Speech Prosody 7, May 22, 2014, Dublin Implicit Prosodic Priming and Autistic Traits in Relative Clause Attachment Sun-Ah Jun & Jason Bishop jun@humnet.ucla.edu; jbishop@gc.cuny.edu UCLA & CUNY
Introduction: ambiguous RC attachment Ex. Someone shot the servant of the actress who was on the balcony . NP NP NP1 PP NP1 PP Rel.Cl servant P NP P NP of NP1 of Rel.Cl. actress Low attachment High attachment e.g., Spanish, Dutch, German, French e.g., English, Arabic, Romanian 2
Introduction: The Implicit Prosody Hypothesis • The attachment decision has been shown to be influenced by the length of RC in various languages regardless of their attachment preferences. (e.g., Fernandez & Bradley 1999 on Spanish, Quinn et al. 2000 on French, English, Arabic, Lovric et al. 2001 on Croatian, Hirose 1999 on Japanese, Wijnen 2004 on Dutch, Vasishth et al. 2004 on Hindi, Jun & Kim 2004 on Korean). ex. Someone shot the servant of the actress who was on the balcony. Someone shot the servant of the actress who was on the balcony of the Hotel Romania. => suggests prosody is involved. 3
Introduction: The Implicit Prosody Hypothesis The Implicit Prosody Hypotheses (IPH) (Fodor 1998, 2002) • In silent reading, a default prosodic contour is projected onto the sentence and influences syntactic ambiguity resolution. Other things being equal, the parser favors the syntactic analysis associated with the default prosodic contour for the construction. Claim: speakers interpret a prosodic break before an RC as a marker • of a stronger syntactic boundary, which then prompts high attachment // NP1 NP2 RC 4
How can we access “ Implicit Prosody ” ? • Is Implicit prosody the same as explicit or overt prosody? • Fodor and her colleagues assumed that implicit prosody and explicit prosody are equivalent, and examined the phrasing of explicit prosody (i.e., out-of-the-blue reading). • But, recent production studies (e.g., Bergman et al. 2008, Jun 2010) suggest that the explicit prosody, in the form of out-of-the- blue reading, is not a suitable way to access implicit prosody. • Can we manipulate implicit prosody and see its effects on processing? 5
Can we observe implicit prosody at work? • The effect of implicit prosody in RC attachment has been shown via visual display (e.g., Swets et al. 2007, Traxler 2007, 2009) When the entire sentence “ NP1 NP2 RC … ” is presented on a single line, e.g., The sister of the writer that had blonde hair arrived this morning. English subj preferred low attachment and Dutch subj high attachment; However, English subj with low working memory capacity (WMC) preferred high attachment, suggesting (NP1 NP2)(RC) phrasing. But, when the RC is presented on a separate line from the two head nouns, i.e., The sister of the writer that had blonde hair arrived this morning high attachment is preferred regardless of native language or WMC. => suggest the line break introduced an implicit prosodic boundary before RC and affected RC attachment. 6
Can we prime implicit prosody? • Structural priming : tendency to better process the current sentence because of its structural similarity to a previously experienced ( “ prime ” ) sentence (Bock 1986) e.g., RC or PP attachment primes RC attachment (Scheepers 2003, Locncke et al. 2011) • Prosodic priming by explicit prosody : Jun & Bishop (2013) Explicit phrase boundary before RC, i.e., (NP1 NP2) // (RC) primes High attachment but, syntactic structure could have primed the attachment 7
Experiment: Priming by implicit prosody Present study: manipulate an implicit prosodic break by a visual cue, a comma, (with no possible syntactic priming) and see if it primes implicit prosody of a new sentence, thus affecting RC attachment resolution. Restrictive relative clause vs. Non-restrictive relative clause • with a single head noun • e.g., The newspaper reporter phoned the secretary who was annoyed. The newspaper reporter phoned the secretary, who was annoyed. Non-restrictive RC (cued by a comma) is often produced with a big • prosodic boundary before RC (e.g., Watson & Gibson 2004) Prediction : Non-restrictive RC prime sentences will increase high attachment responses in the target S, compared to restrictive RC primes. 8
method: stimuli • 30 prime sentences, in two RC types -- all had a single head noun and 4-6 σ RC 15 Restrictive RC prime sentences (without a comma) ex. The newspaper reporter phoned the secretary who was annoyed. 15 Non-restrictive RC prime sentences (with a comma) ex. The newspaper reporter phoned the secretary, who was annoyed. • 16 target sentences (4-6 σ RC with two head nouns, unbiased) An RC attachment question for each target sentence 9
method: stimuli (cont ’ ) Fillers • 30 filler prime sentences • 15 sentences without a comma • 15 sentences with a comma • A question for each filler prime sentence • 28 filler sentences and a question about each sentence 10
method: Subjects • 120 native speakers of American English • UCLA undergrads 11
Procedures and experiment design • Main experiment (15~20 min) : subjects read 3 prime sentences of the same RC type on a computer screen, then read a target sentence, at their own pace, and answered an attachment question (i.e., choose NP1 or NP2). demo: 12
Experiment: “demo 1” – prime 1 My uncle had to repair the bicycle tire, which my mom purchased.
