subtests along
trith the other independent variables, R25 for the full equation before
deletion of any variables were .43 for group A and .43 for group B.
Fer group A, all variables except vocabulary and recoding were deleted
one at a time without a significant drop in R2. With only V0 and RE
left in the regression equation, the R2 was .40.

A somewhat similar pattern occurred when the RNM reading compre-

hension test was used as a criterion measure. This time the R2 for the

76

full equation for form A was .67 and for form.B .51. After deletion of
all but the significant contributors to the variance, age, vocabulary,
and recoding remained for group A. These three variables yielded an 32
of .67. IQ, V0, and D8 were the variables remaining for group B and
the R2 for the equation with these three independent variables was .50.

The least squares-add program yielded the following results. For
group A the first variable to be added was vocabulary and the regression
equation yielded an R2 of .35. Receding was added next, increasing the
R2 to .41. This increase in variance was significant at the .01 level.
No other variables were added to the equation.

A similar pattern occurred for group B. This time recoding was the
first variable added and an R2 of .34 was obtained. Next, vocabulary
added a significant amount of variance and an R2 of .40 was obtained.
Again, no other variables were added.

Using reading comprehension test RNM as the criterion variable,
similar results were obtained. For group A the variables accounting for
a significant amount of variance were vocabulary, age and recoding; and
they were added in that order. For group B vocabulary, IQ and digit
span were added to form the final regression equation.

When the least squares delete and add programs were used to analyze
the data in which the five recoding subtests were separately entered in-
to the regression equation, the important independent variables were
vocabulary, semantic-syntactic conflict, grammaticality and self-embedded-
ness for group A and vocabulary, IQ and transformational complexity for
group B. These variables were left after all non-significant variables
‘were deleted in the least-squares delete program and were the first vari-

.ahles to be added in the least squares-add program.

77

The least squares ordered delete program permitted control over
the order in which the variables were deleted and allowed several
variables to be deleted simultaneously. By deleting a specific variable
and then noting the drop in the R2 value from the full to the restricted
model, the proportion of variance accounted for which is unique to the
deleted variable(s) could be determined. Two different full models were
constructed for use in analyzing the data from groups A and B. The
first full model, referred to as FM I, included the predictor variables:
IQ, V0, and RE. Regression equations were developed to predict both
RM and RNM scores. The second full model, referred to as FM II, con-
tained the following predictor variables: IQ, V0, GR, SE, SS, AM, and
TR. Again, regression equations were developed to predict both RM and
RNM scores.

The R2 values for FM I are presented in Table 19. Several re-
stricted models (designated R21, R22, etc.) werecnnstructed, and these
are described in Table 19 along with their corresponding R25. By com-

2 of the corres-

paring the R2 of any restricted model (R2) with the R
'ponding full model (FM), the contribution to the prediction of the
criterion variable can be ascertained for the deleted variable(s).

By subtracting the a? for the restricted model from the R2 for the
:full model a proportion indicating the unique variance accounted for by
'the deleted variable can be obtained. This proportion will be referred
‘to as the predictability index for the deleted variable(s). For example,
the R2 for R23 (see Table 19) predicting RM is .36 for group A and .35
:for'group B. The R25 for the corresponding full models are .#l and .uo.

'fhis indicates that RE (the variable deleted to create R23) accounted

78

for a 5% drop in R2 for both groups A and B. Thus, RE's predictability
index for predicting RM is .05. It can be stated that recoding (RE)

makes a unique contribution to the prediction of reading comprehension

and is an important variable in the regression equation. An F value

could also be obtained to test the significance of the drop in a? from
full model to restricted model, but these F tests are not reported because
they are a function of the number of cases and tend to be meaningless with
large sample studies. The comparison of importance is a comparison of the
absolute size of the drops in R2 from one equation to another.

An examination of Table 19 reveals the fact that IQ makes no unique
contribution to the prediction of reading comprehension involving memory.
Deleting IQ from the equation predicting RNM, however, causes a drop in
R2. The predictability index for IQ predicting RNM is .10 for group.A
and .03 for group B. IQ is important in the prediction of.RNM. It is
more important when RNM is based upon the complex sentences than when it
is based upon simple sentences. Vocabulary makes a unique contribution to
the prediction of reading comprehension. When RM is the criterion pre-
dicted, vocabulary's predictability index is .09 for group A and .04 for
group B. When RNM is the criterion predicted, vocabulary's predictability
index is .18 for group A and .07 for group B. RE also is an important
predictor of reading comprehension. The index of predictability for RE
predicting RM is .05 for both groups.A and B. When RNM is the variable
being predicted, the predictability index for RE is .17 for group A and
.00 for group B. RE is not an important predictor of RNM when the

passage is written in simple sentences.

R25 for Full and Restricted Models Using IQ, V0, and RE as

79

TABLE 19

Predictors of Reading Comprehension for Groups A and B

 

 

 

 

 

 

 

l
Model 1 Independent Vari- Variables De- Criterion ‘ R2 Value
5 able in the leted from Predicted i
2 agpégjgn the eguatipn # iGroup A B
FM-I IQ, v0, RE None ; an 9 .ul .uo
_____________ 1,____-__;___iure___4_...61_ p.27..-
RZ-l g v0, RE ; IQ 3 RM ’ .41 :.uo
...—...."— _________ é. ....... L_-*{EI'I_-4_;52-1L_'_M__
RZ-Z IQ, <3 i v0 ’ an ' .32 .36
----_______-_i_-_-___;_JBL---fifi-b@__
R2-3 IQ, v0 RE an .36 .‘.35
_ . MM .% fa?
____________________ T“"—”"“ _ _
37-h as IQ, v0 1 my .32 (5.34
' ’6‘ £
________________ T . ._ .. .. .. _. .. t .... ...“.L‘j— .. .... .. £5 .. r'lo_ .—
na-5 JO 7 Id, Re i an i .35 ;.33
' ..‘t, _ i
_________ ._ _,_._.+-_...- _ - ....r _ _A:}_._._i iy' ..inflz -
{z-G Id 5 we, do } n5 l .11 $.25
; x b .0; rs}?

 

80

Vocabulary and recoding seem to be the important predictors of
reading comprehension when memory is involved (RH used as criterion)
and IQ becomes an important predictor along with V0 and Rs when memory
is not involved (RNM used as criterion).

Further examination of Table 19 discloses the fact that the unique
contribution of two variables, taken together, is pg; the simple sum-
mation of the unique contribution of each variable taken separately.
The predictability index of RE and V0, taken together, is .30 for
predicting RM, .16 units higher than the sum of their individual
indices.

The R25 for the second full model (FM II) and the corresponding
restricted models are presented in Table 20. An examination of this
table indicates the relative importance of the subtests of recoding
as predictors of reading comprehension. This information is useful in
answering three important questions not answerable from the results
presented in Table 19: "Which of the recoding subtests is most im-
portant in predicting reading comprehension?"; "Why does recoding (RE)
predict RNM differentially for groups A and B, but predict RM equally
as well for groups A and 8?"; and "Do certain combinations of subtests
predict better than some other logical combinations of subtests?"

The predictability indices for each of the variables included in
Table 21 is presented in Table 22. These indices were computed by
finding the difference between the h2 for FM II and the R25 for R2-7

through 32-13.
Semantic-syntactic conflict (88) was a better predictor of RNK

for group A than any of the other subtests (predictability index = .12).

81

TABLE 20

R25 for Full and Restricted Rodels Using 14, v0, GR, Sc, 33

 

 

 

 

 

 

 

 

 

 

AM, and TR as Predictors of Reading Comprehension for Groups A and 8

Model Independent Varir Variables De- Criterion 4 R? Value
able in the leted from Predicted L
equation the e nation 1 Group A B
FM-II IQ, V0, GR, SE, None Rh 3 .50 .h9
... _ _ 1,. §SL éML Iii... _. _ _ r. _______ “. .. ...-“£39"... ... _. L62 _1u.'5.1_
Rz-7 V0, «:11, 32., SS, IQ rd». y .50 .49
-. _ _ .14... IR. ._ _ ..‘_ .. ... ....... 1p _ {£3}; _ L _. i6}. _, ~39.
Rz-B 1 IQ, GR, SE, 35, v0 3 if} ; .uu .uu
:—-—wea_——e—-r-—:--—-e-J¥--;—¢u-tu-
12.2-9 It}, V0, SE, 03, . 11’. I, [M 5 .LF9 1 .%
___r§gyt ______ L _______ L_J@L_L_fi2_by—
RZ-IO IQ, v0, 31, as, . 3s } iM -; .49 i.ee
__ _ __ . g3; TR ______ - _________ i u_.~-1_ _ I _ A62 __ : .51_
Rz-ll {'10, v0, *1, SE, T’ .3 f’ in T' .3 flue
___'BAQ- ______ L _______ J_:@L a_Jfi-L5L
32-12 F IQ, v0, G1, SE, 2 Ah [' Rn € .50 s.u8
__.__‘,_§.S.._'ILR ...... L. _______ L_.fifll__;.__s__L-il_._
RZ-lB ; IQ, V0, GR, SE, 2 Ti 1 KM ! .50 5.42
.—-—-’ §Silfi__.__-_,___.‘.l__j_-L__3_?~§_¥1‘1__‘.._.-_62-._-50__
Rz-lu { IQ, v0, AM, FR ca, Sh, so 1 4M g .37 1.44
_ _ _ ________________ I __ ...—1i;— _ L __ 4.6; __ : ell—£7_ _
RZ-15 f IQ, vo, GR, as t' AR, Tn {' in j .50 TLAO
---ias _______________ ..--ilu.._2_.._6£z_‘.io__
R2-16 F 3R, SE, 33, AR r' 1;, v0 3 AM f’ .uh. flag
_____£_T_R__. ______ L__:_____;___n_:yv:__;__g5 _;_.35__
Rz-17 , v0, AR, TR 2 IQ, eR, SE, , Rh 3 .37 g.n3
---; ......... 3% ______ ‘_;@L-3-Lfl_fl%i_
Rz-IR ; v0 GR, SE, so {'Iu, A., TR F" RR ?' .50 T.
__-‘ _________ ‘ _______ 5_J%__i_£fi-M%__
R2-19 s IQ, TR, AM I'ro, 31, SE, {' as. f' .23 TL36
_._._.’ _________ 5 £5 ...... 3 _.__§L_._.' _.13§._.’-39_._
Rz-zo IQ, Gt, SR, 33 T'vo, AE, T” f" an 7' .u3 ffl3u
_____________ L_______;____,_’;;,1’__;_;LL_L_;__L._L£2__
Rz-zl 1., V0 - SR, 83, 33, 3 gm .36 3.35
i_-____, ...... pfiufiL:__p_QflL__-J2_LML_
rtz-22 AR, IR . IQ, V0, cu, ,~ .43; .20 ', .26
--- __________ L¥i$____;_jfl____éfl_;¥L_
22-23 SR, SL, 35 ; IQ, vo, A? 3 a" .42 .31
1 TR ' .42 I .32

 

8"
,I
5'

TABLE 21

Predictability Indices 32 for FM-II Minus R2 for R2 of
Is, V0, GR, SE, SS, AN, and TR when used to Predict
RM and RNM for Groups A and B

 

 

 

 

 

 

Independent R4 RNM
Variable Group A Group B Group A Group 3
IQ .00 .00 .04 .01
v0 .06 E .05 .12 .08
sit .01 k .03 .00 .02
33 .01 i .03 .00 .00
as .12 1 .01 .05 .02
AM .00 i .01 .03 .00
TR .00 .07 .00 .01
x

 

 

 

fl

Transformational complexity (TR) predicted RE best however, for group
B (predictability index = .07). SS was also the best single subtest
predictor of ENE for group A (predictability index = .05). For group
B SS had a predictability index of .02 which was the same as that for
Git .

Notice that the predictability index of SS (Table 21) is higher in
all cases for group A than for group B. This test not only is a better
predictor on the whole than the other subtests, but it is the only one
of the subtests which consistently predicts better for the complex
reading passages (group A). GR, SE, and TR all have higher predicta-
bility indices than SS for the simple reading passages (group B).

.Ambiguity (AN) contributes to the prediction of RH for group B and RNM

83

for group i but does not contribute to the prediction of an for group A
or uNh for group D.

A comparison of all the independent variables presented in Table
21 indicates that if an is being predicted for group A the variable with
the highest index of predictability is $8 (.12). If, however, in is
being predicted for group 3, then V0 has the highest predictability
inder (.05). When RNM is being predicted, V0 has the highest pre-
dictability indev for both groups (.12 for group A and .OR for group B).
IQ is not an important predictor of RH (predictability indices = .00);
and although it is an important predictor of YER, it has lower pre-
dictability indices than either VC or 33 (.04 for group A and .01 for
group B).

The predictability indices of some of the important combinations
of the independent variables in FH-ll are presented in Table 22.

