Verhul conrprehe~rsron refers to a person’s ability to understand linguistic materials. such as neldspapers, magazines, textbooks. lectures. and the like. Verbal comprehension has been recognized as an integral part of intelligence in both psychometrtc theories (e.g.. Guilford 1967; Thurstone 1938; Vernon 1971) and information-processing theories (e.g.. Carroll 1976: Hunt 1978: Sternberg 1980). and has, under a variety of aliases, been a major topic of research m differential
and experimental
The theoretical a number
of different
vocabulary, lary
reading
has been
psychology
construct
of verbal
ways.
Most
comprehension,
recognized
for many
often,
can be operationalized
it is directly
and general
not only
years.
comprehension
measured
information.
as an excellent
measure
in
by tests of
Indeed,
vocabu-
of verbal
compre-
level of intelligence (e.g., Jensen 1980: Matarazzo 1972). Verbal cl mprehension is also measured indirectly by tests intended to measure constructs other than verbal comprehension, but that have a heavy verbal load. Examples of such tests include verbal reasoning tests such as analogies and classifications, tf the vocabulary level is relatively high, and mathematical reasoning, if the test items are presented as word problems. The importance of verbal comprehension to
hension,
l
but also as one of the best single indicators
Preparac~on
Research
of this XII&
to Rober.
Foundataon psychology
department
psychology
department
Sternbq.
Dept.
was supported
J. Sternberg.
predoctoral
fellow at Yale
and
Daniel
Umversity;
at the Unwers~ty
of P\ychology,
by Conlracl
The article
Yale
NOlXJl478C0025
was wntle” Kaye
of Cabforma
Unwers~ty.
while
was
Dame1
Box
of a person’s
an
Kaye
Yale
0304-422X/82/0000-0000/$02.75
0 1982 North-Holland
of Nr\al Scnence
fellow
m
the
a faculty
member
m
the
Malhng
Slatmn.
USA
the Offlce
was d National
postdoctoral
IS currenlly
al Los Angeles. I IA.
from
Janet Pouell NIMH
overall
New
address. Hdven.
Robert CT
J.
06520.
the assessment as well as the theory of intelligence is shown by the fact that almost all of the major tests of intellectual ability contain verbal tests of the kinds noted above, and that those that do not include such tests are self-consc~ously nonverbal in order to minimize measurement of abilities through content that is culture-laden (e.g.. Cattell and Cattell 1963: Raven 1965). We present here a discussion of some ideas we and others have had about an important aspect of verbal comprehension - learning from context. Our discussion is dtvided into five parts. First. we provide a capsule literature review on past work dealing with learning from context. Second, we briefly present our own developing theory of learning from context, and attempt to relate our ideas to the past ideas :hat have helped give rise to them. Third, we provide a summary of some data that lend empirical support to our theoretical Ideas. ,Fourth, we attempt to place our theory and data on learning from context into a broader conceptual framework for understanding the nature of verb21 comprehension. Fifth. we discuss some psychological and educational imphcations of our theory ancd data.
Rewsrch on learning from colnte\t: a brief literature review
Research on learning from context can be divided into two streams - research dealing with context external to a given word or concept to be learned and research dealing with context internal to the word or concept.
Modern research on verbal learning from external context at least to Werner and Kaplan, who proposed that rhe chdd acquwe\
the meanmg
verbal
or OhJeLtIve.
ohJrcls
or through
the meanwg the wrhal
he learns wrbal
01 ucrds
prmc~pall>
to understand
delmwon
of a uord
IS graped
COBICXI. (1952
3)
The
I” two ways
verbal
cecond
qmhols
~.a)
One
through
IS t’,rough
m the course 01 conwrsauon.
can be traced
IS by expbat the adult’s
~mpbc~l
IL.11 I\
reference dmxl
or conlexlual
mferred
from
back
ather
nammg
of
reference. the cwb
01
Werner and Kaplan were especially interested in the second way of acquiring word meanings. They devrsed a task in which subjects were presented with an Imaginary word followed by six sentences using that word. The subjects’ task was to guess the meaning of the word on the basis of the contextual cues they were given. One example (from the twelve imaginary words they used) is “contavish”. which they intended to mean hole. They did not, of course, tell :he children in their study the meaning of the word, but rather presented them with six sentences (1952: 4):
(I) (2) (3) (4) (5) (6)
You can’t fill anything with a contavish. The more you take out of a contavish the larger it gets. Before the house is finished the walls must have contavishes. You can’t feel or touch a contavtsh. A bottle has only one contavish. John fell into a contavish in the road.
Children ranging in age from 8 to I3 years were tested in their ability to acquire new words presented in this way. Developmental patterns were analyzed by a number of different means. Werner and Kaplan found that (a) performance improves gradually wtth age. although the various processes that underlie performance do not necessarily change gradually - whereas some change gradually, others change abruptly; (b) there is an early and abrupt decline in signs of immaturity that relate to inadequate orientation toward the !ask itself: (c) the processes of signification for words undergo a rather decisive shift between approximately IO and I I years of age; and (d) language behavior shows different organizations at different ages. Daalen-KapteiJns and Elshowt-Mohr (1981) pursued the Werner-Kaplan approach by having subjects think aloud while solving Werner-Kaplan type problems. They proposed an ideal strategy for learning from context whereby subjects could form a model (provisional representation) of the meaning of a new word. In this strategy. (a) the sentence is reformulated so that it can be brought to bear directly upon the neologism. for example. in Sentence 1 above. the strategy might yield the statement that “a contavish is not a substance that can be used to fill anything”; and (b) the reformulated information is transformed into an aspect of the meaning of the neologtsm. for example. “a contavish may be some kind of absence of substance”. Using ingenious protocol-analysis techniques, the investigators found that (a) word acquisition is guided by models, with an initial model chosen on the basis of the interpretation of the new words meaning in the first sentence. and wrth subsequent processing guided by this model; (b) the processing of each new word presentation in context can lead to the filling of “slots” in the model. to adjustment of these slots, or to the formation of a new model altogether; (c) if the model is not sufficiently well articulated to permit active search and evalua:ion of possibly relevant info:mation, as tends to be the case for low-verbal subjects, model-guided search can be replaced by sentence-based processing (step (a) of the ideal strategy without the subsequent use of step (b) of thts strategy); and (d) high- and low-verbal subjects learn word meanings differently, with high verbals generally using both steps of the ideal strategy and low verbals generally using only the first step (sentence-based processing). Whereas Daalen-Kapteijns and Elshout-Mohr specified strategy for word acquisition in some detail but the representation (what they referred IO as the model) of information about the word in only minimal detail. Keil (198lb) has
specified representation in considerable detail but strategy in only minimal detail. Keil presented children in grades kindergarten. 2, and 4 with simple stories in which an unfamiliar story was described by a single predicate. An example of such a story is “THRtXTLES are great. except when they have to he fixed. And they have to be fixed very often. But it’s usually very easy to fix throttles”. Subjects were then asked what else they knew about the new words (here. “throstle”) and what sorts of things the new words drscribed. Keil found “erences about the that even the youngest children could make sensible general categories denoted by the new terms and about the properties the terms might reasonably have. Errors were systematic and in accordance with Keil’, (1979. 198la) theory of the structure of ontological knowledge. This theory provides a powerful basis for inferring a possible structure for storing (at lea!.t ontological) information about the meaning of a new word and for inferring the posstble predicates of the word. Jensen has suggesied that vocabulary is such a good measure of intelligence “because the acqutsition of word meanings is highly dependent on the educrron of meaning from the contexts in whtch the words are encountered” (1980: 146). Marshalek (1981) has tested thts hypothesis by using a faceted vocabulary test, although he did not directly measure learning from context. The vocabulary test was admmtstered together with a battery of standard reasoning and other tests. Marshalek found that (a) subjects sometimes could give correct examples of how a given word is used in sentences, despite their having inferred incorrect defining features of the word: (b) subjects with low reasoning abihty had major difficulties in inferring word meanings; and (c) reasoning was related to vocabulary measures at the lower end of the vocabulary difficulty distributton but not at the higher end. Together. these findings suggested that a certain level of reasoning ability may be prerequisite for extraction of word meaning. Above this level. the importance of reasoning begins rapidly to decrease. It has been assumed in the above review that the ability to learn from external context leads to higher vocabulary. It should be pointed out. however, that the relationship between learning from context and level of vocabulary is probably bidnectional (Anderson and Freebody 1979; Sternberg 1980): learning from contex: can facilitate vocabulary level. at the same time that a higher vocabulary Ieve! can facilitate learning from context. The theoretical ideas expressed above have been embodied in psychometric measurements of verbal ability. The Scholastic Aptitud1.t Test (SAT) and the Graduate Record Examination (GRE). for example, both have sentence completton items requiring examinees to fill in a blank occurring in a sentence the word that is most appropriate to the context; the Degrues of Reading Power Test (DRP) measures verbal comprehension exclusively by measuring ability to understand use of words in context, as does the Words-in-Context Test (WING) (Cook et al. 1963; Heim 1970).
in vocabulary trammg These ideas have also been implemented 1974). and (see Gipe 1979; Jolmson and Pearson 1978; O’Rourke
programs have been
used in second-language as well as first-language teaching (see, e.g.. Shumway and Forbes in press) The basic idea has been to present new words in a variety of contexts, letting the students infer the meanings of the words. The rationale has been that in comparison to rote learning. this method of presentatton facilitates deeper levels of processing (Craik and Lockhart 1972) and hence more meaninglul learning and better retention. To summarize, learning from external context has played an important part in theorizing about, measuring. an.J iraining verbal comprehension. The review presented here is of necessity sketchy ,tnd incomplete. but shows. we believe. the broad range of psychologists and educators believing in both the theorettcal and the practical importance of contextual learning of verbal concepts.
