A rose by Any Other Name: Long-Term Memory Structure and Sentence Processing. Kara D. Federmeier

1
A rose by Any Other Name Long-Term Memory
Structure and Sentence Processing. Kara D.
Federmeier Marta Kutas

• presented by Nataliya Chabanyuk
• LIN7912, Instructor Nina Kazanina
• University of Ottawa

2

• Getting both himself and his car to work on the
  neighboring island was time-consuming. Every
  morning he drove for a few minutes and then
  boarded the
• ferry
• gondola
• plane

3
Hypothesis and aims of the study

• The N400 component is sensitive to semantic and
  category membership manipulations.
• The N400 is used to examine to which extend the
  structure of long term memory interacts with
  contextual information during on-line sentence
  processing.

• How the readers processing is affected by memory
  structure even when that structure is irrelevant
  to the language comprehension task.
• What is the role of memory structure in reading
  by comparing its influence when sentence contexts
  are strong versus when they are weaker.

4
Background

• Psycholinguistic research Effect of context
• Words that are predictable in a sentence context
  are perceived and processed more rapidly and
  accurately than the same words when they occur
  out of context or in incongruent contexts.
• Ehrlich Rayner, 1981 Morris, 1994 Zola, 1984
  Duffy, Henderson Morris, 1989 Hess, Foss
  Caroll, 1995 McClelland ORegan 1981
  Stanovich West, 1983 Fischler Bloom, 1985
  Kleiman, 1980 Schuberth, Spoehr Lane, 1981
  Duffy et al., 1989 Morris , 1994 Ratcliff, 1987

• Electrophysiological
• support psycholinguistic findings.
• Contextual information is used early and builds
  continuously over the course of processing a
  sentence.
• Kutas Dale, 1997 Rugg Coles, 1995 Kutas
  Hillyard, 1984 Van Petten Kutas, 1990
  Schwantes, 1985 Styanokvich West, 1983
  Fischler Bloom, 1979 Kleiman, 1980
  Schwanenflugel LaCounbt, 1988

5
Psycholinguistic research Effect of context

• Contextual information decreases the duration of
  readers eye fixation.
• Congruent contexts facilitate the time to
  pronounce sentence/phrase final words.
• Congruent contexts facilitate the speed of
  lexical decision.

6
Electrophysiologicalsupport psycholinguistic
findings.

• N400 first observed in 1980 by Kutas Hillyard
  in the task of reading for comprehension
  (semantically anomalous sentence final words)
• Contextual information facilitates processing of
  highly predictable words (cloze probability)
• and of unexpected but contextually congruous
  words ( semantically related)

7
EEG with epileptic subjects

• Anomalous sentence endings were associated with
  large potentials in the left and right anterior
  medial temporal lobes. Those potentials are most
  probably generated in anterior fusiform,
  parahippocampal gyri and the hippocampus proper.

parahippocampal gyri
hippocampus
8
Language areas
9
Long-Term Memory and N400Language comprehension
crucially relies on information stored in
long-term memory, thus, the structure of
long-term memory affects word processing during
reading.

• Sensitive to
• Semantic, syntactic, contextual manipulations
• Word and picture recognition memory
• Category membership

• Insensitive to
• No semantic manipulations of context
• Grammatical or morphological violations
• Deviations in non-linguistic stimuli

10
Categorization researches

• Long term memory is structured on the basis of
  perceptual and functional similarities captured
  by semantic categories (taxonomic hierarchy).

a plant


a tree
a flower
a palm
a tulip
a pine
a rose
11
Experimental studyEffects of contextual
violations

• Within category violation.
• The item comes from the same semantic category
  as the contextually predicted item and thus share
  many features in common with it.

• Between-category violation.
• The item comes from the different semantic
  category and thus shares far fewer features in
  common with the predicted item.

12
Within category violation

• She felt that she couldnt leave Venice without
  the experience. It might be a touristy thing to
  do, but she wanted to ride in a
• ferry
• Expected
• gondola
• Between-category violation
• helicopter

13
Between category violation

• The patient was in critical condition and the
  ambulance wouldnt be fast enough. They decided
  they would have to use the
• ferry
• Expected
• helicopter
• Within-category violation
• plane

14
Your guess

• Amy was very anxious about traveling abroad for
  the first time. She felt surprisingly better,
  however, when she actually boarded the
• plane
• ferry, helicopter, gondola, boat
• helicopter
• ferry, gondola, boat
• gondola

15
ERP expectations

• In the same time window,
• best completion will elicit P300-P500
• between category violations N400
• What about within-category violation?

16
How the level of processing indexed by the N400
could be interpreted.

• 1. The system is sensitive to a general feature
  match between an item and a sentence context
• Similar amplitude to expected items and
  within-category violations.

• 2. The system is sensitive to specific contextual
  information
• Similar amplitude of N400 to both within and
  between-category violations.

17
How the level of processing indexed by the N400
could be interpreted.

• 3. The system is structured by feature similarity
  as reflected in in semantic categories
• N400 will elicit smaller amplitude to
  within-category violations related to
  between-category violations.