TABLE 22

. . . . . -. . a2 2 . i
Predictability lndices (Difference between “FE and u_,) of important

Combinations of the Independent Variables 1;, VC, 3T, SJ, SS, AK
and TR when used to fredict HF and kNH for Groups A and B

 

...-o

 

 

 

 

Independent Rh ENE
Variables lroup A Group B Group_& _ Group 8
GR, 3x, 35 .1? .05 .06 .04
V0, :1}, Sill, -J‘\.: .27 F 010' 099 ‘ 01?
S
l
I?, 3?, 3K, 33 ,1? ,04 _00 S .07
l v
-\ ~ 'n I ' ‘ \ I . ‘
;:,:mq s), Ar, r< .14 ; .14 .15 l .14
i 3
Ar, rm .00 , .09 . .03 t .01
l - 1 i i..~

 

 

 

 

 

The decision was made to combine the subtests GR, SS, and SS. The
intercorrelations among these three subtests are fairly high (Table 9),
and all load heavily on the same factor (Tables 14, 15, 16, and 17).
The predictability indev for these three variables combined is higher
than for any one of them alone. Adding IQ to the group does not in-
crease the predictability indices appreciably for ER but does cause
some increase for RNM. Adding V0 to these three, as would be expected,
increases the predictability indices considerably. a particularly
interesting result was obtained when all five subtests were combined
and then deleted 1x1 form a restricted model. The addition of an and
TR to the 13, SE, 83 group only increased the predictability index from
.13 to .14 for group A when the criterion was RM. However, for group B
(criterion (am) the predictability index increased from .05 to .14. The
increases in predictability for criterion RNfl are even greater (group.x,
from .06 to .15 and group B .00 to .14). As can be seen in Table 21,
when all five subtests are combined, the predictability index is high
and approximately the same across all four conditions. The combination
of AN and TR does not contribute to the prediction of Rh when the
passage is complex (group A) but does when the passage is simple (group
B; predictability index = .09). Although not large, the predictability
indices for Am and TR combined do indicate that they contribute to the
prediction of RNA for both a complex passage (group A) and for a simple

passage (group B).

CE‘IIXT’T‘Q 'l - .DIFSCUStIlOII

In this chapter each of the major objectives and hypotheses will
be discussed in light of the results reported in Chapter IV and implica-
tions for further studv will be noted.

912.199 tiYLLl

The first objective of this study was the design and develop-

ment of tests which would reliably measure individual dif-

ferences in various sentence skills which have been postu-

lated as necessary to the recoding process.

Five tests of sentence skills were developed and a measure of internal
consistency, huder-dichardson 620 (RH a20), was obtained for each test.
Given the fact that each of these tests were fairly short, the n: .20
reliability coefficients which were reported in Table 9 are all rela-
tively high. It should be noted, however, that the tests for ambiguity
and transformational complevity have lower reliability coefficients

on the average than the other three tests.

deplication is a built-in feature of this study because two ran-
domly assigned groups were used. All subjects were tested under the
same conditions at the same time on the same tests. The only tests
which were not the same for both groups were the criterion tests. It
seems reasonable to conclude that if similar results were obtained for
the two groups, then some added confidence in the stability of the
measures would be gained.

Whereas most of the subtests yielded reliability coefficients

which were approximately the same for both groups, the test for trans-

formational complexity yielded less internal consistency, but it also

85

86

proved to be less stable than the other tests. This fact may account

for the low correlations between transformational compleyity and other
variables as well as for the limited usefulness of this test as a pre-
dictor of reading comprehension.

It is worth noting that the subtest with the highest internal
consistency (KR §20 = .89) is the semantic-syntactic conflict test. An
indication of the stability of this reliability coefficient comes from
the fact that it replicated from group A to group B. fhe Ll fZO was
.8? for both groups. This test is also the test which accounted for
more variance in reading comprehension than any of the other subtests;
and in fact for both Rb and RNM reading comprehension measures, this
test accounted for more unique variance than 1).

Objective 2

The second objective of this study was to determine the eytent

to which the recoding subtests all measure the same general

sentence skill factor and the extent to which each measures

some specific sentence skill.

90 conclusive evidence was uncovered ia attempting to meet this
objective. The bulk of the evidence seems to suggest, however, that

hese tests are measuring the same factor with one erception, the test
of ability to recall transformational cues. fhe transformational com-
plexity test had the lowest correlations with the other subtests and
in the various factor analyses, tended to load on a separate factor
from the other recoding subtests. It should be pointed out, however,
that this test also is the test with the lowest h; #20 reliability

coefficients and the obtained coefficients do not replicate well from

group A to group 3. The low correlations with the other subtests may

37

be a function of the lower reliability of this test. The fact that
this test tends to load on its own factor in the factor analyses can,
in turn, be a function of the low intercorrelations. Transformatioaal
complerity, furthermore, did not correlate as well as did the other
subtests with the major independent measures nor with the dependent
measures in this study. These low correlations might also be a func-
tion of the lower reliability of the test.

Because of the nature of this test it might be e7pe-ted to load
heavily on a short term memory or a Speed factor. The As needed to
recognize in eyact detail, sentences which they had just previously
read and studied carefully. A similar taSk was required of the 3 in
the digit span test and a high correlation might be expected between
TR and OS. The correlation was -.22 for group A and .01 for group 5.
Neither correlation is high and the first is in the Opposite direction
from that expected. There also seems to be no replication across the
two groups. The 35 were given only 20 seconds to study five sentences
in this task. Perhaps the results would agree wita the results of the
reading rate measure. The correlation between mi and Ti was -.0° for
group A and -.03 for group a. This time the results replicated across
the two groups but the correlations were again in the opposite direc-
tion from expected and were essentailly zero. Further investigation
of this test's relationship with the other tests used in this study
should be made after the test items have been rewritten and the test
lengthened in an attempt to increase its reliability.

The tests, JR, SE, and 35 all tended to correlate very highly with

each other and the intercorrelation pattern of these three tests

88

replicated across the two groups. These tests also tended to load
together on the same factor in the factor analyses and this pattern
replicated across groups. Replication across chances in the criterion
measure (Rh or HRH) also occurred.

It is unclear whether or not the test for sensitivity to ambiguity
(l%) is measuring the same factor as the other subtests. In some of
the factor analyses it tends to load on the same factor as GR, 33, and
83 (cf. Table 10, lb, 16). In the factor analyses on the data from
group 3, however, AH tends to load on the same factor as Tu. Further
study of the kinds of skills required to perform well on the AS test
are needed. It correlates most strongly With vocabulary (at least for
group A; r = .52) and is perhaps a skill which depends upon or involves
a 3'5 ability to conjure up quickly and accurately many shades of
meaning for the words in the sentence. It probably involves more than
just sensitivity to the ambiguity of the structural cues in the sen-
tence. How much more is a question for further research.

Obi” C tiVG 3

The third objective of this study was to determine the rela-

tionships among the various sentence skill tests and the

following variables: IQ, vocahulary, reading rate, digit

Span and reading comprehension.

fioderate to high intercorrelations were found among the variables:
IQ, vocabulary, the recoding subtests, and the two reading comprehension
tests. It should be noted that the intercorrelations among most of the
variables replicated acorss the two groups. The correlations between

IQ and the other variables did not replicate, however, as an examination

of Tables 6 and 7 discloses. For group B, the correlations between 1Q

89

and the two readinz tests we 9 ,5]- with _i and .61 with 15.7“ . i‘hese r8
are within the range usually found between IQ and reading comprehension.
For group A (the complex passages) however, low correlations were ob-

.13). Interestingly enou5h, the

tained (r,_ ., = .33 and r1

.L‘g’.-.. "

\)"‘..':
I". .L.

'll.‘

only situation in which IQ makes a unique contribution to the variance
accounted for in reading is the situation in which Tit, form A, is the
criterion measure.

The particular IQ test used in this study is not generally con-
sidered to be the best group intellig,nce test. Its reliability,
especially for the narrow range of slightly above average 35 used in
this study is not known. An examination of the items seems to indicate
that a high degree of verbal competency is needed to score high on
this test. In many cases vocabulary is important and in many others
the question is not difficult if the S can understand what the ques-
tion is asking. For example, the following question taken from the
test is really very simple once it is understood what is being asked:
"Find the two letters in the word UATJR that have just as many letters
between them in the word as in the alphabet. Which one of these two
letters comes first in the alphabet?" Perhaps, different results would
be obtained if a different IQ test were used and certainly this would
be appropriate in further research. It must be noted, however, that
the correlations between this IQ test and the vocabulary test (.41 for
qroup A and .71 for group 8) are in line with those reported in the
literature between these two variables. Also in line with previous
findings are the correlations between the reading tests and Id for

group 3 (simple passazes). One final indication that the results

90

obtained in this study may be due simply to the particular IQ test

used, comes from one of the early pilot studies. A different pattern
of results were obtained from an early pilot study in which the IQ
scores were obtained from the school records. The correlations between
IQ and RM (form.A) was .62 and between IQ and RM for form B was .26.

The RNM reading test was not used in the pilot study but notice that the
correlation between IQ and reading comprehension was considerably higher
for the complex version than for the simple version. It seems that the
specific IQ test used may have influenced the results of this study and
further studies should be conducted using a variety of IQ tests which
have known reliability and validity.

Vocabulary. the best single predictor of reading comprehension in
this study, correlated moderately to strongly with the other predictor
variables. It also correlated strongly with the criterion measures.
(for group A: rVO,RM = .59 and rVO,RNM = .70; for group B: rVO,RM = .57
and rVO,RNM = .65) Note that these correlations indicate that vocabu-
lary's relationship to reading comprehension is not affected very much
by differences in the level of complexity of the sentences within the
reading passage (the correlations are approximately the same for group
A and B).

The three subtests which are the best predictors of reading compre-
hension (GR, SE, and SS) correlate more highly with vocabulary than
the remaining two subtests. AM, however. does correlate fairly well
with vocabulary in group.A and in that group is a better predictor of
reading than it is in group B. Although success on the ambiguity test

may require the S to have a vocabulary of many shades of meaning.

91

success on the 33, SL, and 33 tests does not seem to require any un-
usual vocabulary skill. The GE test essentially eliminates vocabulary
as a factor because the only English words used are simple function
words such as "is, was, the," etc. The Sh test also includes no
unusual vocabulary items but rather requires the S to utilize word
order information. The semantic-syntactic conflict test, by far the
best predictor of reading and the subtest most highly correlated with
vocabulary, requires the S to ignore the cues of semantic meaning and
to rely on position and word form cues. It is postulated, therefore,
that 3H, 33, and 83 are measuring a different skill from that measured
by VO. It is further postulated that high intercorrelations exist
among these tests because they all measure language skills which are
probably obtained simultaneously through many language eyperiences and
especially through the language experience we call reading. Those

85 who have wide and varied reading experience would be evpected to
exhibit a high degree of skill on reading, vocabulary, grammaticality,
self-embeddedness, and semantic-syntactic conflict tests. ferhaps in

‘
7

future studies vocaoulary and recoling skills (as measured by semantic-
syntactic conflict, for example) could be independently manipulated
experimentally in an effort to demonstrate that recoding is a separate
skill from vocabulary (even though both may be related to some more
general language factor).

Yone of the subtests correlated well with reading rate or digit
Span. Reading rate, furthermore, consistently loaded on a unique

factor in the factor analyses. Digit span and reading rate seem to he

measuring something different from the other independent variables in

92

the study, but they do not contribute anything to the prediction of
the criterion variables used and so no useful positive information
about these measures was obtained from this study.

The correlations between each of the various subtests and the two
types of reading comprehension measures indicate that all of the sub-
tests correlate moderately with both types of comprehension test.

The strongest correlations were found between 83 and criterion measures
RE and REM. Although the memory, non-memory dimension did not seem to
make much difference for the 33 test, a much higher correlation was
obtained between AK and RNE than between if and l? for group A and

slightly higher correlations between Ab and RE than between AH and

31:
for qroup B. As was noted earlier, in is a test whose pattern of
intercorrelations varies widely with different types and different
forms of the readinz comprehension tests. Further study of the eyaet
nature of the skill(s) involved in this test is in order. it would
be interesting to find out if in further studies A; continued to cor-
relate higher with the simple passaee when memory is involved but
higher with he complex passage when memory is not involved. At
present, there seems to be no lorieal reason for this pattern of
intercorrelations,and it is probably due to the lack of reliability
of Afi and/or a result of the particular reading tests used in this
study. lenlication usina a variety of readina tests and lender, more
reliable A} test would help to discover if that is indeed the case.
Chi”, ctive 11/
The fourth objective of this study was to determine the

proportion of variance in reading comprehension scores
uniquely accounted for by each of the following variables:

93

Ii, Josibulqry, readini rate, and the sentence skill

measures: :rammiticality, self-embeddedne s, semantic-

syntactic conflict, ambiduitv, and transformational com-

DIOXity.

Readinr rate was one of the first variables deleted in the least
squares delete analysis and was not added fit all to the rerression
equations in the least Sduares add analysis. This information tended
to confirm what the ndioritv of researchers have indicated: readinr
rate is not 1H inportant contributor to the prediction of reading
comprehension. Consequently, reading rate was omitted fron the least
squares ordered delete analysis.

The least squa es delete and least squares add protrams are dif-
ficult to interpret, but the pattern was consistent for both procedures:
The innortant predictors vere It, V0, and i3 (and when the subtests
were used instead of lb, 33 was nest important\. Chose variibles were
deleted last and added first. In order to obtain a more accurate
assessment of the unioue contribution of edeh of these variables, the
least squares ordered delete prOTr1fi qu used. One of the Droblens of
the least squares delete and add proTrdms arises from the fdct that the
viriahles are deleted from or added to, the reeression enuation in a
stenvise fashion and the order of deletion or addition makes an impor-
tant difference in the erooortion of vnriance credited to a particular
variable. fhis problem can be overcome in the least squares ordered
delete or ~rqm because sinfle variables or trouus of variables can be
deleted from enecified, full model equations and essentially 1 systema-

tic deletion "with replacement" can be accomplished. Any variable or

combination of verinbles can be deleted.

9a

When the full model for predicting reading comprehension included
the predictor variables I}, V0, and g:, the variance uniquely accounted
for by I) was less than that accounted for by either V0 or Rm. When
RUN was the criterion variable, I; accounted uniquely for 103 of the
variance for group A and 3fl for qroup 5; whereas when KL was used as
criterion, 13 contributed no unique variance for either group A or
group 5. Similar results were obtained when the full model included

1, V0, and the five separate subtests of YE (GR, SE, 83, AH, and Pd).
Under these conditions I: again accounted for no unique variance in RH,
and it accounted for an even smaller proportion of QUE than it accounted
for when 13 was used in place of the five subtest scores (I"s pre-
dictability indices for :51 were now .04 for group i and .01 for group
B). Uhether these results would replicate if a different I) test
'would be used is a question that can only be answered by further re-
search.

