Beruflich Dokumente
Kultur Dokumente
Jennifer Weiss
Senior Thesis Project
for
Cognitive Science
Vassar College
1994-1995
ACKNOWLEDGEMENTS
Special thanks to Janet Andrews, my advisor, for her encouragement and guidance,
to Doug Winblad for being my second reader, Carol Christensen for her
emergency lessons, Debbie Ratchford, Richard Lowry for his numbers,
Marta Kutas, all my gracious volunteer subjects, my housemates for their
support and patience, Marisa Arias for her knowledge of Spanish,
and lastly to my parents for more than just the tuition payments.
1
Jennifer L. Weiss
ABSTRACT: Studies have shown the effect of semantic priming in monolinguals. A recent area of study
isthe use of ERP's in order to note the neurological activity involved in semantic priming. This experiment
investigateddifferences in bilinguals (early vs. late vs. learning) with respect to semantic priming. The
studytested across language priming as well as within language priming. The hypomesls was that early
bilingualsprocess on a semantic level so they prime more easily, while late bilinguals and learners do
surfacelevel processing and therefore show less of an effect of semantic priming. Therefore, it was
expectedthat the early bilinguals would display a more significantly reduced N400 than the late bilinguals
andboth would show a greater reduction of N400 than the learners.
INTRODUCTION:
This study intends to investigate the effect of semantic priming within and across-
languages for Spanish-English early bilinguals, late bilinguals, and students who
haveonly recently begun stUdying Spanish. The idea behind the study is twofold. If
achieved, the expected results would support a theory of how a person encodes two
languages at an early age versus how they encode a second language at a later age.
The study would also encourage investigations with larger implications, results that
wouldtell something not only about a critical period in bilingual language acquisition,
butalso critical period and language acquisition in general. The current study is
basedon the idea that there is a critical period for language acquisition and that
languages, whereas people who learn a second language at a later age are, in a
sense,having to work with what is already set up by the first language and by other
vacillation back and forth between the neurobiology and the psychology of language.
apparentbehavior of a child. The current study is one which intends to aid in the
2
investig~tionof more of the details of how the brain actually encodes language based
onphysiological phenomena not just behavioral ones. It is important to note that these
modelsof behavior are certainly not insignificant and can prove useful when used in
been:how rigid is the brain with regard to particular parts of the brain being
designatedfor particular tasks? levy and colleagues (1992) studied a child who had
a localized lesion in the left hemisphere as an infant and found evidence that supports
the"plasticity model". The plasticity model is a theory that the brain has great plasticity
forlanguage and if there is damage to one part of the brain, often other parts of the
brainare able to make up for the loss. The child studied by levy et at, apparently had
childrenhis age (levy et aI., 1992, p 23-24), despite the damage to parts of the brain
thatare known to be important for language and do cause language disorders when
damagedin adulthood. The biggest problem with the levy study is that the child was
studiedat an early age and so it is difficult to know, without waiting until he is older, if
hisabilities will consistently be at about the same level as those of his peers. But the
caseis initial evidence for plasticity of the brain with respect to language acquisition.
Another important issue within the field has been the question of lateralization.
Manyhave tried to find out which cerebral hemisphere is involved in which parts of
language. Especially with regard to bilingualism there has been much bickering back
andforth about whether or not bilinguals have the same proportional usage of the left
1992). For the purpose of the current study, the importance of lateralization is only
pertinentas it may provide clues to how language is encoded in the brain of an early
Mateer(1979) begin to give insight into how and where the brain stores language.
Mateer,1979, p. 1403). The brain is not made of two hemispheres that mirror one
anotherand language involves many other functions such as memory and motor skills.
Thus,lateralization and other issues of localization are important topics in the field of
othertasks are involved in language so that we can pinpoint what parts of the brain
areused just for linguistic tasks and what parts are used in coordination with other
areaof study is the critical period. Most investigators agree that timing is important in
JamesHurford suggests that the sensitive period for language acquisition ends as late
exposureto language is most important because that is the time when language
developmentlater on, the majority takes place in the first years of life. The complex
structuresof the brain are formed early; fundamental connections are established
beforebirth, while after birth there is more of the same kind of neuronal development
layeringeffect where the base layer is determined before birth. Fox discusses this in
involving the present as layered on top of the recent past and both being layered on
top of the more distant past. Similarly, with imprinting the animal is born with a base
on which they record the experience. Thus, a duckling is born with the materials
necessary to imprint the image of its mother (who is presumably the first thing it is
going to see when it is born) and all it needs is that visual experience to complete a
very major connection. Fox compares this process to learning and development in
general: a child enters the world ready to add experiences to its pre-established
functions. In this way, we learn by adding experiences one on top of the other. With
language acquisition, there is much debate about whether this is the actual process
that goes on in development but it seems to me that this is at the least a good way of
talking about language acquisition, however abstract the parallel may actually be to
"
the developmental process. It is a springboard from which we can dive into the
sensitive period in an effort to avoid aligning myself with either theory, although they
are not exactly the same, is certainly a psychological theory but corresponds nicely
withthe biological data. (A theory of an existing critical period is in a sense more strict
than that of a sensitive period, strict in that a sensitive period implies a desirable or
even ideal time to learn whereas critical period labels an essential or imperative
period during which one must learn.) Hurford reports that some research on aphasia
provides additional evidence for a critical period because it appears that children can
recover to a great extent whereas adults never fully recover their language abilities.
This would make sense given the hypothesis proposed earlier that children have more
mayhave to rely on the remaining healthy connections and use them as much as
neuronal connections.
-"~---i
acquisition may in fact limit people's abilities. In order to show that a critical period of
language acquisition is something that would develop naturally over time he creates a
set of simulated individuals, attributing specific characteristics to each, and runs them
through cycles that simulate changes such as reproduction, accidents, etc. In this
programmed evolutionary microcosm, Hurford finds critical period effects for all of the
individuals before 1000 generations. Thus, for each of the different conditions, he still
findsthat within 1000 generations each individual has had offspring who display some
kindof critical period. He uses this as evidence to show that a critical period of
language acquisition does stand the test of evolution and perseveres as an evolved
trait.