• 4. System is sensitive both to specific
  contextual information and to the relationship
  between concepts in long term memory
• N400 will be of the intermediate amplitude.

18
Study Design and Technique

• Materials
• 132 pairs of sentences with three target words (3
  lists of 44 sentences)
• The expected exemplar, the highest cloze
  probability
• The within-category violation, an unexpected
  (cloze probabilitylt0.05) exemplar from the same
  taxonomic category as the expected exemplar
• The between category violation, an unexpected
  (cloze probabilitylt0.05)exemplar from a different
  category than the expected exemplar.
• There were no lexical associates of any of the
  possible endings within the sentence containing
  the target word.

19
Study Design and Technique

• Material.
• Target items were rotated across the stimulus set
  such that each item appeared three times, once as
  each kind of ending.
• Target conditions were controlled for length,
  frequency and concreteness context sentences
  were controlled for length and grammatical
  complexity.

20
Study Design and Technique

• Cloze procedure cloze probability (CP) for a
  given word in a given context was calculated as
  the proportion of individuals choosing to
  complete that particular context with that
  particular word.
• Mean CP for the expected exemplars 0.74
• Mean CP for within category violations 0.004
• Mean CP for between category violations 0.001

21
Study Design and Technique

• Constraint.
• High constraint (one single preferable ending
  sentences with CP from 0.784 to 1.0
• Low constraint sentences (several compatible
  cloze endings) with CP from 0.17 to 0.784

22
Study Design and Technique

• Plausibility ratings.
• Does the ending make sense?
• Mean PR for expected exemplars95.6
• Mean PR for within-cat. violations28.3
• Mean PR for between-cat. violations15.3
• ANOVA confirmed the significant plausibility of
  Ending Type.

23
Study Design and Technique

• Participants.
• 10 men 8 women, 18 to 24 years of age,
  right-handed, monolingual English speakers

24
Study Design and Technique

• Experimental Procedure
• At the zoo, my sister asked if they painted the
  black and white strips on the animal.

25
xxxxxxxxxxxxx
26
I
27
explained
28
to
29
her
30
that
31
they
32
were
33
natural
34
features
35
of
36
a
37
zebra.
38
(No Transcript)
39
Study Design and Technique

• Experimental procedure.
• Sentence final words were presented for a
  duration of 500 ms
• I explained to her that they were natural
  features of a
• donkey.
• (200 ms)
• I explained to her that the were natural features
  of a
• poodle.
• (200 ms)

40
Study Design and Technique

• Experimental Procedure.
• Recognition memory test, consisting of 50 sets of
  sentence pairs 10 new ones, 20 unchanged
  experimental pairs, 20 modified sentence pairs-to
  be classified as new, old or similar.

41
Study Design and Technique

• EEG Recording Parameters.
• The electroencephalogram was recorded from 26 tin
  electrodes embedded in an Electro-cap, referenced
  to the left mastoid.
• Data Analysis.
• Trials contaminated by eye movements, excessive
  muscle activity, or amplifier blocking were
  rejected off-line before averaging.

42
Schematic of the electrode array used in the
experiment. In all, 26 scalp electrodes were
employed, arranged in a series of four equally
spaced concentric rings.
43
Results

• Behavior.
• Correctly classified 88 of the items on the
  recognition memory test.
• The most common mistake old classified as
  similar and similar as old.
• Conclusion
• The experimental sentences were attended during
  the recording session.

44
Results

• ERPs
• Early components in all conditions
• at posterior sites P1, N1, P2
• at frontal sites N1, P2

45
Results

• ERPs
• In the expected exemplar condition
• broad late positivity largest over central and
  posterior sites
• In two violation conditions
• Negative peaking around 400 ms (N400) largest
  over central and pariental sites.
• The N400 is followed by an extended late
  positivity of similar amplitude to that observed
  for the expected exemplars.

46
(No Transcript)
47
Results. Effect of ending type.

• Effect of ending type, shown at the right medial
  central site. A three-way split can be observed
  in the amplitude of the N400 response. N400
  amplitude was significantly larger for
  between-category violations and significantly
  larger for within-category violations than for
  expected exemplars.

48
Results. Distribution of the N400. Difference
waves showing N400 effect to within-category
violations and between-category violations. The
waveforms at the 16 electrode sites illustrate
the distribution of the N400 effect. For both
conditions, the N400 effect was larger over
medial posterior sites and slightly larger on the
right than on the left.
49
Results. Constraint effect.

• Effect of constraint on the N400 response, shown
  at the right medial central site. Constraint did
  not affect the response to expected exemplars or
  between-category violations. Within-category
  violations in high constraint sentences elicited
  smaller amplitude N400s than within-category
  violations in low constraint sentences.