Vocabulary accounted for a :reater proportion of variance on a
Ivhole than any other independent variable in the study. Only when the
«complex version of an was the criterion variable did any other variable
luave a hizher predictability index than V0. The complex version of
Rri'was best predicted by SS (predictability index = .12 whereas the
ccnnparahle index for V0 was .05).

The pronortion of variance accounted for by each of the sentence
skill subtests varied greatly depending upon which version and type of
Pqfissaqe was used as the criterion measure but overall, the best pre-
dicrtor amond the subtests was the semantic-syntactic conflict test.

n - 7.. -.,'. .
‘iulis test predicted the complex verSions of both he and Run eSpecially

1: I -.ll.-.f .

95

well (predictability indices = .1? and .05 respectively). The simple
version of dNH was predicted equally well by 33 and 33 (predictability
indices = .02). The simple version of WE was predicted better by GR,
SE, and T? than by SS, and equally as well by Al. T3 was an eSpecially
wood predictor of the simple version of TE. It had a hiaher predict-
ability index for that criterion measure (.07) than any other sinale
variable in the study.

Perhaps the most interesting finding related to the specific
objective under discussion is the fact that each of these sentence
skill tests predicts a specific criterion better than it does the
other criteria. "2 and SE both predict the simple version of RR
better than they predict any of the other reading tests. AM predicts
the complex version of RNM better than it predicts any other tests,
and TR predicts the simple version of Rh better than it predicts the
other tests. TR, it should be noted, has the hivhest predictability
index (.07) of any of the predictor variables for the simple version
of RF. SS predicts the complex versions of both Rh and HEM (indices
of .12 and .05 respectively) better than it predicts the simple ver-
sions of those two tests. ”hen all five subtests are combined (i.e.
deleted toacther to form the restricted model) the predictability
indices are higher for all criterion measures than for any single pre-
dictor in the study (includinz IQ and V0). Furthermore, the 3m pre-
dictability indices are approximately the same for all four criterion
measures (.lb for the simple version of both Ru and NHL; .14 for the
complex version of it; and .15 for the complex version of REL).

Further study of the nature of these tests and how they predict

96

different types of reading tasks needs to be made.

Objegnyg 5

The fifth objective of this study was to increase the pre-

dictability of reading comprehension over and above that of

the traditional predictors of IQ and vocabulary by the

addition of the sentence skill recoding tests as predictor

variables.

In accordance with this objective it was hypothesized that when
recoding (RE) is included in a multiple regression equation along with
IQ and vocabulary, a greater proportion of variance in reading compre-
hension (measured by both memory and non-memory comprehension tests)
will be accounted for than if recoding is deleted from the regression
equation.

This hypothesis was confirmed for all situations except where the

2 for the full

simple version of.RNM was the predicted variable. The R
model using RNM as the criterion for group B (simple version) was .h7.
When RE was deleted,the R2 remained .47. For group A, however, the

R2 for the full model predicting RNM was .63 and when RE was deleted,
R2 dropped to .fi6. Deleting RE from the regression equation used to
predict RM caused a drop in R2 for both group A and group B. For
group A the R2 dropped from .41 to .36 and for group B it dropped from
.#0 to .35.

This general hypothesis was tested again, using the five subtests
of recoding in place of the single RE score in the full equation. The
only subtest that caused a drop in R2 from full to restricted model
in all situations was SS. This variable added to the prediction of
reading comprehension above that accounted for by IQ and V0 for both

versions of each of the two types of tests. When it was deleted from

97

the full equation predicting the complex versions of the reading tests,
greater drops in R2 occurred than when it was deleted from the equa-
tions predicting the simple versions. The R2 went from .50 to .38 for
RM (version A) and from .67 to .62 for RNM (version A). On the other
hand, the R2 only went from .#9 to .48 for RM (version B) and from

.51 to .h9 for RNM (version B).

Whereas SS was a better predictor of the complex versions of the
reading tests, the tests GR, SE, and TR all predicted the simple
versions better than they did the complex ones. Judging from the
factor analyses one would expect GR and SE to best predict the same
kinds of tests which are predicted best by 38 because these three
tests loaded consistently on the same factor. TR on the other hand
consistently loaded on a different factor from GR and SE. Why GR and
SE best predicted the simple versions as TR did, rather than the com-
plex versions, as SS did, is not known. Further study of the relation-
ships among these various subtests is in order. It would be useful,
for example, to attempt to discover Specifically what kinds of behavior(s)
a 8 must engage in in order to successfully complete each of these sub-
tests. Perhaps quite different, but highly correlated skills are in-
‘volved in the different tests. Replication of this study with improved
versions of GR, SE, and TR would be another step toward the solution
of this problem.

Further confirmation of the above hypothesis comes from the drops
‘which occurred in R2 from.full to restricted models when the five
subtests were dropped simmltaneously. The R2 drapped from .50 to .36

:for criterion RM, group A; from .49 to .35 for RM, group B; from .67 to

98

.52 for RNM group.A; and from .51 to .h? for RNM, group B. As was

hypothesized, a greater preportion of variance in reading comprehension
was accounted for by the predictor variables IQ, V0, and the sentence
skill tests than that accounted for by only IQ and V0.

The use of the sentence skill scores increases the accuracy of
predicting reading comprehension. Furthermore, information concerning
a 8'3 sentence skill scores is more important in predicting reading
than is information concerning his IQ score. To demonstrate this fact,
Table 23 was constructed.

TABLE 23

Predicted RM and.RNM Scores for Four Hypothetical Ss
Differing on IQ and.RE Scores

 

 

 

 

 

Predicted
Subject Predictor Variables Reading Scores
IQ 0 GR 3;: 33 AM TR RM RNM
#1 130 20 18 20 i 20 18 8 17.3 18.7
2 90 20 18 20 20 18 8 17.0 17.3
3 115 20 1 22 25 26 | 23 10 18.5 15.6
! I
a 115 20 14 10 g 8 16 a g 14.3 10.5
1 1 :1 ,__- I

 

 

 

 

 

 

 

Table 23 presents the scores on the predictor variables for four hypo-
thetical 35. Two of 85 have the same vocabulary and sentence skill
scores but different IQ scores. The other two 83 have the same IQ and
Vocabulary scores but different sentence skill scores. For each student,
RM and RNM reading scores were predicted using the regression coefficients

from Full Model II (group A). Notice that although students #1 and #2

have IQ scores that are approximately three standard deviations apart,

99
there is very little difference in their predicted reading scores
(especially for RM). By contrast, the differences of approximately
three standard deviations in the sentence skill scores for students #3
and #u are more accurately reflected in the predicted reading scores.
Very different reading scores are predicted for students #3 and #h for
both RM and.RNM.

923m

The sixth objective of this study was to determine if sen-

tence skill measures were better predictors of reading com-

prehension when the reading passage contained complex sen-

tences than when the passages contained simple sentences.

In accordance with this objective the following hypothesis was
tested: When recoding is deleted from a regression equation which also
included IQ and Vocabulary as predictor variables, the drop in R2 will
be greater for group A than for group B. This hypothesis was confirmed
for the prediction of RNM. R2 dropped from .63 to .#6 when RE was
deleted from the FM I equation predicting RNM (version A). When RNM
(version B) was being predicted, however, the deletion of RE from the
equation caused no drop in R2. The practical value of this finding is
demonstrated in Table 2“. Scores for four hypothetical 85 are presented.
.All four 35 have the same IQ and V0 scores but differ on the RE scores.
Using the regression weights developed from.FM I for groups A and B
respectively, the Ss' scores on both the simple and the complex versions
of’RNM were predicted. Notice that for the simple version very little
difference in predicted reading comprehension scores occurred. However,
.as RE scores increased, there was a corresponding increase in the pre-

dicted reading scores for the complex version so that the difference in

predicted RNM (complex version) between S¥2 and S#5 is six points. The

100

difference between the predicted simple versi’on RNM scores for these

same 35 is only 1.1 points.
TABLE 24

Predicted RNM Scores (Simple and Complex Versions) for Four
Hypothetical Ss Differing on RE Scores Only. Predictions
Were Made from FM I Regression Equations.

 

 

 

 

 

 

Sgbjegt Pzggigtgz_1gtiaglgs Predicted Reading Scores
12 v0 an. MWW
#1 115 20 52 lh.4 10.1
2 115 20 68 14.7 12.1
3 ‘ 115 20 an 15.1 14.1
4 I 115 20 100 15.5 16.1
.

 

 

 

 

Although the RE variable is more important in the prediction of the
complex version of RNM than it is for the simple version of RNM, it
seems to make the same contribution to the prediction of RM, no matter
enrich version is being predicted. Perhaps this is due to the fact that
the subtests of RE predict differentially. The SS subtest clearly
predicts the complex versions of both RM and RNM better than the simple
xnersions. TR, on the other hand, predicts the simple versions of the
two tests better than it does the complex ones. For the prediction of
RNM, AM is an important predictor of the complex version along with $8;
arxi the strength of GR, SE, and TR as predictors of the simple version
is: not nearly as great as it is when RM is being predicted. It may be,
then, that when RM is being predicted, the influence of the various

subtests balance out, as it were, their predictive power: the SS test

101

predicting the complex versions best and the other subtests predicting
the simple versions best. In fact, when the subtests are entered into
the full model regression equation (FM.II) and then all deleted at
once (R2 21) the drop in R2 from full to restricted model is almost
exactly the same for both versions of both criterion tests. The inter-
relationship of these various subtests with each other and with various
kinds of reading tests needs further study. This information, along
with added information on the kinds of responses Ss must make to suc-
cessfully complete each subtest, would help to further improve the pre-
diction of reading. Once the skills involved in the performance of
each of these subtests can be specified and the importance of these
skills in reading be more precisely determined, then a program of re-
search can be established to investigate how these skills are learned.
Furthermore, it may then be va1uab1e to begin to seek ways of teaching
these skills and the transfer of them to the reading task.

Objectivg Z

The seventh objective of this study was to determine if

comprehension of a passage including complex sentences is

more difficult than comprehension of a passage containing

simple sentences.

In accordance with this objective it was hypothesized that group A
would obtain lower mean reading comprehension scores for both RM and
FUWM than group B. This did not occur. There was no significant dif-
ference between the mean reading scores of the two groups. This was an
unexpected finding for two reasons. First of all there are numerous

.studies which indicate that complex sentences are more difficult to

ilearn than simple sentences, and it was expected that if given a large

102

number of complex sentences in a prose passage, SS'WOUId have more dif-
ficulty learning the material than if they were given a passage contain-
ing almost all simple sentences. Secondly, one of the last pilot runs
indicated that the complex version was indeed more difficult. Forty-
one high school 35 were assigned randomly to the complex and simple
versions of RM. The mean for group A (complex version) was 16.0 (S.D. =
5.33) and the mean for group B was 13.35 (S.D. = 3.94). The difference
between these two means was significant at the .05 level (t = 1.82

with 40 d.f.).

First of all, it should be noted that even though the two means in
the pilot run were significantly different, the absolute magnitude of
the difference was not extremely large. Perhaps this difference was a
chance occunmnce (it happens one time in twenty at the .05 level) and
did not reflect any real difference in ability to comprehend and recall
the passage. Another reason for the difference obtained in the pilot
run not replicating in the final experiment may be that the 35 were
different. The 83 in the pilot groups were about the same age (i’= 16.5)
as the Ss in the final study (i’= 16.8) but the mean IQ of the pilot 83
‘was not known (no IQ was given). The differences between the two groups
on recoding ability cannot be determined because the recoding subtests
‘were changed somewhat from this pilot run to the final data collection.
The RE scores obtained by the pilot 33 were considerably lower however,
than RE scores of the 83 in the final study (Pilot study mean RE = 68
for group B and 72.5 for group A). If the RE tests were fairly compar-
.able in difficulty from pilot to final study, then it may be that the

lower recoding ability of the pilot Ss is what caused the difference in

103

the mean scores on the two versions of the recoding passage. This
interpretation would have to rest upon the argument that if 85 have
relatively strong recoding skills they will not be hindered in reading
comprehension by the presence of complex sentences in the passage.
Furthermore, it would need to be argued that 35 having relatively weak
recoding skills would be hindered in reading comprehension by the
presence of complex sdntences in the passage. This notion was tested
by examining the mean.RM and RNM scores for the $3 from groups A and
B who scored in the upper and lower 20% on recoding. It was hypothe-
sized that a greater difference in the mean scores between groups A
and B would occur for those 35 who were low in recoding than for those
88 who were high in recoding. The means and standard deviations of the
RM and RNM scores for the S3 who scored in the upper and lower 20%
(approximately) on recoding are presented in Table 25. The differences
'between the means for the lower 20% are not much larger than the cor-
responding differences for the upper 20%.

TABLE 25

The Means and Standard Deviations of the RM and RNM.Scores
for the 35 Who Scored in the Upper and Lower 20% on Receding (RE)

 

 

 

 

 

Lower 20% Upper 20%
RM RNM RM RNM
Groum Z §.DL i §.D. i S.D. 3(- 8.1).
.A 12.56 2.90 9.38 n.13 19.19 5.17 16.56 2.57
.B 1h.06 5.2a 10.96 0.67 20.93 2.19 16.19 b.88

 

 

 

 

 

104

None of the differences were significant and so these data offer no
support for the hypothesis.

