This study examined how syntactic and semantic processing interact during sentence comprehension using event-related brain potentials (ERPs). The researchers recorded ERPs as participants read sentences varying in the consistency between syntactic and semantic cues. They found that syntactic cues were more likely to be overridden by semantic cues when the cues were partially consistent, eliciting a P600 effect, but resisted semantic cues when inconsistencies required multiple edits, eliciting an N400 effect. The findings provide evidence that syntactic and semantic processing can interact adversarially and independently during sentence comprehension.
1. CONFLICT AND SURRENDER DURING SENTENCE
PROCESSING: AN ERP STUDY OF SYNTAX-
SEMANTICS INTERACTION
ALBERT KIM & LES SIKOS (2011)
2. INTRODUCTION
Human language processing requires a rapid extraction and coordination of
syntactic and semantic cues from linguistic input.
In serial architectures syntactic constraints strictly determine the possible
interpretations (Frazier, 1987; MacDonald et al, 1994).
In parallel architectures syntactic and semantic analyses occur partially
independent and in parallel processing streams in which interact in an
adverserial relationship (Bornkessel & Schlesewsky, 2006; Kim & Osterhout,
2005; Kolk & Chwilla, 2007; Kuberberg, 2007).
Studies indicate that semantic cues can sometimes drive interpretative
commitments even in the case of direct opposition from unambigious
syntactic cues (Kim & Osterhaut, 2005).
3. INTRODUCTION
Syntactic analyses vary from fragile to robust in their ability to form
semantic processing.
The level of robustness of syntactic analyses depends on syntactic cues
and types of context.
This robustness may be modulated by inter-individual differences
(Nakano, Saron, & Swaab, 2010).
The robustness may also be modulated by cross-linguistic differences in
the priority of specific linguistic features (Bornkessel & Schelesewsky,
2006).
4. METHODOLOGY
Example Sentence:
The hearty meal was devouring. Strong Semantic Attraction
Agent
Sentence above elicits a P600 effect at the verb but no N400 effect occurs.
Interpretation: Semantic attraction opposes and dominates syntactic cues, causing
the well-formed syntactic cues to appear ill-formed.
P600 effect is consistent with structural reprocessing in response to the perception
of syntactic structural anomaly.
5. METHODOLOGY
In another study:
The dusty tabletops were devouring Absence of Semantic Attraction
Agent
Sentence above elicits N400 effects, not P600 as tabletops are implausable as
agents or themes for the verb ‘devour’.
Interpretation: In the absence of semantic attraction, the processor pursues the
syntactically signalled but semantically implausable analysis.
According to Kim and Osterhout (2005) semantic processing can sometimes
dominate interpretation including unambigious syntactic cues.
6. THE PURPOSE OF THE STUDY
Kim and Sikos (2011) examined:
1. Whether supposedly unambiguous syntactic cues varied in their
susceptibility to challenges from opposing semantic attraction.
2. Whether supposedly unambiguous syntactic cues were more vulnerable
to opposition from semantic attraction when the syntactic cues were
partially consistent with the semantic attraction.
3. Whether syntactic cues would be more resistent to challenge from
semantic attraction when a partial consistency was reduced.
7. THE PURPOSE OF THE STUDY
Example of a stimuli sentence:
1a. The hearty meal was devoured control
1b. The hearty meal was devouring single-edit repair
1c. The hearty meal would devour multiple-edit repair
Kim and Sikos (2011) assumed that:
The ability of semantic attraction to control interpretation of their stimuli was
facilitated by the ease with which syntactic cues might be repaired with a single morpho-
syntactic edit devouring – devoured
The multiple-edit-repair sentences would enhance N400, reflecting the ability of
repair-resistant syntactic cues to support the syntatically licenced but implausable
interpretation winning the conflict opposing semantic cues.
8. METHODOLOGY
PARTICIPANTS
55 students from University of Colarado
15 participants were excluded due to
excessive electrophysiological artifacts
or for behavioral response accuracy
40 randomly assigned participants (20
male/20 female) were aged between
18-25 (mean:20).
All participants were right-handed
native English speakers with normal or
corrected to normal vision.
Test was a single session, lasting about
90 mins including 30 mins of
experiment preparation.
9. METHODOLOGY
STIMULI
96 stimulus items were created (control,
single-edit repair, multiple-edit repair).
Single-edit repair sentences could be
fixed by changing the verb (-ed/ing).
