Martin J. Pickering, Simon Garrod, Behavioral and Brain Sciences,
2012.
An Integrated Theory of Language
Production and Comprehension
Computer Science & Engineering
2012-20835
Sang-Woo Lee
Background - Aphasia
Broca’s Aphasia
Expressive aphasia
Agrammatic aphasia
Understand what other people say, but cannot speak the
sentence well.
Caused by damage to, or developmental issues in the anterior
regions of the brain
Including (but not limited to) the Broca’s area
3
Wernicke’s Aphasia
Also known as Receptive Aphasia
Fluent apahsia, or sensory aphasia
Speak the sentence fluently, but not well-organized sense in their
speech
Traditionally associated with neurological damage to Wernicke’s area
in the brain
(Actually it is not just simply associated to Wernicke’s area in current experimental
result, but anyway…)
4
There are modules which specify some function
perception
action
The “classical Lichtheim-Broca-Wernicke” Model
5
Contents
Background - Aphasia
Traditional independence of production and comprehension
Interweaving in action and action perception
Perception process in action
Predict next action of other
Joint Action
Interweaving in Production and Comprehension
Comprehension process in Production
Predict next speech of other
Interactive Language
Professor’s Question
6
Traditional independence of
production and comprehension
Traditional model of communication
Discrete stages
A produces, B comprehends
B produces, A comprehends
8
Horizontal Split
Assumes “horizontal split” between production
and comprehension
Arrows-within-arrows indicate feedback (in
interactive accounts)
But this feedback is internal to production or
comprehension
It may involve “general knowledge”
But production does not involve comprehension processes
And comprehension does not involve production processes
9
Example of predict in Behavioral Instance
Interlocutors are not static, as the traditional model assumes, but
are “moving targets” performing a joint activitiy (Garrod &
Pickering, 2009)
10
Example of predict in Neuroscience
(Pickering & Garrod, 2007)
Big(neuter)
Painting
(neuter)
große Gemälde
Big (common)
Bookcase (common)
großen Bücherschrank
11
Other Counterexample
Also, many experiments demonstrate effects of
one on the other
Picture-word interference (Schriefers et al., 1990)
Word identification affected by externally controlled
cheek movement (Ito et al., 2009)
And strongly overlapping neural circuits for
production and comprehension (e.g.,
Pulvermüller & Fadiga, 2010; Scott et al., 2009)
12
Result
Prediction process could be naturally understood with production
module
When they comprehend the utterance, they also use production
model internally.
- There is forward model
To predict perception
caused by their own
utterance
- Fast alert when you say
something wrong
13
Interweaving in action and action perception
Perception process in action
Predict next action of other
Joint Action
Interweaving in action and action perception
Close links between action and action perception,
e.g.
participants’ arm movements affected by observing anot
her person’s arm movements (Kilner et al., 2003)
And making hand movements can facilitate concurrent vi
sual discrimination of deviant hand postures (Miall et al.,
2006)
Such links could have various purposes
Supporting overt imitation
facilitating memory or understanding (“postdictively”)
But authors propose that they aid prediction of ow
n and others’ actions, by use of a forward model
Based on computational neuroscience (Wolpert, 1997; se
e Grush, 2004)
Forward modelling in action
In our terms, the action command causes the action implementer
to move the hand and the perceptual implementer to construct t
he percept
And the efference copy causes the forward action model to gener
ate the predicted hand movement and the forward perceptual m
odel to construct the predicted percept
Forward Modeling in Action
- Just act
- Feel involved percepts
of your own act
e.g. Own coordination
info
Feeling of wind blowing
to your arm
Gravity info …
Efference Copy
18
Forward Modeling in Action
- Predict perception
caused by their own action
e.g. Own coordination info
Feeling of wind blowing to your arm
Gravity info …
Prediction-by-Simulation
- Predict perception of
other’s next action
by seeing other’s current
action
Joint Action
People are highly adept at joint activities (Seba
nz et al., 2006).
