Rutgers Laboratory for
Developmental Language Studies
The Role of Prosody
PARTICIPANTS 18 undergraduate students recorded in a sound-­‐proof recording booth. Data from 2 par+cipants excluded (N=16) STIMULI 32 sentences, presented as responses to one-­‐sentence prompts with verb of saying/repor+ng (key word underlined) Target items 1.  Predicate of Personal Taste (PPT), POS/NEG perspec+ve (That sounds like fun.) 2.  Copular asser+on, POS/NEG perspec+ve (Yeah, they were mums.)
Minimal pairs across (1) and (2) created by matching key vowel and prosodic structure Control items 3.  Scalar contrast (She’s not just nice – She’s lovely.)
4.  Declara+ve vs. Interroga+ve (She’s moving to Iowa./?)
5.  Informa+on vs. Contras+ve focus
(This weekend I’m going hiking with Jane. / His wife’s name isn’t Joan, it’s Jane.) Prosody can also convey verbal irony. However, experimental evidence across studies have produced mixed results, because they have conflated speaker nega+vity, sarcasm, and irony, relied on spontaneous speech, or situated target uUerances in biased contexts Georgia Simon: [email protected] Perception Experiment
IRONY INTERPRETATION PERCEPTUAL VALIDATION Were speakers successful at portraying +/-­‐ perspecCve? Can listeners judge irony from prosodic cues alone? METHOD: 8 research assistants blindly rated all target items PARTICIPANTS: 15 undergraduate Linguis+cs students (128 minimal pairs) as posi+ve (1), nega+ve (-­‐1), or neither par+cipated for extra credit. STIMULI: 39 PPT pairs (see (1) above), presented in random (0). Random order in Superlab, blocked by speaker. order in Superlab Means taken across raters for each item. PROCEDURE: harsh cri+cal maUer-­‐of-­‐fact happy excited said
said
biUer angry bored earnest kind That s
ounds l
ike f
un =
≠
Excluded both members of minimal pair if: meant
meant
1: Average ra+ng for either member was on the unan+cipated side of 0, or ra+ngs for pair not significantly ParCcipants success rates: different from each other Ironic: 88.6%, Sincere: 91.9% 2: Average ra+ng for either member not significantly Average d’ score 2.75 different from 0 Analyzing Verbal Irony
DuraCon of Key Syllable 300.0
Intensity of Key Syllable Intensity (dB)
200.0
150.0
100.0
50.0
127.9
146.5
Sincere
PPT
Insincere
PPT
(PPT, Positive) (fun) (PPT, Ironic) (fun)
Pos
Ironic
118.2
127.7
CSincere
opula
(Inform.) (mums)
Pos
CInsincere
opula (Negative) (mums)
Neg 0.0
247.5
153.4
Sincere
PPT
Insincere
PPT
(PPT, Positive) (fun) (PPT, Ironic) (fun)
Pos
Ironic
227.6
160.1
CSincere
opula
(Inform.) (mums)
Pos
CInsincere
opula (Negative) (mums)
Neg 74.0
73.0
72.0
71.0
70.0
69.0
68.0
67.0
66.0
65.0
71.5
68.1
Sincere
PPT
Insincere
PPT
(PPT, Positive) (fun) (PPT, Ironic) (fun)
Pos
Ironic
72.7
70.0
CSincere
opula
(Inform.) (mums)
Pos
CInsincere
opula (Negative) (mums)
Neg Irony longer than all other categories (p<.0001) Max F0 of Key Syllable 250.0
Max F0 (Hz)
150.0
140.0
130.0
120.0
110.0
100.0
160.0
Irony lower Max F0 than all Posi+ve categories (p<.0001) Irony less intense than all other categories (p<.0001) % Perceived Falling Sentence-­‐Final IntonaConal Contour Control Items 100.0%
100.0%
80.0%
60.0%
40.0%
8.3%
20.0%
0.0%
87.5%
97.9%
SScalar
calar Statement
Declar.
