By Peter Dang and Kyle Dennison
 Ability
to acquire language is a unique
and essential human trait
 Chomsky’s
Universal Grammar
• The human brain contains a limited set of rules
for organizing language
• Assumption that all languages have a common
structural basis
 Purpose: To
investigate the system
underlying the acquisition of new
linguistic competence based on
Chomsky’s UG
 Alternative View: The
human mind is a
general learning tool without specialized
structures
 Two
fMRI experiments
• Both consisted of native German speakers
• Difference was whether participants learned
Japanese or Italian
 Subjects
were to learn 3 grammatical
rules in foreign language and 3 artificial
rules of an ‘unreal’ language
 Prior
to the experiment, participants
were given a list of words to memorize
 Avoid
confounding syntactic rule
acquisition with word learning
 Recall
ability was tested over words
 Subjects
learned real and unreal
grammatical rules during pauses of 3
minutes between sessions in the form of
slides
 Subjects read 6 sentences in real and
unreal languages and judged their
syntactic correctness
• Correct – Left middle finger
• Incorrect – Left index finger
 Null-subject
parameter
• German: “I eat the pear”
• Italian: “Eat the pear”
 Subordinate
construction
• German: “Pia says that Paolo the pear eats”
• Italian: “Pia says that Paolo eats the pear”
 Negative
Construction
• “Paolo eats the no pear”
 Interrogative
Construction
• “Pear the eats Paolo”
Eight subjects from each experiment were analyzed:
 In judging the syntactical correctness of sentences the
subjects achieved the same percent correct by the end of
the experiment.
 Real Italian: 97 ± 4.1 %
 Unreal Italian: 91.6 ± 10%
 Real Japanese: 97 ± 3.2%
 Unreal Japanese: 94 ± 4.68%
 The increase in performance between the acquisition of
real vs. unreal rules was not significantly different.
 T7 = 0.1, P = 0.57 in the first fMRI experiment
 T7 = 0.5, P = 0.32 in the second fMRI experiment
Figure 1 Behavioral measurements. (a) Performance (mean % correct) in
judging the syntactical correctness of the sentences presented in each of the
five fMRI sessions per learning task (data averaged across subjects).
Performance showed an improvement across sessions during the learning
tasks. Performance increase was not significantly difference between the
acquisition of real versus unreal Italian (left) or Japanese (right) grammatical
rules.
 Over the course of the sessions subjects responded
faster (reduced RT) and improved in accuracy.
 There were four of these experiments: Real vs. Unreal
Italian and Real vs. Unreal Japanese.
 In every reaction time experiment the faster
responses were significant:
 T7 = 0.79, P < 0.0001
 Subjects were able to make faster grammatical
judgments in the real learning tasks than in the unreal
learning tasks (Both Italian and Japanese).
 Real vs. Unreal Italian: T7 = 0.79, P < 0.0001
 Real vs. Unreal Japanese: T7 = 1.9, P < 0.03
 These results are significant.
Figure 1 Behavioral measurements. (b) Mean reaction times (RT) for the correct syntactical
judgment of the presented sentences in each of five fMRI sessions per task. Subjects showed a
significant RT reduction (P < 0.0001) across sessions in all grammatical rules learning tasks. Reaction
times were faster during the grammatical learning tasks (either Italian or Japanese) than during the
artificial learning tasks (P < 0.03).

There were two conditions:
1. Looking at a black screen (the baseline task)
2.

Judging grammatical correctness of sentences (classification task)
With the condition of judging grammatical correctness fMRI data
showed activation in:
 Anterior Cingulate

Occipital Cortex

Cerebellum

Inferior and middle temporal gyrus

Prefrontal parietal

P < 0.05 and corrected for multiple comparisons
 fMRI measured blood-oxygen level dependent (BOLD) signal
 A random effects model was used to identify the activation pattern
specific to the interaction between the change in BOLD signal and type of
rule learning (real or unreal) reflected by accuracy of performance.

No specific pattern of brain activation was found for unreal
grammatical acquisition (Italian or Japanese).

There was an interaction between real and unreal grammatical
acquisition in Broca’s Area (pars triangularis of the left inferior
frontal gyrus).

