Word category and verb-argument structure information in the dynamics of parsing

1
Word category and verb-argument structure
information in the dynamics of parsing

• Frisch, Hahne, and Friedericie (2004) Cognition

2
Word class information

• Major word categories nouns, verbs,
  prepositions, adverbs, etc.
• Necessary for telling whether a concatenation of
  words is legal or not in that language
• e.g.
• the doctor (Determiner noun) is okay
• the of (Determiner determiner) is not

3
Phrase structures
?Different members of the same category can have
different, lexeme-specific relationships to other
elements e.g (1a) Anne visited the doctor last
summer. (2) Anne sneezed the doctor last
summer.
4
Different views

• Both phrase structure and argument structure
  information are used to restrict the number of
  structural alternatives
• Phrase structure preferences alone are used for
  initial sentence processing

5

• Visit
• Usually appears after a grammatical subject (can
  be a noun, can be a gerund)
• Usually doesn't appear after a preposition or a
  determiner
• Needs to be conjugated for distinctions like
  number, person, voice, mood, and tense, etc.
• Usually takes two arguments (the one who is
  visited and the one who visits)
• The visitor is usually animate
• Work
• Usually appears after a grammatical subject (can
  be a noun, can be a gerund)
• Usually doesnt appear after a preposition or a
  determiner
• Needs to be conjugated for distinctions like
  number, person, voice, mood, and tense, etc.
• Usually takes one argument (the one who works)
• The argument is usually animate
• Doctor
• Usually appears after a determiner or an article
• Usually takes the thematic role of agent
• ..
• .

6
Different views

• Both phrase structure and argument structure
  information are used to restrict the number of
  structural alternatives
• Phrase structure preferences alone are used for
  initial sentence processing

7

• Does word class information processed prior to
  argument information?
• What happens if there are double violations?
  Additive or not?

8
ERP components

• Qualitative differencedifferent latencies,
  spatial distributions or polarities in different
  conditions
• Quantitative differenceamplitude modulations
  without changes in latency or topography
• (E)LAN
• N400
• P600

9
N400
10
left anterior negativity (LAN)
Expect the Unexpected Event-related
Brain--Response to Morphosyntactic Violations
Coulson, King and Kutas 1998 Language
and cognitive processes, 13 (1), 21-58
11
P600
Expect the Unexpected Event-related
Brain--Response to Morphosyntactic Violations
Coulson, King and Kutas 1998 LANGUAGE
AND COGNITIVE PROCESSES, 13 (1), 21-58
12
Conditions 2 (phrase structure) 2(argument
structure)

• Correct
• in the garden was often worked and
• (Work was often going on in the garden.)
• Phrase structure violation only
• in the garden was on-the worked and
• Argument structure violation only
• the garden was often worked and
• Phrase structure argument structure violation
• the garden was on-the worked and
• 160 critical items160 filler items

13
Experiment 1

• Procedure
• Word by word visual presentation
• Duration 400ms
• Inter-stimulus interval 100ms
• Subjects were asked to perform an acceptability
  judgment 800ms after the final word of each
  sentence.
• ERP recordings
• Data analysis
• Trials with incorrect responses and/or ocular
  artifacts are excluded from the averages.
• Time window 300-600 (for N400 effect)
• 600-1200 (for P600 effect)

14
(No Transcript)
15
Experiment 1

• Procedure
• Word by word visual presentation
• Duration 400ms
• Inter-stimulus interval 100ms
• Subjects were asked to perform an acceptability
  judgment 800ms after the final word of each
  sentence.
• ERP recordings
• Data analysis
• Trials with incorrect responses and/or ocular
  artifacts are excluded from the averages.
• Time window 300-600ms (for N400 effect)
• 600-1200ms (for P600 effect)

16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
Prediction
LAN N400 P600
Phrase structure violation (?) ?
Argument structure violation ? ?
Double violation (?) ? ?
20
Results Exp 1

• Exper 1(visual)

Fig. 1.
21
Results Exp 1
22
Results Exp 1
23
Summary
LAN N400 P600
Phrase structure violation ?
Argument structure violation ? ?
Double violation ?
24
Experiment 2

• Procedure
• Auditory presentation
• Normal speech rate
• Subjects were asked to perform an acceptability
  judgment 800ms after the final word of each
  sentence.
• ERP recordings
• Data analysis
• Trials with incorrect responses and/or ocular
  artifacts are excluded from the averages.
• Time window 200-400 (for ELAN effect)
• 300-500 (for N400 effect)
• 700-1200 (for P600 effect)

25
Prediction
LAN N400 P600
Phrase structure violation ? ?
Argument structure violation ? ?
Double violation ? ? ?
26
Results Exp 2

• Exper 2(auditory)

Fig. 4.
27
Results Exp 2
28
Results Exp 2
29
Summary
LAN N400 P600
Phrase structure violation ? ?
Argument structure violation ? ?
Double violation ? ?
30
comparison

• Exper 1(visual)

• Exper 2 (auditory)

Fig. 4.
Fig. 1.
31
comparison
comparison

• Exper 1(visual)

• Exper 2 (auditory)