Moreover, Mayberry (1993) found the same kind of critical period effect in her
Mayberry's study, however differs from the previously described data because it
studiedsubjects who were deaf from birth as well as subjects who became deaf in late
childhood. The subjects who were born deaf learned ASL during infancy to late
childhood. Those who became deaf had learned English before they lost their hearing
so ASL was considered their second language. The results were such that those
whosesecond language was ASL performed at a much higher level than those who
acquiredASL at the same age but as their first language. In other words, the deaf
subjects who began learning sign language from an early age were native ASL
signers. The subjects who learned English before they became deaf and began to
learnASL later in their lives, did not have this same mastery of the language. This is
strongevidence for a critical period of language acquisition and also enhances the
ideathat we have language abilities with some predesigned structure that are general
evidence from deaf subjects indicates that we have certain deep structures of
language that are across modalities. In any case, the study shows that there is also a
critical period for signed language. This kind of evidence is another approach to the
study of a critical period and can function as more behavioral data to supplement the
evolutionary story, which is pertinent to language in humans and aids in the effort to
Marin notes that "observation of evolutionary trends in the organization of the cerebral
behavior along the phylogenetic scale, and, most significantly, the intense studies of
clinicopathological correlations have given us the basis for most of our present
Morere, 1990). They recognize the many instances of data from deaf aphasics but
they also criticize the lack of study as to whether there is a critical period for language
the left hemisphere. They found, at least in an initial investigation of the phenomenon,
thatthe age of onset of deafness had a great effect on the specialization of the left
hemisphere for speech. They used adults who had become deaf or nearly deaf after
thirty-six months of age (who were learning to speak as normal children before their
hearing loss) and found that their left cerebral hemispheres were not significantly less
specialized for processes involved in speech than those of normal adults, whereas
those adults who had become profoundly deaf before thirty-six months showed a
language development. It also raises the question of how the two hemispheres of the
7
braintake part in language. Much of the physiological data indicate that particular
Learning more about the way language is organized in the brain is crucial to
understanding human language. Yet language is not a separate function; there are a
varietyof processes involved both directly and indirectly with language. Thus,
distinctions must be made in order to determine what parts of the brain are used
Thestudy showed evidence for the distinction between word-memory sites and
areasbeing the posterior part of the language cortex, while the anterior language
1978, p. 337). In a later study by Fried, Ojemann, and Fetz, similar investigations were
performedand it was concluded that "a major difference between overt speech and
(Friedet aI., 1981, p. 355). The brain is a small space in which a great many events
experimentbe controlled for all variables, and all kinds of brain phenomena involved
begiven proper attention. The idea here is that there are many different processes
thatare intertwined with language and in order to find which parts of the brain do what
withrespect to language, one must also consider the many processes involved in
8
language. If, for instance, one wishes to look for particular areas that are active in a
certain language task, the other processes involved in the task should also be
considered. It is clear that reading involves more than just the mental comprehension
of words, and with our understanding of which parts of the brain are active for ocular
occurs. Similarly, a better understanding of memory that does not involve language
(such as non-language tasks using purely visual memory) can help us to pinpoint
specific areas where language is involved during a task such as one involving word
rnernorization. In other words, it is essential that language be isolated and the way to
do so is to gain knowledge of the other processes involved and how they are
organized apart from language. The more we know about the brain and its functions
that do not involve language, the more we will also know about the parts that do
involve language.
As stated earlier by Marin, much of what we know about language has come
from patients who have language-related disorders. There are a huge number of
different kinds of disorders and many of them overlap greatly. Often it is difficult to
classify a patient as having one kind of disorder. Some of the most common language
disorders, however, include Broca's aphasia and Wernicke's aphasia, which are
Hedescribes Broca's aphasics as people who do not have a grasp of the grammatical
rulesof their language. They also experience difficulty with word production, selection
of words, initiation of motor activity, etc. Wernicke's aphasics, on the other hand, have
display difficulty with phoneme sequence and with their representations of semantic
categories. Much of the actual description of the full functions of these areas is still
general story about what is happening where in the brain, but we still cannot describe
.. , -
9
the actual processes that are taking place with much detail, especially because in
studying disorders or brain damage, it is not always clear that there are not other
Bilingualism is a topic that, like aphasia and evolutionary models, can provide
great insight into the way the brain handles language. One of the biggest questions is
howthe brain encodes two languages. Are both languages separate? Is there an
"English language" box and a "Spanish language" box which have connections
betweenthem but are pretty much kept separate? I tend to think not. Bilingualism is
notso simple as having two languages in little storage boxes. First it is imperative to
definebilingualism. Part of the motivation for this study is to do this. Someone is often
definition may refer to fluency as being able to speak and understand both languages
as if the person were monolingual. In this case, if the person's languages are at about
the same level of competency, then he or she is considered bilingual (Hamers and
Blanc,1983, pp. 15-17). Hamers and Blanc note that there are different kinds of tests
of bilingual competence and as they point out, defining bilingualism varies depending
inwhich field it is measured. There are various aspects that make up the bilingual
otherbehavioral data, etc. All of these are ways of determining not only a bilingual's
languagecapabilities, but also her identity as a bilingual. The intention of this paper is
bilingualis one who has learned two languages from an early age while people who
learnlanguage after the critical period of language acquisition will only ever become
"fluent"in their second language, not fully bilingual. The distinction between "fluent"
andfully bilingual would, in this case, be a physiological one rather than a behavioral
one. Someone who is fluent can have great control over her second language, but
perhapsis slower at some language tasks than bilinguals who learned their second
---,
10
language before the critical period. That is not to discourage anyone from learning a
second language; it is clear that language can be learned after childhood and it can
be learned well. Yet I propose that what makes someone a true bilingual is a
physiological difference in how his two languages are encoded. The word "bilingual"
implies two equal languages, which late bilinguals presumably do not have according
to this biological definition of bilingualism. If we could show that people who learn a
second language an earlier age handle their languages differently than do people
who learned a second language at a later age, not only would it support the
hypothesis that there is a critical period for second language acquisition, but it would
With these predictions, and basing our growing understanding of the brain and
initial hypothesis about bilingualism. If indeed particular areas of the brain have
designated functions as the evidence seems to indicate, and there is a critical period
(atthe "right" age) has a particular development parallel in many senses, to the
developing structure of the first language in the brain. What I mean by this is that the
foundation that is the newly formed brain. Certainly there are many innate processes;
infantsare not born without anything in their heads. Infants are born with the ability to
perform many functions, but with respect to language they are born with the potential
to acquire language and they are born with many preset structures for language. Their
language abilities depend on the stimulation received from their surroundings which
cerebral cortex by dynamic selection", a process which declines gradually after age
-----, ,
11
two (Breathnach, 1993, p. 45). This implies that both a first language and a second
language would develop together in the brain if learned at an early age but a second
language learned at an older age must be forced to fit into the neural structure
designed in early childhood when the child learned its first language. This theory is
Evidence for this theory comes again from disorders such as aphasia. The most
significant support of this theory is an eighteen year old patient who had brain damage
at age four (Van Lieshout et aI., 1990). This young girl had a radiation lesion in her left
thalamic and temporal region which caused problems with short-term auditory
namingdifficulties (Van Lieshout et aI., 1990, p.184-185). The most interesting feature
ofthis patient's deficits is that she had the same problems with both of her two
languages (English and Dutch) despite the fact that she learned Dutch after the
radiation lesion. This could be used as evidence for the theory of bilingualism just
proposed. The girl began to learn Dutch at the age of six and a half, so she would be
considered an early bilingual. It seems that if she had begun to learn a second
language at this early age, she would have been able to compensate for the damage
because of the plasticity of the brain described earlier, assuming the brain is still
developing at that age (which is in fact not certain). Yet she could not; the Dutch she
learnedsuffered the same problems as her English suffered from the brain damage,
which suggests that there were fundamental language processes destroyed which
couldnot be overcome by learning a second language after the damage was done.
Thisis a good indication that there are designated areas for language where, for
instance,grammar and vocabulary are stored. It is also contrary to the plasticity model
stoppedaltogether earlier than age six). The patient's "assigned language areas"
12
were what suffered from the lesion, thus she lost some of her abilities in English and
could not later succeed with some aspects of Dutch because those overall language
processes were no longer functioning properly. The greatest information learned from
this subject is that there are indeed shared language "areas" for the different
The patient described above is one subject who begins to help us paint a
picture of language acquisition and of how it is that the brain handles language. We
startto wonder about these possible differences between early and late bilinguals.
physiological, between people who learn a language at an early age and those who
learna second language at a later age. Again, it is assumed that this has to do with
"
biological development and some kind of critical period of language acquisition.
