Deconstructing Discourse

1
Deconstructing Discourse

• A PET Study of Narrative Production

From the lab of Dr. Allen Braun, NIH/NIDCD
2
Narrative and the Human Brain

• Why Look at Narrative Production/Story Telling?
• Narrative Production incorporates all of the
  levels of linguistic capacity and performance
  
• Phonetic / Phonological (sounds their
  organization)
  
• Lexical (word meaning and grammatical class)
• Sentential (sentence meaning and structure)
• Discourse Pragmatic (context, conventions of
  language use, speaker goals)
  
• Each of these levels can break down due to brain
  damage, e.g., in Aphasia and Traumatic Brain
  Injury (TBI).

2
3
Goals of this Investigation

• The purpose of this neuroimaging study of
  narrative production was to examine, in an
  exploratory fashion, the generation of separate
  components of language within spontaneous
  narration, using the technique of Positron
  Emission Tomography (PET).

This was accomplished by Identifying and definin
g the linguistic variables that signal these
components of language Linking these variables t
o changes in regional Cerebral Blood Flow (rCBF)
as measured by PET.
3
4
Methodology Narrative Production with PET
Participants produced spontaneous (not
rehearsed) narratives about personal topics
(e.g., weddings, vacations) while undergoing PET
scans.
The PET scans were performed using Scanditronix
PC2048-15B tomography (Uppsala, Sweden). This
machine had an axial and in-plane resolution of
6.5mm.
Participants eyes were patched and head motions
were restricted by using a thermoplastic mask.
Each scan required that 30mCi of H215O be
injected intravenously.
The narration task was initiated 30 seconds
prior to scanning and continued throughout the
scan. Scans began when the count rate within the
brain reached a threshold value and then
continued for 4 minutes.
4
5
Behavioral Task Personal Narrative

• Participants were asked to relate an event or
  series of events from personal experience, e.g.,
  weddings, vacations. This task required that the
  participants use normal speech rate and
  intonation.

Participant speech output was recorded and a
computerized signal signified the start of the
scan The bolus of radioactive water takes 5 seco
nds to reach the brain after injection.
The bolus takes about 20 seconds to circulate
through the brain. As a result, the most useful
information is between 5-25 seconds
post-injection. However, in order to include com
plete utterances, approximately 35 seconds of
narrative was submitted to coding and analysis
for each subject.
5
6
More Details

• Participants were 18 normal volunteers
• 7 women (mean age 36 SD9.84 range 24-50 yrs)
  
  
• 11 men (mean age 33 SD8.15 range 23-47 yrs)
  
  
• All participants were right-handed and were
  devoid of any medical or neuropsychological
  illness that could affect the outcome of the
  resultant scans.

PET Data Analysis The 18 PET scans (one per sub
ject) were analyzed using MEDx (Medical Numerics)
3.4.3 image processing and analysis software to
motion correct and intensity-normalize raw
images. The resultant images were normalized to
a common stereotaxic space and smoothed to
accommodate for anatomical differences with
Statistical Parametric Mapping (SPM99) software.
6
7
Linguistic Analysis of Narratives

• Data were tape-recorded, transcribed and verified
  for accuracy. Recordings of narratives last
  approximately 35 seconds.
  
• Each of the narratives was broken down into
  sentences or T-units, defined as an independent
  clause plus its dependent modifiers.
  
• In order to narrow the scope of this
  presentation, 8 variables were chosen as measures
  of production of different components of language.

7
8
Linguistic Variables
I. Phonetic/Phonological Level Speech Sound
Complexity Speech Rate of Syllables in duratio
n of narrative (35sec) Fundamental Frequency (F0
) Range Range between lowest and highest
fundamental frequency, i.e., pitch Measures
intonation.
II. Lexical Level Semantic Complexity
Type-Token Ratio of Different Words / Total
of Words Measures lexical semantic richness.
Proper Names Names of specific individuals or
places per T-unit.
III. Phrasal Level Syntactic Complexity
Mean Length of Utterance (MLU) Words per
T-unit. Clauses Main, relative, coordinate, co
mplement, adjoined clauses per T-unit.
IV. Narrative Level Pragmatic Complexity
Inclusion Ratio of T-unit words to maze items,
e.g., pause fillers, false starts, repetitions,
reformulations Measures fluency.
Cohesion The grammatical means by which
sentences and paragraphs are linked and
relationships between them established, e.g.,
pronouns, determiners, conjunctions, conjuncts
and adverbials.
8
9
A Word about Scoring of Cohesion

• Narrative Cohesion was rated on a scale from 0 to
  9
  
• 0 Absence of Cohesion Interconnections among
  propositions are completely lacking.
  