32
comparison
comparison

• Exper 1(visual)

• Exper 2 (auditory)

33
Summary

• The integration of major category information and
  of lexeme-specific argument taking properties of
  verbs elicit qualitatively different brain
  responses.
• Phrase structure violation elicit (a LAN followed
  by) a P600 in the ERP
• The LAN effect is more robust in the auditory
  modality, and is only marginally significant in
  the visual modality.
• Argument structure mismatches is associated with
  larger N400-P600 responses.
• Sentences containing double violations (phrase
  argument violation) elicited similar responses as
  sentences containing phrase violation along
• (a LAN followed by) a P600 in the ERP
• BUT no N400 effect!!
• The P600 doesn't seem to be additive when
  compared with two other conditions

34
General discussion

• The early phrase structure violation correlate
  LAN
• The presence of a LAN is independent of an
  additional argument structure violation
• supports the view that there exits a correlate
  for initial phrase structure processing.

35
General discussion

• The lexical integration effect N400
• The larger N400 in this biphasic response
  indicates the semantic/thematic problems which
  arise when a NP argument cannot be assigned a
  thematic role by the verb
• The absence of N400 effect in the double
  violation condition
• supports the structure-first approaches to
  parsing
• There is a functional priority of word category
  integration (phrase structure processing) over
  the integration of all other information types
  (e.g. argument structure)
• Failure to integrate phrase structures would
  block the following argument structure
  integration.

36
General discussion

• The post-initial evaluation effect P600
• The P600 elicited in the two different kind of
  violation may reflect different kind of syntactic
  repair due to differences in the nature of the
  violations
• Phrase structureconcatenating items to derive a
  phrase structure representation
• Argument structurematching process between a
  (legal)structure output and more detailed
  information from the specific lexical entry
• The non-additive P600 in the double violation
  condition
• Ceiling effect, or
• The revision processes are primarily determined
  by phrase structure violations, and are
  independently of other types of information.

37
Conclusion

• Mismatch of major category information leads to
  an enhanced LAN as well as P600.
• Lexeme specific argument taking properties of
  verbs is associated with a less reduced N400 and
  an enlarged P600.
• The successful integration of word category
  information typically precedes the application of
  verb-argument structure information.

38
Questions?

• All critical words in the experimental item as
  well as filler items are repeated for 4 times.
• Some researchers had found larger N400 for Phrase
  structure violations
• A word may have more than one argument structure,
  like
• Frank sneezed the napkin off the table.
• Is the distinction between major category
  information and argument information theory (e.g.
  GB) specific? Could it be
• If failure of integrating word category
  information blocks integration of other
  information, we will never be able to pick up new
  grammatical use of words and wont be able to
  understand sentences like this one Dont you try
  to blue pin me.

39
Questions?

• All critical words in the experimental item as
  well as filler items are repeated for 4 times.
• Some researchers had found larger N400 for Phrase
  structure violations
• A word may have more than one argument structure,
  like
• Frank sneezed the napkin off the table.
• Is the distinction between major category
  information and argument information theory (e.g.
  GB) specific? Could it be
• If failure of integrating word category
  information blocks integration of other
  information, we will never be able to pick up new
  grammatical use of words and wont be able to
  understand sentences like this one Dont you try
  to blue pin me.

40

• Federmeier, Kara D Segal, Jessica B Lombrozo,
  Tania Kutas, Marta. Brain responses to nouns,
  verbs and class-ambiguous words in context.
  Brain. Vol 123(12) Dec 2000, 2552-2566.

41
Questions?

• All critical words in the experimental item as
  well as filler items are repeated for 4 times.
• Some researchers had found larger N400 for Phrase
  structure violations
• A word may have more than one argument structure,
  like
• Frank sneezed the napkin off the table.
• Is the distinction between major category
  information and argument information theory (e.g.
  GB) specific? Could it be
• If failure of integrating word category
  information blocks integration of other
  information, we will never be able to pick up new
  grammatical use of words and wont be able to
  understand sentences like this one Dont you try
  to blue pin me.

42
Questions?

• All critical words in the experimental item as
  well as filler items are repeated for 4 times.
• Some researchers had found larger N400 for Phrase
  structure violations
• A word may have more than one argument structure,
  like
• Frank sneezed the napkin off the table.
• Is the distinction between major category
  information and argument information theory (e.g.
  GB) specific? Could it be
• If failure of integrating word category
  information blocks integration of other
  information, we will never be able to pick up new
  grammatical use of words and wont be able to
  understand sentences like this one Dont you try
  to blue pin me.

43
Questions?

• All critical words in the experimental item as
  well as filler items are repeated for 4 times.
• Some researchers had found larger N400 for Phrase
  structure violations
• A word may have more than one argument structure,
  like
• Frank sneezed the napkin off the table.
• Is the distinction between major category
  information and argument information theory (e.g.
  GB) specific? Could it be
• If failure of integrating word category
  information blocks integration of other
  information, we will never be able to pick up new
  grammatical use of words and wont be able to
  understand sentences like this one Dont you try
  to blue pin me.

44
Early left anterior negativity
Adapted from