There is evidence from all different realms that support this and provide evidence for
different kinds of bilingualism that depend on learning the second language during the
investigation of visual field asymmetries to argue that late bilinguals process words in
their second language at a more surface level (Vaid, 1987). Vaid cites a study by
Genesee et al. where the investigators found that late bilinguals could identify in which
language a word was presented faster than earlier bilinguals could. The late
bilinguals displayed earlier neural responses in their right hemisphere during this
language identification task while early bilinguals showed faster responses in the left
hemisphere. Vaid's study finds a right visual field superiority for late bilinguals while
nosuch effect was found for either early bilinguals nor for the rnonolinqual group
(Vaid, 1987, p. 273). This evidence was interpreted as supporting the hypothesis that
late bilinguals processing at a more surface level while early bilinguals are perhaps
the level of the representation or symbol as opposed to processing at the surface level
_.---.i1
.~.•
..!~
13
which refers to a focus on the syntactic features of an object or word. Vaid points out
that "this orientation to meaning may have developed in early bilinguals as a way of
overcoming the potential interference created by having to attribute two symbols to any
Johnson and Newport (1991) approach the issue of a critical period from a
different angle. These authors attempted to investigate critical period and maturational
effects on the ability to learn universal properties of language. In their conclusion they
argued for a critical period, claiming, "linguistic universals such as subjacency become
less accessible to the language learner with increasing maturation" (Johnson and
Newport, 1991, p. 254). Meanwhile, Magiste (1992) offers additional evidence for a
children, as measured by a picture naming task. This study was of German children
who came to live in Sweden. Other studies of different languages include Johnson
and Newport's (1989) reference to Korean and Chinese speakers who entered the
United States at an early age as having a significant advantage over those who
entered after puberty (Johnson and Newport, 1989). In their study they noted several
things, one of which is that children are better at learning a second language than
adults (Johnson and Newport, 1989, p. 89). The second point was that their data
indicated that "subjects who arrived in the United States before the age of seven
reached native performance on the test" where as "for arrivals after that age, there was
90). Yet the subjects who arrived after puberty did not seem to show this decreasing
abilitywith age; they simply all performed at a much lower level than those who had
displayed aphasia with respect to both of his two languages after having diluted
.." ...........•••••...
:::1,
14
amobarbital injected into his left middle cerebral artery (Berthier et aI., 1990, p. 452).
The authors explain that by the way the patient reacted, first speaking in English and
then only being able to speak his native language Spanish after 52 seconds after the
injection, implies that in the subject "English was represented in the central Sylvian
core of the left hemisphere, while Spanish was better represented in the more distal
perisylvian regions" (Berthier et aI., 1990, p. 452). The interesting part of this study
was that the subject was a late bilingual who had learned English as his second
language while in high school. Thus, the separation of the two languages in the brain
inthis case was an actual physical one for this late bilingual. It is not clear that this is
the case for all late bilinguals and it would also be important to do research to see if
early bilinguals display a different reaction to this kind of injection. Unfortunately, the
cases in which this kind of injection is necessary are not abundant. Yet it is
encouraging that actual biological data such as this one subject indicate a separation
of the two languages, and it can be used as a stepping stone towards a theory of late
Thus, the evidence seems to indicate that there is a critical period for language
critical period for language in general and leads to more interesting questions
semantic priming. The idea behind the current study is that perhaps the semantic
priming effect can be used to investigate differences between early and late bilinguals.
Semantic priming was first measured by reaction time where subjects were given a list
of word pairs and were asked later to recognize words that were in this list in another
list. The research showed that people's reaction times were faster for words which
were semantically related to the words they followed than for words which followed
unrelated words. This effect has been labelled the semantic priming effect. The idea
isthat people are better at learning words if they are "primed" by a related word.
•
> ---
15
partially,from the spread of activation from a recently accessed lexical item (cat) to the
detectorof the target item of interest (dog), so that when the target word is actually
presenteda short time later, its detector is "primed" or closer to its recognition
threshold. An implicit assumption of this type of contextual priming effect is that the
semanticpriming is simply making it easier to access one word by first stating another
thatis related. It is as if by saying the word "cat", the ANIMAL "box" is opened so that
whenthe next word is "dog", it is easier to get at because the "box" is already open.
potentials(ERPs), have given new insight into the field of psychology. We now have
theability to better investigate patients with disorders to discover what the brain is all
about,and we can also study healthy subjects as well. We can study normal
functioningof the brain in ways we could not just a few years ago. We can study
somestudies there are differences between men and women, while other studies add
Ojemannuses patients who suffer from epilepsy. This is a variable that may have had
aneffect on the results and it is difficult to say that the results can indeed generalize to
therest of the population. For these kinds of reasons, it is difficult to study language
functionsin the brain overall because of the variance among individuals. Yet many
16
distinctions can and have been drawn. Studies such as those using
electrophysiological data, for example, provide better and more controlled evidence
because they are more flexible as to who may be tested. That is, a very particular
group can be used or a wide variety of individuals can be tested. And the data is more
withthe ability to take a giant leap forward in our endeavors within the field of
electric field which the brain produces in fixed time-relation to an event" (Brandeis and
Lehmann, 1986, p. 151). They are useful because they can be used to determine
wherethe brain is active for different kinds of processes. They can also be recorded to
determine whether a person responds more to one kind of stimuli than another. ERPs
are measured from electrodes placed at different sites on the scalp, attached with a gel
that is both cohesive and conductive. Subjects are presented tasks while the
electrodes are on their heads, and as they participate in the activities the electrodes
canmeasure how much activity there is in the brain at particular times relative to, for
example,the presentation of a stimulus. In the case of the current study, the subjects
listenedto a list of words while the ERP measured their brain activity at seven different
electrodesites: Fz, Cz, Pz, P3, P4, T5, and T6 (see Appendix A).
To interpret ERP data, many different kinds of tests can be applied. For
instance,some studies require measures such as peak latency or area under the
curve. Some compare peak amplitude at a particular point in time while others
averageover a period of time. This study used mean amplitude as the measure; mean
amplitude is the average amplitude over a particular time period, called an epoch.
Thiskind of measure can be used to compare across subjects or for different kinds of
stimuli. ERP data can be interpreted in many different ways depending on how one
---- ----
17
choosesto. analyze it and what kinds of variables are involved. The study of language
hasmost definitely benefitted from this kind of technology. ERPs have often been
1952, can now be tested with more concrete investigations (Brandeis and Lehmann,
1986, p. 156). Brandeis and Lehmann explain that in a study using ERPs, Chapman
universality of the Osgood scales at the behavioral level" (Brandeis and Lehmann,
1986, p. 156). Similarly, behavioral studies of memory and language led into later
ERP studies of the same phenomena. From ERP studies it was determined that a
accordingto the probability of the stimulus as expected by the subject. In other words,
changesin P300 are often the result of some kind of unexpected perceptual stimulus.
Inone study cited by Brandeis and Lehmann (1986) incongruous words were
presentedat the end of a sentence and instead of finding an expected effect of P300,
theauthors found a a negative effect at about 400 ms after the stimulus was presented
(N400) (Brandeis and Lehmann, 1986, p. 156). This finding has lead to a great deal of
furtherstudy of N400.
In the past it was recognized that in certain word recognition tasks it takes
peopleless time to recognize a target word that has previously been presented
followinga related word than for a target word that has been presented following an
unrelatedword. In other words, "RTs [Reaction Times] are slower to name or classify a
targetword (e.g., dog) when it is preceded by an unrelated context (e.g., the word car)
thanwhen it is preceded by a related context (e.g., the word cat)" (Holcomb, 1993, p.