Research on the use of Internal context has proceeded along somewhat different lines from research on the use of external context: the major theoretical issue seems to have been whether afftxed \sords (such as preduposed) are stored in memory in unitary form (i.e.. as predrsposed). m a set of lexically decomposed forms (i.e., as pre-drs-pose-d ). or in both of these kmds of forms, with the form that is accessed in a given instance depending upon the task and task context. From the point of view of this article (although certainly not from all points of view). research on internal context is less advanced than research on external context, in that the question of how internal context IS used by comprehenders in vocabuary devr!opment has barely begun. whereas the question of how external context is used in vocabulary development has received at least modest attention, as shown above. But recent research on hoa affixed words are stored m memory is relevant to our concerns: If such words are stored only as single lexical entries, then one might expect comprehenders to have considerable difficulty in :he use of internal context for Inferring word meanings; if, on the other hand. affixed words are stored as sets of separate morphemes, either instead of or in addition to lexical entries for the complete words, then one might expect comprehenders to be able to use internal context fairly freely in inferring word meanings. Evidence for the view that affixed words are stored as sets of separate lexical entries corresponding to their individual morp. -mes can be traced back at least to Taft and Forster (19’75). These invest.adtors performed three exper,ments employing a lexical-decision task. In this task. subJects are presented with a string of letters and have to decide as quickly and as accurately as possible whether the letter string does or does not constitute a real English-language word. Three major results of interest words that were stems of prefixed words (e.g.,Juoetrure.
emerged. First. nonwhich is the stem for
rcjhgoenare)
took longer IO recognize
as nonwords
than did nonwords
that were
not stems of prefixed words (e.g., perrorre. for which the re- in repmorre not function a!. a prefix). This result suggested that the nonword stem directly represented in the lexicon and that one or more extra steps needed to identify the stem as a nonword. Second. words that could occur as free morphemes - which can stand by themselves as words (e.g.. oenr word) - and a:; bound morphemes - which cannot stand by themselves
does was were both as a as
words (e.g., oen: when serving as the stem of the word i~toenr) - took longer to identify as words when the bound form was more ftsquent in the English language than was the free form. This result suggested that the existence of vent (or any other comparable letter string) as a bound ~norpheme and possibly as a salient and separate nonword lexical entry in mewScry may interfere with or in some other way lengthen the latency for recogm of uenr (or any other comparable letter string) as a free morpheme that cat. stand as a word irt its own right. Third. prefixed nonwords took longer to identify as nonwords when they coniained a real stem (e.g., dquoenare. in whichJuoenare is a real stem) than did contra) nonwords that did not contain a real stem (e.g., deperrorre, in which pertowe i.5 not a real, i.e., separable. stem). This result again suggested that the stem may have been stored separately and hence interfered with recognition of the total letter string as a nonword. Taft and Forster presentecl a model of word recognition that incorporated the notion that morphemes are each represented as separate lexical entries. Further support for a notion ,f separate storage of individual morphemes can be inferred from a study *If Murrell and Morton (1974). although the purpose of this study and tlte theoretical framework in which it was placed were different from those th,rt concern us in the present review. The Taft and Forster view of storage of affixed words by their individual morphemes was challenged by Manelis and Tharp (1977). who interpreted da1.a from two of their own experiments as supporting the view that words am-e stored only as single umts. These investigators also employed a lexical-decision task. Certain controls were introduced in stimulus selection that had not been implemented in the Taft and Forster (1975) studies. Two main results emerged. First. in a cructal comparison of latencies for words that were affixed with latencies for words that were not affixed, no significant difference emerged for time to recognize the two respective kinds of letter strings as words. Although this result supported the notion that affixed and nonaffixed words are stored in the same way, other results from this experiment strongly suggested that the two kinds of words are nor identically processed, so that there was at least some ambiguity in the results. Scrcond, a test condition designed to encourage subjects to use lexical decomposition failed to show the significant difference that would havd: supported the conclusion that decomposition was used. On the basis of these results, the investigators accepted a simple made1 in which affixed words are stored unitarily in memory in the same form that nonaffixed words are stored.
Subsequent research (Stanners et al. 1979; Taft 1979) has attempted to reconcile the two views presented abov: by suggesting that both forms of representation may be used and accessed at different points in information processing or in different tasks or task contexts. Of particular interest in this latter regard is a study by Rubin et al. (1979). which was interpreted by its authors as suggesting that decomposition is used only if special strategies are evoked as a result of a stimulus list that is heavily biased m favor of affixed words. Taft (1981) has recentI!: countered this vmw with further experiments suggesting the generality of lexical decomposition in word recognition. At this point, the issue of just what differences, if any. exist between the representation and processirg of affixed and nonaffixed words remains unresolved. The studies desribed above have all used a lextcal-decision task. or some variant of it, to test the form in which affixed words are represented m memory. Obviously, any one paradigm is limited in the information it can provide regarding a given psychological issue. Holyoak et al. (1976) and Kintsch (1974) have done exporiments related to those reviewed above that address the issue of lexical decomposition from a more semantic point of view. These investigators used semantic-decision and memory expertments to mvesttgate lexical decomposition. Because the hypotheses they investigated. and hence the experimental paradigms, did not bear on exactly the issues discussed above (namely, the representation and processing of affixed versus nonaffixed words), the research will not be described in detail. It is worth noting, however. that the results of these experiments generally suggest that representation of complex words is unitary, and that lexical decomposition is not routinely done. But some ?f the experiments suggest that lexical decomposition is possible under at least some circums!ances. The theoretical ideas discussed above have not been directly embodted in tests of verbal comprehension. Such tests do not separately measure people’s knowledge of prefixes, stems, and suffixes, or people’s ability to integrate these kinds of knowledge. But in standard vocabulary tests, in which words are presented for definition in the absence of any external context, the use of internal context provides the only viable means of figuring out meanings of words that are unknown or scarcely known. Introspective reports of vocabulary-test takers suggest that they use their knowledge of prefixes, stems. and suffixes to attempt to figure out meanings of at least some words. Internal context, like external context, can on occasion impede attempts to infer wc rd meanings. For example, mehorufe and amelrorare have essentially tht satne meaning, despite the addition of the prefix (I- in the latter form. Psychologists and educators interested in vocabulary training have recognized the importance of internal context in vocabulary-skills training p;ograms. Both Johnson and Pearson (1978) and O’Rourke (1974). for example. have incorporated training on intraword cues into their vocabulary development training programs, and Carroll (1964) has also stressed the importance of
such cues in generalized
reading
rcndmg
mstructton
he viewed
students
on the use of mternal
To
can
summarize.
affixed
words
stmilar
to
that
of
the subjective morphemes
evtdence
ts mixed
regarding
.tllon
people
context
WI
the meanm;;s
Theoq Two ;I
actLatly
part.
in inferring
of affixed
of learning
from
learning factlnation verbal
concepts
a neu
or
inhibtt
(or
tnhibit)
,.ttuations.
The
learning verbal abihties
later
individual
idea
abilities
meanings.
the results of previous
or under
what
circumstances
of word stems and affixes
to figure
words.
The
first
than others. part
verbal
pertains
It is that
elements
that
learning:
are hypothesized verbal
are
part
hypothesized
IO
to new better
at
differences
learning,
in
in people’s
and to be wary
the same or very similar in people’s
and to !ransfer
of that
alsc
are
individual
to differences
facilitate
of
the new
elements
and also its transfer
to be involved
concepts
in which
some individuals
as
IO why
the degree
contextual
is that
i.1 large
we view
the difftculty of
by the context
to why II
which
idea pertains
a function
concept
are others.
that inhibit
to retrieve
from context.
of the concept
be traced
contextual
elements
later
in inferring a
provided
a new
than
can
differences
Such
in
is in large
retrieval
concepts
IO exploit
:hat
of the critical
decomposition
the same or very similar
learning
comprehension
of contextual
have suggested
in terms
has its own entry.
require
of learning
of learning
second
verbal
of
mterest
comprehtnsion.
concept
s embedded;
facilitate
processing
In terms of our present
word
are easier to learn
(or in;libition)
fdcthtate
to
training
such use would
complex
our theory verbal
verbal
concept
of
contelt
of general
some verbal
or other
and
who
that
of whether
do use thetr knowledge
basic Idea< underlie
suhtheory
(1976). each word
to use lexical
~n\e~g,~t~ons Irave open the question tndtvtduals
approach
places us in a position
is organized
at the same time
on the individual’s
the use of mteroal
representanon
of the evidence
even though
of new uards.
extra effort
the
and Johnson-Laird words.
the phonics to a program
clues.
of each individual
would
the meanings chst~nct
Miller
lexicon
Indeed.
as preparatory
Our own reading
of derived
an orgamration
instruction.
contextual
sources of differential
these concepts
to new
situations. The
theory
dtstinguishes
mance.
The
use in
inferring
applies effort.
the
to peoplr’s
model
meaning
word
particular
between occurs task
that and
model
individuals’
specifies of
capabilities,
7 he performance
relationship that
competence
between
the contextual
a previously
given
those
unknown
and their word.
of and
The
performance
model
can model
time
imposed
the context
how well a given set of c!ues will
situation.
This
unlimited
constraints
word
perfor-
clues individuals
unknown
the expenditure
specifies
the previously affect
competence
by
be utilized
applies
to
and the
in which in a
people’s
utilization of their competence, @en the constraints imposed by the particular task and, situation on people’s information-processing capabilitIes. We shall consider in turn the application of each of these models to external and internal context.
During the course of one’s reading (01 other encounters with words). one commonly comes upon words whose meanings are unfamiliar. When such words are encountered. one may attempt to utilize the external context In which the words occur in order to figure out the meanings of the words. Our theory specifie5 external clues that influence the likelihood that these meanings Gil be correctly inferred. Conlperence nlodel The competence model specifies the particular kinds of context clues that individuals can use to figure out the meanings of new words irom the contexts in which the words are embedded. Not every clue can be applied m everi situation, and even when a given clue can be utdized. its usefulness wtll he medinted by the performance-model considtrations to be desc:ihed in the next section. Context clues ~;e hints contamzd in a passage that facditate (or. in theoq and so.metilnca in practice, impede) deciphering the meaning of an unknown word. We propose that cont^xL clues can be classitied into eight categories. depending upon the kind of information they provide. Alternative and related classification schemes have been proposed in the past by Ames (1966). McCullough (1958). Miller and Johnson-Laird (1976). and Sternberg (1974). among others. The eight categories in our classification scheme include: (I)
Ten~poral clues. Clues regarding the duration cr frecuency of X (the unknown word), or regarding when X car. occur: aiternntively. clues describing X as a temporal property (SULT as duration or frequency) of some Y (usually a known word in the passage).
(2) Spurid clues. Clues regarding the general or specific location of X. or possible locations in which X can sometime.. be found: alternatively. clues describing X as a spatial property (such as geIl=ral or specific location) of some Y. (3) Value clues. Clues regarding the worth or desirability of X. or regardmg the kinds of affects X arouses; alternatively, clues describ ng X as a value (such as worth or desirability) of some Y. (4) Srufrve descrrprwe clues. Clues regarding properties of X (such as size, shape, color, odor, feel. etc.); alternatively, clues describing X as a statlye descriptive property (such as shape or color) of son>; Y.
