Aphasia: Comprehension II

1
Aphasia Comprehension II

• 9.56/24.907
• November 2nd, 2004
• Christopher Hirsch

2
Administrivia

• Exam solutions to be posted
• tonight
• Papers (Nov 16th)
• 2 pg write-up (due in class)
• Next week Genetics
• And in what really matters . . .

3
Get out and
4
Grodzinskys TDH
5
Grodzinskys TDH
6
TDH (Review)

• Traces are deleted
• All traces of movement are deleted from
  syntactic representations
• Nonlinguistic, linear default strategy
• Every non-theta-marked DP is assigned a thematic
  role according to the default strategy
• No other syntactic deficits exist
• Brain-behavior mapping
• All Brocas aphasics are subject to TDH
• Damage to Brocas area leads to TDH problems
• Brocas area is solely responsible for licensing
  traces

7
Today

• Problems for the Trace Deletion Hypothesis
• Problems from previous lectures
• Non-predicted movement problems
• Problems for default strategy
• Brain-behavior mapping
• Non-predicted non-movement problem
• Conceptual issues

8
1. Previously Seen Problems

• Unrelated productions problems
• Passives
• Above chance performance
• Psychological verbs
• Referential NPs
• Which NP-questions
• Quantified subjects
• Object-extracted questions

9
Tree Pruning Hypothesis (TPH)

• Production problems in Brocas aphasia are due to
  pruning the syntactic tree above some (arbitrary)
  level
• Allows for variation
• Hebrew/Arabic data (pruning above AGR)
• Japanese data (pruning above TNS)
• Requires distinction of AGR and TNS node
• Problem under modern syntactic theory?
• No support for distinction from comprehension?

10
Passive Review

• Hirsch Wexler patients do not all show chance
  performance on actional passives (as predicted by
  TDH)

11
Passive Review
12
Passive Review

• All patients, however, performed worse on
  nonactional passives
• Nonactional passive performance was at chance
  level (49), not below chance (as Grodzinsky
  predicts)

13
Passive Review
14
Referential NPs

• Why does the default strategy apply only to
  referential NPs?
• Why can aphasics comprehend sentences involving
  non-referential NPs?

15
Referential NPs
16
Referential NPs
17
Wh-questions
18
Wh-questions

• CP Which applesj did IP the mani VP ti eat
  tj
• TDH would appear to predict below chance
  performance here, not the chance performance
  found by Hickock and Avrutin
• The first DP which apples would receive an agent
  role, while the second DP the man would receive a
  theme role

19
2. Non-predicted Movement Problems

• Psychological passives
• Fear-type
• Frighten-type
• Druks Marshall
• On-predicted patterns
• Meta-analyses
• Does an overarching agrammatic pattern exist?
• Verb movement problems

20
Psychological Verbs

• Fear-type active
• The dog feared the cat.
• Fear-type passive
• The cat was feared by the dog.
• TDH predictions
• Fear-type active AC
• Fear-type passive BC (???)
• Relevant Studies
• Grodzinsky (1995)
• Balogh Grodzinsky (1996)
• Hirsch Wexler (2003)
• Piñango (2000)

21
Grodzinsky (1995)
percent correct n per cell 14 (7 verbs per
condition, presented 4 times, 2 each for actives
and passives
22
Balogh Grodzinsky (1996)
passives only n 4
23
Hirsch Wexler (2003)
chance
24
Piñango (2000)
number and percent correct
25
Fear vs. Frighten

• The woman feared the man.
• Experiencer
  Theme
• The man was feared by the woman.
• Theme
  Experiencer
• The woman frightened the man.
• Theme Experiencer
• The man was frightened by the man.
• Experiencer
  Theme

26
Data
number and percent correct
27
Data Summarized
percent correct
28
Exceptions to the Generalization

• Hirsch Wexler
• Difficulties finding agrammatic patients
• Druks Marshall
• Meta-analyses

29
Hirsch Wexler
30
Difficulties

• Its hard to find agrammatics
• Beretta Munn (1998) screened 15 to find 6
• Hagiwara (1993) screened 70 to find 10
• Grodzinsky does not report number of patients
  screened
• Furthermore, Grodzinskys studies use passive
  comprehension problems as an inclusionary
  criterion