47). This difference was named the "semantic priming effect" and is still used to
~
-- ,
18
semantically related words than when they are followed by unrelated ones. Holcomb
thespread of activation from a recently accessed lexical item (cat) to the detector of the
targetitem of interest (dog), so that when the target word is actually presented a short
timelater, its detector is "primed" or close to its recognition threshold" (Holcomb, 1993,
p.48). As Hillyard and Picton tell us, a study was done where the authors were
thelist of words in which the incongruous word was presented (Hillyard and Picton,
1986, p.568). What Hillyard and Picton report is that the investigators found a greater
N400for these unrelated words and thus proposed that N400 was a measure of
reactionto semantic relatedness. Using this finding along with the study cited earlier,
researchbegan into how N400 was related to semantics. Hillyard and Picton continue
totell us of a study, similar to the task described earlier where sentences were
presentedwith unexpected words at the end of them, only in this new study there was
acomparison of N400 amplitudes for unexpected words that were unrelated to the
sentencecompared with amplitudes for unexpected words that were actually related to
theword that was expected. What they found was that "the N400 wave was smaller
whenthe unexpected words were semantically related to the most likely completion of
thesentence" (Hillyard and Picton, 1986, p. 568). Thus there is a greater reduction in
N400for semantically related targets than for unrelated words. This phenomenon was
subjectswith related and unrelated word pairs and measured reaction time as well as
reductionin N400. He found both results: reaction times were shorter for related
targetsthan unrelated targets and reduction of N400 was greater for related targets
r;
19
These results all indicate that there is a reliable method of recognizing semantic
a reduction of N400, but they present the stimuli auditorily. The study shows that a
greater reduction of N400 amplitude occurs for auditorily presented related pairs than
subjectswith two different tasks. The first task, called the Memorize task, required that
subjectssimply listen to a list of words and memorize as many as they could. The
secondtask, called the Count Nonwords task, required that subjects count to
hypothesized that "the N400s in the recognition memory task would be more sensitive
tothe activation of the semantic.information both within and between words relative to
thelexical decision task in which nonwords were counted" (Sentin et aI., 1993, p. 162).
There were 16 subjects with normal hearing, all right-handed. The authors
testedthese subjects, measuring their ERPs while the subjects were performing on
eachof the tasks. In addition to left and right mastoids for a reference, and vertical and
horizontaleye electrodes for eye movements, they used electrode sites Fz, Cz, Pz, P3,
P4, T5, T6, as well as six other specially selected sites. The data was collected in the
formof an EEG and later averaged over epochs. The subjects were also given a word
thatwere related to old targets, and new words that were unrelated to words in the
previouslist. Subjects had to determine whether the words were old or new and had
Sentin and colleagues summarize their data by stating, "In general, these
-
20
component in the auditory modality" and "the amount of attention directed to semantic
analysis appears to be important in determining the size of the N400 priming effect"
(Bentin et aI., 1993, p. 167). Their data also indicates that auditory presentation of
stimuli shows the same semantic priming effect as visual presentation. As far as the
related versus unrelated pairs, they saw a difference in mean amplitude between the
related and the unrelated pairs. Their related pairs were divided into categorically
related pairs and antonym pairs and although the semantic priming effect was a bit
largerfor the antonyms, it was not significantly greater than that of the categorically
related pairs. As far as the effect of electrodes, the authors found that the electrodes
that were more frontal and more central had greater negativity. In order to perform
more analyses, the EEG data was divided into three epochs: 300-400 ms, 400-700 ms,
and700-900 ms. The semantic priming effect proved greatest in the middle epoch of
400-700 ms. In the recognition task there was a significantly greater percentage of
falsealarms for new words that were related than there was for new words that were
unrelated to words in the previous list. The Bentin et al. study is useful because it
expands the field of semantic priming to include ERP analysis of the auditory modality.
Based on the information provided by the Bentin et al. study, the current
investigation was an effort to discover if our knowledge of semantic priming and N400
could perhaps provide insight into the critical period of bilingual language acquisition.
The main focus was how early bilinguals prime under these conditions versus how
latebilinguals and learners do. The hypothesis was that early bilinguals would show
a greater effect of semantic priming, thus a greater reduction at N400, than would the
latebilinguals or the learners. The benefit of using ERP measures for examining the
semantic priming effect is that the three groups' N400 amplitudes could easily be
contrasted.
Semantic priming across languages is not such a new idea; it has recently been
effect of semantic priming found between languages where the subjects were required
to pronounce the target words. The authors do not specify all of the ages of acquisition
forthe subjects, but most of them appear to have learned both of their languages at an
early age. The study investigates the theory of "automatic spread of activation", which
is a way of describing the way words are connected in the brain. They found that both
prime language and target language have effects on the priming effect, that is whether
the prime was in the same language or not was important in the priming effect. They
claimthat these results are "most consistent with the notion that the between-language
semantic priming effect reflects a structural feature of the semantic network-it is not
organised in terms of language specific areas" (Tzelgov and Eben-Ezra, 1992, p.266).
Also,Chen and Ng (1989) found better lexical decision performance when the prime
wasa translation than when the prime was a semantically related word. They also
effectsof semantic facilitation. They interpret their results as being support for a
concept-mediation model where "words in the two languages, and pictures, are all
ofthese links to concepts, subjects can actually respond to various kinds of tasks
involving the processing of pictures and words, such as lexical decision and naming"
(Chenand Ng, 1989, p. 460). The biggest problem with the Chen and Ng study was
thatthey did not specify whether the bilinguals were early or late bilinguals, although
fromtheir description of the subjects as having studied English in school for twelve
years,it is assumed that they must be late bilinguals considering the mean age was
twenty. Finally, a third bilingual study used N400 to note differences in bilinguals and
monolinguals (Ardal et aI., 1990). They did not, however, find differences due to age
of acquisition. These studies are all beginnings of new kinds of investigations into
bilingual language processing from which much can be said about the way language
is encoded.
22
early bilinguals, late bilinguals, and early bilinguals with respect to semantic priming.
The null hypothesis is that if indeed there is no difference in the organization of two
languages for those who learned both languages at an early age versus those who
learned a second language after early childhood, then one would not expect that there
is many difference in the semantic priming effect for these two groups. In addition, one
would not expect that the subjects would perform any differently on the recognition
task.
PROCEDURE:
This study was modelled after the study by Sentin, Kutas, and Hillyard who
investigated semantic priming with auditory stimuli. Their study found a reduced
N400,thus an effect of semantic priming, when stimuli were presented auditorily. The
current study uses Sentin, Kutas, and Hillyard (1993) as a basis to investigate
semantic priming in bilinguals. Thus, the main added variable was the use of two
different languages. The study used targets in English primed by English words (E-E),
targets in English primed by Spanish words (S-E), Spanish targets primed by English
words (E-S), and Spanish targets with Spanish primes (S-S). The purpose of using
these four types was to find if early bilinguals, late bilinguals, and people just learning
a second language show semantic priming within languages (S-S and E-E) and
across languages (E-S and S-E). The hypothesis was that early bilinguals would
show greater semantic priming than the late bilinguals and similarly, the learners
would show even less of an effect of semantic priming than the late bilinguals.
The second part of the study was designed to test how we" the subjects did
indeed learn the words, a sort of recognition task. It also was designed to see if
perhaps early bilinguals code their words at a semantic level while late bilinguals do
more surface processing. The list presented included target words that were on the
previous list, target words that were not, and some words that were translations of
---1 _
23
targetwords on the previous list. The hypothesis here being that the early bilinguals
wouldmake more mistakes if they were "paying more attention to" the meaning of the
words rather than which language they were in, thus suggesting that they process on a
semantic level rather than a surface level. Again, processing at the semantic level
indicatesthat there is some feature about an object or idea that is represented in the
theother hand, has to do with grammatical and other such organizational features of
an idea.