(5) Fnncrro~~l clescrrprroe &es. Clues regarding possible X can perform. or potential uses of X: alternatively. possible purpose, action. or use of Y. (6) C~~~rso//Btah/en~ent rlrres. Clues regarding possible conditions for X: alternatively. clues describing X enahhng condition for Y.
purposes of X. actions clues describing X as a causes of or enabling as a. possible cause or
(‘I) Clrrss nwn~her.~hrp clues. Clues regarding one or more classes to which X belongs. or other members of one or more classes of which X is a member; alternatively, clues describing X as a class of which Y is a member. (8) Eqrrroaletrce c/rre.c. Clues regarding the meaning of X. or contrasts (such as antonymy) to the meaning of X; altermuvely. clues describing X as the meaning (or a contrast in meaning) of sorre Y. An example of the use of some of these clues in textual anaiysis might help concretize our descriptive framework. Consider the sentence. “At dawn, the BLEN arose on the horizon and shone brightly”. This sentence contains several external contextual clues that could facilitate on,‘, Inferring that BLEN probublv means SUN. “At dawn” provides a temporal clue. describing when the arising of the BLEN recurred: “arose” provides a functional descriptive clue. describing an action that a BLEN can perform: “on the horizon” provides a spatial clue. describing where the arising of the BLEN took place: “shone” provides another functional descriptive clue. describing a second actton a BLEN can do; finally. “brightly” provides a stative descriptive clue. describing a property (brightness) of the shining of the BLEN. With all these drfferent clues. it is no wonder that most people would find it very easy to figure out that the neologism BLEN is a synonym for the word. fun. We make no claim that the categories we have suggested are mutually exclustve. exhaustive, or independent in their functioning. Nor do we claim that they in any sense represent a “true” categorization 01 context clues. We hove. however. found this classification scheme useful in understanding subJWIS strategies in deriving meanings of words from context. Collectively, they provtde a basis for a person to exercise his or her competence in inferring woru meanings.
The performance model specifies relations word and the passage in which it occurs that Thus, whereas the competence model specifies cues that Individuals can use IO figure out the
between a previously unknown mediate tht usefulness of cues. the particular kinds of context meanings of unfamiliar words,
the performance model spectfies the medtatind variables that can affect, either posttively or negatively, the application of these possible cues i,l b grven situation.
(I)
Number of occurrences of the unknown word, A given kind of cue may be absent or of little use in a given occurrence of a previously unknown word. but may be prtsent or of considerable use in another occurrence. Multiple occurrences of an unknown word increase the number of available cues and can actually increase the usefulness of individual cues if leaders integrate information obtained from cues surrounding the multiple occurrences of the word. For example. the meaning of a gtven temporal cue may be enhanced by a spatial cue associated with a subsequent appearance of the unknown word, or the temporal cue may gain in usefulness if it appears more than once in conjunction wi)ii the unknown word. On the other hand. multiple occurrences of an unfamiliar word can also be detrimental if the reader has difficulty integrating the information gained from cues surrounding separate appearances of the word. or if only peripheral features
of the word are reinforced being of central importance (2)
Vanuhilrty
of conte.rts
and are therefore incorrectly interpreted to the meaning of the unfamihar word.
rn whrch mulrrple
0ccurrence.v
of rhe unknown
as I$ord
Different types of contexts, for example. different kinds of subject matter or different writing styles, and even just different contexts of a given type, such as two different illustrations withm a given text of how a word can be used, are likely to supply different types of informatton about
uppeur.
the unknown word. Variability of contexts increases the likehhood that a wide range of types of clues will be supphed about a given word. and thus increases the probability that a reader will get a full ptcture of the scopr of a given word’s meaning. In contrast, mere repetition of a given unknown word in essentially the same context as that in which it previously appeared is unlikely to be as helpful as a variable-context repetition. because ieu or no really new cues are provided regarding the word’s meaning. Variabihty can also present a problem in some situations and for some individuais: if the information is presented in a way that makes it difficult to integrate across appearances of the word, or if a given individual has difficulties in making such integrations. then the variable repetitions may actu,tlly obfuscate rather than clarify the word’s meaning. In some situations and for some individuals, a stimulus overload may occur, resulting in reduced rather than increased understanding (3) Imporrunce o/ the unknown word to ‘uaierstandmg rhe comexr m whrch it IS embedded. If a given unknown word i.\ judged to be necessary for understanding the surrounding material in which it is embedded. the reader’s incentive for figuring out the word’s meaning is increased. If !he word IS judged to be unimportant to understanding what one is reading (or hearing), one is unlikely to invest any great effort in figuring out what the word means. Whereas in explicit vocabular,f-training situations. the irdividual may always be motivated to infer word meanings, in real-world situations, this will not be the case. Thus. a +estion of interest from the
perspective
of our
recognize
which
model
words
some cases, it really given word’s different and
is the extent
meaning.
It is possible
levels,
that
to distinguish
is, the importance
the meaning
of the sentence
the meaning
of the paragraph
the importance
rls one form
of a word
of comprehension
man (1977.
1979.
1981).
monitoring
Flavell
(1982).
to under-
and
to under-
The
ability
it occurs.
to understanding
context
of the form
Collins
and
at
the sen:tince
word
tt occurs,
in which
In
importance
between
of a given
in which
can
are not.
time to Ergure out a
between
We distinguish
standing
reader
and which
the individual’s
standrng
recognize
an individual
to a passage.
levels of text organization.
paragraph
to whrch
are important
may not be worth
studied
Smith
to
may be seen
(in
by Markpress).
and
others. (4)
tMp/liltlrv.\
o/
~rrrrXmm~
~twrd.
A given
of
the word
the
nature
Incatron
~‘011I(‘.V1
whose
Consider
ftrst
meaning
In contrast,
temporal
temporal
to expect
a given of
all
kind
of
words.
can affect
unhnown
that
word’s
a substantial recognized:
meaning.
portion indeed.
unknown
word
the
text, cue
to which
unknown
word.
Rubin
before the placement out what
word
placement De!rary
unknown other
unknown
unwilling
the density
(1976).
means
helpful
!nan
a
It is unrealistic
in figuring
an example
of
out
the
how
the
whose meaning
is to be
as relevant
may
relevance be
from
of
the
the known cue
misinterpreted The
found
may
helpfulness before
that
than
to
by be an
of context or after
context
in a cloze test was more helpful
go in the blank
the
word -lever
as relevant
the cue comes
for example.
a relati*:ely
to inferring
the
occurring to figuring
was con text occurring
after
the
of the blank.
of U&KJVVI
prevrously herself
should
the
by whether
of a blank
“diurnal”
then there is probably
be recognized
it is more proxrmal.
cues may also be mediated
can event
where
helpfur
If the cue is separated
of
the word
a Jrunrul
pasture.
now
the
ts to be
cue describing
be more
to the word
is unknown.
the cue will
upon
If a given cue occurs in close proximity
cue helpfulness.
whose meaning
htgh lihelihood
probably
Consider
Inferred
of
when
IO lrgure out that
of cue to be equally
kinds
of the cue in the text relative
IO the word
the nature
out that rrrg is a low-lying
location
and
IlW
upon
whose meaning
than a spatial
cue would
o/
depending
is to be inferred
of how
an individual
a spatial
rr!ewlr,lp
helpful
cue describing
be more helpful
cue in figuring
meanings
rlw
to the word
an example A
probably
the event occurs in aiding
(5)
ulldersrot7tiltlg
cue can be differentially
cue helpfulness.
occurs would
&r!s.
,)I
of the cue m the text relative
Inferred. ,tffect
.wrrolitlci~rlg
words. If a reader is confronted words.
or unable
of unknown and
words.),
therefore
with
he or she may be overwhelmed to use available
words
is high,
unhelpful
and it can be difficult
and find
cues to best advantage.
relatively
words
a high density
(for
to dtscern
When
more text is occupied figuring which
out
of
him or
by
meanings
of
of the cues that
are
avallable
apply
IO which
of the
words
that
are
unknown.
In
such
a
situation, utilization of a given cue may depend upon figuring out the meaning of some other unknown word. in which case the usefulness of that cue (and very likely of other cues) is decreased. (6) Concreteness of the unknown word and rhe surroundrng con~rx~. Concrete concepts are generally easier to apprehend, m part because they have d simpler me.ining structure. Familiar concrete concepts such as tree. churr, and pencrl are relatively easy to define in ways that wouk’ satisfy ~tost people (with the exception of those interested in philosophical discusston of how words mean); but familiar abstract concepts such a5 /rut/t, /me. and Jusrice arc extremely difficult to define in ways that would satisfy large numbers of people. Indeed. each of these concepts has been the subject of multiple books, none of which have provided “definitive” definitions. Moreover, the ease of inferring the mcdnmg of the word will depend upon the concreteness of the surroundmg descriptton. A concrete concept such as tng might appear more opaqu? embedded in a passage about the nature of reality than it would emboddt d in a passage about the nature of food sources; stmilarly, an abstract concept such as pulchrrrude (i.e.. heourr) might be more easily apprehended m a passage about fashion models thr in one about eternal versus ephemeral qualities.
I
(7)
t
Usefu1ttes.r of previouslv known ~nformmon itr cue u~rlrxrron. Inevit,tbly. hr usefulness of a cue will depend upon the extent to which past knowledge can be brought to bear upon the cue and ,ts relation I(> the unknown vvord. The usefulness of prior informatton \: ‘11depend in large part upon a gtvcn individual’s ability to retrieve the informatmn. to recoPmze Its relevance. and then to apply 11 appropriately.