31
Druks Marshall (1995)

• Issues
• Present two case studies that do not fit with
  paradigmatic BA results
• Attempt to explain results as function of damage
  to Case system
• Subjects
• MH
• 42, female, left-frontal infarct, 6 years
  post-onset
• BA by BDAE
• BM
• 68, male, fronto-temporal lesion, 4 years
  post-onset
• not BA by BDAE (but BA is closest category)

32
Experimental Design

• Experiment 1 Picture Matching Task
• Active The cat is chasing the dog.
• Passive The dog is being chased by the cat.
• Experiment 2 Picture Question Task
• Active Who is pinching somebody?
• Passive Who is being kissed?
• Experiment 3 Existential Picture Matching Task
• Actives There is a chicken eating in the garden.
• Passive There is a chicken being eaten in the
  garden.

33
Data
correct ratio / percent
34
Data Summarized
conditions showing pattern / total conditions
35
Data Summarized
conditions showing pattern / total conditions
36
Edgar Zurif (1996)

• MH
• AC for actives expected
• BC for passives is odd, but notes that this still
  maintains finding of passives worse than
  actives also points out BC only for 2/3
  conditions (other was at C)
• BM
• not even BA by BDAE DM dismiss BMs melodic
  line and interrupted word run findings of
  better than the upper limit for BA
• lesion not even isolated to BA (frontal-temporal
  lesion) might include WA

37
Druks Marshall (1996)

• In response to Zurifs criticism that BM failed
  to match any of the categories delineated by the
  BDAE
• No more than twenty or thirty percent of
  dysphasic patients will fit neatly into one of
  the specific dysphasia syndromes (Albert,
  Goodglass, Helm, Rubens, Alexander, 1981)
• In response to Zurif noting that BM had a large
  frontal-temporal lesion (possible including WA),
  DM claim to be more concerned with behavioral
  characteristics than lesion localization

38
Berndt et al. (1996)

• Meta-analysis Findings
• 64 cases from 15 studies (1980- 1993)
• Only 36 cases had pattern where active sentences
  were performed better than chance level and
  passive sentences were understood at a level no
  different from chance
• Of remaining 64, 30 of the patients performed
  above chance in both conditions, whereas 34
  exhibited chance performance for both conditions
• Empirical demonstration against group studies
  in neuropsychology

39
Berndt et al. (1996)
MH
Active Passive
Active Passive
Passive Active
BM
Passive 40
Grodzinsky et al. (1999)

• At chance, Brocas aphasics responses must be
  treated like coin tosses
• Single-case approach is invalid
• Carry out a meta-analysis attempting to show that
  all Brocas aphasics fit the agrammatic profile
• Selection Criteria (1980-1996)
• Subjects must be Brocas aphasics
• (only 21 / 42 BA in Berndt et al.)
• Each data entry must be independent

41
Data

• Central Tendencies
• Actives 86
• Passives 50
• Simulate Passive Scores
• Unbiased coin tossed repeatedly for each of 42
  subjects, the length of the runs equal to the
  diverse number of trials in the tests (6-48)
• Results pooled from many simulations

42
The distribution of responses to active and
passive (vs. simulation of chance performance)
Chance performance is equivalent to flipping a
coin
43
Berndt Caramazza (1999)

• Criticize Grodzinsky et al. (1999) for patient
  selection bias (doctored data)
• Excluded data
• Obvious agrammatics
• New data
• Preselection bias

44
Zurif Piñango (1999)

• Excluded Data
• Those patients not diagnosed as Brocas aphasics
  (agrammatism tied to BA has lesion localizing
  value, whereas agrammatism, on its own, seems not
  to)
• Taking coarse-grained clinical instruments to
  establish diagnostic categories in face of
  detailed speech descriptions

45
Zurif Piñango (1999)

• New Data
• Added results from 21 cases not included in
  Berndt et al. (1996)
• A few cases of preselection on the basis of
  agrammatic comprehension pattern (10 / 42
  questionable)

46
Zurif Piñango (1999)

• Reanalysis of data (using 32 / 42 original
  subjects)
• Actives 84.2 (86)
• Passives 57.2 (50)
• Nothing has changed!