In this area the increase in BOLD signal was correlated with the
increase in accuracy during the acquisition of natural rules for
Japanese and Italian.
Figure 2 Result of the interaction between
performance and type of rule learning (real
versus unreal Italian). Left, the activation
specific to real language acquisition
resulting from the random effects analysis
is displayed on selected slices of the MRI
template available in SPM99. The threshold
was set at P < 0.05 (corrected for multiple
comparisons). Right, plots of changes in
BOLD (blood oxygen-level dependent)
signal in the left inferior frontal gyrus
(Talairach coordinates x, y, z: –45, 21, 6) for
the five sessions containing ‘real
grammatical rule’ trials and the five
sessions with ‘unreal grammatical rule’
trials are shown as a function of accuracy
within sessions. The distances between
individual subjects’ peak voxel and the
peak voxel derived from the group analysis
using a random model were 28, 23, 3, 24,
13, 11, 23 and 15 mm, respectively, for
subjects 1–8.
Figure 4 Results of the interaction between
performance and type of rule learning (real
Italian versus unreal Italian in yellow; real
versus unreal Japanese in red) resulting from
the random effects analysis are shown on
selected slices of the T1 template,
thresholded at P < 0.001 (uncorrected) for
visualization. Right, plots of individual
changes in BOLD signal in the left inferior
frontal gyrus during the acquisition of
grammatical rules of either real or unreal
Japanese are shown as a function of accuracy
within sessions The distances between
individual subjects’ peak voxel and the peak
voxel derived from the group analysis using a
random model were 12, 21, 25, 9, 9, 9, 6 and
11 mm,
respectively, for subjects 1–8.
 The correlation coefficient between parameter estimates in Broca’s
Area and the accuracy of performance showed:
1.
2.
A significant positive correlation between BOLD signal and accuracy
in the real grammatical task.

Italian experiment: r = + 0.66, P < 0.001

Japanese Experiment: r = + 0.47, P < 0.001
A significant negative correlation between parameter estimates
(BOLD signal) and learning unreal rules.

Italian experiment: r = - 0.50, P < 0.001

Japanese experiment: r = - 0.32, P = 0.021
 There was a significant increase in BOLD signal in 2 parts of the right
inferior frontal gyrus related to better performance (accuracy).

This positive correlation between BOLD signal and accuracy occurred
for both the real (Italian and Japanese) and unreal (Italian and
Japanese) tasks.

Real Italian r = .37, P = .008; Real Japanese r = .21, P = .009

Unreal Italian r = .33, P = .01; Unreal Japanese r = .040, P = .004
 There was also activation in the left lingual gyrus
 The Japanese study showed further activation in the:
 Left cingulate, Right Insula, Right supramarginal gyrus, and Left
cerebellum
 Adults acquire language by learning explicitly about grammar.
 The inferior frontal gyrus is part of a neuronal system that
deals with episodic memory retrieval.
 It’s generally accepted that the job of the right inferior frontal
gyrus in learning “real” and “unreal” rules is to monitor
information independently of its character.
 It retrieves either hierarchical or linear information from
episodic memory when making a response.
 Explicit retrieval of information does not seem to be enough
to acquire proficiency in a new language.
 This study presents neurophysiological evidence that learning UG based
and Non-UG based languages involves different brain systems.
 Broca’s Area plays a significant part in learning “real” language rules.

Because languages from different families were used general
statements can be made about acquiring language.

The study points to a universal syntactic specialization among “real”
languages because activation of Broca’s area occurred regardless of
language of the subjects.

This study combined with previous research leads the authors to
think that this region is specialized for acquiring and processing
hierarchical structures (common character of every known grammar)

Additional support is provided by the negative correlation between
BOLD signal and learning “unreal” grammatical rules.
 Members of the KE family are prone to dysphasia (difficulty
understanding/speaking language, either written or spoken).
 There are structural and functional abnormalities in the left
prefrontal cortex, which includes Broca’s area.
 Problems with the FOXP2 gene are the likely cause of KE
family’s disorders
 This is consistent with the observation that language
impairments can run in families and are more concordant in
identical than fraternal twins.
 This shows that genetic material can influence language
ability.
 The target of selection in humans appears to be the FOXP2 gene.
 Results seem to support a common origin for human speech and
possibly some communication in non-human primates.
 Chimpanzees and gorillas learn words, but they don’t have the same
syntax abilities as humans nor are their sign motions as well defined.
 Monkey’s show left hemisphere asymmetry similar to humans, but the
left inferior frontal lobe has less cellular density and is less differentiated
 This study in particular demonstrates that the left inferior frontal gyrus is
centrally involved in acquiring new languages based on UG.
 The authors speculate that the complexities of Broca’s area may be what
makes us different from the other primates.