SUBJECTS
Thesubjects were college students who volunteered for the study or were given
laboratorycredit for their Introductory to Psychology course. All subjects were right-
handedwith normal hearing. There were five pilot subjects whose data were not used
inthe study but rather were simply test subjects used to insure that the study would
proceedsmoothly. There were seven subjects in the early bilingual group, seven in
thelate bilingual group, and four in the learner group. Subjects in the early bilingual
groupwere mainly those who had been brought up speaking both English and
Spanishor had studied it from an early age (at least before age 9). Late bilinguals
werethose who became fluent after age 10. They had either studied or moved to or
froma Spanish-speaking country. Learners were those who had possibly been
exposedto Spanish earlier but had not studied it or had not become familiar with the
STIMULI
Thewords used were recordings of a female voice digitized by the NeuroScan STIM
Soundprogram. Each word was a separate Sound file and each file was baseline
corrected and filtered in order to control for irrelevant sounds recorded. Each word
thefile so that all of the stimuli onset were equivalent. Words were no longer than 700
e
24
ms. The stimuli were presented biaurally through foam insert earphones at a preset
90.0dB. The stimulus onset interval was set at 2000 ms, 250 ms longer than the SOA
usedby Sentin et al. The reason for the difference was due to difficulties with the STIM
The first task was to listen to a list of word pairs based on a list provided by the
authorsof the Sentin, Kutas, and Hillyard study. Their list consisted of 128 word pairs
butthe list used here was expanded to include 144 word pairs. The added pairs were
basedon word pairs considered, but not actually used by Sentin et al. Sentin and
colleagues had chosen only words that ranged in frequency from 10 to 2,110 per
millionoccurrences as noted in Francis & Kucera (1982), and they cited a mean
frequency of 83.7. They also used word pairs that had been previously ranked,
accordingto relatedness, by eighty students. They only chose word pairs that were
givena mean rating above 2.95 for semantic relatedness. These were the word pairs
alsoused in the current study to form a general list of 144 semantically related pairs.
Thelist of semantically related words was then broken into four categories (S-S, S-E,
E-S, or E-E). This was accomplished by assigning numbers 1-4 to the list of word
pairswhich were ranked in order of semantic relatedness, the rating assigned from the
Sentinet al. study. The word pairs of each of the four categories were then assigned
eitheran R or a U. In order to form two lists from the 144 related pairs, the list of words
wasdivided up such that for one list the pairs with a U were kept related while those
withan R were made unrelated, while in the other list the U pairs were made unrelated
whilethe R pairs were kept with their related primes. Thus there were two final lists,
ListA and List S, where List A was a randomly assigned list of word pairs in which the
R targets were kept with their related prime while the U targets were randomly
assignedan unrelated prime that was of the same category type (S-S, S-E, E-S, or E-
E). List B consisted the same list of word pairs but the U pairs were kept with their
relatedprimes while the R pairs were made unrelated. The random reassignment of
...
--
25
primesfor each list, in order to make the unrelated pairs, was within type, so that for
examplea S-E target was sure to have an unrelated prime that was in Spanish. As in
theBentin, Kutas, and Hillyard study, there were pairs of categorically related words as
wellas antonyms. In order to increase the number of pairs, both categorically related
pairsand antonyms were added, although more antonyms were added so that the
proportionof antonyms to categorically related pairs was greater than in the 8entin et
al.study. This was deemed acceptable because 8entin et al. did not find a significant
differencebetween the types and collapsed their data. They found that priming was
slightlybetter for antonyms but it was not statistically significant. Therefore, there was
nothreat of affecting the data by adding more antonyms than categorically related
pairs;the only effect would be an increased overall semantic priming, but this would
befor all three groups (early, late, and learners) so would not affect the results of this
particularstudy.
FIGURE 1
In both List A and List 8 there were thirty-six across-language unrelated pairs
unrelatedpairs and thirty-six same-language related pairs (see Figure 1). See Figure
2fora division of the categories into the two lists. Both lists had the same order for
theirtargets, thus only the order of the primes was different for the two lists.
-~
26
-----"-----------------------------------------------------
FIGURE2
The second task consisted of a recognition task using a list of 144 words. The
listwas simply the list of targets from the previous task divided up into three categories
andput in a new randomly assigned order. One third of the words in this list were the
sametargets presented again, one third were a translation of targets (that is, if the
wordwas previously DOG, the new word presented was that same word but in
Spanish, PERRO), and one third were completely new unrelated words. These
unrelated words were assigned so as to have the same amount of Spanish and
Englishwords as there was in the one third of the list of targets. Then this new list was
randomized in order to intersperse the three categories (Old, New- Translation, and
New-Unrelated). The New- Translation group was a substitution for the New- Related
groupused in the Sentin, Kutas, and Hillyard study. From here on the "New" category
willbe used to refer specifically to the New- Unrelated group, which "Translation" will
beused to refer to the New- Translation words. See Figure 3 for a division of the three
categories.
27
FIGURE3
METHOD
Inthe first task, the subject was asked to listen to a list of words, some of which were in
English and some in Spanish. The subject was told that he would see a gray screen
witha light blue cross in the center and was asked to fixate on the blue cross, moving
hishead as little as possible and blinking at a comfortable rate but trying not to blink
toooften. Then he was asked to memorize as many of the words as he could. Finally,
the list was presented and the ERP was measured while the subject was listening to
After the list was finished the subject was instructed that she would now hear
another list of words, some of which were on the previous list and some of which were
not. She asked to place her right hand on the mouse, with her index finger on the key
onthe left, and her middle finger on the key on the right. Then she was asked to
respondto each word she heard, that if the word she heard was on the previous list, to
pressone of the keys and if it was not on the previous list, to press the other key.
Subjects were counterbalanced for which finger they used for new and old words.
Thus,the subjects who heard List A were divided in half, half of them were required to
pressthe left key for old words and half were required to press the right key for old
words. Subjects who heard List B were also divided in this way. In the second task,
ERPwas not measured. The point of the second task was two-fold. The first object
wasto make sure the subjects did in fact memorize the words in the first list. The other
..oj
- -- I
28
goalof the second task was to measure the reaction time. It was predicted that the
semanticlevel than late bilinguals, who may process at the surface level. Similarly, it
waspredicted that the Early bilinguals would actually make more mistakes on the
New-Translation words than the Late bilinguals. This is because if Early subjects
wereprocessing the words at a semantic level, perhaps their coding for language was
lesssignificant and would be more readily ignored when they heard the translation of
theword in the previous list. In other words, they were processing the words
semantically so when recognizing words, the translation of the target would would call
upthe same semantic representation and these subjects would mistakenly claim the
Finally, each subject was asked to fill out a questionnaire to rate his language
anyother languages.
recordingsites used were Fz, Pz, Cz, P3, P4, TS, and T6. In addition, vertical eye
movementsand blinks were recorded from one individual electrode placed above the
eyeand another just below. Additional individual electrodes were used, one behind
eachear over the right and left mastoids, for the purpose of off-line averaging. Again,
theuse of the mastoids (as opposed to using earlobes) was due to the attempt to
replicateas closely as possible the Sentin et al, study. While each of the lists were
presented,the subjects were asked to fixate on a small light blue cross which was
presentedon a dark gray screen. This was in an effort to limit movement of the head
andeyes of the subject while the data was being recorded. Each word was
considereda separate sweep, lasting 1300 ms. The program for the lists of words was
writtenon Gentask, a program within the STIM system of Neuroscan. The Gentask file
29
communicates with the SCAN system, which records the EEG. Each sweep began
exactly 100 ms before the stimulus was presented and ended at 1200 ms after the
onsetof the stimulus. This period was used to allow for later analysis of the data
wherethe total EEG would be divided into epochs as in the Sentin, Kutas, and Hillyard
study.