A variable that has not been considered so far in this discussion
of the relative difficulty of the two versions of the reading passages
is reading time. It might be argued that given enough time, an average
or above average reader can correctly recode the sentences in the com-
plex version. Once the sentences have been recoded they can then be
stored in memory as easily as simple sentences which need little or
no recoding. Thus, if given unlimited time to read the passages, the
85 could comprehend and recall the major ideas of the complex passage
equally as well as they could those of the simple passage.

It might be expected that if verbatim recall were the response
asked for from the 83, that those given the complex version would re-
call less. Perhaps if that were the experimental criterion task, the
TR subtest would be a good predictor. This kind of task may be inter-
esting to study in future research, but it does not have the practical
significance that is dervied from a study which examines the variables
affecting "understanding" of a prose passage. Verbatim recall is far
less often an educational objective than is comprehension of major
ideas.

It seems that further research on the effect of reading time on
different levels of difficulty (judged by sentence complexity) would
be worthwhile. A study which would pace readers through the use of
mechanical aids and inveStigate the effect of increased reading speed
on passages containing various levels of sentence complexity would be

in order.

105

I ca s Su

Some of the implications to be discussed in this section have
already been touched upon in the above discussion but will be included
here so that a unified description of further research can be presented.
The practical applications of the findings of this study are wide and
varied. These results indicate that sentence skills may be very im-
portant in reading at the high school level. It may be useful for the
remedial reading teacher to try to discover any weaknesses her students
may have in sentence skills and drill them on these skills along with
vocabulary. Or, perhaps an approach to remedial reading which might
be useful would be one in which students were given reading passages
which at first included only simple sentences and then gradually intro-
duced complex ones. This procedure might be important for beginning
reading also. Vocabulary is introduced slowly and at a carefully con-
trolled pace in the elementary readers. Perhaps sentence complexity
should also be as slowly and as carefully introduced in these readers.
Because both vocabulary and recoding seem to be important variables in
reading comprehension, it might be educationally sound to introduce new
vocabulary and new sentence structures at different times so that the
8 doesn't have both to handle at once. This notion also suggests that
learning of subject matter that has a great deal of new terminology
(e.g. biology) might be facilitated if the sentences are kept simple
so that the student does not struggle with new terminology and complex
sentence structure both at once.

Although the above applications of the findings of this study to

teaching help to point out the importance of this kind of research to

.....-

106

education, none of them should be implemented until further research can
be done. ‘With the goal in mind of eventual application to the teaching
of reading in particular and prose learning in general, the following

outline is presented as a suggested program of research for the further

study of recoding as an individual difference variable in prose learning.

I.

II.

III.

IV.

VII.

VIII.

XI.

XII.

Replication of the present study using a variety of IQ and
reading comprehension tests.

Rewriting the sentence skill subtests in an effort to increase
their reliability and in an effort to make them useful for
testing at lower age levels.

Investigation of the interrelations among the recoding tests,
IQ, and reading comprehension tests at various age levels.

Development of a taxonomy for classifying the many levels of
sentence complexity.

Investigation of the relative reportance of the various kinds
of sentence cues which help a S to understand sentences of
various levels of complexity.

DevelOpment of tests which result in "pure" measures of the
S's ability to utilize each of the four kinds of sentence cues.

Investigation of the importance of these sentence cues in
listening comprehension. Note that other cues (e.g. intona-
tion, facial and body gestures, stress) are also important in
listening comprehension and the relationship among these cues
and the sentence cues must be studied.

An investigation of the memory, non-memory dimension of read-
ing comprehension and the interaction of this variable with
the various predictors of reading.

The determination of the relationship between recoding (RE)
and other language tests (e.g. the ITPA).

An investigation of the effect of the interaction of reading
rate and level of sentence complexity in prose learning.

An investigation of the effect of the interaction of level of
sentence complexity and level of vocabulary in prose learning.

A survey of the level of sentence complexity found in elemen-
tary readers.

107

XIII. An investigation of the effect of introducing sentence com-
plexity at differing rates in elementary reading material.

CHAPTER VI - SUMMARY

The major purpose of this study was to utilize the transformational
model of sentence structure to investigate individual differences in
the skills of recoding and to examine the importance of these individual
difference variables in prose learning. Recoding was defined in accord
with Miller's (1962) definition: "a process of utilizing transforma-
tional rules to 'break-down' each sentence into its kernel elements".
According to Miller, the process he calls recoding involves the assign-
ment of a constituent structure to each sentence so that it can be
stored in kernel form. Evidence for the psychological reality of re-
coding was discussed and then an attempt was made to identify the
specific psychological processes which enable a person to "break-down"
a sentence, assigning a constituent structure to it. It was postulated
that there are certain cues which a person gets from the sentence which
help him make the appropriate associations among the various sentence
parts. The first kind of cue was entitled "position-cue". The posi-
tion of a unit within an English sentence or phrase gives a hint as to
its function as well as to its meaning in that particular sentence. A
second kind of cue postulated as useful in understanding which elements
of a sentence function in which capacity was "word-form" cue. The form
of the word (especially the endings) is an indication of how it is
functioning in a sentence. "Semantic'cue" was a fourth one postulated
as important in sentence understanding. This cue is essentially a
referent or meaning cue. The fifth one was termed "transformation cue."

It was postulated that the 8 must recognize which transformtion(s)

108

109

have been made on a sentence so that he can correctly identify the
various sentence parts on the basis of position and word-form cues.
The relative positions of the basic elements in a sentence (subject,
verb, and object) changes depending upon which transformations have
been made. ‘Word-form changes also occur as different transformations
are made. Using these four sentence cues as guides, a person must
decide which element is the subject and which the verb: which the
modifier and which the modified. Next, he must decide which functional
units are to be associated together. Finally he can then "break-down"
the sentence into its major groups of functional units (kernel sen-
tences).

The five aspects of sentence structure were identified which can
in some way make it difficult to fully utilize the sentence cues dis-
cussed above. These five aSpects of sentences (grammaticality, sen-
tence depth, semantic-syntactic conflict, structural ambiguity, and
transformational complexity) were used as the basis for the construc-
tion of five sentence skill tests. These tests were to be used as
measures of S's recoding skill and were to be correlated with measures
of reading comprehension.

Because memory was considered to be an important variable in prose
learning, two reading comprehension tests were developed to be used as
criterion measures in this study. The first test called for the 33 to
read a passage and answer a number of multiple choice questions without
looking back at the passage. (This was termed a memory test.) The
second reading comprehension test encouraged the 83 to look back at

the passage when answering questions about it. (This was called a

110

non-'memory test . )

Because the sentence skill tests proportedly measured a 3'5
ability to comprehend sentences of various kinds and various levels
of difficulty, it was decided to manipulate the level of difficulty
of the sentences within the prose passage. Accordingly, two forms of
each reading test were constructed. In one form of the tests the
prose passages were written in highly complex sentences and in the
other form the passages were written in simple, active declarative
sentences. It was postulated that recoding would be a more important
variable in the comprehension of a passage containing complex sentences.

The first major objective of this study was the design and develop-
ment of sentence skill tests which would reliably measure individual
differences in sentence skills that have been postulated to be important
in recoding. A second major objective was to determine the importance
of these tests in the prediction of reading comprehension. It was
hypothesized that these tests would predict reading comprehension
better if the passage used in the comprehension test contained complex,
difficult to recode, sentences than if the passage contained mostly
simple, active declarative sentences. It was further hypothesized
that a comprehension test which contained complex sentences would be
more difficult than a similar test containing simple sentences. Another
objective of this study was to determine whether or not these sentence
skill tests would add any precision to the prediction of reading com-
prehension over and above that already available through the use of IQ
and vocabulary as predictor variables. It was also an objective of

this study to determine whether the recoding subtests were measuring

111

the same general factor or several unique factors. Finally, it was an
objective of this study to determine the relationships which exist among
the sentence skill tests and the following variables: IQ, vocabulary,
reading rate, digit Span, and reading comprehension.

In order to accomplish the above objectives, 154 high school 35
were administered the following tests: IQ (Otis Quick scoring form EM,
Gamma series), vocabulary (V0), Digit Span (D8), reading rate (RR),
grammaticality (GR), self-embeddedness (SE), Semantic-Syntactic Conflict
(SS), Ambiguity (AM), Transformational complexity (TR), reading compre-
hension-memory (RM), and reading comprehension (RNM). All of these
tests except the IQ test were constructed by the author of this study.
The lSh'Ss were randomly assigned to one or the other of the two forms
of each reading comprehension test and all the other tests were exactly
the same for all 154 35. The complete battery of tests took approxi-
mately two hours to administer.

After the data were collected several statistical analyses were
made.

Means and standard deviations were computed for all variables and
the intercorrelation matrix of all variables was obtained. In an
attempt to better understand the nature of the skills being measured
‘by the various tests, factor analyses were performed on the major pre-
dictor measures and the two reading comprehension tests. Factor analyses
‘were also performed on the five sentence skill tests in order to deter-
xnine the factor structure of these tests.

In order to determine which of the several independent variables

.served as the best predictors of reading comprehension, all the

112

independent variables were entered into least squares delete and least
squares add multiple regression programs.

Once the major predictor variables were identified by using the

least squares add and delete programs, a more precise method of deter-
mining the amount of variance accounted for by a given independent
variable or group of independent variables was used. This was a method
suggested by Bottenberg and Ward (1963) and is referred to as the least
squares ordered delete program. The research strategy utilized in this
procedure involved determining the multiple regression equation which
contains the major predictor variables for a given criterion measure.
This equation was referred to as the full mode]. (FM). Next, a number
of restricted models (RZs) were constructed. Each restricted equation
‘was constructed by deleting a given variable or group of variables.
'This was accomplished by setting the Beta weights for the variable(s)
of interest at zero. Various hypotheses were then tested by comparing
'the R2 (multiple R squared) between the full model and any given re-
stricted model. Two basic full models were constructed. The first con-
tained the independent variables of IQ, vocabulary, and a total recoding
score. The second contained the independent variables of IQ, vocabulary
grammaticality, self-embeddedness, semantic-syntactic conflict, ambiguity
arxi transformational complexity. Each of these full models were used
for groups A and B, when reading comprehension involved a memory factor
arui when reading comprehension did not involve memory. Several re-
stricted models were constructed and were compared to these full models.
The major findings of this study are summarized below:

1) Reliable tests which measure 85' ability to understand and

2)

3)

it)

5)

6)

7)

8)

9)

10)

11)

113

recall English sentences can be constructed.

Individual differences in the sentence skills measured by these
tests do exist.

These sentence skill tests correlate with vocabulary, IQ, and
reading comprehension.

Factor analyses of the sentence skill tests yielded two factors,
the first loaded on heavily by GR, SE, SS, and AM and the
second loaded on by TR almost exclusively.

Vocabulary (V0) was the single most important predictor of
reading comprehension of all the predictors used in this study.
The next most important predictor was Semantic-Syntactic Con-
flict (SS).

IQ contributed less unique variance in reading comprehension
than either vocabulary or recoding.

Recoding added to the variance in reading comprehension above
that already accounted for by IQ and vocabulary both when a
reading test involving memory was the criterion and when one
not involving memory was used.

Recoding was a more important contributor to variance in read-
ing comprehension (not involving memory) when the reading
passage contained complex sentences than when it contained
simple ones.

When memory was a factor in the reading comprehension test (RM),
recoding was important in predicting both the complex and the
simple versions.

The sentence skill tests each predicted different types of

12)

114

reading tests better than some other type; but when all five
tests were deleted from the regression equation together, they
were demonstrated to be important to the prediction of both
versions of each of the reading tests used.

No difference was found in the mean reading comprehension test

scores between the simple and the complex versions.

BIBLIOGRAPHY

Artley, S. A., "General and Specific Factors in Reading Comprehension,"

WW 19%. 1.6.. 181-188-

Blumenthal, A. L., "Observations with Self-Embedded Sentences,
W. 1966. 6 (10). 453-451».

Bormuth, J. R., "Mean Word Depth as a predictor of Comprehension

difficulty, " aliflgzgia Jggmal 91 Egngaflgna], Reseazgh, 19614,
15 (5) 226-231.

Bottenberg, R. A. and Ward Jr., J. B., Applied Mgtiple Linear Reggession,

Technical Documentary Report PRL-TDR-63-6, U. S. Department of
Commerce, Office of Technical Services.

Carlson, T. R., "The Relationship between Speed and Accuracy of Com-

prehension," 9mm]. of Educatignal Research, 1949, 32,, 500-512.

Carroll, J. B., "On Learning from Being Told," Edugtional Psychologist,
1968.

Carroll, L., "Jaberwock'y," in Main C. F., Seng P. F. eds., Poems, San
Francisco: Wadsworth Publishing Company, Inc., 1961.

Chomsky, N. Smggtic Staggtures, The Hague: Mouton and Company, 1957.
Chomsky, N., "Language and the Mind," Psyghglggy Today, December, 1967a.

Chomsky, N., in Appendix - Lenneberg, E. H. Biflggical Foundagggs gf
La a e, New York: John Wiley and Sons Inc., 1967b.

Chomsky, N ., Aspects 91: a Thegm 91: SEER. Cambridge, Massachusetts:
The M.I.T. Press, 1965.

Chomsky, N. and Miller, G. A., "Introduction to the formal analysis of
natural languages." In Luce, R. D., Bush, R. and Galanter, E.,

eds. , ngdbggk gf Majhgmatigal Esychglogy, New York: Wiley, 1963,
2,, 269‘322.

Clifton, C. and Odom, "Similarity Relations among Certain English Sen-

tence Constructions," Ps h ca Mo a hs, 1966, 59 (5).
Whole No. 613.

Coleman, E. B., "The Learning of Prose Written in Four Grammatical

Transformations," Jggmal 9f Applied Psychology, 1965, 54:9, 332-341.