Contrast
85.4%
68.8%
IQuestion
nterrog. Information
Inform. Contrastive
Contrast. Focus
% Falling Contour
Contact
n=4 n=6 (3 in 2 versions) n=6 (3 in 2 versions) Sentences presented in pseudorandomized order in Powerpoint. Par+cipants emulated ajtudes of two different characters throughout session. (See “Recording Session” to right.) Target sentences excised using Praat speech analysis socware. Falling v. non-­‐falling contour coded by ear (MIN.3 blind coders); Acous+c analysis using Praat scripts % Falling Contour
THE PRESENT WORK controls for (a)  the posi+ve and nega+ve valence of the speaker’s perspec+ve, and (b)  veridicality (intended truth value of proposi+on expressed by the uUerance : literal meaning or opposite truth value) to determine the prosodic correlates of verbal irony, as opposed to a mere “nega+ve” uUerance, AND determine whether listeners can use these cues to iden+fy an uUerance as ironic or sincere. n=8 (4 in 2 versions) n=8 (4 in 2 versions) • 
• 
• 
• 
Duration (ms)
Verbal Irony
Verbal Irony: The speaker’s inten+on is at odds with the literal interpreta+on of the uUerance. Speaker flouts the Maxim of Quality (Grice), intending that the listener calculates the opposite truth value of the proposi+on expressed by the uUerance. Sarcasm: subtype of verbal irony; harsh and cri+cal PREVIOUS WORK •  Spontaneous speech •  Bryant & Fox Tree 2002: Par+cipants were able to iden+fy sarcasm from radio excerpts, but not acer the uUerances had been content-­‐
filtered. Reliance on local rather than global prosodic features. •  Bryant & Fox Tree 2005: Sets of par+cipants rated uUerances for sarcasm, as well as anger, inquisi+veness, authority, new v. given info. Items rated more sarcasHc also rated more inquisiHve and more angry. •  Lab speech •  Rockwell 2000: Speakers read target sentences in biasing contexts (nonsarcasm and ‘spontaneous’ sarcasm), then asked to read target sentence “sarcas+cally” (‘posed’ sarcasm). PercepHon parHcipants able to perceive posed but not spontaneous sarcasm. •  Cheang & Pell 2008: Target items preceded by biasing context for four ajtudes (sincere, sarcas+c, jocular, neutral), and varied presence of enan+oseman+c terms (yeah right). Sarcasm may be encoded similarly to other negaHve affecHve states. •  Across studies: average pitch (usually lower), pitch span (varyingly wide or narrow), speech rate (slower), intensity (higher or lower). Frequent, nonspecific men+on of importance of stress. Recording Session
Production Experiment Speakers can manipulate prosody (i.e. intona+on, stress) to encode implicit communica+on •  Highlight informaCon structure (Baltazani 2002, Koizumi & Bradley 2007, SyreU et al 2014, Ward & Hirschberg 1985, Wagner et al 2013) •  Resolve structural ambiguiCes (Baltazani 2003, Beach 1991, Cooper & Paccia-­‐Cooper 1980, Hirschberg & Avesani 2000, Koizumi 2009, Nagel et al. 1994, Nagel et al. 1996, Price et al 1991, Speer et al 1993,) •  Communicate emoConal states (Banse & Scherer 1996) excluding ‘fun’ Target Items 90.0%
80.0%
70.0%
60.0%
50.0%
85.9%
100.0%
Sincere
PPT
PPT,
PPositive
os
92.2%
100.0%
96.9%
Insincere
Insincere
PPT CSincere
opula C
opula Ironic Informative
PPT,
Ironic
Pos Negative
Neg GOAL Elicit two versions of the same sentence from the same speaker – one nega+ve and one posi+ve – while holding s+muli and context constant METHODOLOGY Chris and Taylor (gender neutral names, allowing for any parHcipant to assume roles) •  Chris is genuine and sincere. Chris is always honest and pleasant. When Chris speaks, you can interpret the words literally. What Chris says is what Chris means. •  Taylor is harsh and cri+cal. Taylor has a consistently nega+ve ajtude. The words Taylor uses do not necessarily reflect how Taylor feels. •  Pat: Experimenter Predicate of Personal Taste Copular Asser+on Pat: Hannah says her date is taking her to mini golf. Chris/Taylor: That sounds like fun. Pat: Laura said her boyfriend brought her flowers yesterday. Chris/Taylor: Yeah, they were mums. Conclusions
•  Verbal Irony exhibits a disCnct suprasegmental signature different from posi+ve, veridical asserted content, and from nega+ve speaker perspec+ve on the topic •  This prosodic informa+on is retrievable and used by listeners tasked with iden+fying speaker meaning. •  Novel methodology allowed elicita+on of different prosodic contours without appeal to context. Acknowledgements
We gratefully acknowledge a Rutgers startup grant to K. SyreU, funding from the Aresty Research Center at Rutgers, and the assistance of the research assistants from the Rutgers Laboratory for Developmental Language Studies. Selected References
A"ardo et al (2003). Mul+modal markers of irony and sarcasm. Humor. Bryant (2010). Prosodic contrasts in ironic speech. Discourse Processes. Bryant & Fox Tree (2002). Recognizing verbal irony in spontaneous speech. Metaphor & Symbol. B & F T (2005). Is there an ironic tone of voice? Lang and Speech. Cheang & Pell (2008). The sound of sarcasm. Speech Comm. Cutler (1974). On saying what you mean without meaning what you say. CLS 10. Fodor (2002). Prosodic disambigua+on in silent reading. NELS 32. Gibbs (2000). Irony in talk among friends. Metaphor & Symbol. Price et al (1991). The use of prosody in syntac+c disambigua+on. JASA. Rockwell (2007). Vocal features of conversa+onal sarcasm: A comparison of methods. J Psycholing Res. Speer et al (1993). Prosodic structure and sentence recogni+on. JML. Syre" et al (2015). Speakers’ produc+on and listeners’ percep+on of scopally ambiguous sentences in a discourse context. JoL.