Multiple-edit repair sentences could be
fixed with more morphosyntactic
changes: changing the verb and adding
‘be’ to the verb.
As participants read plausable control
(1a), single-edit repair (1b), and
multiple-edit repair sentences (1c), their
ERPs were recorded.
Multiple-edit repairs contained the
same content words as single-edit
repairs but NOT the same the syntactic
cues.
The sentences were psedo-randomly
ordered and intermixed with filler
sentences.
10. METHODOLOGY
Example filler sentences:
1a. Well-formed and plausable sentence:
Seattle is famous for its rainy weather and pleasant temperatures.
1b. Semantically anomalous sentence:
*The old blender does not beg ice cubes anymore.
1c. Syntactically anomalous sentence:
*The angry driver will honks the horn at pedestrians.
11. RESULTS
Acceptability Judgment: Target sentence acceptability agreed with intended
judgments; controls are normal and single-edit and multiple-edit repair
sentences are unusual.
ERPs: All waveforms showed a clear negative-positive complex at first.
1. Single-edit repair wave forms showed a large positive shift, relative to
control, while multiple-edit repair sentences did not contain this positive shift
and instead contained a widely distributed negativity.
12. RESULTS
Broad Negativity:
Pairwise comparison showed the multiple-edit repair condition was more
negative-going than in the control condition, but the single-edit repair
condition did not differ significantly from control.
Examining each of the lateral channel groups individually, the multiple-edit
repair condition was more negative than control condition, while the single-
edit repair condition did not differ from control condition at any channel
groups.
13. RESULTS
P600 effect: (650-900 ms) Voltages in the single-edit repair condition were
more positive than in the control and multiple-edit repair conditions, reflected
in a main effect of sentence type in all groups.
Pairwise comparisons showed that the single-edit repair condition was more
positive-going than both the control and multiple-edit repair condition.
The multiple-edit repair condition was marginally more positive than control
condition.
The P600 effect was larger, which reflected in an interaction between position
and sentence-type.
14. DISCUSSION
When syntactic cues are nearly consistent with semantic attraction to an
anomalous syntactic analysis (single-edit repair), the combination of strong
semantic plausability and partial syntactic support generates an analysis that
challenges the single-edit repair condition.
When syntactic cues are more distinct from a configuration that would
accomodate the semantic attraction (multiple-edit-repair), they can resist the
alternative combinatory analysis, eliminating the P600 effect.
In situations involving conflict between a semantically attractive interpretation
and unambigious syntactic cues, the syntactic cues can vary in their ability to
resist vs surrender to challenge from semantics.
15. CONCLUSION
Findings contributed to the evidence that combinatory sentence processing
operates independently, and can sometimes dominate syntactic analysis.
In many sentence processing situations, the eliciting conditions of P600 are
consistent with reprocessing focused on lexico-syntactic analysis, while difficulty
with semantic access or integration manifests in N400.
N400 and P600 may both reflect domain-general mechanisms, which are
selectively recruited by semantic and lexico-syntactic processing difficulty,
respectively.
16. REFERENCES
Bornkessel-Schlesewsky, I., & Schelesewsky, M. (2008). An alternative perspective on ‘semantic P600’ effects in
language comprehension. Brain Research Reviews 59(1), 55-73.
Bornkessel-Schlesewsky, I., & Schelesewsky, M. (2006). The exyended argument dependency model: A
neurocognitive approach to sentence comprehension across languages. Psychological Review, 113(4). 787-821.
Frazier, L. (1987). Theories of sentence processing. In J. L. Garfield. (Ed.) Modularity in Knowledge
Representation and Natural Language Processing. (pp:291-307) Cambridge MA: MIT Press
Kim, A., & Osterhout, L. (2005). The independence of combinatory semantic processing: Evidence from event-
related potentials. Journal of Memory and Language, 52(2), 205-225.
Kolk, H., & Chwilla, D. (2007). Late Positives in usual situations. Brain and Language, 100(3), 257-261.
Kuperberg, G. R. (2007). Neural Mechanisms of language comprehesion: Challenges to syntax. Brain Research,
1146, 23-49.
MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S. (1994). The lexical nature of syntactic ambiguity
resolution [corrected]. Psychological Review, 101(4), 676-703.
Nakano, H., Saron, C., Swaab, T. Y. (2010). Speech and span: Working memory capacity impacts the use of
animacy but not of world knowledge during spoken sentence comprehension. Journal of Cognitive
Neuroscience, 22(12), 2886-2898.