Epstein, W., "The Influence of Syntactical Structure on Learning,"
WM. 1961. .73. 80-85.

115

116

Epstein, N., "A Further Study of the Influence of Syntactical Structure

on Learning," American Jguznal 9f Psychology, 1962, 25, 121-126.

Epstein, W., "Some Conditions of the Influence of Syntactical Structure
on Learning: Grammatical Transformation, Learning Instructions,

and 'Chunking,'" Jggynal 9f Verbal Leayning and Verbal thavioy,
1967. .6_. ins-419.

Enrich, A. C., "The Relation of Speed of Reading to Comprehension,"
W. 1930. 32. 402-406.

Fodor, J. A. and Bever, T. G., "The Psychological Reality of Linguistic
Segments," Journal of Verbal Learning and Verbal Behavior, 1965, &,

414-420.

Gough, P. B., "Grammatical Transformations and Speed of Understanding,"
Journal of Verbal Learning and Verbal Behavior, 1965, &, 107-111.

Gough, P. B., "The Verification of Sentences: the Effects of Delay of
Evidence and Sentence Length," a V a n a Ve -

W. 1966. i (5). 492-496.
Johnson, M. G., "Syntactic Position and Rated Meaning," Jogynal of

Veybal Learning agd Verbal Behayigy, 1967, 6, 240-296.

Johnson, N. F., "The Psychological Reality of Phrase-Structure Rules,"

Jggynal 92 Verbal Leaynigg and Vegbal Behavior, 1965, 3, 469-“75.
IKrumboltz, J. D., Learning and the Educational Prgcess, Chicago: Rand

McNally, 1965.

ILenneberg,.E. H., Biological Foundatigns of Langgage, New York: John

Wiley & Sons, Inc., 1967.

LLindquist, E. F., and Hieronymus,.A. N., Igya Tests of Basic Skills,
Boston: Houghton Mifflin Company, 1955.

bflacKay, D. G., "To End.Ambiguous Sentences," Peyceptign and Psychophysics,
1966. 1 (12), 426-436.

jMarks, L. E. and Miller, G. A., "The Role of Semantic and Syntactic
Constraints in the Memorization of English Sentences," Journal of

Verbal Learning and Veybal Behayigr, 1964,‘3, 1'5.
Martin, C. J ., and Alonzo, L., Comprehension of Full Length and Tele-

a a a B i C n, U. S. Department of Health,
Education and welfare, 196?.

Martin, E. B., and Roberts, K., "Grammatical Factors in Sentence Reten-

tion," Jggrga; of Vezba; Leayning and Veybal Behavior, 1966, 5,

211-218.

117

Mahler, J., "Some Effects of Grammatical Transformations on the Recall

of English Sentences," ngal 9f Verbal Learning and Verbal
M0 1963' 2.! 3146-351-

Mehler, J. and Carey, P., "Role of Surface and Base Structure in the
Perception of Sentences," J 0333a]. of Vega} Learning_and Verbal
m. 1967.6 _. 335- 338

Miller, G. A., "Some Psychological Studies of Grannnar," American

W. 1962. 12. 7"8'752-

Miller, G. A., Heise, G. A., and Lichten, W., "The Intelligibility of
Speech as a Function of the Context of the Test Materials," J__q_____urna1

W 1951. .1. 329935.

Miller, G. A. and Isard, 3., "Some Perceptual Consequences of Linguistic

Rules," a o Ve ba Lear n a d Ve ba Behavio , 1963, 2,,
217'228.

Miller, G. A. and Selfridge, J. A., "Verbal Context and the Recall of

Meaningful Material," Ameyigan Jgurnal of Psychology, 1950, 63,
176.185.

Osgood, C. E. and Sebeok, T. A., PsycholingyListics, Bloomington:
Indiana University Press, 1965.

Rinne III, C. R., "Improvement in Reading Comprehension through Increased
Awareness of .ritten Syntactic Patterns," 1967 Doctoral Disserta-
tion Stanford University, Dissertation Abstracts, 1968.

Robeck, M. C., "Intellectual Strengths and Weaknesses Shown by Reading
Clinic Subjects on the WISC," Journal of Developmental Reading,
1964, 2, 120-129.

Roberts, P. , English Sentences, New York Harcourt Brace and World, Inc. ,
1965.

Rothkopf, E. 2., "Some Theoretical and Experimental Approaches to
Problems in Written Instruction," in Krumboltz, J. D. ed., Learning

and the Egugaugna], Eygcess, Chicago: Rand McNally and Co., 1965.

Savin, H. B. , and Perchonock, E., "Grammatical Structure and the Im-

mediate Recall of English Sentences," J ou_rgg], of Verbal Learning
a d be B havior, 1965,11, 3118-353.

Slobin, D. I. , "Grammatical Transformations and Sentence Comprehension
in Childhood and Adulthood," Journal of Verbal Learning and Verbal

thavior, 1965, 5, 219-227.

118

Stolz, W. 3., "A Study of the Ability to Decode Grammatically Novel

Sentences," ngyng; gf Verbal Learning and Verbal Behavior, 1967, 6,
867-8 73 .

 

Strang, R., "Problems in the Improvement of Reading in High School and
Colleges," Lancaster, Pennsylvania, The Science Press Printing
Company, 1940 .

Strang, R., "Relationships Between Certain Aspects of Intelligence and

Certain Aspects of Reading," Educational and Psychglogical Measgye-
916111. 1943. 3. 355-359.

Tinker, M. A., "The Relation of Speed to Comprehension in Reading,"
Schggl and Socigyy, 1932, 36, 158-160.

‘Wheeler, L. R. and Wheeler, V. D., "The Relationship Between Reading
Ability and Intelligence Among University Freshman," Journal of

Egygatiggal Psychglggy, 1949, 99, 230-238.

‘Yngve, V. H., "A Model and an Hypothesis for Language Structure," Pro-
ceedings of the.American Philosophical Society, 1960, 19$, (5),
14144459.

Zipf, G., The Psygho-biology of Language, Boston: Houghton-Mifflin,
1935.

AHEMHXOL.

Mean item difficulty and mean item discrimination indices for the
sentence skill tests, the reading comprehension tests, and the vocabu-

lary test for groups A and B:

 

 

 

 

 

 

 

 

Mean item dif iculty index; Mean iteg_gig t de

‘ngt: Group A Group B Group A Group B
v0 38 as 53 3 50

GR g 33 1 32 36 g 36

SE 3 28 5 32 as 3 A2

53 E 29 32 i 50 54

AM :28 1+6 ; 3L. as

m. ; so ' no 1‘ 1+3 33

w 33‘ >9 ; 37 } 37

RN ? )3 29 g 51 3 50

 

119

1.13a- |.I.WILI|| .—
h- i ... u

APPENDIX 'IWO

Va a Tes

120

VO CAB ULARY TES T

Directions: Look at the first word in each line and then find

another word in the same line which means the same or nearly the

same as the first word. Note the letter of this word and fill in

that letter on your answer sheet.

Examples:

1. automobile A. fish B. car C. wagon D. blue E. engine
2. gay A. sad B. unusual C. happy D. yellow E. fine
3. pepper A. ball B. spice C. color C. house E. fun

121

A9.
50.
51.

52.
53-
54.

55.

560

59.

60.

61.
62.
63.

6h.
65.
66.

67.

68.

boat A. animal
feline A. female
gra- A. happy
cious
play A. solo
create A. food
redun- A. re-
dant leased
con- A. separ-
verge ate
bland A. edge
condone A. prepare
expe- A. get rid
dite of
enjoin A. take
part
elon- A. stretch
gate
ensue A. doctor
peer A. equal
eradi- A. build
cate up
ramble A. car
bouyant.A. boyish
itiner- A. hOSpi-
ary tal
infam- A. bad
ous
enigma A. design

122

ordinary C.
. dog C.
sad C.
pretend C.
. enlarge C.
reasoned C.
try C.
tool C.
put down C
. put out C.
require C.
shrink C
follow C.
look C.
knock C.
down
gaze C.
silly C.
menu C.
. good C.
puzzle C.

ship
tiger

kind

children

consume

re-
peated

religion

taste-
less

. go along

with

help out

add to

. magnify

law
occur

get rid
of

store
float

factory

burned

disease

. sea

poetry

playful

game
monster

redone

unite

blonde

sorrow

turn
down

seek out

. control

Chinese

. catch

let off

. talk

fun

program

. praised

charm

F3

.fmut

. pretty

helpful

. T.V.

design

retold

tolerate

sense-
less

. consider

play Up

. strike

deserve

force
dock

let
loose

wander
large

contents

exploited

book

69.

70.

72.

73.

74.

75-
76.

77.

78.

79.

80.

mon- A.
strous
quake A.
.nmfi- A.
nal
tact A.
exag- A.
gerate
kindle A.
yield .A.
laby- A.
rinth
uncan- A.
ny
omi- A.
nous
dex- A.
trous
volumi-A.

nous

huge 8.
fear B.
small B.
sharp- B.
ness

push in B.
wood B.
protect B.
labora- B.
tory

dog B.
inclu- B.
sive

soapy B.
large B.

horrid

doctor
elected
slow-
ness

knock
down

bizarre

process

bassi-
net

dumb

fulfil-
ling

sugar

loud

123

C.

C.

C.

C.

C.

C.

moist D.
queer D.
money D.
aggres- D.
siveness
blow up D.
fire D.
give D.
maze D.
cat D.
threat- D.
ening
agile D.
high D.

certain E. stormy
fake E. shake
pointed E. living
polite- E. stickiness
ness

pin- E. break up
point

point E. childlike
out

gain E. bake

cork E. experiment
clever E . open
vehicle E. electrical
timely E. catastrophic
awful E. occasional

...:

lsLnLable

l-‘H
HO

H
N

\0 CD \1 0\ U1 {1' U) N
e o a e e e e e

looLoloobleolm

loxleLn
LALDIN

leb'oxlookloloo

EL»

'0 LA
I: IN

loo

looloolooLuqu»

loxlolelHloxL-‘Lvlw

ble

I:

IOLAINLILnlf-‘IOIH

lmL»

L2
IH

'cLo Ln
loo

IV'VLnluLoLn

APPENDIX THREE

DIGIT SPA N TEST

LobL)
klwls

lm I.»

l~|oolele
LoLnIH
|+~|~Lo|~

121+

APPENDIX FOUR

W

125

PART ONE

This is a test designed to see how aware you are of the way
English sentences are constructed. You will read several "sentences"
which will have little or no meaning because most of the words will
be nonsense words. However, some of these "sentences" will have
the structure of English sentences (i.e., they will be grammatically
correct) and will sound like complete English sentences. The rest
of the sentences will 2 L have the structure of English (i.e., they
will be grammatically incorrect) and will not sound like complete
English sentences.

Your task will be to read each sentence and'decide whether or
not it sounds like a complete English sentence. If it does sound
like English, you will fill in the Space on your answer sheet marked
"T" (for "True - this sounds like English."). If it doesn't sound
like English, you will fill in the space on your answer sheet marked

"F" (for "False - this does not sound like English.").

T F

Examples: A. Faps are vumpy. II I l
B. Zips Gabble Gimple Saber dip is. j I II

C. Some geeps. I f -

DO NOT WRITE IN THIS BOOKLET

11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.

25.

127
mmpzmpmimHMe.
When the klappy, the snordle blumped along.
harper in dOOple did.
Sally picked up it and nordly zooped it away.
Antic noomers was very curzy.

A garply zumap vluped.

His flizziest gormuth a very Slippy swinker on the toodle.

The goozey fladders was vaping.

Nany fleepers within a scapper and poofing the girge.
As twardle footed, numer zallied.

These zord am in his orkle then.

The crooty zorgle, that might have zumped away.

Some vouples when the korps gramed.

These forpals corply veeped many zoogs in foone.
Some famlier gleep frambled.

Zurpy gazzle tamps were sloopy.

Few plams around a toodle so klappy bapped anyway.
My soodly, when gorples cam and garbled me.

When kimp frambled, sooly vrampt.

A glapiest bloup are Spoofing.

The vroomer in a clap.

Even though zorgle, parsy vorpled glapily.

Those skeezy zams, some plooky doomy carp in tipple.
Two zweep which were galumphing away.

A warplier burdle are scorping.

GO ON TO THE NEXT P

A '1"

rt-Jib "'"

 

1...... [U

128

26. The durly zamer, that turkled daply.
27. Their fladders when the turldes vapped.

28. The zampy diply a frumed glip.

STOP

DO NOT TURN THE PAGE UJTIL TOLD TO DO SO

APPENDIX FIVE

SelfitEmbegded Sentence Test

129

PART TWO

This is a test designed to see how well you can understand
sentences which have phrases embedded in them. Your task will be
to read a sentence and then reapond to a number of statements which
are true or not true according to the initial sentence. For each
numbered statement you must decide whether or not it is true accord-
ing to the sentence above it. If it is true, mark "T": if it is

not true, mark "F" (false).

Examples:

The child that the mother spanked cried loudly.

A. The child cried. I | l

B. The mother cried. I I I.

DO NOT WRITE IN THIS BOOKLET

130

The birds that the fence that the men put up held sang a song.

29.
300
31.

the birds sang a song
the men held the fence

the

men sang a song

The car that the mechanic that the garage hired raced ran into the

stands.

32.
33.
34.
35.

the
the
the
the

The girl that

36.
37.
38.
39.

the
the
the
the

car ran
mechanic hired
mechanic ran
car raced

the boy that the students liked teased loved to read.

boy liked the girl
students teased the girl
boy loved to read
students liked the boy

The elephant that the tiger that the hunter scared saw ran.

“‘00
41.
42.
43.
4h.
45.

the
the
the
the
the
the

The deer that

forest.