Decoding G/ rnternu/ conlexl By internal context. multiple morphemes
we refer to the morphemes +vithm a word constituted that combine to give the word its meanmg.
of
Competence model The competence model specifies the particular kinds of context cues that individuals can use to figure out the meanmgs of new words. Because internal context is much more impoverished than is external context. the diversity of kinds of cues is much more restricted (see. e.g.. Johnson and Pearson 1978: O’Rourke 1974). The four kinds of cues constituting our scheme are (I)
Pre/rx cues. Prefix cues generally facilitate decoding of a Occasionally, the prefix has a special meaning or what prefix really is not; in these cases, the perceived cue may (2) Srem cues. Stem cues are present in every word. m the
word’s meaning. appears to be a be deceptive. sense that every
word
has a stem.
multiple (3)
Again,
meanings
Su/fi~
meaning:
such cues may be deceptive
and the wrong
c’rtes. Suffix
cues.
in unusual
too.
cases where
o, in cases where what appears
of a given stem has
one is assigned.
generally
facilitate
the suffix
dec0dir.g
of
a word’s
takes on a11 atypical
to be a suffix
really
meaning.
isn’t. the perceived
cue
may be deceptive. (4)
Inrerucrrtv word
cues. Interactive
parts described
conveyed
cues are formed
above convey
The usefulneLs
of these kinds
Suppose
/WII~~.WVX~~
(set Just and Carpenter
10 most people.
task is to infer
But many
tlon
people
bnowledge derives property
relation
abstract
an interaction
would for
suggest
inferring
Again. internal
we make
parsing.
we
think
Again. unknown
(I)
These
model
Number
of
contextual
claims
from
heat
cues might
Note
on one’s
that this cue
would
element
be a relevant
be combmed
of light
that
this
simple
represents
the
only
represents
kinds
emission
of
(and
to infer
from
cues
unoriginal)
possible
one,
in inferring
meaning to onz’s
embedded.
incltider,
occurrences
appears just Importonce
model
but
provide word
specifies
in which
heated
parsing
classification not
the
a basis
for
of
scheme.
onlv.
plausible
a person
to
meanings.
of
the
the context word.
is likely
relations
it occurs
five variables
but not iaentical
analysis,
a given unknown
(2)
interpreta-
be correct.
and the context
are similar
relative
a reasonable
and the stem: Neither
to the property
probably
performance
word
word’s
the prefix
meaning.
these
refers to heat.
and that the suffix
model the
cues. Our
variables
Moreover.
word
no
rhenno-
by
rhermr+
unfamdiar
and lumrtlesce woulcl draw
emitted
exercise his or her competence
Performance
the prefix
the light
cues
Collectively.
nouns.
can be shown
of the word is probably
:c~ults in some degree of light.
would
it
meaning
word
that is not
the rest of the word.
“to give off light”.
that
inference
contextual
althdough
The
that
rltemm-
betv,een
then,lolunlr,,e.~sc,erIre refers
ObJecls. This
of
know
betwen
that heat typically
from
in Itself that
used to form
of a possible
from
the meaning
1980).
the root /W~IUW.~W is i verb meaning
-e~ce IS often
in combination
in Isolation
of cues in decoding
an example.
that
one’s
when two or even three of the
information
by a given cue considered
mediate
affect
unknown
word.
a previously the usefulness
CI!C useiu!ness.
to those considered
In
for externa!
the
case
of
These
context:
internal
cues are the same on every presentation
However,
one’s, incentive
to be increased
incentive
that
between
that
for a word
to try to figure
to try to figure
of
OUI the
that keeps reappearing
out the meaning
.~f a word
that
once or very few times. o/ :he unknown
Again.
a word
word to undersrumlmg that
is importtint
the context
for understanding
m which II is the context
in which it occurs is more likely io be worth the attention it needs for decontextualization. One can easily skip unimportant words, and often does. As betore, importance can be subdivided into the importance of the unknown word lo the sentence in which it is embedded and the importance of the word to the paragraph in which it is embedded. (3) Deprsrry 01 unknown .vor& If unknown words occur at high density. one may be overwhelmed at the magnitude of the task of figuring out the meanings of the words, and give up this task. Yet, it is possible that the greater the density of unfamiliar words in a passage, the more difficulty the reader Nili have in applying external context cues, and hence the more important will be internal context cues. A high density of unfamiliar wcrds may encourage word-by-word processing and a greater focus on cues internal lo unfamiliar words. This performance variable interacts with the next one to be considered. (4) Density of decomposable unknown words. Because internal decontextuahzation may not be a regularly-used skill in many individuals’ repertoires. individuals may need lo be primed for its use. The presence of multiple decomposable unknown uords car, serve this function, helping the individual become aware that internal dczontextualization is possible and feasible. In this case, the strategy is primed by repeated cues regarding its app!icability. (5) Usefulness of preorous!v known informarron rn cue uttlrrarron. Again. one’s knowledge of words, word cognates. and word parts will play an Important part in internal decontextualization. The sparsity of information provided by such cues (in contrast to external cues) almost guarantees an important role for prior information. To summarize, our theory of learning from con:ext specifies kinds of external and internal cues that Individuals can use m inferring meanings of previously unknown words, and variables that affect how well these cues can be utilized in actual attempted applications. The competence model outlines the full set of variables that (according lo our theory) might be present in a given external or internai context. These variables are not sufficient for describing actual context utilization. however. Our performance model specifies the additional variables that will determine the differeniial application of the set of “competence cues” both across individuals.
Initial
texts
empirical
for a single
individual,
tests of the theory
and
of learning
empirical of these
within
from
a single
text
across
contest
tests of the theory presented tests are encouraging for
m the further
empirical
exploration
some of the data for elsewhere
The
theory
school
of
32
37 distinct
writing
addtttonal
sample
The students’
literary.
within
a given passage. The
which
123
high
length
that
low frequency
passages.
words.
divided and
As a
passages.
among
four
historical.
An
in the hter,Ary style.
for
reqmred
to earlier
following
but not both.
in the various
sctentific.
multtple
occurrences
passage (the sample
illustrates
the experimental
materials.
in context
that
were required
the students
of a single
only a single definition). passages and definitions
ment)
Ltke
in
appearing
words defined
asking
words
as best they cculd each of the low-frequency
(except
to look back
responses.
multtple
tokens
newspaper.
task was IO define
each passage
not permitted
by
125
types. Passages were equally
passage was written
wrthin
191’1).
tested
roughly
I to 4 extremely
of
of the c:ata analysis
more than once. in whtch case it was repeated
of 71 word
styles:
here a ,iummary
details
and Sternberg
was
oi
them from
word
words
current
passages
within
there were a tot,tl
dtfferznt
; Powell
a gtven passage or between
based upon
We present
so far. leaving 198 I
decontextualization
read
could be presented
utthin
rc\ult.
to
embedded
A given word crther
development.
and Sternberg
external
students
contained
and theory
we have collected
(Kaye
and
passage
were their
from
particularly. to define.
word
Students in making
the experi-
the
The
kinds
words
of
to be
are Italicized:
the words
of sufficiently the words hnow
in this passage. low frequency
before.
in advance
but that what
all to-be-defined
to assure that
words
if they had. the subjects
the words
meant
were nouns.
most subjects
and were
had never even seen
were extremely
(oti,rt
means “steam”
mternal
and external
unlikely
and a cerhdh
to IS
a
“visit”). Two
kinds
of data
Ideas
about
main
dependent
the 71 v,ord
learmng
analysis from
variable
tokens. Mean
sought
context.
In
was the quality
Qualities inter-rater
the internal-va!idation of the students’
of definittons
trained
raters.
ratmgs
was used as a defimtion-goodness
reliability
were rated was 0.92.
validation
procedure.
definitions
the
of each of
independently
so an average
score for each word
of our
by three
of the three
for each subject.
These averages were then averaged over subjects to obtain a mean goodness of defimtion rating for each word. The main independent variables were ratings of the number or strength of the occurrences of our competence apd performance cues (with the exact nature of the rating depending upon the cue) with respect to their roles in deciphering the meaning of each low-frequency word token. Ratings for competence cues were provided by one of the investigators (J.S.P.). who was blind to the data at the time. Ratings for performance cues were provided by 128 students from the same high school as the students who actually defined the words; there was no overlap in subjects between the t\vo tasks. In the external-validation procedure, the ratings of definition got>dness served as independent variables to predict students’ score5 on standardlred tests of intelligence, vocabulary. and reading comprehension. These score\ acre available from school files. Prediction was accomplished via simple corre atlon. The logic of our procedure was based on the belief that individl;al differences in measured verbal intelligence and comprehension derive from mdi :Idual differences in individuals’ abihties to ascertain meanmgs of words presented m context. Obviously, though, our experimental and correlational procedurC:s did not permit establishment of a directed causal link. and any obtamed correlations could equally well be interpreted as suggesting that the ability to tlef:ne words in context is derivative from verbal abihties. or as suggestmg that decontextualization ability and verbal ability are both dependent upon some third. higher-order variable. For the internal-validation procedure, separate regressions Uere computed for each 01 the four passage styles. because preliminary ana!ysea shoned the relevant variables and the regression weights to differ frcm one style to another. We used a stepwise procedure in which he allowed only three variables plus a regression intercept to enter into our fmal model.\. The decision to limit the number of variables was made on the basis of our Judgment of the degree of refinement of our data. and in the hope ol minimizing the risks of capitalization upon chance that inhere m stepuise regression. Because of multicollinearity (correlation among) Independent mariables. it was not possible to make strong inferences regarding the “true” subsets of variables that were differentmlly relevant from one passage style to the next. The correlation between predicted and observed goodness ratings was 0.92 for literary passages. 0.74 for newspaper passages. 0.85 for science pas,:vzes. and 0.77 for history passages. All of these values were sta!lstlcall> bgmflL rnt. We conclude on the basis of these data that subsets of our model can prove jr good prediction of the goodness-of-definition data. although we certainlq do not believe that our model accounts for all of the reliable variance. Indsred. the square roots of the internal-consistency reliability coefficients (based on all possible split halves of sub;ects) for our four data sets. which place upper limits on the values of R. were all 0.98 or above. showing that therr: was a
considerable
amount
fits of the model our
initial
of reliable
variance
MI
accounted
subsets seemed sufficiently
attempts
to understand
for
high to merit
differential
word
Nevertheless.
optimism
difficulty
the
regarding
in learning
from
context. For the external-validation of
all written
with
scores
on
tvpes. Not cantly
Correlations
scores (mean
with
the
for
only were these obtained
greater
correlation
test for
that
psychometric
verbal
in a reasonable
is fairly
between
tests (see, e.g.. Hunt
position
somethmg
tests of verbal
typical
to conclude
closely
intelhgence
akin
that
in general.
the individual
the
quite
passage
statistically
greater
measures
signifi-
than 0.30.
of cognitive
et al. 1975).
to that
passage
:scores were
coefficients
than 0, but they were also significantly
correlated
the various
and 0.65 with reading
psychometric
scores computed
rated goodness
task were
tests. Correlations
learning.from-context
of correlation
measuring
subjects’
the learning-from-context
were 0.62 with IQ. 0.56 with vocabulary.
comprehension. for
on
the psychometric
types combined similar
procedure,
definitions)
Thus,
a level
tasks and
we believe we are
learning-from-context
which
is measured
and of verbal
task
by stand
comprehension
is
.dized
in particu-
lar.
Our
goal
extent
m our
to which
study
of very
frequently
used prefixes
students while
students’ in defining
tested itself,
we plan
tested
I08
equally
our
students,
of
!5 different prefix
used Latin
and each
received
halt of the wcrds
use internal
context
our study
than
the theory
in secondary
school
(approxi-
I I)
of whom
rather
research.