RESULTS
Each ERP was baseline corrected for all the of the electrodes, using the
prestimulus interval (-100 to 0 ms) as the baseline measure. Then each ERP was
artifactreduced to correct for eye blinks, usually using around a 450 millisecond
sweepduration and a trigger threshold of ten percent. Finally, the data was run
throughan artifact rejection process. Once the data was collected and corrected,
averageswere taken for each of the types one through twelve for each subject. Type
assignments can be seen in Figure 4. Averages were also taken for each of the
targets)regardless of which kind of pair (S-S, S-E, etc.). This was to done so that it
couldbe analyzed whether or not there was indeed a reduced N400, an effect of
FIGURE4
The ERP averaged files were then used to measure the mean amplitude
between400 and 700 ms. This epoch was used because in the Sentin et al. study,
30
thiswas th.e epoch in which they found the greatest effect of semantic priming. This
The ANOVA consisted of a dependent variable of mean amplitude with one between
subjects measure (Group) and two within groups measures (Electrode site and
Relatedness). The repeated measure of Group had three levels: Early, Late, and
Learner. Electrode site was divided into the seven electrodes: Fz, Cz, Pz, P3, P4, T5,
T6. Relatedness of targets also had three categories: Antonym, Categorically related,
andUnrelated. The ANOVA showed the following significant effects: Group: F (2,15) =
6.254, P = .0106; Electrode site: F (6,90) = 33.626, P = .0001; Relatedness: F (2,30) =
6.473, P = .0046. There was also an interaction effect of Relatedness and Group: F
(4,30)= 6.427, P = .0007. No other effects were significant. Further comparisons were
performed in order to view the effects within the variables. Early and Late groups were
contrasted (F [1,15] = 3.325, P = .922) as were Late and Learner groups (F [1,15] =
3.951, P = .0654) and Early versus Learners (F [1,15] = 12.401, P = .0031). There was
a significant interaction effect of Relatedness with Early versus Late subjects (F [2,30] =
5.440, P = .0096), a significant effect of Relatedness with Late versus Learners (F
[2,30]= 5.223, P = .0113), and a significant effect of Relatedness with Early versus
Learner(F [2,30] = 8.888, P = .0009). See Appendix B for means tables. The
significant effect of Relatedness also called for more detailed analysis such as Related
Antonymcategories to form the Related group. Then Category and Antonym were
The overall performance of all subjects on the recognition task 49.0% correct.
Subjectsdid better on the New words (57.2%) than they did on the Translations
31
(50.8%) while scoring the lowest on the Old words (45.2%). The mean latency overall
was 1.1768 ms and was divided as such: 1.1878 for the New words, 1.1677 for the
Oldwords, and 1.1829 for the Translation words. It should be noted that the
recognition task data for one of the eighteen subjects was not used because it was
incomplete due to a computer error in presentation. This subject was one of the four
Learnersubjects. The recognition task data was also subjected to an ANOVA with
repeated measures. The latency and percent correct measures were individually
compared to two between-subject measures, Group (Early, Late, and Learner) and List
(ListA and List B), as well as one within-subject variable which was the Type. Type
FIGURE5
131: Translation from S-S Related 151: Old from S-S Related pair
132: Translation from S-E Related 152: Old from S-E Related pair
133: Translation from E-S Related 153: Old from E-S Related pair
134: Translation from E-E Related 154: Old from E-E Related pair
135: Translation from S-S Unrelated 161: Old from S-S Unrelated pair
136: Translation from S-E Unrelated 162: Old from S-E Unrelated pair
137: Translation from E-S Unrelated 163: Old from E-S Unrelated pair
138: Translation from E-E Unrelated 164: Old from E-E Unrelated pair
141: Spanish New Unrelated
142: English New Unrelated
The results of the ANOVAs showed no significant effects for latency. The only
significant effect for percent correct was List: F (1,13) = 7.147, P = .0191 as seen in
Appendix C. More detailed analysis of the types, which required collapsing across
significant effect of type for Old versus New words (F [1,221] = 5.000, P = .0264).
Translation versus New, Related Translation versus Related Old, and Unrelated
Translation versus Unrelated Old, all proved insignificant. Other contrasts were
~
32
DISCUSSION:
Just from glancing at the grand averages for the related targets versus the
unrelated targets, one can see a decreased negativity for the related words around
400 ms for most of the electrode sites (See Appendix D, parts I-XII). Similarly, the
effectof relatedness was a significant one. This indicates a replication of the semantic
priming effect found in the Sentin et at. study. Looking again at the grand averages,
thepeak is clearly delayed, usually greatest at around 430 ms, an effect also found in
theBentin study. Sentin et al, also found a negative wave at about 100 ms after
presentation of the stimuli which can also be seen in the grand averages for the
currentstudy. The other important result to note in the grand averages is the
difference between the electrodes. Fz, Cz, and Pz were clearly more negative and
seemto show greater semantic priming (see Graph 1). Similarly, P3, P4, T5 and T6
whichare more posterior and parietal sites, did not show as great of negativity as the
threecentral and more anterior locations. This is consistent with the results noted by
Bentinand colleagues who found that "during the 300-900-ms epoch the ERPs were
significantly more negative frontocentrally than at the parietal and temporal sites"
'-
33
GRAPH 1
Interaction Plot Effect: ELECTRODE SITE
Dependent: MEAN AMPLITUDE
with standard error bars
w
P
::J
r-
::s~ -0
~
.: 1
~
~ ~2
4-;
0
<Il
.3
~
IlJ
~
:::l -4
IlJ
U
.5
-6
FZ cz PZ P3 P4 T5 T6
ELECTFODE SITE
\'
\ 34
\
'RAPH 2
Interaction Plot Effect: GROUP
Dependent: MEAN AMPLITUDE
with standard error bars
-.6
-.8
~
(:4
::> -1
E-<I
::1
Po.
-1.2
~ -1.4
~
-1.6
~ -1.8
~
.•... -2
0
II)
-2.2
~
C1i
-2.4
~
...•
...•
C1i -2.6
o
-2.8
EARLY LATE LEARNER
GROUP
PH 3
Interaction Plot Effect: RELATEDNESS
Dependent: MEAN AMPLITUDE
with standard error bars
-.4
~ -.6
(:4
::>
E-< -.8
::1
Po.
-1
~
~ -1.2
~ -1.4
.•...
0
II) -1.6
~
C1i -1.8
~
....•
....•
C1i
-2
U
-2.2
CATEGORY i- ANTONYM UNRELATED
RELATEDNESS
34
\
RAPH 2
Interaction Plot Effect: GROUP
Dependent: MEAN AMPLITUDE
with standard error bars
-.6
-.8
~
0 -1
;:J
£-<1
•....• -1.2
...l
~ -1.4
4;
-1.6
~ -1.8
~
.•.. -2
0
(/)
-2.2
~
aJ -2.4
~
....
....
aJ -2.6
U
-2.8
EARLY LATE LEARNER
GROUP
PH 3
Interaction Plot Effect: RELATEDNESS
Dependent: MEAN AMPLITUDE
with standard error bars
-
-.4
~ -.6
0
;:J
E-< -.8
::1
~
-1
~
-1.2
~
~ -1.4
.•..
0
(/) -1.6
§
aJ -1.8
:?:
....
.... -2
aJ
U
-2.2
CATEGORY ANTONYM UNRELATED
.
RELATEDNESS
35
In Graph 3 we see the effect of relatedness. Although there was not an overall
difference between the related words and the unrelated ones, the categorically related
targets words displayed significantly less negativity than antonyms and unrelated
targets. There was a significant contrast between category and antonyms as well as
between category and unrelated. Antonym and unrelated targets did not show any
significant difference. In the Bentin et al. study, the antonyms actually showed a
greater effect of semantic priming for antonym targets than for categorically related
targets, although this effect was not a statistically significant one. There are several
possible explanations for why the antonyms did not show the same effect in this study.
The expectation had been that the antonyms would be approximately equivalent to the
categorically related words. This was not the case overall. By looking at the
interaction effect of relatedness and group (see Graph 4), we can see that for the early
groupthe category type was much higher than the antonym while results for the late
groupshowed that they were equivalent. The most bothersome evidence comes from
thelearners, who show the opposite effect than was expected. The fact that the
antonyms for the learners was so significantly different is reason for concern. One fact
significant difference in the recognition task and if this is representative of this data,
someunderstanding may evolve from a comparison of the two lists for the ERP data.