• 1 Weak Cohesion Narratives with fairly
  imprecise or erroneous referential ties Not a
  logical succession of propositions.
  
• 3 Reasonable Cohesion Rather logical
  succession of propositions but includes some
  erroneous referential ties and intrusions, or
  lacks information critical to the
  macrostructure.
• 6 Good Cohesion Narratives with cohesive and
  logical succession of propositions but possibly
  lacking explicit referential ties among
  macropropositions.
• 9 Strong Cohesion Narratives with cohesive
  macrostructure in which referential ties are
  explicit and accurate on all textual levels The
  narrative follows the standard form of setting,
  action, and resolution and/or conveys a sense of
  purpose.

9
10
Illustrations of 8 Linguistic Variables (I)
Low Type-Token Ratio .570 Subject 16
This was one of the things that was practiced at
the rehearsal. The bride was there, the groom was
not there, so I had to stand in for the groom and
walk down the aisle with the bride-to-be. That
was rather nice, even though I didn't know the
bride. A lot of people were at the rehearsal.
After the rehearsal there was a rehearsal dinner
at the uncle of the bride's house. He had a very
elaborate house, a very big house across the
street from UCLA, uh which was also near the
church.
High Proper Names 2.25 PNs/T-unit Subject
10 I got a phone call one day from a good buddy
of mine at Reuters News Pictures in Guatemala
City and I went up, uh, clears throat wanted me
to come up to Guatemala City and, uh, visit him
about seeing a particular lady who was with the
FMLN, the leftist rebel group in, in Salvador,
fighting against the military there. So I flew up
to Guatemala the next day and we arranged to
drive down uh from uh Guatemala City to Porto San
Jose, which was on the Pacific coast of uh, San-
Guatemala.
10
11
Illustrations of 8 Linguistic Variables (II)
Low MLU 15.57 Words /T-unit and Low Clauses
1.86 clauses/T-unit Subject 11 ?MLU and Clauses
highly correlated! The uh, the choir sang for
us, which was very nice, they sang Mozart's Ave
Verum and then other things by Mozart uh, a very
nice service. The best man was Al Schreyer, who
was a post-doc with me at Emory uh University. He
had since moved to-to uh Montreal. Uh he came
down for the ceremony. We then after the uh
wedding proceeded on quite a lengthy uh-uh
honeymoon. We went first to a place uh north of
Georgia, Helen, a very nice little town, for two
or three days. And then we drove to uh to
Saratoga which was quite a drive, uh where we uh
we uh we played the horses and uh we uh we heard
music uh the Philadelphia Orchestra.
High MLU 24 Words /T-unit Subject 9
During the first semester of the program, we
took four classes. Uh, one of them was an
assessment class, a class that taught us some of
the psychological tests used in psychological
assessments, like the Rohrschach and the Brace,
which is a measure of kind of more cognitive
types of development. Another class we took was
in learning and memory, which was a basically a
class that covered a lot of the research that had
been done in that area. The class that we took um
with the learning and memory professor, involved
a lot of learning and memory itself.
11
12
Illustrations of 8 Linguistic Variables (III)
High Clauses 3.4 Clauses/T-unit Subject 17
He enjoys it with some of his close friends.
It's very different than the shooting that takes
place here. And every year he's there for about a
week before either I leave or Sean and I leave
ourselves. Uh we will go, or either myself or he
will go with us about the first week in January
and get on a plane, fly to Miami and then fly on
over to San Jose, arrive in the airport, make our
way through customs as fast as possible. And
usually when we do go we take all kinds of things
with us that we try to sneak through customs
because theyre so, the dutys so expensive
there.
Low Inclusion 90 Words/(Words Maze Words)
Subject 8 And he uh, studied there and uh was
in the Navy. And he graduated from the Naval
Academy and, uh, was in an a- he actually uh was
in Viet Nam, and he served there, and, several
times. And he uh graduated from the Naval
Academy. And um then he met my mom about that
time and they got married. And uh, he was in the
Navy for a long time and uh we all came along not
too long after they were married. But uh he so
hes a military man.
12
13
Illustrations of 8 Linguistic Variables (IV)
Low Cohesion Rating 3.0 (Subject 18)
Uh I guess the thing to do in Atlanta if I was
going to be a tourist guide would be go see an
Atlanta Braves game, which is amazing, that
they're starting to win now, winning
consistently, thirty-four about thirty-three of
my years they didn't win at all. Atlanta Falcons,
they apparently have a new stadium, which is
called the Georgia Dome. Supposed to be really
nice. Then theres this thing called Underground
Atlanta which was built by the Ralph Company
which is a lot like the uh Harbor Place or
something like Harbor Place in Baltimore and
Faneuil Hall in uh Boston built by the same
company.
High Cohesion Rating 9.0 (Subject 5)
Um, but then the next day it's a twelve-mile
hike, minimum, uh if you don't take any side
trips and that's to reach the summit and the
caldera and to go around and-and hit the second
uh cabin. And that's important, to hit the cabin,
because we were going in winter and there is snow
sometimes up there. So we were at about 11,000
feet and then Rick started feeling ill. And by
the time we hit 13,000 feet he just really
couldn't go on, it was pretty obvious. And we
were, could tell that we werent gonna make it
all the rest of the way anyway cause wed been
hiking pretty slowly. So we decided to set up
camp.
13
14
Results Correlating Brain Activity with Each of
the 8 Measures