50
Summary of main results.

• Expected exemplars elicit a late positivity in
  the 350-to 400-ms time window.
• Both within-category violations and
  between-category violations elicit a qualitively
  similar N400 response with a medial,
  posterior-central, right hemisphere distribution.
• The amplitude of the N400 is bigger for
  between-than for within-category violations.
• The amplitude of the N400 is bigger for
  within-category violations in low than in hight
  constraint sentences.

51
Discussion

• The aim of the study was to find out
• To which extend the processing of the final word
  in the sentence is affected not only by specific
  information directly activated by prior words
  (context) in the sentence, but also by more
  general, context-independent information
  (semantic feature overlap) indirectly deriving
  from the structure of real-world knowledge in
  long-term memory.

52
Discussion

• If the on-line processing of two items sharing
  significant numbers of semantic features in
  common differed when the preceding context more
  consistent with one than the other.
• If the on-line processing of two items neither
  of which is especially consistent with the
  context, nonetheless differed as a function of
  their semantic similarity to the most probable or
  expected ending.
• If the impact of either of these variables would
  be modulated by the degree to which the context
  anticipated a particular exemplar versus several
  possibilities.

53
Discussion

• Expected exemplar elicited a late positivity.
• Within-category violation elicited a moderate
  N400 between 300 and 600 ms
• Conclusion
• Context processing serves to make available
  specific feature information, not only general
  feature information.

54
Discussion

• The amplitude of the N400 is larger in
  between-category violations than in
  within-category violations.
• Conclusion
• The language processing system is sensitive to
  the categorical relationship.
• The influences of context and semantic feature
  overlap on a words processing are of the same
  kind.

55
Discussion

• What was confirmed
• Best completions enjoy greater processing
  benefits than semantically related but
  contextually unexpected endings, although both
  typically show facilated processing relative to
  contextually and semantically unrelated endings.

56
Discussion

• What was found
• The smaller N400 to within-category violations
  compared to between-category violations reflects
  an influence of semantic memory structure, built
  of real-world experience, on on-line language
  processing.
• The processing systems sensitivity to the
  feature overlap affords within-category
  violations a processing benefit relative to the
  between-category violations.

57
Discussion

• Contra arguments
• The results reflect lexical associative priming
  from a word in the sentence context to the
  expected exemplar and, by extension, to the
  within-category violation.
• No target sentences did not contain any lexical
  associates to any ending type
• distance

58
Discussion

• Contra-argument.
• Plausibility within-category violations elicited
  smaller N400 than between category violation
  because they were more plausible, i.e., actually
  did fit the context better.
• Plausibility alone does not suffice N400
  amplitudes are not monotonically related to rated
  plausibility. Broken down by contextual
  constraint.

59
Discussion

• Effect of constraint on the N400 response, shown
  at the right medial central site. Constraint did
  not affect the response to expected exemplars or
  between-category violations. Within-category
  violations in high constraint sentences elicited
  smaller amplitude N400s than within-category
  violations in low constraint sentences.

60
Discussion

• Contra-argument.
• Plausibility.
• Among the within-category violation endings, the
  more plausible endings elicit larger N400 than do
  the more implausible endings.

61
Discussion

• Plausibility.
• Checkmate! Rosaline announced with glee. She
  was getting to be really good at baseball. (H)
• elicit a smaller N400 than
• She keeps twirling it around and around under her
  collar. Stephanie seems really happy that Dan
  gave her that earring. (L)

62
Discussion

• As lexical associative priming and plausibility
  are not enough to account for the N400 amplitude,
• the explanation is inherent in the structure of
  information in long-term memory
• Context sets up very specific expectations for
  the expected exemplar and provides the
  within-category violations with the greater
  facilitation.
• Conclusion a functional link between the
  expected endings and the within-category
  violations, reflecting memory structure.

63
Discussion

• N400 responds to the degree of mismatch of every
  word related to the context.
• Overlap between of semantic features of the
  within-category violation and expected exemplar
  determines the size of the observed N400
  response.
• Thus, long-term memory structure has an inherent
  effect on sentence processing in real time.

64
Discussion

• What was known
• Semantic memory has a categorical structural
  component.
• What was observed
• A reliable category-based effect during the
  processing of the sentence final word outside of
  a categorization task.
• Conclusion
• Category-based structure of long memory
  influences language processing.

65
Conclusion

• Language comprehension system is sensitive to
  specific contextual information and to the
  consistency between that specific information and
  the meaning of a target word by around 375 ms
  into word processing.
• In the same time window the influence of
  contextual information on word processing and
  influence of semantic overlap is observed.
• It is an inherent influence of long-term memory
  structure on language processing.

66
Conclusion

• The processing of a sentence context results in
  the activation of a set of semantic features
  associated with the word or words that are likely
  to come next.
• Semantic feature (mis) match determines the
  difficulty pf processing.
• Stronger contexts allow better predictions and
  greater facilitation for items that share
  features with the predicted word.

67
Conclusion

• Context and long-memory structure have a dynamic,
  mutually dependent relationship with one another
  and contribute jointly to the process involved in
  making sense of what read.

68
The End!!!
Thank you!!!