Forinstance, in the recognition task, the percent correct was much lower for List A than
forList B. This may be interpreted as those subjects who were presented List A for
data. If this is the case, there may be something wrong with the structure of List A and
if only List B were analyzed for relatedness * group effects, perhaps there would be
resultsmore consistent with the Bentin et al. study. Another possibility is that
antonyms across languages are just not the same as antonyms within languages.
Thiscould be for a number of reasons. One reason is that oftentimes antonyms are
-- .-
. "'..•.
••••
36
culturally defined, or at least emphasized more in some societies than in others, such
as BLACKIWHITE. Perhaps trying to prime across languages for these kinds of pairs
does not prove to be consistent with a semantic priming effect. Also, using
BLACKIWHITE again as an example, perhaps these are not in fact coded as antonyms
across languages but rather as categorically related words under the category
"COLORS". Another possible explanation is the fact that words do not always have
conceptually accurate translations from the English words provided by Sentin and
colleagues as is possible; however that still does not rule out the possibility of a word
in one language meaning something slightly different in the other language. Similarly,
homophones were avoided as much as possible, but it is clear that some words can
initiate several different semantic codes. The word "sink" can be interpreted as the
object in a kitchen or it can be the verb of an object not staying afloat. Perhaps primes
that could have more than one meaning were misleading, especially in the case of
antonyms. For List A and List S word pairs, see Appendix E I-II.
GRAPH 4
Interaction Plot Effect: RELATEDNESS * GROUP
Dependent: MEAN AMPLITUDE
with standard error bars
~ .5
0
;:J 0
5~ -.5
~ 0 CATEGORY
i f •
-1~
~ -1.5 ANTONYM
~ Ii UNRELATED
'+-< -2
0
III
-2.5
§
Cl)
~ -3
~
Cl) -3.5
U
-4
EARLY LATE LEARNER
GFOJP
".
-
37
The recognition data provided results that were quite curious. First of all, there
was no significant effect of latency. Second of all the percent correct overall was only
49.1 percent, significantly lower than that of the Sentin et al. study, and only at chance
level. See Graph 5 for percent correct by group. One interpretation of this result is that
the subjects did worse because of the added language. In other words, perhaps there
was an overall effect of language being a confusing factor in the task, causing all of the
subjects to perform worse. One way to investigate this would be to contrast the mean
percent correct for the bilingual groups, the Early and the Late groups, with the
Learner group. If indeed the mean was significantly higher for the bilingual groups
than for the learning group, then it may be that the early bilinguals did not display any
significant number of false alarms on the translations because the overall performance
on this recognition task was so poor. This could be further investigated by examining
the data to see whether the bilinguals did better overall. One would expect that the
bilinguals would have done better than the learners if just the added language was the
factor that caused the subjects to perform so poorly on the recognition task. If the
bilinguals did better than the learners, then it is possible that overall performance was
skewed by the data from the learners. It would also be important to investigate
whether the two bilingual groups had the same success rate on the translations.
•...
-----. •••••
38
GRAPH 5
Interaction Plot Effect: GROUP
Dependent: PERCENT CORRECT
with standard error bars
53
52
51
I-
o 50
w
a:
a: 49
0
o
48
~
'0
(/)
47
c
<ll
Q) 46
~
Qi 45
o
44
43
42
EARLY LATE LEARNER
..., Group
The most problematic result of the recognition task, as well as the study overall,
is the difference found in List A and List 8 (see Appendix C for means table and
graph). Again, this could be due to an uneven distribution of problematic words like
homophones or a poor translation. Tzelgov and Eben-Ezra (1992) note that abstract
and concrete words yield different results in semantic priming. Perhaps there was a
trend for more abstract words to be found in List A. The difference between the two
lists may also be a problem relating to the distribution of subjects who were presented
each of the lists. There were several subjects in List 8 who did exceptionally well on
the recognition tasks, so perhaps this caused differences in the data. Similarly,
because one of the subject's data was not used, there was only one subject in the
Learner group that was presented with List 8; if this subject happened to do very well
on the task, the entire learner group as well as the overall data may have been
affected by this. A way to investigate this further would be to test more subjects so as
<
--
39
to decrease the influence each subject has and to increase chances of getting a more
Other problems with the study include the small number of subjects, the limited
kinds of subjects used, etc. In an effort to replicate the Sentin et al. study, the lists were
taken from their list of English pairs. This may have not been the most beneficial way
to form pairs that involved a different language as well, with a different cultural
background that has linguistic implications. Yet the overall semantic priming effect
found in the Sentin et al. paper was also noted in this study and there certainly was a
difference in the groups (early, late, and learner), which, with further investigation, may
lead to a better understanding of bilingualism. Another problem was that the number of
words in the lists for both tasks was increased. Perhaps the larger number of words
had an effect on the ability of the subjects to memorize and or just process the words.
There are a great number of implications that can be drawn from this study.
There are also more analyses of the data that could be performed to further investigate
the results of the study. For instance, the whole issue of between-languages versus
within-languages is an important one but one which could not be analyzed in this
project because of time constraints. Thus, comparing the E-E and the S-S pairs with
the E-S and the S-E pairs would be one way to see the overall differences between
dividing up the groups first so as to find any difference in the between-language pairs
and within-language pairs according to group. This could be done for both the ERP
data and the recognition task data and would tell us a great deal about the overall
results as well as lead to implications of differences between early and late bilinguals
and learners.
investigate some of the results found here. One such study would be one that
in early bilinguals versus late bilinguals. Another study could investigate between-
language pairs and within-language pairs, including the differences these kinds of
pairs show in early bilinguals versus late bilinguals versus learners (as described
above). Other studies that would provide interesting results would be ones that looked
intodifferences in parts of the brain active for early bilinguals, late bilinguals, and
learners. Perhaps there are some differences between the three groups and where
they are processing words. There are a number of ideas that can be drawn from the
research presented here. Although the results were not clear enough to draw any
difference between early bilinguals and late bilinguals that is worth investigating.
IMPLICATIONS:
One general question to be asked is: Why does all of this matter? The answer
is not clear because the data are not clear. As we encounter and interpret more and
moredata on bilingualism, we will know whether or not it is a field of study that can in
fact help us understand language processing. The motivation for the above paper was
an interest in language learning and a curiosity about when the best time to learn a
languageis. The fact that early bilinguals have in previous research shown to be
more"truly" bilingual, and in this study there is at least some difference between the
languageprocessing of early bilinguals and late bilinguals and learners, leads us to
believethat indeed there is something different about early bilinguals. If second
languagelearning is similar to first language learning, then the best time to learn a
secondlanguage would be right along with the first language, in which case there
shouldbe pressure to integrate bilingual programs into school systems. Here in the
languages. In Valencia, Spain most children study Spanish, Valenciano, English and
.. --
41
usually French as well, from a very early age. If we can show that there is something
very different about learning a second language at a young age, perhaps we too will
be encouraged to learn more than one language right away rather than wait until we
~
\ \.
@ ®
d '~Er
/ '
@
® ® @
@ @
®
®
(e6BluOW eLH uo
@
JBedde l,uSeoO) P
9,
pendix
B
: GROUP
ndent: MEAN AMPLITUDE
Count Mean Std. Dev. Std. Error
: ELECTRODE SITE
:endent:MEAN AMPLITUDE
: RELATEDNESS
:endent:MEAN AMPLITUDE
Count Mean Std. Dev. Std. Error
CATEGORY 126 - .887 3.075 .274
ANTONYM 126 -1.788 2.521 .225
UNRELATED 126 -1.836 2.433 .217
a: RELATEDNESS • GROUP
endent:MEAN AMPLITUDE
•...