• The 18 subjects narratives contained on average
  5 T-units, and 13 words and 2 clauses per T-unit.
  Mean rate was 4 syllables/sec.
  
• Cross-correlations of the linguistic measures
  revealed that MLU, Speech Rate and Clauses were
  highly linked, as were Clauses and Type-Token
  Ratio.
• The 8 linguistic variables were correlated with
  PET scans acquired during this particular
  language task. The correlation technique takes
  into account intersubject variability and assumes
  a linear relationship between language score and
  rCBF values.
• Correlations were performed with MEDx software,
  by running multiple regression analyses on the 18
  brain images with the 18 scores derived from each
  of the 8 chosen variables.
• Shown are resultant z-maps overlaid upon a
  standard MRI template, thresholded at 1.96, with
  the exception of Proper Names, which is
  thresholded lower at 1.00.

14
15
Activation Correlated with Speech Rate

• Left Right Frontal Lobe
• Left Right Insula
• Left Middle Temporal Gyrus
• Left Right Occipital Lobe, Cuneus
• Right Superior Temporal Gyrus

15
16
Activation Correlated with Fundamental Frequency

• Right Left Frontal Lobe
• Right Caudate Claustrum
• Right Left Insula
• Right Limbic Lobe, Cingulate Gyrus
• Right Parietal Lobe, Angular Gyrus
• Right Left Superior Parietal Lobule

Left Limbic Lobe, Parahippocampal Gyrus
Left Occipital Lobe, Cuneus
16
17
Activation Correlated with Type-Token Ratio

• Left Inferior Temporal Gyrus
• Left Superior Temporal Gyrus
• Left Insula
• Left Occipital Lobe, Cuneus
• Left Parietal Lobe, Postcentral Gyrus
• Left Inferior Parietal Lobule
• Left Claustrum
• Right Occipital Lobe, Lingual Gyrus
• Right Occipital Lobe, Cuneus
• Right Inferior Parietal Lobule

17
18
Activation Correlated with Proper Names

• Left Middle Frontal Gyrus
• Left Medial Frontal Gyrus
• Left Inferior Frontal Gyrus
• Left Occipital Lobe, Lingual Gyrus
• Left Middle Occipital Gyrus
• Left Limbic Lobe, Cingulate Gyrus
• Left Right Superior Temporal Gyrus
• Right Superior Parietal Lobule