58 were
8, IO. and
1of
and
was exposed
IS commonlv
50 were
to 58 prefixed
used Latin
prefixes
stems. Smce there were I5 different
prefixes
Stem appeared
Pplf of the words
between
performance
in this area (as described
stems, there were a total of 30 different
were of very low frequency
or stem)
Thus.
Each subject
that were selected each to contain
of
these
(prefix
use such context.
to be applicable,
grades
whether
relationships
wherher individuals
whom
the
the correct
knowledge
and their actual
state of research
theory
among
constituents
of such words
how individuals
derive their
to determine
We also examined
at a state university.
I of IS commonly
sought
to test in subsequent
balanced
and
other
for
could
the basis of
of the words’
the present
was to determine
students
on
We
the words.
in detail
and Each
to either
them. Given
undergraduates
words
stems.
knowledge
a prerequisite
mately words
words
we felt there is a need to know examining which
We
and
to define
metacognitive
earlier).
decontextualization and college
low-frequency
were attending
attempting
before
of internal
secondary-school
defimtions
in fror?
and of 2-3
in a multiple-choice
in a word-rating
two
individual
word
to six different
syllables
in length.
word-definitions
task. Words
presented
parts.
words.
All
Each subject task and
the
in each of the
two tasks were counter-balanced across subjects. In the word-defmitions task, each word was paired with four possible definitions, one of which was correct and three of which were incorrect. One of t’.te incorrect definitions retained the meaning of the prefix only, one retained meaning of the stem only, and one retained the meanmg of nei!her the prefix nor the stem. An example of such a prohlem is
tk
EXSECT (a) to cut out
(totally
correct)
(b) to throw out
(prefix
only correct)
(c) to cut against
(stem only correct)
(d) to throw against
(totally
incorrect)
In the (metacognitive) word-rating task, each word was paired wrth four questions: (a) How familiar is the word? (b) How easily can you define the word? (c) How similar is the word to another word you have sern or heard? (d) How similar is the word to another word you can define? Subjects responded by circling numbers arrayed con a 7-point scale, with more positive responses to the questions answered in terms of higher-numbered values on the scale. All subjects were also asked to rate all 3@ word parts for their meaningfulness (i.e.. familiarity of meaning). This task, which occurred at the end of the experiment. also involved a 7-point rating scale. A hierarchical muhiple regression procedure was used to predict scotes both on the learning-from-context cognitive task and on the metacognitive ratings of words and word parts. In such a procedure. sets of indepenoent variables are entered into the regression in a fixed order and in successive steps. The variables entered at the first level of the hierarchy were always “dummy variables” for age. test form (i.e., which set of test words was received in which task), and age X test form. Thts level was used IO control for the effects of those variables that might be expected to affect test performance. but that Here not directly relevant to the question of how subjects answered the test Items or made their ratings. The particular independent variables entered at the second level of the hierarchy were the theoretically relevant ones. and varied from one regression to the next. Consider, for example, the question of whether subjects are using prefixes in order to figure out word meanings. If they are. then performance on a word with a given prefix, such as ex in the abo’e example. should be better predicted by performance on other words sharing this prefix than by performance on words not sharing this prefix. If subjects are not using prefixes, then performance on words sharing the same prefix should be no better predictors of performanze on the given word than shouid performance on words not sharing the same prefix. The same logic applier for word stems. Thus, a typical independent variable at the second level of analysts would be
performance on other words sharing the same prefix (for determining whether subjects IJSCZprefixes m figuring out word meanings) or sharing the bame stem (for determining whether subjects use stems in figuring out word meanings). The variable entered at the third level of the hierarchical regression was an interaction The
term between
results
suggested
w’err able to use internal corrrlation wcrc
between
generally
school
of then
determining
and observed
both
cognitive
that
ranged
suggested
Interesttngly.
the much
larger
number
We conclude gives greater tratnmg tton
at the lower
meanings. in both
verbal
word
fuller
understandrng to tram
theory
subtheory attempt
and
of learning
that
from
we believe
various divides
metacomponents. berg 1980). last kind.
younger
the preftx
students
that
of 0.63.
modifying than
knowlto
used interappear
not to
there is a real need for
for decontextualization
program
such a theory
that
combines
et al. in
is that a theory of
part
of a full
can pave
of
instruc-
(see Sternberg
the way
of verbal comprehension
theory both
of
for
a
and for a means by
their comprehension.
context
comprehension
we have presented
much
comprehension. structure
This
to the field
components.
Our own work on learnmg
of
represents of our
context
the work
framework
of verbal
processes
into
and learning
from context
only
article.
a we
by proposing currently
a
being
serves to organize
that may help eluctdate
comprehension
performance
In this part
into a broader
encompasses
contributions verbal
that
in general.
and research
and data into a coherent
framework
values of
focus for
framework for rerbal comprehension
of verbal
ships among
was
was better
is an important
to improve
done on the topic of verbal theory
That
a traming
context
that
to place our theory
framework
with
decontextualization
of the na!ure
individuals
A componential
The
from
had
knowledge
a median
from our own data and that of others
comprehension.
which
with
high
in the language.
utilization
meanings
for
students
at least. spontaneously
levels m word-part and Internal
not
result may be attributable
arguments
press). Our concluston learnmg
This
(1974)
We have proposed
external
to 0.78
meant.
sample.
but
and college
of prefixes
students.
force to O’Rourhe’s
students
stem was the central
of stems than of prefixes
college-level
of R (the
vice versa). Significant
0.53
words
to ftgure out word meanings.
nal context
word
that
students.
Values
metacognitive
the word
knowledge
edge of stems. at least for our word
school
students
(and
from
that
each of the vartous
this stem meaning.
not
college
is. their
performance
regressions
what
for
high
meanings.
scores on each of the test items)
high school
knowledge,
of results
but
to help infer word
significant
However.
R for the various pattern
students,
college
predicted
metacognitive
predictive
The
at the first and second levels of analysis.
that
statistically
students.
accurate
the variables
context
interrelation-
comprehension. three general components
The kinds: (Stern-
addresses processes of the
Metacomponents are higher-order control processes ning and decision-making in cognitive performance
used for executive planin gene:al. and verbal
performance in particular. Metacomponents functionmg colle:tively are what have sometimes been referred to as the “e.recutive” or the “homunculus”. Several different metacomponents have heen Identified in previous work (Sternberg 1979. 1980; see also Brown 1978. for a related pc,lnt of view): (a) deciding on the nature of the problem, (b) de:idmg on proce’.\es for solvmg a problem, (c) deciding on a strategy for comhmtng these proccbbes. (d) decidmg on how to allocate processing resources. (e) ceciding on ho\T, to rrpresent the information upon which processes and stratee;ies act. and (I) momtorting one’s solution of a given problem. We view several lines of n:search on verbal comprehension processes as addressing metacomponential ilving rl problem .md deciding on a strategy for combining these proces\es (‘XC) are addressed by information-processing models of reading such as those of Fre.ierlksrn (1980). Just and Carpenter (1980). and Kmtsch and van DiJk (I 378). w hiL h attempt to account for real-time processing of information while r:admg is tahlng place. These models specify with considerable precision the dl,clsions that the reader has to make in order to process incoming text. Te>t-processing computer programs (see. e.g.. Schank 1975; Schank and Abelson 1977) also specify these decisions in great detail for the particular classes of text each of them is able to handle. The metacomponent of deciding on how to allocate procez,slng resources during reading (d) seems to be addressed to at least some txtent both by eye-movement methodologleb (e.g.. Carpenter and Just 1977: Just and Carpenter 1980: McConkie 1976: Rayner 1978) and lry dual-tdsk methouoiogies (e.g., Hunt and Lansman 1982). The former mt thod is able to suggest where in a given text a reader is placing his or her attention (or proces,ing resources more generally) at a given time during reading. The latter method is able to suggest the extent to which processing resourcr:s are being drawn upon at a given time during reading. Each method. then. addresses in a different way
the question of how readers budget their nme and efforts. The metacomponent of deciding on how to represent information mentally (e) can he seen as the topic of what has become a large volume of research on the representation of meaning (e.g.. Anderson 1976; Hampton 1979: Katz 1972: Kcil 1979; Kintsch 1974: Medin in press: Rosch 1978; Smith et al. 1974; see Smith and Medm 1981. for a review of this literature). Some of these models apply most clearly at the level of individual words, others at the level of semantic categories. and others at the level of sentences. In this research, alternattve models of meaning. such as necessary-and-sufficient (defining) feature models. prototypical (characteristic) feature models, and models based upon various ways of combinmg defining and characteristic features. are tested for their abtlity to account for different types of language-understanding data. These representations are then assumed to be accessed during the reading process. Fmally. the metacomponent of solution monitoring (f) can be seen as addressed by work such as that of Collins and Smith (1982). Flavell (1981). and Markman (1977. 1979. 1981) on comprehension monitoring. This work is particularly relevant IO our present concerns. because it deals with the role of context in understandmg whole passages, and not Just single words. Markman has suggested some signals people can use to detect failure to compre-
(1981)
hend verbal materials. These signals use surrounding context, and constitute part of what might be construed as means of effective comprehension monitoring. One signal is perceived absence of structure. If one finds it difficult or impossible IO tmpose a structure on verbal materials. then this failure should serve as a stgnal that the intormation is not well understood. A second signal is multtple perceived structures. In the sentences “John and Bill went to the store. He bought some bread”. at least two possible structures can be imposed. Ggnalling the difficulty one has in understanding the message the writer intended to communicate. A thtrd signal is the discovery of inconsistencies. which has been the topic of Markman’s recent research (e.g.. Markman 1977. 1979). Inconsistencies may indicate mis-structuring of information comprehended earlier, such that the imposed structure does not work for informanon that is comprehended later. A fourth signal is inability to use structure to formulate expectations. Except in the case of highly novel materials, if one cannot formulate plausible expectations about what is to come next on the basis of what has come already. this failure may indicate a lack of comprehension of the text.