Appendix
C
Means Table
Effect: List
Dependent: % CORRECT
Interaction Plot
Effect: List
Dependent: % CORRECT
With Standard Error error bars.
58
56
~ 54
o
w
a: 52
a:
0
o 50
~
0
"0
!/l 48
c
<1l
Q)
:::2 46
Qj
o 44
42
40
A B
List
GRAHD AVERAGE
Related •....
5 Unrelated APPENDIX D
(red) (green)
5.80 part I.
4.20
2.88
1.40
~~
o
J
o
b -1.40
~
-2.813
-4.20
-5.60
-v.eal I I I I I I I I I
-100 30 160 2913 4213 550 680 810 940 113713 12013
11i I I i seconds
GRAHD AVERAGE,
Related vs Unre La t e d
5.60
(red) (green) APPENDIX D
part II
4.2121
2.80
1.40
0.013
l~
+'
o
:>
o
b -1.4121
I:
-2.813
-4.20
-5.613
-7.00' , I I I ! I I I I I
-1013 30 1613 291.3 420 550 680 81121 94121 11.371.3 121210
11i I I i seconds
GRAHD AVERAGE.
2.40
1.6B
0.80
0.00
lQ
t-
oJ
'J
s,
(J
-0.80
I:
-1.80
-2.413
-3.213
-4.00' , , I I I I I I ! I
-100 30 160 290 420 550 68B 810 940 1070 1200
11i I I i seconds
GRAND AVERAGE.
3.20 APPEtiDIX D
Related vs Unrelated part IU
(red) (green)
2.40
1.60
0.80
0.00
...,l~
"0
:>
.J
s, -10.80
(J
I:
-1.6£1
-2.40
-3.20
-4.00' , I I I I I I I I I
-100 30 160 2'~0 420 550 680 810 940 1070 1200
11i I I i seconds
GRAND AVERAGE.
APPENDIX D
4.10 Related vs Unrelated
part V
(red) (green)
3.20
2.30
1.40
0.513
l~
+'
o
:>
o
b
I:
-1.30
-2.20
-3.10
-4.1301 I I I I I I I I I I
-1013 30 1613 290 420 550 680 810 940 1070 1200
11ill i seconds
GRAHD AVERAGE
4.10
APPENDIX D
Related us Unrelated part VI
(red) (green)
3.20
2.313
1.40
0.50
~~
o
J
o
b -13.40
~
-1.30
-2.20
-3.10
-4.013 I I I I I I I I ! I I
-1130 313 160 290 4213 5se 6313 810 '~40 un0 1200
11i I I i seconds
GRAND AVERAGE
4.10 APPENDIX D
Related vs Unrelated
part VI I
(red) (green)
3.213
2 .3~3
1.413
lq
+'
-
o
::>
.J
b
I:
-1.30
-2.213
-3.113
-4.1313' , I I I I I I I I I
-100 30 160 2913 4213 550 680 810 940 un13 12013
11j 1 1 j seconds
APPENDIXE - part I LIST A
:=unrelated, r=related)
APPENDIXE - part I LIST A
(u=unrelated, r=related)
APPENDIXE - part I LIST A
SHORT 3.46 4 u 4
107 A ANTES AFTER 3.66 2 r 6
108 C WARM BRASS 3.11 4 u 4
109 A BOTAS DEVIL 3.46 2 u 2
110 C FLOWER GALLETA 2.95 3 u 3
1111 C ENOJADO ENFADADO 3.07 1 r 9
.~
'u=unrelated. r=related)
APPENDIXE - part I LIST A
u=unrelated, r=related)
APPENDIXE - part II List B
(u=related, r= u n re lated)
1
APPENDIXE - part II List B
~=related,r=unrelated)
APPENDIX E - part II List B
(u=related, r=unrelated)
APPENDIX E - part II List B
(u=related, r=unrelated)
.. "---.
BIBLIOGRAPHY
Ardal,S., Donald, M.W., Meuter, R., Muldrew, S. and Luce, M. (1990). "Brain
Responses to Semantic Incongruity in Bilinguals." Brain and Language, 39,
pp. 187-205
Bentin,S., Kutas, M., and Hillyard, S.A. (1993). "Electrophysiological evidence for task
effects on semantic priming in auditory word processing." Psychophysiology,
30, pp. 161-169.
Carlson,N.R. (1991). Physiology of Behavior, 4th ed. Needham Heights, MA: Allyn
and Bacon.
Chen,H.-C. and Ng, M.-L. (1989). "Semantic facilitation and translation priming effects
in Chinese-English bilinguals." Memory & Cognition, 17 (4), pp. 454-462.
fox, J.L. (1983). "Debate on Learning Theory is Shifting." Science, vol. 222, 16
December, pp. 1219-1222.
rancis, W.N. and Kucera, H. (1982). Frequency Analysis of English Usage: Lexicon
And Grammar. Boston: Houghton Mifflin.
ried, I., Ojemann, G.A. and Fetz, E.E. (1981). "Language-Related Potentials Specific
to Human Language Cortex." Science, vol. 212, 17 April, pp. 353-356.
amers, J.F. and Blanc, M.H.A. (1989). Bilinguality and Bilingualism. Cambridge, UK:
Cambridge University Press.
Hillyard, S.A. and Picton, T.W. (1986). "Electrophysiology of cognition." From
Handbook of Physiology: A critical, comprehensive presentation of
physiological knowledge and concepts (Section 1: The Nervous System.
Volume V.) Stephen R. Geiger (Ed.). American Physiological Society.
Holcomb, P.J. (1993). "Semantic priming and stimulus degradation: Implications for
the role of N400 in language processing." Psychophysiology, 30, pp. 47-61.
Hurford, J.R. (1991). "The evolution of the critical period for language acquisition."
Cognition, 40, pp. 159-201.
ohnson, J.S. and Newport, E.L. (1989). "Critical Period Effects in Second Language
Learning: The Influence of Maturational State on the Acquisition of English as a
Second Language." Cognitive Psychology, 21, pp. 60-99.
ohnson, J.S. and Newport, E.L. (1991). "Critical Period Effects on Universal
Properties of Language: The status of subjacency in the acquisition of a second
language." Cognition, 39, pp. 215-258.
evy, Y., Amir, N. and Shalev, R. (1992). "Linguistic Develoment of a Child with a
Congenital, Localised L.H. Lesion." Cognitive Neuropsychology, 9(1), pp. 1-32.
arcotte, A.C. and Morere, D.A. (1990). "Speech Lateralization in Deaf Populations:
Evidence for a Developmental Critical Period." Journal of Brain and Language,
39, pp. 134-152.
arin, O.S.M. (1982). "Brain and Language: The Rules of the Game," Chapter 3. In
M.A. Arbib, D. Caplan, and J.C. Marshall (Eds), Neural Models of Language
Processes. New York: Academic Press, Inc.
Paradis, M., Goldblum, M.-C. and Abidi, R. (1982). "Alternate Antagonism with
Paradoxical Translation Behavior in Two Bilingual Aphasic Patients." Brain and
Language, 15, pp. 55-69.
Vaid,J. (1987). "Visual Field Asymmetries for Rhyme and Syntactic Category
Judgements in Monolinguals and Fluent Early and Late Bilinguals." Brain and
Language, 30, pp. 263-277.
Ian Lieshout, P., Renier, W., Eling, P., de Bot, K. and Slis, I. (1990). "Bilingual
Language Processing after a Lesion in the Left Thalamic and Temporal
Regions." Brain and Language, 38, pp. 173-194.