#6.
’47.
’48.
49.

the
the
the
the

tiger ran

hunter saw the tiger
hunter scared the elephant
hunter scared the tiger
elephant saw the tiger
tiger scared the hunter

the man that the bear feared avoided lived in the

deer feared the man

deer lived in the forest
man avoided the bear
bear feared the man

60 ON TO THE NEXT PAGE --

132

The votes that the people that the candidates represented cast
indicated a choice.

50. the people indicated a choice

51. the candidates represented the votes
52. the peeple cast the votes

53. the candidates indicated a choice

The cop that the thief that the judge that the people hired sentenced
shot changed his ways.

54. the cop changed his ways
55. the people hired the cOp
56. the judge sentenced the people

STOP

DO NOT TURN THE PAGE UNTIL TOLD TO DO SO

3

APPENDIX SIX

-S c C

133

PART THREE

This is a test to measure how well you can understand sentences

in which the structure and the word meanings do not agree very well.

(e.g., The balloons unified some glorious water fountains.)

At the top of each page you will read a number of sentences

which are grammatically correct but which may have unusual meanings.

Below these sentences you.will find a number of multiple choice items.

In each case you will be asked to choose from among three choices,

the phrase which is either ggzrggt or ingozzect according to the

sentences at the top.

Examples: People fondled funny boats.
Zoos caught total fits.

1 2 3
Choose the correct phrase I I - I l
1. boats fondled people
2. people fondled boats
3. people are funny
Choose the correct phrase l 2 3
1. boats are funny ' 1 1 i I
2. people fondled funny
3. boats fondled funny
Choose the inggzzggt_phrase 1 2 3
l. zoos caught fits l l I 1 l

2. fits were caught
3. zoos caught total

DO NOT WRITE IN THIS BOOKLET

134

135

Spilled chaos saves baseball love.
The bottled young rhythm drank grape wonders.

Hunters simplify motorists across the soup.

1. Choose the correct phrase
1. chaos saves baseball
2. chaos was saved
3. love was saved

2. Choose the incorrect phrase
I. chaos saves love
2. chaos was saved
3. chaos was spilled

3. Choose the correct phrase
I. the young drank wonders
2. the rhythm drank grape
3. the rhythm was bottled

4. Choose the incgrzect phrase
1. the rhythm drank wonders

2. the rhythm drank grape
3. the rhythm was young

5. Choose the correct phrase
I. hunters simplify the soup
2. motorists simplify the soup
3. the motorists are being simplified

GO ON TO THE NEXT PAG --

136
Loud jungles use sudden trout.
Jellied freedom creates bubblegum joy.
The trash was built by a daring fortune fighter.

6. Choose the correct phrase
I. trout use jungles
2. trout are sudden
3. jungles use sudden

7. Choose the inggrrect phrase
1. jungles use trout
2. jungles are loud
3. trout are loud

8. Choose the correct phrase
1. freedom creates joy
2. freedom creates bubblegum
3. jelly creates bubblegum

9. Choose the incgrrgct phrase
1. freedom creates bubblegum

2. joy is created
3. freedom was jellied

10. Choose the correct phrase
1. a fortune was built
2. the trash was daring
3. a fighter was daring

GO ON TO THE NEXT PAGE --

137
Rusted laughter pardons banana hate.
The lowly thirsty aroma sang players.
The cleverly colored snow needed cancer.

A witness appraised the shocking company dragon.

11. Choose the correct phrase
1. laughter pardons hate
2. hate is pardoned
3. banana pardons hate

12. Choose the incgrrect phrase
l. laughter is rusted
2. hate is pardoned
3. banana is hated

13. Choose the correct phrase
1. the thirsty sang
2. the aroma sang
3. the players sang

14. Choose the inggrregt phrase
1. the aroma was lowly
2. the lowly were players
3. the aroma was thirsty

15. Choose the inggrrggr phrase

1. snow is cleverly
2. snow is colored
3. cancer is needed

16. Choose the correct phrase
1. witness is shocking
2. dragon is shocking
3. witness appraised the company

GO ON TO THE NEXT PAGE ~-

1.11,...“ 0.3""

138

The slim fast gargled syrup pickles.
Tired popsicles enjoy chocolate time.
A firmly bubbled sheep spoke garbage.

Furious mustard charges city folly.

17. Choose the correct phrase
1. pickles were gargled
2. slim gargled
3. syrup was gargled

18. Choose the correct phrase
1. slim was fast
2. fast gargled pickles
3. slim_gargled fast

19. Choose the incorrect phrase
I. pOpsicles enjoy time
2. popsicles enjoy chocolate
3. pOpsicles are tired

20. Choose the incgrrect phrase
1. sheep are firmly
2. sheep are bubbled
3. garbage was spoken

21. Choose the correct phrase
1. city was furious
2. folly was charged

3. city was charged

22. Choose the incorrect phrase
l. mustard is furious
2. mustard charges folly
3. mustard charges city

30 ON TO THE NEXT PAG --

139

Bubbled rain catches minnow 'ce cream.
Intelligent bacon tossed slowly scorched dreams.
Lany catsup toys surround double oranges.

.ine watered freedoms polished sugar pop.

dlurpy fab es carried candy elephant stores.

23. Choose the incorrect phrase
1. rain was bubbled
2. rain catches minnow
3. ice cream is caught

24. Choose the correct phrase
1. bacon scorched dreams
2. dreams were tossed
3. bacon was scorched

25. Choose the incorrect phrase
1. bacon is intelligent
2. bacon tossed dreams slowly
3. dreams were scorched slowly

26. Choose the correct phrase
1. toys are catsup double
2. uwssmwmmd
3. oranges are surrounded

27. Choose the incorrect phrase
1. sugar was polished
2. freedoms were watered
3. pop was polished

28. Choose the incorrect phrase
l. fables carried candy
2. fables carried elephant stores
3 fables carried candy stores

‘.
I 0

STOP

DO NOT TURN THE PAGE UNTIL TOLD TO DO SO

APPENDIX SEVEN

Ambiguity Test

140

PART FOUR

This test is designed to measure your ability to recognize
when a sentence has more than one meaning. A good English sentence
has only one meaning and can only be interpreted in one way. Some
sentences are hard to understand because they have more than one
meaning. Your task will be to read a sentence and decide whether
or not it has only one meaning. If it does only have one meaning,
you will mark 1 for -- "it has only one meaning". If it has more
than one meaning, then you will mark 2 for -- "it has 2 or more
meanings”.

RELEMBER: "l" for -- it has only one meaning

"2" for -- it has 2 or more meanings

Examples: 1 2
1. I am going away toda‘ . I I A
2. He took the stone from the cloth II I.
and washed it.
3. The book in the car that is red. l I '
a. He bought the boxes on the train I] II
5. He swallowed a whale. I I 1

E0 NOT EARN IN THIS BOOKLET

141

1:6 .
“r7.
1+8 .
49.
50.
51.

142

The requirements of the courses which are listed below
The peOple who saw the play frequently praised it
Hashing children can be interesting

The boy gave his parents a dollar to Spend

The plates fell to the floor and broke

Ship sails today

The fool taught Lear all he knew about the world
A boy on the slide who fell

The boy looked fast

Flying planes can be dangerous

The dog's nose smells

Bill had his yard finished

A criminal lawyer

he decided to vote

Call me a taxi

I had a book stolen

Alice was given John as a partner

I read the end of the story that was interesting
He decided on the boat

John found the boy studying in the library

She has her hair sprayed

Some models look better than others

The letter was weighed by the postmaster

GO ON TO THE

NEXT Pile --

fl «divine-:11

1N3

You follow that car

The song was sung by the band

I pushed the kid around the corner

The committee's appointment was announced

"‘1

choose me a winner

STOP

MO POT TUgfi THE 1A}; UNTIL TOLD

TO

JOSO

APPENDIX EIGHT

{Eransfgrmétional Cgmplefitz Tgst

141+

 

THAT FIVE.

This is a test designed to see how well you can recall sentences
exactly. You will read a number of statements and then you will
turn the page and without looking back, choose from among a number
of possible statements the one which is exactly like one of the
statements you had read before.
Example:
Hasn't John going away?

Sally played the game.

1 2 3 4
1. 1 Sally did play the game. A A I I l I '
2 Sally played the game.
3 Didn't Sally play the game?
4 Sally didn't play the game.
1 2 3 4
2. 1 Has John going away?
2 John was going away. ' I II II II
3 John wasn't going away.
4 None of the above

LC TOT LAfih IN THIS BOOLLET

1&5

114-6

The ball was hit by Pete.

Has the fish seen by the man?
Didn't the girl chase the dog?
The teacher didn't grade easy.

Are pictures painted by artists?

DO NOT TURN THE PAGE UNTIL TOLD TO DO SO

57.

53.

59.

61.

P‘bJNH Cle-J 4?me {2'me

PWNH

147

DO NOT TURN bACh

Did the girl chase the dog?
Didn't the dog chase the girl?
Didn't the girl chase the dog?
Hone of the above

The pictures are painted by artists.
Are pictures painted by artists?
Artists didn't paint the pictures.
.Aren't pictures painted by artists?

Pete did hit the ball.
Wasn't the ball hit by Pete?
The ball wasn't hit by Pete.
None of the above.

The teacher did grade easy.
Did the teacher grade easy?
Didn't the teacher grade easy?
The teacher didn't grade easy.

hasn't the fish seen by the man?
has the fish seen by the man?
The fish was seen by the man.
None of the above

DO NOT TURN THE PAGE UNTIL TOLD TO DO SO

1&8

The soldier didn't dig the trench.
Did a monkey climb the tree?
Baseball is enjoyed by many fans.
Weren't the toys broken by the dog?

Some doughnuts were not eaten by the boys.

DO NOT TIRES THE PASE Ul-ITIL TOLD TC DO SO

62.

63.

64.

CWNH {EWNI—J (2'me {TKONH

C'KDNH

149

00 NOT TURN RACK

Weren't the toys broken by the dog?
fiere the toys broken by the dog?
The dog had broken the toys.

None of the above.

Here some doughnuts eaten by the boys?
here not some doughnuts eaten by the boys?
Some doughnuts were eaten by the boys.
None of the above.

Didn't the soldier dig the trench?
Did the soldier dig the trench?
The soldier did dig the trench.
None of the above.

hany fans enjoyed baseball.

Baseball is enjoyed by many fans.
Baseball isn't enjoyed by many fans.
15 baseball enjoyed by many fans?

Didn't a monkey climb the tree?
The monkey didn't climb the tree.
Did a monkey climb the tree?

The monkey did climb the tree.

STOP

DO NOT TURN THE PASE UNTIL TOLD TO DO SO

150

here some crayons broken by the child?
Hadn't the machine hit the rock?

Did the cup fall on the saucer?

A grape was swallowed by the bird.

Hasn't the man fired by his boss?

DO NOT TURN THE PAGE UNTIL TOLD TO DO SO

67.

69.

70.

FWJtka 43k0h3k‘ €“flrvf4 {?\OBJF4

CWJEOFJ

151

DO NOT TUEh BACK

Had the machine hit the rock?
has the rock hit by the machine?
Hadn't the machine hit the rock?
None of the above.

Has the man fired by his boss?

‘wasn't the man fired by his boss?

The man was fired by his boss.
The man wasn't fired by his boss.

Some of the crayons were broken by the child.
Some of the crayons weren't broken by the child.
The child had broken some of the crayons.

none of the above.

The cup fell on the saucer.

The cup had fallen on the saucer.
Didn't the cup fall on the saucer?
Did the cup fall on the saucer?

The bird swallowed a grape.

A grape wasn't swallowed by the bird.
Jasn't a grape swallowed by the bird?
Hone of the above.

STOP

:0 hOT TURN THE PASS UNTIL TOLD TO DO SO

APPENDIX NINE
Reading Comprehension Test (Memory):

Forms A and B

152

In the year 1800 on the continent of Africa, two unfriendly
nations existed, Mambo and Yam. The Nile River separated them. Yam
was richer.

King Koko began to plan the defeat of Yam. Koko was a military
man; most of his subjects were expert warriors; none were merchants.
They liked war and often fought another tribe's battles for money
and glory.

However, Koko was getting old. If Koko conquered Yam, he would
be rich and control thousands of people; he could rest his body, live
in luxury, enjoy personal wealth and power; and be the strongest
ruler on the continent.

Koko felt the time for war against Yam was suitable because his
Mambo warriors disliked King Lester of Yam. Koko knew his warriors
would welcome the chance to defeat Lester and his followers because
of their dislike and jealousy of Lester plus their greed for the wealth
of Yam.

Disliking Koko's plans, however, and comparing Koko and Lester, a
small band of expert Mambo warriors wanting to serve the kind of king
that Lester was, decided to be traitors to Koko. As a result they
Secretly met with a small group of Yam merchants who wanted to he
trained as warriors and who agreed to pay them money and supply them
with weapons so they would defend Yam. Thus, a not so large band of
Strong, fierce, traitor I-Iambo warriors and a small group of weak, fat

merchants formed the Pro Lester Union.

153

 

 

154

Never thinking much about war because none were expert warriors
and being mostly merchants, Lester and his followers were busy - the
plentifUl goods of Yam being traded by them all over the continent.

Most important in the preparation of King Koko's army for war
was the worship of the Sun-God, Solmambo. The witch doctor looking
directly into the sun at high noon and chanting, asked all the Mambo
warriors to do the same killing any who fainted from the heat or
looked away from the sun. Fifty warriors were killed by him. To
give them courage, the Mambo warriors' faces were painted with goat's
blood.