Performance components are the processes used in the execution of various strategies for task performance. They are the processes that the metacomponents decide to execute to attain the goal of task completion. Many of the
research programs described in the preceding section on metacomponents deal with performance components as well. For example. the reading models of Frederiksen (1980). Juct and Carpenter (1980). and Kintsch and van DiJk (1978) address not only questions of how and on what bases subjects decide what to do, but also what, exactly, they do as they re,ld. Computer models also must address both kinds of issues. But the researc,h that has most directly addressed the question of the role of performance components in verbal comprehension is that which has emerged from the cognitive-correlates tradition of research initiated by Earl Hunt (e.g., Hunt et al. 1975; Hunt 1978) and followed up by a number of other investigators (e.g., Jackson and McClelland 1979; Keating and Bobbitt 1978; see also Perfetti and Lesgold 1977. for a elated approach). According to Hunt (1978). two types of processes underlie verbal comknowledge-based processes and mechanistic prehension ability -
I
(information-free) processes. which are what we refer to as performance components. Hunt’s approach has emphasized the latter kind of process. Hunt et al. (1975) studied three aspects of what they celled “current information processing” that they believed to be key determinants of individual differences in developed verbal ability: (a) rensnwcy of overlearned codes lo arousal by mcommg wth which temporal lags can be assigned. and hence order the speed with whwh the mternal reprerentac~ons m STM memory for events occurrm~ over mmutes) can he created.
stmwIus ~~~~ormauon. (b) the accuracy mformatlon can be processed. and (c) and mtermedlate wm memory (KM. mtegraled. dnd altered. (1975. 197)
The basic hypothesis motivating their work IS that Intelligence tests indirect11 measure these information-processing skills by directly measuring the products of these skills. For example, higher vocabulary as measured by a vocabulary test can be interpreted as deriving from increased information-processing skills that earlier in life er.abled acquisition of this vocabulary: conversely. reduced vocabulary reflects earher deficits in hese skills. In a typical experiment, subjects will be presented the Posner and Mitchell (1967) letter-matching task. The task comprises two experimental conditions. a physical-match condition and a name-match condition. In the physical-match condition, subjects are presented with pairs of letters that either are or are not physical matches (e.g., “AA” or “bb” versus “Aa” OI “Ba”). In the name-match condition, subjects are presented with pairs of letters that either are or are not name matches (e.g.. “Aa”, “BB”. or “bB” versus “Ab”. “ba”. or “bA”). Subjects must Identify the letter pair either as a physical match (or mismatch) or as a name match (or mismatch) as rapidly as possible. The typical finding in these experiments is that the difference between mean name match and physical match times for each of a group of subjects is correlated about -0.3 with scores on tests of verbal ability. The theore:ical interpretation of this
I
finding
is that
physical
speed of lexical
match
The tation
described
is a matter
Sternberg tlons
We
are rather
weak
1981;
Ilogaboam
and
remain
concerned
that
abundant
ha\e
u\ed
.Idequate
u\ed.
heen done
.md
ferencc
discriminant
perceptual
perceptual
abilities
at lea\t
to us that
sion. and what
Learning
components
component5 retrieval
it 1s potentially
IS
prefix
combined
an afflxed learning verbal part
word
and transfer
from
with
context
of this article
third
parts
of
Obviously. central
meaning directly
other
Johmon
linrs
and what research
of learning
taxonomies 1977;
of
Rieger
(e.g..
Collins
to new contexts for
of a stem to Infer unknown.
further
this
of on
components
in
in the first
literature
both
and here.
in this literature
as of
and believe how learning
comprehension.
see as especially on story
Lebowitz
provide
of a
literature
comprehension,
relevant grammars
1980;
1975: Stein and Glenn
and taxonom:es
meaning
to it in the second
studied
are those
Punc-
the meaning
The
understanding
et al. 1975;
in
processes
example.
this literature
discuss
of verbal
transfer
in a text as a basis for
its role IS in verbal
components
comprehen-
and
the known
reviewed
we
more seem.
components
as when.
of learning
that
with
of new information.
the roles of learning
not
1975: Rumelhart
al. 1979). These grammars
retention.
or when
has been made toward
takes place.
role in verbal
our own contribution will
m the understanding
progress
understanding inference
we
we view the components
context
Two
addresses
shared
this role is.
IS recognized
meaning
it is
the obtained
task would
as learning
word.
speed than
variance
of old information
but since we have article.
\om(
was previously
and presented
the
importance
considerable from
Information
are
this dif-
may be the case here. it seems
of information.
the known
comprehension.
to perceptual
acquisition.
We identify
have not
ability.
et al. 1981). Thus.
at least in part
is Just what
of an unknown
whose
related
as verbal
are used in the storage
relevant.
the meaning
such procedures
called
1980)
or stative-descriptive
Inferring
When
as well
access plays
ubed in the decontextualization tional
procedures.
Blft whatever
(separately
in Sternherg
of old information.
which
difference
that the letter-matching
to be clarified
a rather
match
may reflect
speed of lexical
literatures
provide
physical
minus
IO mcasurt.
remains
and
most of the studies
et al. in press: Willis
of the kind
on the surface.
the personality go).
1978: correla-
is that
is considered
abihty
Pellegrino 0.3~level
concern
more strongly
(Lansman
with verbal
and
correlations
validation
speed
seems to be much
to verbal ability
the ahllities
A further
on the name
correlation
hkely
in both
low as ability
Inference.
ability.
but its interpre-
(Carroll
they are ratha’
mmus
level of verbal
replicable,
have been and
hasis for causal
that
hy the name match
above seems to be widely
of dispute
1981).
(rndced.
access, as measured
time. IS m some sense causal of acquired
finding
to
an
and
Mandler
on and
1977; Warren
schemes for understanding
et
how context is crganized, and the ways in which it can and cannot facilitate learning. In order to understand learning from context, it seems reasonable that first one should have a structural model for context: we vtew these grammars and inference schemes as providing alternative structural models for the global contexts from which so much of learning takes place. Inlerucrrons
of componenrs
A recursive model for how the various kinds of components described above interact both in adult learning and in children’s cognitive development has been provided elsewhere in some detail (Sternberg 1980, 198la). and so will not be provided here. The basic idea is that metacomponents directly activate performance and learning components, which in turn provide direct feedback to the metacomponents about the state of the organism. Performance and learning components can activate each other indirectly, but only through the mediation of the metacomponents. Similarly, feedback between performance and learning components is only indirect, since it must first pass through the mediating influence of the metacomponents. Concludmg
remarks
To summarize, we have attempted to describe a componential framework for interrelating a variety of theoretical and empirical endeavors in the field of verbal comprehension. We make no claim that our assignment of -esearch enterprises to aspects of our framework is uniquely correct. Nor do we claim that our framework is the only one possible or even the best one possible. But we believe that what we have attempted, at least so far. is a unique endeavor. and therefore may at least serve as a basis upon which others can build in the attempt
to understand
Psychological
verbal
and educational
comprehenston
in a unified
and systemattc
way.
implications
We have attempted in this article to present an overview of the nature of verbal comprehension, with particular attention to the role of learntng from context in verbal processing. We conclude our overview with a discussion of what we see as some of the implications of the theory and research we have reviewed.
Absence o/ rheones oj verbal
comprehensron
Although there exist a number of subtheories of verbal comprehension that deal with phenomena such as learning from context. lexical access. comprehen-
sion monitoring,
and real-time
of verbal
comprehension
absence:
we are unaware
phenomena of
this
aubtheortes
suggests
and
data
such a theoretical
one
from and
A unified
theory
considered
a unitary
of there
(see. e.g.. Cattell 1971).
simtlarly Horn
and
flutd
abilities
1971:
Horn
ahthties. attempts
respecttvely.
(e.g..
Pellegrmo
19Xlb:
Whitely
1980).
of for verbal
ia the most attention. sion
To date. sion
viewed
differences
here.
approach
and
Mttchell 198lt’) tasks
1980:
on simple
on complex
seeks to understand in
terms
of
the
several
1980.
that we are
Although
this theory
deserves someone
pre-
of fhud
Sternberg
very
aspects of verbal for
of a
Cattell
serious
comprehento extend
that encompasses
an
even most of
we have discussed.
g -ferbal
extensive Hunt’s
ability
literature
ability theory
1981b).
and
source
of and
by which
(Hunt
task considered performance
and
cn complex
processes
that
of
individual
for studying
approach an attempt
An
Hunt
comprehen-
method
(see
Pel-
is made
to
tasks in terms of
tasks (such earlier). Glaser
1978:
on verbal
information-proccessing
(Pellegrino
component
skills.
attempt
(1978).
1938:
as crystallized
been
1981;
therefore
remains
information-processing
appr jach
Thurstone
theories
his “cognitive-correlates”
Sternberg
1967. letter-matching
tive-components”
it
to understandin the fairly
reflects
performance
performance
and
a new theory
performance.
legrino
Glaser
of Hunt
informatton-processing
in verbal
comprehension
understand
Thus.
phenomena
has dominated
verbal
have
1980.
several of the many
or to develop
as an
that
one we have.
only
considered
Hunt’s
SnoK
is strongly
the existence
constructs
there
be
case. wh;t
comprehension
and mferential
whereas 1980;
comprehension,
the verbal-comprehension
ct al. 1975)
1966:
suggested
informatmn-processing
and Glaser
broad-rangmg
subtheory
But
could is the
way,
of verbal
Cdttell
reasonrng
unified
such
there has been only one comparable
it deals with
we have
extsting
and
goal were there
for suggesting
in a unifted
has stnlngly
of abstract
that
useful
construct
Horn
are related
comprehension
to these two psychologrcal
to establish
ahiltties
aware
1968:
of
provide
how compre-
context
he a futile
helteve
together
a unified
from
verbal
not
it does not
in each.
to he quite
cluster
same evidence
construct
do
tends
but
part
integration
for example.
would
in whtch
We
which
have referred
Itminary
\ense
theories by their of verbal
in the preceding
theoretical
of words
involved
comprehension
in fact being
The
untfied
and
meanings
construct.
evidence.
the
of perspectives.
processing
reading,
the range
provided
for
It does not indicate,
of verbal
hands of psychological
Vernon
basis
psychological
Factor-analyttc suggestrve
framework
learning
during
arc most conspicuous
encompasses
The
T vartety
tn terms of the information
no meaningful
that
possible
integration.
monitoring
processing
construct
of any theory
we have discussed.
article
hension
information
a., a unified
as the Posner alternative
1980;
Sternberg
and
“cogni1977.