18
19
Activation Correlated with MLU

• Left Superior Temporal Gyrus
• Left Inferior Parietal Lobule

Right Parahippocampal Gyrus Right Thalamus
19
20
Activation Correlated with Clauses

• Left Inferior Frontal Gyrus
• Left Superior Temporal Gyrus
• Left Inferior Parietal Lobule
• Right Thalamus
• Right Limbic Lobe, Cingulate Gyrus
• Right Superior Parietal Lobule

20
21
Activation Correlated with Inclusion

• Left Right Fusiform Gyrus
• Left Limbic Lobe, Parahippocampal Gyrus
• Left Right Superior Temporal Gyrus
• Left Right Occipital Lobe, Lingual Gyrus
• Left Right Inferior Frontal Gyrus
• Left Thalamus
• Left Right Frontal Lobe, Precentral Gyrus
• Left Limbic Lobe, Posterior Cingulate
• Left Superior Frontal Gyrus
• Left Right Parietal Lobe, Precuneus
• Right Middle Frontal Gyrus
• Right Caudate
• Right Occipital Lobe, Cuneus
• Right Inferior Superior Parietal Lobule

21
22
Activation Correlated with Cohesion

• Right Inferior Frontal Gyrus
• Right Superior Frontal Gyrus
• Right Left Middle Frontal Gyrus
• Right Left Medial Frontal Gyrus
• Right Temporal Lobe, Fusiform Gyrus
• Right Left Limbic Lobe, Anterior Cingulate
  Cingulate
  
• Right Left Parietal Lobe, Precuneus
• Right Left Occipital Lobe, Cuneus

Left Superior Temporal Gyrus Left Limbic Lobe, P
arahippocampal Gyrus Left Frontal Lobe, Precentra
l Gyrus
22
23
Summary of Results

• Left-Lateralized Measures
• Speech Rate (frontal, temporal, occipital)
• Type-Token Ratio (temporal, occipital)
• Proper Names (frontal)
• MLU (temporal)
• Clauses (frontal, temporal, parietal)

• Right-Lateralized Measures
• Fundamental Frequency Range (frontal, medial)
• Cohesion (frontal, medial/limbic)

Bilateral and Medial Measure
Inclusion (Occipital, Temporal, Parietal, Limbic)
24
Speculations about Underlying Operations
Performed by Distinct Brain Areas

• Left Frontal Phonological syntactic
  computation
  
• Left Temporal Lexical retrieval
• Left Right Parietal Episodic memory,
  Selection
  
• Right Frontal Affective expression (e.g.,
  through intonation), Event ordering
  
• Bilateral Occipital Imagery
• Medial/Limbic Causal ordering, Theory of Mind,
  Selection

24
25
Conclusions

• These findings suggest that the phonetic/
  phonological, lexical semantic, syntactic and
  pragmatic components of language production in a
  narrative context can be related to specific,
  distinct areas of activation in the brain, even
  though all are being generated simultaneously.

The findings on localization of semantic,
syntactic and pragmatic components agree with
neuroimaging studies of comprehension, which are
more plentiful (e.g., Xu et al., 2005).
These results are consistent with lesion
literature on deficits in narrative construction
(see Mar, 2004).
25
26
Future Research

• Experiments with new Methodologies
• With fMRI Better temporal and spatial
  resolution New techniques can cancel out head
  motion artifacts caused by overt speech.
  
• With patients as subjects in neuroimaging
  experiments Can discover compensatory mechanisms.

Experiments with new Linguistic Behaviors
With more controlled behavior, e.g., Story
Retelling, Picture Description
With other types of discourse, e.g., Expository
26
27
References

• Braun AR, Guillemin A, Hosey L, Varga M (2001)
  The neural organization of discourse an H2
  15O-PET study of narrative production in English
  and American sign language. Brain 124(Pt
  10)2028-44.
• Mar RA. (2004) The neuropsychology of narrative
  story comprehension, story production and their
  interrelation. Neuropsychologia 42(10)1414-34.
  
• Xu J, Kemeny S, Park G, Frattali C, Braun A.
  (2005) Language in context emergent features of
  word, sentence, and narrative comprehension.
  Neuroimage 25(3)1002-15.

27
28
Heres to Stories with Happy Endings!