Climbing into their war canoes and using their heavy spears as
paddles, the Rambo warriors, being ready to fight now, paddled down
the river toward Yam. Useful not only as paddles and being well-
balanced with wide iron tips which had very sharp edges which made
them useful as swords or as double-edged axes these Spears could be
used as clubs in close fighting due to their heaviness.

When Yam was approached by the war canoes, the alarm was shouted
by Lester's lookouts. In a religious ceremony in which drums were
beat, ram horns were blown, and rattles were shaken while a native
tribal dance was being done by Yam musicians, any evil Spirits who
‘might harm Lester during the battle were chased away. All the Lester
followers, consisting of all Yam merchants and other Yam subjects and
the Pro Lester Union prepared to meet the invaders.

Lester's camel cavalry, wielding heavy, difficult to use iron
tflvords, being led by King Lester into battle and being outnumbered by

f3;ve to one with his fat merchant warriors being no match for the tough

155

expert Mambo warriors, were losing the battle. Deciding that he
could only win if he could fight Koko and kill him in personal combat,
Lester charged at Koko with his sword but before he could strike a
spear was sent by Koko into Lester's chest with such force that it
came out through Lester's back and the battle had been won by Mambo.

King Koko, now the master of seven thousand slaves, two hundred
thousand cattle and sheep, and the business wealth of Yam, was proud
of his victory. Because his wealth had to be protected and the Yam
citizens had to be kept under control by Koko, hambo's original social
structure, with most of the Mambo citizens still being expert warriors,
remained the same. Yam's social structure, however, with all of the
Yam people and the traitor Mambo warriors now being slaves of King
Koko, changed.

Some of Lester's faithful subjects, after King Lester was killed
by King Koko, in order to prevent the traditional burning up of killed
enemy leaders that the Fambo warriors liked to perform, succeeded in
rescuing and secretly burying his dead body. This wasn't ever re-
vealed by them but they claimed, instead, that Lester had been taken
away to safety by some friendly spirits so he could recover and return
to free his people. This, being believed by a rather sizable number
of Yam slaves, and being no comfort to Koko who is beginning to wonder
about it himself even though he feels certain that Lester, whose dead
body has never been found, was killed, is known as the Lester Legend.

If he knew about the Freedom Group which is made up of some legend
believers and the traitor Rambo warriors who fought for Lester but

don't believe the legend, Koko wouldn't worry about the legend so much.

156

Secretly hiding weapons in the jungle and planning for the overthrow
of Koko, this Freedom Group is keeping the Lester Legend alive. From
time to time a skull is found by Koko in front of his but or a dead
Mambo warrior. Shipments of goods to other nations often never reach
their destinations. Often Koko's food is not without poison. Grouchy,
irritable, and suspicious, Koko's relation with his troops is becoming
unfriendly because he blames them for not keeping control of the
slaves. He sometimes thinks they are plotting against him with the
slaves. Perhaps for Koko, whose future is not promising, victory

was not so sweet after all.

In the year 1800 on the continent of Africa, two unfriendly
nations existed, Mambo and Yam. The Nile niver separated them.
Yam was richer.

hing Koko began to plan the defeat of Yam. Hoke was a mili-
tary man; most of his subjects were expert warriors; none were
merchants. They liked war and often fought another tribe's battles
for money and glory.

However, Koko was getting old. If Koko conquered Yam, he would
be rich and control thousands of people; he could rest his body, live
in luxury, enjoy personal wealth and power; and be the strongest
ruler on the continent.

koko felt the time for war against Yam was suitable because his
Rambo warriors disliked kin? Lester of Yam. Loko knew his warriors
would welcome the chance to defeat Lester and his followers because
of their dislike and jealousy of Lester plus their greed for the
wealth of Yam.

Iowever, a small band of expert hambo warriors disliked hoko's
plans. These hambo warriors compared hoko and Lester. 'hey decided
that Lester was the kind of king they wanted to serve. They would
be traitors to koko. As a result, this small band of traitor Mambo
warriors met secretly with a group of Yam merchants. The group of
Yam merchants was small and they agreed to pay the warriors money
and supply them with weapons. The warriors would defend Yam. The

small group of Yam merchants wanted to be trained as warriors. Thus,

157

 

158

the Pro Lester Union consisted of a small band of traitor Mambo
warriors and a small group of merchants. The warriors were strong
and fierce. The merchants were fat and weak.

Lester and his followers never thought much about war. They
were too busy trading the plentiful goods of Yam all over the con-
tinent. Hone of Lester's followers were expert warriors and most of
them were merchants.

King hoko prepared his army for war. The most important event
was the worship of Solmambo. Solmambo was the sun-god. The witch
doctor looked directly into the sun at high noon. He chanted. All
the Lambo warriors did this also and any warriors who fainted from
the heat or looked away from the sun were killed. The witch doctor
executed fifty warriors that day. The Rambo warriors painted their
faces with goats' blood to give them courage.

how the Fambo warriors were ready to fight. They climbed into
their war canoes and paddled down the river toward Yam. Their
paddles were also their spears. These spears were heavy and well-
balanced. They could be used as clubs in close fighting because they
were heavy. They could also be used as swords or double-edged axes
because they had wide iron tips. The iron tips had sharp edges.

The hambo war canoes approached Yam. Lester's lookouts shouted
the alarm. Yam musicians beat drums, blew ram horns, and shook
rattles while doing a native tribal dance. This religious ceremony
was to chase away any evil spirits who might harm Lester during the
battle. All the Lester followers prepared to meet the invaders.

The Lester followers consisted of all Yam merchants and other Yam

159
subjects as well as the Pro Lester Union.

hing Lester led his camel cavalry into the battle. However,
Lester and his forces were losing the battle. The Yam warriors were
no match for the mambo warriors, Lester's army was outnumbered by
five to one, and Lester's cavalry wielded heavy iron swords which
were difficult to use. Lester decided he could only win if he could
fight Koko in personal combat and kill him. Lester charged at Koko
with his sword. Before Lester could strike, hoko sent his spear in
Lester's chest. He threw it with such force that it came out through
Lester's back. Rambo had won the battle.

King Koko was proud of his victory. Now he was the master of
seven thousand slaves, two hundred thousand cattle and sheep, and
the business wealth of Yam. The social structure of Mambo remained
the same. host of the mambo citizens were still expert warriors.
Koko had to control the Tam citizens and protect his wealth. The
social structure of Yam changed, however. All of the Yam people and
all the traitor Rambo warriors were new slaves of king Koko.

After king hoko killed hing Lester, some of Lester's faithful
subjects succeeded in rescuing his dead body. They secretly buried
it to prevent the hambo warriors from burning it. Traditionally,
Mambo warriors like to burn up enemy leaders that have been killed.
The Yam subjects never revealed their deed. They claimed that some
friendly spirits had taken Lester away to safety. They said Lester
would recover and return to free his people. This is known as the
Lester Legend. Loko was disturbed because a large number of Yam

slaves believe this legend. He is even beginning to wonder about it

 

160

himself. He feels certain that he killed Lester but Lester's dead
body has never been found.

but if hoko knew about the Freedom Group, he wouldn't worry
about the legend so much. This jroup is made up of some Legend
delievers and the traitor hambo warriors who fought for Lester but
don't believe the legend. The Freedom Group is planning for the over-
throw of Koko. They are secretly hiding weapons in the jungle keeping
the Lester Legend alive. From time to time Koko finds a skull or a
dead mambo warrior in front of his hut. Shipments of goods to other
nations often never reach their destinations. Koko's food often con-
tains poison. Koko is grouchy, irritable, and suspicious. His rela-
tions with his troops is becoming unfriendly. He blames them for not
keeping control of the slaves. He sometimes thinks his troops along
with the slaves, are plotting against him. The future for Koko is

not promising. Perhaps his victory was not so sweet after all.

MULTIPLE CHOICE

CHOOSE THE ITEM THAT 358T COfiPLzTES Tdfi STAThAENT OR UiST ANSHhKS TEE
QUESTION. MOdh A3 QUICKLY AS YOU CA3, BUT NOT SO QUICKLY THAT ICU
DARE UNNECESSARY hISTAhES.

l.

The Pro Lester Union consisted of a small group of Yam merchants
and:

a. a large band of traitor Mambo warriors

b. a small band of traitor hambo warriors

c. a small band of Lester's warriors

d. a large band of Lester's warriors

The Mambo army prepared for war by:

a. dancing and shaking rattles

b. painting their faces with goat's blood
c. beating drums and singing

King Lester's lookouts announced the beginning of the war by:
a. sending mirror signals

b. shouting loudly

c. starting fires

d. none of the above

One group painted their faces with goat's blood. why did they
do,this?

a. to give them courage

b. for religious reasons

c. to protect themselves from the sun

agreed to supply with weapons.
a. Yam merchants, Rambo warriors

b. Yam.warriors, hambo merchants

c. Mambo merchants, Yam merchants

d. hambo warriors, Yam merchants

 

Mambo's weapons were:

a. heavy and difficult to use
b. light and well-balanced

c. heavy and well-balanced

Lester's plan to get rid of Koko was to:

a. catch Koko in a bad position

b. have the traitor Mambo warriors kill him in battle
0. fight Koko in personal combat

Who secretly buried Lester's body?
a. Koko's warriors

b. Lester's merchants

c Lester's subjects

d. hoko's merchants

161

9.

10.

11+.

16 .

162

After the battle the people of Yam were:
a. slaves for their new ruler

b. more prOSperous than ever

c. still expert warriors

d. unhappy but proud of their occupations

Who claimed that spirits took Lester's body?
a. Lester's subjects

b. Koko's witch doctor

c. neighboring tribes

d. traitor Rambo warriors

The attitude of King Loko about the rumor of Lester's return was

that he:

a. felt indifferent since there was not too much harm in it

b. discouraged the rumor to avoid the annoyment of his slaves

c. felt uneasy about the increasing number of believers of this
rumor.

The behavior of the Freedom Group was characterized by:
a. negotiating with other tribes to invade Limbo

b. telling lies about hoko

c. planning the overthrow of Loko

d. helping slaves escape to other countries

The is secretly hiding weapons?
a. the Legend believers

b. the Freedom Sroup

c. he Iro Lester Union

d. the Lanbo traitor warriors

Taich group performed a tribal dance as part of their religious
ceremony?
a. the Yam musicians

b. hoko's army
e. Lester's cavalry
d. none of the above

After the battle the relationship between Hoke and his troops was
one of
a. respect and loyalty
b. friendlines

C. unfriendliness

0. none of the above

The nambo witch doctor killed fifty hambo warriors because:
a. they were religiously unworthy

b. they were old

c. they were traitors

d. they were afraid

163

17. Yoko was worried because believed the Lester Legend.
a. he Freedom Ereup
b. his trOOps
c. the traitor hambo warriors
d. many Yam slaves

 

13. After the battle the social structure of changed.
b. Yam

0. both Rambo and Yam
d. neither hambo nor Yam

19. Nhy was Koko grouchy and irritable?
a. relation with his troops was bad
b. his food was poisoned
c. both a and b
d. none of the above

 

20. Koko thinks that are plotting against him.
1. his troops 2. the slaves 3. the traitor warriors
a. l and 2
b. 2 and 3
c. l and 3

91. The rumor of Lester's return was told by:
a. Lester Legend believers

b. hoko's witch doctor

0. neighboring tribes

d. the Freedom iroup

22. Which of the following events occurred first?
a. looking into the sun
b. painting of faces with goat's blood
c. killin: of fifty warriors

To
\ O

.. Among the unusual sounds of the Yam musicians as the war canoes
approached we e:

a. rattles shaking

b. whistles blowing

c. copper horns blowing

AJAD EACH SflT OF TUO SENTEUCES AND THLH CdOCSi T33 ORE TJAT lS TRUE
AND LARK YOUR ANSUER SHEET ACCORDINSLY. REnEhBLR - ONLY ONE 7? ram
THO SEX Eh ES IX EACI PAIR IS CORRECT.

II THE FOLLOUINS SEKTENCES DESCRIBE LIFE bEFORE THE BATTLE

24. a. Only some of the Lester Followers were expert warriors.
b. All of the Lester Followers were expert warriors.

25.

a.

b.

de

a.
b.

a.
b.

164

All merchants were members of the Pro Lester Union.
Only some of the merchants were members of the Pro Lester
Union

FOLLOJIM} SENTEUCES DASCdle LIED AEfiw r4; dATTLm

Only some of the hambo Citizens belong to the Freedom Group.
All of the Rambo Citizens belong to the Freedom Group.

Only some of the Rambo Citizens are Legend believers.
None of the mambo Citizens are Legend believers.

Only some of the slaves are Kambo Citizens.
All of the slaves are Kambo Citizens

STOP

JD HOT TURN T33 1333 UETIL YOU A1; TOLD TO DO 8

R a

APPENDIX TEN
h 1 Test N

nggg A and B

165

fMe

 

Along Pennsylvania Avenue in Washington, D. C., in the yards of
thousands of American homes, in parks around the country, a graceful
fan-shaped tree with silver leaves - the ginkgo can be seen. A
familiar specimen it is, but as the sole survivor of a group of
plants whose geological ancestry dates ten million years back, it is
truly a "living fossil" - a tree that deserves respect having sur-
vived only because it was able to make a deal with man.

The simple deal between tree and man was that each used something
had by the other. Able to happily live in cities with smoky air and
poor soil underfoot, and returning a decorative appearance and a fine
shield from the sun for these meager commodities, the ginkgo can offer
little else - growing too slowly for its wood to be of much value and
with few people caring to eat the strong-smelling fruit.

That the ginkgo was a city lover was discovered only two centuries
ago. The first of its kind to appear in Europe was not brought from
the Orient to the Netherlands until 1754, becoming immediately popular,
for it was immune to diseases and insect pests, and once started re-
quired little attention. Pot seeming to be of particular importance
is the amount of water it received and in addition it could be culti-
vated from seed. By the end of the century it was hardly respectable
to have a garden without one.