informatton-processing
contribute
to
those
tasks,
without recourse to simple tasks. The learning-from-context approach we and others (e.g., Daalen-Kapteijns and Elshout-Mohr 1981; Jensen 1980: Werner and Kaplan 1952; Marshalek 1981) have advocated provides an alternative way of studying verbal comprehension that is more in line with the cognitivecomponents approach that we happen to prefer for the study of verbal comprehension. We certainly do not propose that this approach should replace the cognitive-correlates approach in the verbal domain, but we vieg, them as useful complemem s to each other. The nature of leart~rngfrom context The research wt: have reviewed and done ourselves on learning from represents what WI: see as only a start toward understanding such learning a?d its role in verbal ability. The use of internal context. in particular, 5eems to be poorly understooo, and we are aware of no work at all that has attempted to account for how internal and external contexts are used in an integrated way in individuals’ attempts to infer word meanings. Our own research on external context has stressed the use of context in learning meanings of single words embedded m large numbers of words whose meamngs are known. Viewed from this perspective, research on learning from context may help one understand the origins of indrvidual differences m the acquisition of new vocabulary m a language with which one IS already quite familiar. We have planned complementary research on the use of context for inferring gist from passages in which most of the words’ meanings are unknown. We view this paradigm as addressing questtons regarding the development of vocabulary and reading comprehension in the learning of new (first or second) languages. In such learmng. one is first confronted wrth texts where most of the words are unknown. ar.d only gradually does one begin to confront texts such as those confronted in our initial experiments. where most of the words are already known. If one views the generalized paradigm of learning both new words and how to extr:act gist from context as measuring antecedents of both vocabulary and reading comprehension skills. one may be able to account in information-process ng terms for why vocabulary and reading comp:chension scores tend to be very highly correlated with e,lch bther (Brown et al. 1976). Educattonal mues Predrcrron As often happens in the psychometrtc testing busmess. practice precedes theory. This seemingly reverse temporal sequence has the advantage of grving us useful technology before we fully understand that technology, at the same time that theoretical
it has the disadvantage Justification will never
of enabling us to use tests for whtch a be found. In the case of using context in
reading.
the College
Board
has been administering
a reading
test that is based
entirely upon a cloze procedure (i.e.. insertton aof missing words into an otherwise intact text). This instrument, the Degrees of Reading Power (DRP) test, is already mandated in the State of New York. The test requires students to infer from context what word best ~IIS into a given text at a given place. and thus draws, we believe. on many of the same skills as the vocabulary-from-ccr,*text task does. As noted earlier. Cook et al. (1963) have actually constructed a test that measures ability to learn vocabulary from context, although this test does not seem to have been w?:ly adopted in the United States. To our hnowledge. there are no standardized tests that measure students’ abilities IO use Internal context. Our theoretic21 position and the data we have collected lead us IO believe that tests that measure individuals’ abilities to infer meanings of words from both internal and external context could provide reliable and valid measures of one important aspect of verbal comprehension. Hence, we would
hke to see stepped
up research
on such instruments.
Our theory of learning from context specifies a number of kinds of cues that individuals can use to infer meanings of words. Cue utilization is theorized to he affected both by an individual’s competence and his or her ability to bring this competence to bear upon specific problems. Because we believe that most vocabulary is in fact learned from context. we see considerable benefit in the diagnosis of individuals’ abilities to use the various cues specified by our theory (or possibly the theories of others). At present, we are not able to model Individual-subJect data reliably enough to permit valid diagnosis at the individual level. But there may be other ways of attaining such diagnosis, such as simply inqlliring of subjects as to whether they are aware of the usefulness of certain cues: and with improved measuring procedures, it might become more feasible to conduct individual diagnoses using our methods (Powell and Sternberg 1982). Such diagnoses would be consistent with the kind of process-based mental diagnosis that we believe will represent one wave of progress in future psychological assessment (Sternberg 1979, 1981 b). Trarnrng Learning from context has been tried before as a means of teaching vocabulary (see Gipe 1979; Johnson and Pearson 1978; O’Rourke 1974: Shumway and Forbes in press), but the existing programs have been short on theory. Typically. a student is taught words in context without being taught a systematic way for using the context to infer word meanings. Thus, although Levin (1981) has claimed that the keyword method (Atkinson 1975), a mnemonic method making heavy use of in.eractive imagery, is superior to the context method, our own view is that unlike the keyword method. the context method hasn’t really been tried m a theoretically informed way. We are
currently context 1982). students At
planning
a large-scale
to infer word We believe
meanings
that
training
in a way that will permit
present,
training.
like
of course. that
on
we have prediction
program
(Sternberg
the kinds
for teaching
their efficient no data and
students
et al. in press: Sternberg
of cues we have identtfied utilization
to support
diagnosis.
in ordinary
this claim.
remains
to use
and Powell
can be taught Research
to be done
to
reading. in
on the
future.
References Amea. W.S. 1966 The development of il cl.~a\~f~cat~on scheme of ~omex~u~l ad\ Rradmg Re\earch Quarterly 2: >7-82 Andwon. J R. 1976. Language, memory. and thought. Hdlcdale. NJ Erlbsum. Anderson. R.C. end W.B. Blddle. 1975. ‘On askmg people que>tlon\ about what they dre reddmg‘ In: G.H. Bower (ed ). The psychology of learnmg and mowat~on. 101 9. Nea York Acrdemw Press Anderson. R C. and P. Freebody. 1979 Vocabulary knoulrdge Techmcnl Report No. 136 Champargn. IL: Center for the Study of Readmg. Umvw~ty of Illmo~r. Atkmson. R.C 1975 Mnemotechnu m second-language !c.rxng Amencdn Pawhologat 30 X21-828. Bldck. J.B l9Rl. ‘The effects of readmg purpose on memory :&jr text’ .,I A Bdddele! and J. Long (eds.). Altenl~on and performance. IX Hdlsdale. NJ: Erlbaum. Brown. A.L. 1978. ‘Knowmg when, uhere. dnd hou to remember: a problem of metacognmon In R Claw (ed.). Advances I” matrucuonal p\>chology. vol Hdladale. NJ. Erlbaum Brown. J.1, M.J Nelson and EC Den>. 1976. Examiner’s manual the Nelson-Dam? readmg test. Boston: Hougl ‘~ffbrl. Carpenter. P.A. md M.A. Just. 1977. ‘Readlog comprehension ti e>es see 11’ In M A JU\I and P.A. Carpenter (eds ). Cogmwe processes m comprehrnslon Hdl\dale, NJ. Erlbdum Carroll. J.B. 1964. Worda. meanmgs and concepts Harvard Educational Re\~eu 34 202 Cxroll. J.B. 1976. ‘Psychomcrnc tests as cognmve tasks. a neu ‘*structure of m~rllect” In L B. Resnlck (ed.). The nature of mtelbgence. Hlllsdalz. NJ: Erlbaum. Carrolll J.B. l9gl. Ablbcy and task dllflculty I” cogmhve psychology Educrllonal Researcher 10. 11-21. CatelI. R.P. 1971. Abdwes lhew structure. growth. dnd xt,on Boston Houghton-M,ffbn Cauell. R B ano A.K.S. Cattell 1963. Test of g. culture far. scale 3. Chrmpalgn. IL Inat~~uts for Personably and Ablbty Testrng. Collms. A. and E.E. Smith. 1982. ‘Teachmg the process of wdmg comprehenblon‘. In D K Deuerman and R.J. Sternberg teds.). Hou od hou much cd” mtelltgence be mcredsrd? Norwood. NJ: Ablex. Colbns. A.. E.H Warnock. N A~zllo and M.L. Mdler. 1975. ‘Reaonmg from mcomplrre knowledge’. In: D. Bobrow and t.. Colbns (eds.). Repreremat~on and understdndmg studlr, III cognmve saeoce. New York. Academic Press. Cook. J.M.. A.W. Helm and K.P. Watts. 1963. The word-m-context: d neu t\pc of terhal reasonmg te:.,. Brwsh Journal of Psychology 54: 227-237. Cralk. F.I.M. a.ld R.S. Lockhart. 1972. Levels of procesamg. a framwork for memorv research. Journal of :‘erhal Learnmg and Verbal Behdwor I I: 671-684. D&de~~-Kapte~~os. M.M. van and M. Elshout-Mohr. 1981. The acqwant~on of word meanmgs .I) .I cogn,t,ve learrung process Journal of Verbal Learnrng and Verbal Behwor 20. 386-399
I
i
Flavell. J.H. 1981. ‘Cogmwe skdls. New York: Fredenksen.
J.R.
through
1980.
learnmg.
In: W.P. Dlckwn
(cd ). Chddren’s
oral communlcatmn
Press.
‘Component
chronometnc
Aptitude.
monltormg’
Academic
analysis’.
bkdls In:
and instruction:
m
R.E.
readmg:
Snow.
cognitwe
measurement
P.-A.
Fedt:nco
proce.rr analyse!.
of and
mdiwdual W.E.
of aptllude.
dnfferences
Montague
~0’.
feds.).
I. Hlllsdale.
NJ:
Erlbaum. Glpe.
J. 1979.
lnvesugatmg
techmques
for teachmg
wxd
meanings.
Rradmg
York.
McGraw-Hdl.
Resexch
Quarterly
14, 624-644. Gwlford.
J.P.
Hampton.
Helm.
A
1967.
J A.
Verbal
1979.
Behawor 1970
The
nature
of human
Polymorphous
mlelbgence.
concepts
m
New
semanuc
memory. Journal of Verbal Learnmg and
18. 441-461.
lnlell~gence
and
personab~y:
thelr
~(sbes~ment and
relalwwhlp.
Hatmondworth:
Pengum. Hogahoam.
T.W
end J.W
Pellegrmo
1978.
Huntmg
for mdwdual
differences:
verbal
proceasmg of p~clurer and words. Memory and Co&:nwm 6: l89- 193. K J . A L. Glass and W.A Mah. 1976 Morphologncal structure and semanw Journal of Verbal Learnmg and Verbal @ehwor IS: 235-247.
ababty
and
~emanuc Holyoak.
re~neval.
Horn. J.L. 1968 Organlzatmn of ab11111es and the development of mtelbgence. Psychologwd Re\,ew 75, 242-259. Horn. J L. and R B Caltell. i966. Refmemenl and &I of thb theory of flud and cryualhzed general mrelhgrnce. Journal of Educatmnal Psychology 57, 253-270. Hunt. E.B. 1978. Mechamcs of verbal abdmy. Paychologncal Rewew 85: 109-130. Hunt. E and M. Lansmas. 1982. ‘Indwdual differences m a~temmn’. In. R.J. Sternberg (ed.). Advance\ m thd psychology of human mtelbgence. vol. I. Hdlsdale: NJ: Erlbaum. Hunt. E. C. Lunneborg and J. Lewis. 1975. What does II mean IO be high verbal? Cogmwe Psychology 7: 194-227. Jackson. M.D. and J L. McClelland. 1979. Processnag delermmanla of readmg speed. Journal of Expenmenral Psychology: General 108: l5l- IRI. Jensen. A.R 1980. Bus m mental testing. New York: Free Press. Johnson. D.D. and P.D. Pearson. 1978. Teaching readmg vocabulary. New York: Holt. Just. M.A. and P.A. Carpemer. 1980. A theory of readmg: from eye fixatmns to compreherwon. Psychological Review 87: 329-.354. Katz. J.J 1972. Semanuc theory. New York: Harper & Row. Kaye. D.B. and R.J Sternberg. 1981. ‘he development of lexical decomposihon ablbty. Unpubhshed manuscript. Kealmg. D.P. and B.L. Bobbnt. 1978. Indwdual and developmental differences m cogmweprocessmg components of mental abdlty. Chdd Development 49: 155-167. Ked. F.C. 1979. Semantic and conceptual development. Cambndge. MA: Harvard Universnty Press Ked. F.C. l%la. Conslram~s on knowledge and cogmuve development. Psych&g& Rewew 88: 197-22.7. Ked. F.C. 198lb. Semantic mferencer and the acqwatmn of word meanmg. Unpubbshed manuscript. Kmtsch. Vt. 1974. The representation of meanmg m memory. Hillsdale, NJ: Erlbaum. Kmtsch. W. and T.A. van DIJk. 1978. Toward a model of text comprehension and productmn. Psych&g& Rewew 85: 363-394. Lansman. M.. G. Donaldson. E. Hunt and S. Yantls. In press. Abnbly factors and cogmtive processes. Inrelbgence. Lebowz. M. 1980. Generabzatlon and memory m an mtegrated understanding system. Research Report 186. New rlaven. CT: Department of Computer Saence, Yale Universtty. Levm. J.R. 1981. i‘he mnemomc ’80s: keywords in the classrwm. Educational Psychologist 16: 65-82.