47
Caramazza et al. (2001)

• Statistical reasoning adopted by Grodzinsky et
  al. (1999) is flawed
• Brocas aphasia is NOT associated with a
  consistent pattern of sentence comprehension
  performance

48
Confusing N of subjects with N of trials
. . . Guessing behavior, which results in
chance performance, cannot, and should not, be
50 correct per subject. Rather, it should be
binomially distributed around the mean of 50
correct level. We can now see why results from
multiple subjects are so important in this
context in such a response-type, each subject
flips a coin and uses it for responding to each
experimental question. A single subject, then,
cannot be used to discern the pattern, if there
are experimental conditions that might result in
chance performance. This is so because the score
of this particular subject may be
located anywhere on a binomial curve. (p.137)
49
Confusing N of subjects with N of trials
. . . Guessing behavior, which results in
chance performance, cannot, and should not, be
50 correct per subject. Rather, it should be
binomially distributed around the mean of 50
correct level. We can now see why results from
multiple subjects are so important in this
context in such a response-type, each subject
flips a coin and uses it for responding to each
experimental question. A single subject, then,
cannot be used to discern the pattern, if there
are experimental conditions that might result in
chance performance. This is so because the score
of this particular subject may be located
anywhere on a binomial curve. (p.137)
WRONG!
50
Limits of Group Mean Analysis

• Grodzinsky et al.s (1999) hypothesis claims that
  agrammatic production is necessarily associated
  with agrammatic comprehension
• Requires each patient categorized as an
  agrammatic Brocas aphasic to also present with
  agrammatic comprehension
• The analysis of group means is therefore the
  wrong test for this hypothesis

51
Remaining Problems for TDH

• Only 5/26 at ceiling for actives
• 8 patients with lower confidence limit at 50 for
  actives
• 4 patients with confidence limits not surrounding
  50 for passives
• 1 patient with passives at 100

52
3. Problems for Default Strategy

• Double agent
• TDH accounts for chance performance on passives,
  object-extracted relative clauses,
  object-extracted clefts, etcetera by relying on
  thematic confusion (double-agents)
• Linear order matters
• Default strategy works over linear sequences,
  hence any movement that does not reorder NPs
  should have no effect upon comprehension

53
Beretta Munn (1998)

• Issue
• Test the default strategy of the TDH to see if BA
  actually invoke double-Agents
• Subjects
• 6 BA (15 subjects made first cut, 6 remained
  after second cut)
• Neurologically intact controls matched for age
  and education level

54
Selection Criteria

• First Cut (n 15)
• Neurological CVA involving Brocas area confirmed
  by CT
• At least 3 months postonset
• Neuropsychologically diagnosed as BA (nonfluent,
  effortful, and telegraphic speech, good
  comprehension at conversational level), as
  confirmed by clinical workup and BDAE
• Between ages of 18 and 80
• Second Cut (n 6)
• Demonstrate relevant agrammatics comprehension
  pattern (since this is a test of the TDH)
• AC for actives
• C passives

55
Screening Test Results

• Standard Two-Actor Pictures

percent correct C 50
56
Experimental Design
E.g. The woman kicked the giraffe. The giraffe
was kicked by the woman.
Correct
Predicted for passive given strategy
57
Results for Double-Agent Three-Actor Pictures
Active Sentences
percent correct C 33 (three options)
58
Results for Double-Agent Three-Actor Pictures
Passive Sentences
percent correct C 33 (three options)
59
Results for Double-Agent Three-Actor Pictures
Passive Sentences
percent correct C 33 (three options)
60
Does Linear Order Matter?

• TDH the default strategy should only lead to
  comprehension problems when a sentence deviates
  from a canonical Agent-Theme linear ordering
• That is, even if movement takes place, as long as
  relative ordering of thematic roles is
  maintained, no comprehension deficit should be
  noted (e.g. VP-internal subject raising)

61
Beretta et al. (2001)

• Scrambling Constructions
• Spanish Actives
• a la girafi a la mujerj tj ti la
  esta empujando
• to the giraffe the woman she is
  pushing
• Spanish Passives
• por la mujerj la girafai esta siendo
  empujando ti tj
• by the woman the giraffe is being pushed
• Korean Actives
• saja-luli key-kaj tj ti mul-eyo
• the lion, the dog bit
• Korean Passives
• key-ekeyi saja-kaj tj ti mul-hi-eyo
• by the dog the lion is bitten

62
TDH Predictions

• Active Below chance
• Two movements, object ends up to the left of the
  subject
• Passive Above chance
• Two movements, oblique ends up to the left of the
  derived subject

63
Results

• Aphasics were at chance on both active and
  passive scrambled constructions
• Fine on other control sentences
• Inconsistent with linear order strategy