ballroom dancing, playing a duet, carrying a large o
bject together
Precise timing is crucial
To succeed, A predicts B’s action and B predict
s B’s action
Joint Action
22
Interweaving in Production and Comprehension
Comprehension process in Production
Predict next speech of other
Interactive Language
Forward modeling in language production
Action implementer production
implementer
Perceptual implementer comprehension
implementer
Action command production command
Drives the production implementer
Efference copy drives the forward models
Comparator monitor
compares the utterance percept and the predicted
utterance percept
Unifying production and comprehension
Production and comprehension are interwoven
Tight coupling in dialogue (Clark, 1996; Pickering &
Garrod, 2004)
Behavioural experiments show effects of
comprehension processes on production and vice
versa (e.g., Schriefers et al., 1990)
Overlap of brain circuits for production and
comprehension (e.g., Pulvermuller & Fadiga, 2010)
Such interweaving facilitates prediction of self
and other’s utterances
Classical modeling in language production
production command
i t
Production
implementer
Utterance
- Just say utterance,
- Listen what you say.
psem, syn, phont
Comprehension
implementer
Utterance percept
csem, syn, phont
Efference
copy
monitor
Predicted utterance percept
cˆsem, syn, phont
Forward
comprehension
model
Forward
production
model
Predicted utterance
pˆ sem, syn, phont
26
Forward modeling in language production
production command
i t
Production
implementer
- Predict perception
caused by their own
utterance
- Fast alert when you say
something wrong
Utterance
psem, syn, phont
Comprehension
implementer
Utterance percept
csem, syn, phont
Efference
copy
monitor
Predicted utterance percept
cˆsem, syn, phont
Forward
comprehension
model
Forward
production
model
Predicted utterance
pˆ sem, syn, phont
27
Self-monitoring
Speaker wishes to say kite
In the past, she has always constructed the
kite-concept and then uttered /k/
She therefore constructs forward model
p^[phon](t) = /k/
If she then incorrectly constructs p[phon] = /g/,
the monitor notices the mismatch
If she believes the forward model, she will
detect an error (and perhaps reformulate)
Otherwise, she will alter her forward model
Prediction-by-simulation
B’s utterance
Person B
B’s utterance
psem, syn, phonB t
psem, syn, phonB t 1
Derived intentional
act of
communication
iComprehension
A(t+1)
implementer
Comprehension
implementer
- Predict perception of
other’s next utterance
by listening other’s current
utterance
Utterance percept
Covert
imitation
csem, syn, phonB t
Utterance percept
csem, syn, phonB t 1
monitor
Inverse
model +
context
Predicted utterance percept
cˆsem.syn, phonB t 1
Person A
Derived production
command
iB t
Overt
Responses
Forward
comprehension
model
Derived production
command
Big(neuter)
Painting
(neuter)
iB t 1
Efference
copy
Forward
production
model
Predicted utterance
pˆ sem, syn, phonB t 1
29
Interactive Language
Joint action involves combining accounts of action
and action perception
Similarly, interactive language involves combining a
ccounts of production and comprehension
Facilitates coordination (e.g., short intervals between spea
kers; Wilson & Wilson, 2005)
Facilitates alignment (developing same representations; Pi
ckering & Garrod, 2004)
Alignment in turn facilitates comprehension (better predic
tion of others)
30
Interactive Language
31
Conclusion
We propose that language production and comprehen
sion are interwoven
It assumes a central role to prediction in production, c
omprehension, and dialogue
Speakers construct forward models to predict aspects
of their upcoming utterances
Listeners covertly imitate speakers and use forward mo
dels to predict the speakers
Our account helps explain the efficiency of production
and comprehension and the remarkable fluidity of dial
ogue
Thank you
Professor’s Question
Comprehension process in Production
Predict next speech of other
Interactive Language
Question 1
Q1: Give the evidence for how language production and
comprehension are tightly interwoven. How does this relate to
the perception-action cycle theory of cognitive systems?
35
Question 2
Q2: Explain and give the evidence for how action, action
perception, and joint action are interwoven. Explain how the
authors use this to develop accounts of production,
comprehension, and interactive language.
participants’ arm movemen
ts affected by observing an
other person’s arm movem
ents (Kilner et al., 2003)
And making hand
movements can facilitate
concurrent visual
discrimination of deviant
hand postures (Miall et al.,
2006)
Action implementer
production implementer
Perceptual implementer
comprehension
implementer
Action command
production command
Comparator monitor
36
Question 3
Q3: Give examples of what behavioral and neuroscientific data on
language processing can be explained by the integrated theory of
language production and comprehension explains, while modular
theory does not.
- Neuroscientific data
(Pickering & Garrod, 2007)
- Behavioral data
37