The rocks that the records that the earliest history of this
strange tree were traced from were etched in were in the earth's crust.
Having the same heart-shaped leaves, the same fan-like form, the first
ginkgo, growing to the same height, existed ten million years ago.

It must not have been distributed in only a few places on the earth,
for fossil records of ginkgos have been found on every continent. But
the ginkgo with its seeds ripening without a hard protective coat is
a gymnosperm. When trees developed that bore protected seeds, the
competition was found by the gymnosperms to be too keen and the sur-
vivors, like the Australian kangaroo, were squeezed into geographical
corners.

Strangely enough, not even in China and Japan is the ginkgo tree
found growing wild. Even in the oldest Oriental manuscripts it is
mentioned only as the sacred resident of the temple precincts or the
family garden. No description or picture of wild ginkgos has ever
been found, and it seems clear that the tree has not grown wild within
the Span of civilized man's years on earth. Only the strange bargain
between man and tree saved the ginkgo from extinction.

166

III. Ll II,”

 

The ginkgo tree is a graceful, fan-shaped tree. It has silver
leaves. One can see this tree in Washington, D. C. slong Pennsylvania
Avenue. One can see it in the yards of thousands of American homes
and in parks around the country. This tree is a familiar specimen
but it deserves respect. It is truly a "living fossil". It is the
sole survivor of a group of plants whose geological ancestry dates
back ten million years. This tree has survived only because it was
able to make a deal with man.

The deal was a simple one. The tree uses something man has and
the man uses something the tree has. The ginkgo can live happily in
cities where there is smoke in the air and poor soil underfoot. The
ginkgo offers something in return for these meager commodities. It
offers a decorative appearance and a fine shield from the sun. It
can offer little else. It grows too slowly for its wood to have much
value. Few people care to eat its fruit. The fruit is strong smelling.

People discovered two centuries ago that the ginkgo was a city
lover. The first tree of this kind appeared in Europe in 1754. Some-
one brought it from the Orient to the Netherlands. It became popular
immediately. It was immune to diseases and insect pests. Once it
got started it required little attention. It did not seem particularly
important how much water it received. In addition, one could cultivate
it from seed. by the end of the century it was hardly reSpectable to
have a garden without a ginkgo tree.

Someone has traced the earliest history of this tree. They
traced its history from the records etched in the rocks of the earth's
crust. The first ginkgo existed ten million years ago. It had the
same heart-shaped leaves. It had the same fan-like form. It grew to
the same height. It must have been distributed all over the earth
for people have found fossil records of ginkgos on every continent.
dut the ginkgo is a gymnosperm - its seed ripens without a hard pro-
tective coat. When trees developed that bore protected seeds, the
gymnOSperms found the competition too keen. The survivors were like
the Australian kangaroo for they were squeezed into geographical
corners.

One does not even find the ginkgo tree growing wild in China and
Japan. This is strange. Even the oldest Oriental manuscripts mention
the tree only as the sacred resident of the temple precincts or the
family garden. No one has ever found a description or picture of
wild ginkgos. It seems clear that the tree has not grown wild within
the span of civilized man's years on earth. Only the strange bargain
between man and tree saved the ginkgo from extinction.

167

 

30.

31.

32.

33.

168

Vhich of these characteristics helped make the ginkgo tree
popular?

A. Its attractive appearance C. Its deep roots
3. Its tough wood D. Its sweet fruit

How did the ginkgos of prehistoric times compare with those of
today?

. They had a different shape.

. They had larger leaves.

. They were very similar to those of today.
. They grew taller and thicker.

DOLL?)

There were several reasons for the quick acceptance of the ginkgo

tree. Which of these is not one of them?
A. It was a hardy tree. C. It was attractive.
B. It needed little care. D. It grew rapidly.

Ancient Oriental peoples probably regarded the ginkgo tree as

A. holy. C. bothersome.
B. useful D. unproductive.

What was the main reason that the ginkgo almost became extinct?

A. The tree fell victim to disease and insect pests.

5. The tree could not adapt to a changing climate.

C. Fewer and fewer seeds took root and grew into mature trees.
D. Great numbers of the trees were out down.

What similarity has the author tried to point out between the
history of the kangaroo and ginkgo?

A. 30th survived only because man protected them.

B. At one time, each could be found only in a few regions.
C. Neither has ever been found living in a wild state.

D. Each had something that man wanted.

What were the "records" from which the earliest history of the
ginkgo tree was traced?

. Ancient Oriental writings

. Drawings of prehistoric cave men
. Bits of wood and leaves

. Fossils

bots?»

a... l .... -..:rfnu';

A/
IN).

37.

169

What is the meaning of the term "living fossil” as used in the
first paragraph?

A. It is a fossil that was formed from a living plant.
B. It is a plant that almost became extinct.

C. It is a plant that is found only as a fossil.

3. It is a fossil that has come back to life.

What did man supply that the ginkgo tree needed?
A. Fertilized soil

3. Protection against disease

C. Care for young seedlings

1)

’. Sufficient moisture

When the author stated that the ginkgo "was able to make a deal
with man", he meant that the tree

could live if left alone by man.
had qualities which caused man to protect it.
developed new characteristics to meet a change in its

environment.
was so hardy that it could not be killed by man.

r)c3:>

S3

Neither you nor I will ever have to go to a doctor to be treated
for bronze disease because it can be caught by no human beings. The
disease is not contracted by anything but ancient bronze objects from
Egypt or other old countries.

For years museum authorities all over the world have been pre-
sented a serious problem by bronze disease. When the bronze object
breaks out with a light green ulcerous-looking rash, its presence is
first noticed growing more or less rapidly into a hard, thick crust,
the object sometimes being covered with such a heavy, spreading over
the entire surface, crust that it cannot be recognized. If the
"victim" happens to be a thin object, the bronze will be often eaten
right through by the sores and destroyed.

Scientists looking for a cure for bronze disease had to find out
what caused it first. It was discovered that when they were dug up,
bronze objects which had been buried for centuries in the desert did
not show symptoms of the disease. These same objects, however, when
they were taken to a humid climate, often became ”infected." Objects,
on the other hand, which were found buried in soil which was damp, at
the time they were unearthed were likely to show signs of the disease.

This indication of the effect on the bronze of dampness gave the
scientists who next discovered that the disease was brought on by a
chloride, a salt of hydrochloric acid, present in the soil which was
moist in which the bronze objects had been buried a starting point.
The combination of different metals in the bronze was caused to con-
vert slowly into crusting sores by the salt and moisture.

The action of the disease was revealed by further scientific
investigation to be due to an electrolytic process closely resembling
the process of rusting. By electrolysis is meant the decomposition,
or breaking down, of a chemical compound by an electric current. In
the case of the ancient bronze, an electric current is set up by the
salt in contact with the moisture which causes the copper and tin in
the bronze to separate and to faster, producing the greenish crust.
Scientists reasoned that if corrosion took place in this manner, the
disease could be curred and the bronze restored by reversing the
electrolytic action.

The garageman uses a method for charging an automobile battery
which is similar to the way the "cure" for bronze disease is now
being accomplished. An electric current produced by batteries to which
one electric wire attached to the bronze object to be cured (which is
placed in an ordinary lye solution) and another electric wire (which
is connected to a strip of sheet metal which is also placed in the
solution) are connected, passes through, at a certain constant voltage,
the bronze object and the lye solution performing a chemical change

170

Ti. aid . _

171

in the bronze which reverses the process of corrosion, detaching the
crust from the bronze, dissolving it in the lye solution, and at the
same time stopping the progress of the disease.

Perhaps "cure" should not be used but instead, "check" for having
discovered as yet only a treatment by which the disease is brought to
a halt, no method which will bring about a permanent cure exists. Just
as certain diseases can be develOped by us, like tuberculosis, more
than once if we are exposed to the germs or after the disease has been
checked a relapse can occur so can bronze disease be developed by
ancient bronze objects again or a relapse happen if they are exposed
to the proper combination of salt and moisture. Permanent cure and
total prevention of bronze disease are problems which must be further
explored by scientists and art scholars.

 

You and I will never have to go to a doctor to be treated for
bronze disease. Human beings cannot catch it. Only ancient bronze
objects can contract the disease. These objects come from Egypt or
other old countries.

For years bronze disease has presented a serious problem for
museum authorities all over the world. One first notices the disease's
presence when the object breaks out with a rash. The rash is light
green and ulcerous-looking. It grows more or less rapidly into a
hard, thick crust. The crust spreads over the surface of the object.
Sometimes such a heavy crust covers the object that the obiect cannot
be reCOgnized. If the "victim" happens to be a thin object, the sores
will often eat right through the bronze and destroy the object.

Scientists were looking for a cure for bronze d'sease. They
first had to find out what caused it. They made a discovery. Bronze
objects were dug up. They had been buried in the desert for centuries.
They did not show symptoms of the disease. However, these same objects
often became infected when someone took them to a humid climate.
Some other objects were found buried in damp soil. Uhen they unearthed
these objects they were likely to already show signs of the disease.

This indication of the effect of dampness on the bronze gave the
scientists a starting point for their research. jext, they made
another discovery. The disease was brought on by chloride. Chloride
is a salt of hydrochloric acid. It was present in the moist soil that
the bronze objects had been buried in. dronze contains a combination
of different metals. The salt and the moisture cause these metals to
slowly convert into crusting sores.

Further scientific investigation revealed the cause of the action
of this disease. it is due to an electrolytic process. The process
closely resembles rusting. electrolysis means that an electric cur-
rent breaks dewn or causes the decomposition of a chemical compound.
This is what happens in the ancient bronze. The salt in contact with
moisture sets up electric current. This current causes the copper
and tin in the bronze to separate and foster, producing the greenish
crust. Scientists reasoned that if the corrosion took place in this
manner, they could cure the disease. They could restore the bronze
by reversing the electrolytic action.

no "cure" for bronze disease is now accomplished by much the
same method that the garageman uses to charge an automobile battery.
One takes the bronze object to be cured and places it in an ordinary
lye solution. He attaches an electric wire to it. He places a strip
of sheet metal in the solution. This strip he connects to another
electric wire. The two wires are connected to batteries. The batter-
ies produce an electric current which passes through the bronze bject

172

 

173

and the lye solution. The electric current remains at a certain con-
stant voltage and produces a chemical change in the bronze. This
chemical change reverses the process of corrosion. The electric
current detaches the crust from the bronze and dissolves it in the
lye solution. This, at the same time, steps the progress of the

C1130 SO.

Perhaps we should use the word "check" instead of "cure". The
treatment halts the disease, but no one has discovered a method which
will bring about a permanent c‘ e. He can develop certain diseases,
like tuberculosis, more than once if we are exposed to the germs. ”e
can have a relapse after the disease has been checked. So can ancient
bronze objects develop bronze disease again. 30 can they have a re-
lapse if someone exposes them to the proper combination of salt and
moisture. Scientists and art scholars must further explore the prob-
lems of permanent cure and total prevention of bronze disease.

 

I‘k‘ v'q -

 

41.

42.

#3.

174

which of the following groups would be likely to be most
interested in the solution of the bronze disease problem?

A. Museum workers

B. Medical doctors

C. Garage mechanics

D. Metal workers

Which of the following would be a 10gical "victim" of bronze
disease?

A. An elderly Egyptian

'B.

C.
D.

An antique flask made of bronze
A museum worker who handles bronze objects
An automobile battery

Which of the following statements describes simply but accurately
the electrolytic process that results in bronze disease?

A.

The combination of different metals in the bronze produces

a chemical compound which turns into a greenish crust on the
surface of the bronze.

The tin and c0pper in the bronze combine to form a chemical
compound which, when moisture is present, causes the polished
surface of bronze to become rusty.

The green crust on the surface of the bronze gives off hydro-
chloric acid that eats right through the surface of the bronze.
A combination of chloride and moisture sets up an electric
current that causes the different metals in the bronze to
separate and form crusting sores.

In order for the electrolytic treatment to be successful, the
amount of electricity passing through the bronze object must be

A.
B.
C.

D.

gradually increased as the decomposition progresses.
gradually decreased as the decomposition progresses.
increased or decreased according to the degree of decom-
position at any given time.

held uniform as long as the treatment lasts.

What is the writer's attitude toward the eventual discovery of
a permanent cure for bronze disease?

A.
B.
C.

D.

He is openly optimistic.

He is decidedly pessimistic.

While he does not express an opinion, he leaves the impression
that he is hopeful.

He contradicts himself so many times that it is impossible

to tell.

#5.

47.

175

How is the relationship between the electrolytic process of the
development of bronze disease and the electrolytic process of
the "cure" for the disease most properly expressed?

A. The two processes are the same.

B. The two processes are entirely different.
C. One process is dependent on the other.

D. One process is the reverse of the other.

In paragraph 7 the author illustrates the temporary nature of
the'bure" by means of which of the following?

A. A detailed, factual explanation

B. A dictionary definition of the term
C. A familiar example in human disease
D. Slang phrases

Bronze objects can become diseased again if they are exposed to

A. hOt, dry 311‘.

B. repeated baths in a solution of lye.

C. diseased objects that have not been cured.

D. the condition that caused the disease to develop originally.

The search for the cause and cure of bronze disease, as reported
in this article, is an example of which of the following?

A. An orderly scientific investigation
B. A trial-and-error experiment

C. Historical research

D. Guess work on the part of scientists

In the last paragraph, why does the writer suggest that another
word be substituted for "cure."

A. He is afraid that some readers may not understand the word
II CW6 . N

B. He thinks that "cure" is not a scientific term.

C. He wants to use a more precise word.

D. He is trying to be original in his writing.