“ demo 1 ” – prime 2 The newspaper reporter phoned the secretary, who was annoyed.
“ demo 1 ” – prime 3 The early sun sparkled on the phone ’ s propeller, which was broken.
“ demo 1 ” – target The lady mended the sleeve of the shirt which had been stained. What had been stained? A B the shirt the sleeve
“ demo 2 ” – prime 1 The coach looked at the varsity players who were very happy.
“ demo 2 ” – prime 2 The lady from London disliked the player who didn ’ t say much.
“ demo 2 ” – prime 3 The article failed to mention the library which had just been built.
“ demo 2 ” – target Jennifer blackmailed the boss of the clerk who was dishonest. Who was dishonest? A B the boss the clerk
Procedures and experiment design (cont) For each experimental trial, the MATLAB script selected one of the 16 target sentences and 3 prime sentences, all from the same RC type condition. The order of the primes was randomized for each trial. To prevent subjects from not reading prime sentences, any of the three filler-prime sentences was randomly followed by a question. The prime-target pair was counterbalanced across subjects, and so was the location of attachment answer (left or right) 21
Procedures (cont ’ ) Subjects also completed a task measuring individual differences • that are known to affect sentence processing Autism Spectrum Quotient (AQ) questionnaire (Baron-Cohen et al. 2001) (10~15 min) 22
The Autism Spectrum Quotient (AQ) (Baron-Cohen, Wheelwright, Skinner, Martin & Clubley, 2001) • A self-administered 50-item questionnaire for measuring “ autistic ” -like personality traits in neurotypical (i.e., non-autistic) individuals • 50 items, composed of 5 subscales (communication, social skills, attention switching, attention to detail, imagination): ex. “ I am often the last to understand the point of a joke. ” (communication) “ I would rather go to a library than a party. ” (social skill) “ I frequently get so strongly absorbed in one thing that I lose sight of other things. ” (attention switching) • Higher AQ score = more prominent autistic traits 23
• Why AQ score? AQ has been shown to affect performance in experimental tasks requiring the integration of various types of linguistically-relevant information, including prosody ( Stewart & Ota 2008, Xiang et al. 2011, Bishop, 2012 & 2013) => Among AQ subscales, Communication subscale has often been found the most influential for tasks incorporating pragmatic information. 24
Experiment: Data Analysis and Results Data collected: RC attachment choice, Response time (RT), AQ-scores • Used mixed effects logistic regression to model the probability of subjects ’ • high attachment responses Random effects: subjects, items • Fixed effects Predictors: • 3 stimulus variables: RC type in the prime sentences Order of NPs in Answers Experimental trial 2 Subject variables: Gender AQ sub-scale scores Various Interactions 25
Result 1: Prime type -- significant main effect (p=.035) More %High attachment after non-restrictive RC prime sentences 56% % High Attachment 54% 52% 50% * 48% 46% 44% 42% 40% Restrictive Non-restrictive (with a comma)
Result 2: Prime type X AQ-Comm : -- s ignificant, p=.02 Subjects with higher AQ-Communication scores were significantly more influenced by the presence of a comma in prime sentences 15% Change in % High 13% Attachment 10% 8% 5% 3% 0% -3% Low Mid High AQ-Communication Score ( ‘ Low ’ and ‘ High ’ includes subjects scoring 1 SD below and above the mean AQ-Comm score)
Recommend
More recommend