64
Bastiaanse et al. (2003)

• Production study
• Not only relevant for TPH, but also relevant for
  TDH
• Dutch background information
• Subject-Object-Verb language
• Object can be moved in front of the adverb
• (object scrambling)
• John has yesterday a book bought
  (unscrambled)
• John the booki yesterday ti bought
  (scrambled)

65
Experiment 1

• 8 Brocas aphasics
• Sentence completion with two conditions
• Basic
• Scrambled

66
Experiment 1 (contd)

• Introductory sentence
• This man cuts the tomato and this man cuts the
  bread.
• Unscrambled condition
• Tester This is the man(s) who today(adv) the
  tomato(o) cuts(v) and this is the man who.
• Patient today the bread cuts
• Scrambled condition
• Tester This is the man(s) who the tomato(o)
  today(adv) cuts(v) and this is the man who
• Patient the bread today cuts

67
Results Findings

• Significant difference between basic and
  scrambled conditions

68
4. Brain-Behavior Mapping

• Grodzinsky (at his strongest / most modular)
  believes that damage to Brocas area leads to
  Brocas aphasia (for him TDH), and that Brocas
  aphasia (TDH) is only due to damage to Brocas
  area
• This leads to the following predictions
• Damage to Brocas area must lead to Brocas
  aphasia (TDH)
• Brocas aphasics (TDH) must have damage to
  Brocas area
• TDH-like problems should not be found in other
  populations
• Brocas area should not be involved in non-trace
  related tasks

69
Dronkers et al. (1992/1999)

• Only 85 of patients with chronic Brocas aphasia
  have lesions in Brocas area
• Only 50-60 of patients with lesions in Brocas
  area have Brocas aphasia
• BUT All of the patients with Brocas aphasia,
  even those without lesions to Brocas area, had a
  lesion that encompassed part of the insula

70
Dronkers et al. (1996)
71
Dronkers et al. (1994)
Anterior area 22 damage associated with poor
comprehension for elaborated morphosyntax.
72
Dick, Bates, Wulfeck, Dronkers (2001)

• Had college students interpret actives, passives,
  subject clefts and object clefts under
  acoustically degraded conditions.
• Able to effectively simulate the agrammatic
  profile.

Percent correct response by sentence type for
students under dual-stress conditions
73
Dick, Bates, Wulfeck, Dronkers (2001)

• Had college students interpret actives, passives,
  subject clefts and object clefts under
  acoustically degraded conditions.
• Able to effectively simulate the agrammatic
  profile.

Percent correct response by sentence type for
students under dual-stress conditions
Presumably we dont want to say that low-pass
filtering and compression induces a transient
trace deletion mode
74
Other

• Brocas area is also implicated in (at least) the
  following cognitive functions
• Mathematical processing
• Music processing
• Working memory (phonological loop)
• Certainly the area isnt only involved in the
  computation of syntactic traces

75
5. Another Non-predicted Problem

• Binding Theory
• Chien Wexler (1990)
• Is Mama Bear touching herself?
• Is Mama Bear touching her? X
• Is every bear touching herself?
• Is every bear touching her?
• Interpreted as showing children (up to 5-6 years
  old) understand Principle B, but are missing
  Principle P

76
Another Non-predicted Problem

• Binding Theory
• Grodzinsky, Wexler, et al. (1993)
• Is Mama Bear touching herself?
• Is Mama Bear touching her? X
• Is every bear touching herself?
• Is every bear touching her?
• Interpreted as processing problem, as it is
  unclear why Brocas aphasia should lead to loss
  of pragmatic knowledge (other pragmatic knowledge
  is spared)

77
6. Conceptual Issues

• Why trace deletion?
• Obviously the theory requires that traces be
  first computed, and only then deleted? What
  exactly is Brocas area doing then? Maintenance
  of traces?
• What motivates the default strategy?
• Why agent first? Why would it be
• universal? Does this fall out of
• anything deeper?

78
Short Big Aphasia Review

• Production problems
• CP TNS AGR mistakes
• TPH?
• Unaccusatives
• Comprehension problems
• Syntactic movement
• TDH?
• Issues
• Knowledge Loss vs. Processing Impairment
• So whats the right theory?

79
Next time . . .

• Parkinsons Disease (Basal Ganglia)
• One reading posted
• Practice exam aphasia problems to be posted this
  week(end)