Mandler, J.M. and N.S. Johnson. 1977. Remembrance of thmgs parsed: \tory strwture and recall. Cognitive Psychology 9: l-151. Manehs. L. and D.A. Tharp. 1977. The proceswng of afluxed words. Memory and Cogmt~on 5. 690-695. Markman. E.M. 1977. Realizmg that you don’t understand: a prehmmary mvewgatlon. Chdd Development 48: 986-992. Markman. E.M. 1979. R&zing that you don’t understand: elementary school chddren’\ wdreness of mconslstencies. Chdd Development 50: 643-655. Markman. EM. 1981. ‘Comprehension momtormg’. In: W.P. DIckson fed.). Chddren’c oral
II
communication skills. New York: Academic Press. Marsh&k. B. 1981. Trait and process aspects of vocabulary knowledge and verbal ablhty (NRl54-376 ONR Techmcal Report No. 15.) Stanford. CA: School of Educaclon. Stanford U”lWStty.
Matarauo. J.D. 1972. Wechsler’s mea urement and appraisal of adult mtelhgace. Baltlmorc Wdhams & Wdkins. McConkie. G.W. 1976. ‘The use of eye-wovement data I” determmmg the perceptual \p.m I” read”@. In: R.A. Monty and J.W. Senders feds.). Eye movements and p\ychologlcal procew\ Hdlsdale. NJ: Erlbaum. McCullough. CM. 1958. Coctext ads I” readmg. Readmg Teacher I I: 225-229. Medm. D.L. I” press. ‘Structural prmapleh I” calegonzation’. In. B Shrpp dnd T Tlghr (eds ,. Development of perception and cognltmn. Hillsdale, NJ. Erlbaum. Mdler, G.A. and P.N. Johnson-Laud. 1976 Language and perceptton. Cambndge. MA Harard University Press. Murrell. G.A. and J. Morton. 1974. Word recognition and morphrmlc structure. Journal of Experimental Psychology 102: 963-968. G’Rourke. J.P 1974. Toward a sctence of vocabulary development. The Hague Mouton Pellegrino. J.W. and R. Glaser. 1980. ‘Components of mductw redsmung’. In: R.E Snou. P.-A Federico and W. Montague feds.). Aptitude. learmng. and mstrucuon cognws process analyses of aptatude. vol. I. Hdlsdale. NJ: Erlbaum. Perfettl. CA. and A.M. Lesgold. 1977. ‘Discourse comprehenslo” dnd mdwdual dlfferwces In P. Carpenter and M. Just feds.). Cognmve processes I” comprehension. 12th Annul Crrnrgle Sympowm on Cogn~uon. Hdlsdale, NJ: Erlbaum. Posner. M.I. and R. MachelI. 1967. Chronometrlc analysts of class~fwat~on. Psqchologul Re\le~ 74: 392-409. Powell, J.S. and R.J. Sternberg. 1981. Acquwtion of \ocabuldry from ccontext Manwxpt submitted for pubhcatlon. Powell, J.S. and R.J. Sternberg. 1982. Conwxt clues used I” mferrmg the meanmgs of unfamdw words from text. Unpubhshed manuscript. Raven. J.C. 1965. Advanced progressive matrices. London. Laws. Rayner, K 1978. Eye movements I” readmg and mformatlon processing Psycholo@cJl Bulletm 85: 618-660. Rleger. C. 1975. ‘Conceptual memory’. In: R.C. Sthank (ed.). Concrp~ual mformatlon processing Amsterdam: North-Holland. RosLh. E. 1978. ‘Principles of categorization’. In: E. Rosch and B.B. Lloyd feds ). Cogmt~on .md categorization. Hillsdale, NJ: Erlbaum. Rubm. D.C. 1976. The effectiveness of context before. after. and around d mwng xord. Perception and Psychophysics 19: 214-216. Rubin. G.S.. C.A. Becker and R.H. Freeman. 1979. MorphologIcal structure and I!$ effecr on visual word recognition. Journal of Verbal Learnmg and Verbal Behawor 18: 757-767. Rumelhart, D.E. 1975. ‘Notes on a schema for stones’. In: D.G. Bobrow and A.M. Cclhns (eda ). Representation and understandmg. New York: Academic Press.
Schank. KC. 1975. Conceptual mforma11on procersmg. Amsterdam: North-Holland. 8.hank. R.C and R Ahclwn. 1977. Scnpl\. plans. goal\. and understanding. Hillsdale. NJ. Erlhrum. Shore\. J 1960. Readmg sctence maler~als for IWO dlfferenl purposes. Elementary En&h 32. 546-552. Shumu~y. N. and R.D. Forbes In press Espaliol en espatiol. New York: Holt. Smlih. E.E. and D L. Mcdm. 1981. Categories and concepl~. CambrIdge. MA: Harvard Unwers~t> prL!\.\. Smith E.E. E.J. Shohen and L.J. RIPS. 1974. Strucrure ancl process m seman~,c memory: a fedtural model for srman,,c decluons. Psychologxal Reweu, 81: 214-241. Smarh. H K 1967. The responses of good and poor readers when asked lo read ior dtiierent purposes. Readmg Research Quarterly 3: 53-83. Snow. R.E. 1980. ‘Aptitude procesreb’. In. R.E. Snow. P.-A. Federlco and W. Montague (eds). Aptnude. learnmg. and ,“struc,,on: cogn11ne process analyses oi aplltude. vol. I. Hillsdale. NJ. Erlhaum. Snou. R E. 19x1. ‘To\rsrd a theory of apcaude ior learnmg: I. Flwd and crystalhzed ahilittes and thur correlate\’ In: M. FrIedman. J. Das and N. O’Connor (eds.), lmelhgence and learnmg Neu York, Plenum. Sl.mner\. R.F. J.J. Newr and S Pal, ton. 1979. Memory representation for preiaxed words. Journal of Verbal Learnmg and Verhll Beha\lor IX. 733-743. Stem. N and C.C Glenn. 1977. ‘An analysis oi story comprehension m elementary school chddren’ In. R. Freedle (ed ). M,ul’~dlsclplmary approaches IO dwourse comprehension. Nonwad. NJ Ahlex. Sternberg. R.J. 1974. Ho\r IO prepare for the Mdler Analogwa Tea Woodbury. NY: Barron’s Educational Sene\. Slernbe~g. R J. 1977. Intelhgence. mformauon procesrmg. and analogul reasonmg: rhe componenunl analysts of human nhdlurs. Hdlsdale. NJ: Erlhaum. Sternberg. R J 1979. The nature 01 mental ahdities. American Psychologlsl 34: 214-230. Sternberg. R.J. 19X0. Sketch of B component~al subtheory of human mtelhgence. Behaworal and Bran Saences 3: 573-614. Sternberg. R.J. 1981~ The evolution of theorues 01 mtelhgence Intelhgence 5: 209-229. Stcrnherg. R.J. 198lh. ‘Toward a umfled componenclal theory of mrelhgence: I. Fluid ahihty’. In. M. FrIedman. J ras and N. O’Connor (eds.). lntelhgence and learning. New York: Plenum. Srernherg. R.J ano; J.S. Powell. 1982. Trachmg vocahuldry acquisiuon sb .s. A program for rrammg mtellc-ual skdls m Venezuela. Stemherg. R.J.. J.S Powell and D.B. Kaye. In pres\. ‘Teachmg vocabulary.huddmg skdls: a contexwd approach’. In: A.C Wdkmson (ed.). Communicating wth computers in classrooms: pro.\pects ior dpphed cognmte aence. New York: Academic Press. Taft. M. 1979. Recognwon of allwed words and 1he word frequency effect. Memory and Cogmtlon :979. 7: 263-272. Taft. M 1981. Prefix strippmg revisited. Journsl of Verbal Learmng and Verbal Behawor 20: 289-297. Taft. M. and K.I. Forster. 1975. Lexwal storage and retrieval oi preflxed words Journal of Verbal Learmng and Verbal Behavior 14: 638-647. Thurrtone. L.L. 1938. Primary mental ahdmes. Chvago: Umverwy of Chlcago Press. Troxel. V. 1959. Readmg eighth-grade mathemaclcal materials for selected purpoxs. Unpuhhshed Ph.D. dlsxrtallon. University of Ilhno~s. Vernon, P.E. 1971. The s1ruc1ure of human ahdi1les. London: Methren. Warren. W.H.. D.W. Nlcholns and T. Trahasso. 1979. ‘Event chams and inferences m understand‘ng narra~wc’. In: R.O. Freedle (ed.). New dIrectIons in dtscourse processmg. vol. 2. Norwood. NJ. Ahlrx.
187
Werner, H. and E. Kaplan. 1952. The acyuislllon of word meamngr: a developmental study. Monographs of the Society for Research m Chdd Dewlopmeni 51. Whitely. SE. 1980. Multicomponent latent tra.t models for abdit) tats. Pjychometrlka 45 479-494. Wdlis. S., S. Cornehus. F. Blow and P. Balles. 1981. Trammg rexarch m .!gmg attent~onal processes.Unpublished manuscript.
Roherr J. Sternberg IS Assoaate Professor of Psychology at Yale Unwcrs~ty Juner S Povell I\ a National ScienceFoundaclon Predoctoral Fellow at Yale. Dunrel B Kqe IS Ass~stmt Professorof Psychologyat the University of Caliiorma at Los Angelts.