Inanimacy as a predictive cue to derived subjects Evidence from the development of the “semantic” P600 Justine VanDyke-Lyon1, E. Matthew Husband1, and Fernanda Ferreira1 1
2. Methods
1. Introduction & Background Main Questions
I
What information does the parser use to predict upcoming structure? I Do children use this information the same way adults do? I
Children may have to rely on a default agent-first strategy which assigns agent interpretations to subjects (Townsend & Bever 2001).
Materials 2 (animacy) x 2 (voice) design; 24 items (96 stimuli total). I Content words matched for frequency across age groups.
Current Study Investigate a structural prediction mechanism with subject animacy as the relevant cue in adults and children.
Background Incrementality and temporary ambiguity: Sentential subjects are initially ambiguous between agent and theme interpretations. I Animacy affects production and comprehension of derived subjects (Ferreira 1994, 2000). I Inanimate subjects must be derived because they are incompatible with agent interpretation. I Theme interpretation is compatible with animate or inanimate subjects (Dowty 1991).
The P600 is typically elicited by structural violations and syntactic reanalysis (Osterhout & Holcomb 1992). I However, semantic violations can also elicit a P600 (Kuperberg, et al. 2003; Kim & Osterhout 2005). I Our take: Semantic P600s reflect a structural reanalysis of an erroneous prediction driven by subject inanimacy.
The use of animacy appears to be later developing (Drenhaus & Frey 2008; MacWhinney, Bates, & Kliegl 1984).
17 adults (18–20 years) and 17 children (7–9 years) I Native monolingual English speakers, Right handed, No history of speech, language, or hearing disorders, No neurological trauma or medications.
General Methods
The Mechanism Step 1. Parse subject and recognize animacy. Step 2. If inanimate, predict verb and derived subject; otherwise, predict verb. Step 3. Parse verb, check morphology against predicted structure. We hypothesize that children have difficulty with Step 2.
I
Adults: 81% Children: 74%
I
Thematic violations elicit an N400 on the verb. I 1 vs. 3 (T414, t(16)=2.198, p<.043). I 3 vs. 4 (not significant, all p>.05).
Discussion Adults use cues to thematic relations between subject and verb. I Children are unable to use these cues. Why? I Lack of cue access. I Lack of cue implementation. I
Adult Passive Control Active Violation
Different Result for Children
Child Active Control Active Violation
Thematic violations elicit a P600 on the verb. I 1 vs. 3 (T882 t(16)=3.912, p<.001). I 3 vs. 4 (T872, t(16)=4.588, p<.001).
Data Visualization Spatial maps plot the standard factor loading of the spatial PCA. I Virtual ERP waves plot the grand average factor scores over time by condition. I
Results
Results
3 vs. 4: P600 (T872, t(16)=4.588, p<.001). I 1 vs. 2: N400 (T394, t(16)=4.669, p<.001).
1 vs. 2: N400 (T416, t(16)=2.043, p=.05). I 3 vs. 4 (not significant, all p>.05).
I
P600
Discussion Inanimate subjects lead to a derived subject prediction (Step 2). I Verbal morphology disconfirms this predication, requiring the system to reanalyze (Step 3). I This reanalysis process is reflected in the P600. I Structural predication for animate subjects is less specified (Step 2). I No reanalysis is required upon reading the verb (Step 3). I This process is reflected in the N400. I
Replication with Adults I
Data were low-pass filtered offline at 30 Hz and epoched into 1000 msec windows (-100–900 ms), time-locked to the onset of the verbs in each condition. Ocular artifact removed (Dien, 2005) and bad channels corrected using spherical spline interpolation (Lou & Feree, 2000). I Each condition and each group averaged separately, rereferenced to linked mastoids, and baseline-corrected -100–0 ms. I Subject averaged ERP variance was analyzed using a two-step, spatial-temporal principal component analysis. (Spencer, Dien, & Donchin, 2001). I
P600
N400
I
P600
Discussion I
N400
Animacy is not used to make structural predications (No Step 2). I No reanalysis, no P600 (Step 3). I N400 may reflect increased processing to integrate event meaning of the passive. Child Animate Active Animate Passive
Overall Acceptability
ERP Analysis
5. Results for Children Adult Inanimate Passive Inanimate Active
Do children show neurophysiological evidence for thematic reanalysis similar to adults?
Adult Active Control Active Violation
I
Sentences were presented word-by-word using rapid series visual presentation, 700 msec per word (650 msec + 50 msec blank screen). I All sentences were followed by a probe question: “Is this a good/not so good sentence?” I 6 blocks of 16 sentences, pseudo-random order presentation. I
Continuous EEG: 64-channel QuikCap; Cz reference; sampled at 500 Hz; 100 Hz corner frequency, low-pass filtered online; Neuroscan SCAN software, SynAmps2 amplifiers.
4. Results for Adults
3. Initial Study Initial Question
Sentence The maid was folding the shirt. The maid was folded by the shirt. The shirt was folding the maid. The shirt was folded by the maid.
I
Structural predication for inanimate subject
I
Voice active passive active passive
Subjects
Structural prediction for animate subject
I
1. 2. 3. 4.
Animacy animate animate inanimate inanimate
Adult Animate Active Animate Passive
I
I
Prediction, Animacy, and the P600
I
Development of Animacy Cues
Apparatus and Recording
I
I
I
Department of Psychology, University of South Carolina
N400
6. General Discussion & Conclusion The Mechanism underlying “Semantic” P600
Development of the Mechanism
Kim & Osterhout (2005) propose that semantic P600s reflect conflict in subject-verb thematic relationships. I We argue for a more precise implementation: I Semantic cues may predict particular upcoming structures. I Inanimacy predicts a derived subject structure because inanimate subjects are not compatible with agent interpretations. I “Semantic” P600s actually reflect reanalysis of a structural prediction triggered by animacy.
The relationship between semantic cues and structure develops over time. I Why so late? Perhaps exposure to more complex sentences requires more efficient processing mechanisms. I Understanding this developmental profile will require better differentiation of access, prediction, integration, and reanalysis mechanisms during language comprehension. I This will also contribute to better understanding of language development disorders.
I
Selected References Dowty, D. (1991). Thematic proto-roles and argument selection. Language 67, 547–619. Ferreira, F. (2003). The misinterpretation of noncanonical sentences. Cognitive Psychology 47, 164–203. Kim, A. & Osterhout, L. (2005). The independence of combinatory semantic processing: Evidence from event-related potentials. J. of Memory and Language 52. 205–225. Kuperberg, G. R., Sitnikova. T., Caplan, D., & Holcomb, P. J. (2003). Electrophysiological distinctions in processing conceptual relationships within simple sentences. Cognitive Brain Research 217, 117-129. MacWhinney, B., Bates, E., & Kliegl, R. (1984). Cue validity and sentence interpretation in English, German, and Italian. J. of Verbal Learning and Verbal Behavior 23, 127–150. Spencer, K. M., Dien, J., & Donchin, E. (2001). Spatiotemporal analysis of the late ERP responses to deviant stimuli. Psychophysiology 38(2) 343–358. Townsend, D. & Bever, T. Sentence Comprehension. The MIT Press.
I
http://sites.google.com/site/justinevandykelyon
[email protected] [email protected] [email protected]