The Processing of Prosodic Cues to Rhetorical Question Interpretation: Psycholinguistic and Neurolinguistics Evidence

Mariya Kharaman, Manluolan Xu, Carsten Eulitz, Bettina BraunInterspeech 2019

TL;DR

This paper investigates how listeners distinguish between rhetorical questions (RQs) and information-seeking questions (ISQs) using prosodic cues like pitch accent, voice quality, and duration. They found that pitch accent is the strongest cue, but combinations of cues are crucial for robust interpretation. Furthermore, using EEG, they demonstrated that the brain shows distinct electrical responses to prosodic patterns that violate contextual expectations for question types, suggesting early integration of prosody and context.

What Problem Does This Paper Address?

In everyday conversation, not all questions are posed to elicit new information. Rhetorical questions (RQs) are used to convey a point or elicit agreement, rather than to seek an answer. Linguistically, RQs often have a different surface structure than information-seeking questions (ISQs), but they can also be ambiguous. This paper tackles the question of how listeners reliably distinguish between RQs and ISQs, particularly when the linguistic form is identical. Previous research has shown that prosody (the rhythm, stress, and intonation of speech) plays a significant role, with RQs often exhibiting longer durations, breathier voice quality, and specific pitch accent patterns (like L+H) compared to ISQs (which often use H+!H accents). However, the relative importance of these prosodic cues and how the brain processes these distinctions in real-time remains an open question.

Methods

The researchers employed a two-experiment approach:

  • Experiment 1: Identification Experiment: This behavioral study aimed to determine the relative importance of three prosodic cues – nuclear pitch accent type (L+H vs. H+!H), voice quality (breathy vs. modal), and constituent duration (lengthened or shortened by 10%) – in distinguishing RQs from ISQs. Participants listened to synthesized German wh-questions with manipulated prosody and indicated whether they perceived them as an ISQ or an RQ. Logistic mixed-effects regression models were used to analyze the proportion of RQ judgments. Reaction times were also analyzed using linear mixed-effects regression.
  • Experiment 2: EEG Experiment: This study investigated the neural processing of these prosodic cues. Participants listened to synthesized German wh-questions in contexts designed to elicit either an ISQ or an RQ interpretation (using visual cues like pictures and text). The prosody of the auditory stimuli was manipulated for pitch accent type and voice quality (based on findings from Experiment 1, where duration was found to be less critical). Event-related potentials (ERPs) were recorded to observe brain activity. The analysis focused on difference waveforms between "incoherent" (prosody mismatching context) and "coherent" (prosody matching context) trials, particularly looking for early ERP components like the Prosodic Expectancy Positivity (PEP).

Datasets

The sentence materials used in both experiments were identical to those used in Neitsch, Braun, and Dehé [7]. These consisted of 32 short German wh-questions. For Experiment 1, these sentences were manipulated in terms of pitch accent, voice quality, and duration. For Experiment 2, duration was not manipulated, and the stimuli were presented with visual cues (pictures and text) to establish either an ISQ or RQ context. No new benchmarks were introduced; rather, existing stimuli were adapted and used in novel experimental designs.

Key Insights and Ideas

  • Pitch Accent Reigns Supreme (But Not Alone): While pitch accent type (L+H for RQs, H+!H for ISQs) was the strongest single predictor for distinguishing RQs from ISQs in Experiment 1, it wasn't sufficient on its own. Listeners relied on a combination of prosodic cues for robust interpretation.
  • Duration Matters for Reaction Time: Longer durations were found to lead to longer reaction times when participants identified a question as an RQ, suggesting duration plays a role in the processing effort, though it was less influential than pitch accent for initial identification.
  • Brain Detects Prosodic Mismatches Early: The EEG experiment revealed that the brain is highly sensitive to prosodic cues that violate contextual expectations. An early positivity (around 50 ms after the critical noun) was observed when the prosody of the question did not match the expected type based on the visual context.
  • Emotional Resonance of RQs: The early frontal positivity observed in response to prosodic violations in RQs might reflect a combined linguistic and emotional processing, hinting that RQs might carry a stronger emotional valence than ISQs.

Key Results

  • Experiment 1: Participants' judgments of RQs were significantly influenced by pitch accent type, voice quality, and duration, with pitch accent having the strongest effect. For example, questions with L*+H accents, breathy voice, and longer durations were most frequently identified as RQs. Reaction time analysis showed that shorter durations led to longer reaction times for RQ choices.
  • Experiment 2: ERPs showed a significant difference between coherent and incoherent prosodic-context pairings. Specifically, prosodic RQ realizations in an incoherent ISQ context (and vice-versa for ISQ prosody) elicited distinct ERP waveforms, with an early positivity over frontal electrodes for RQ prosody mismatches. This difference was observed as early as ~50 ms after the onset of the noun carrying the pitch accent.

Limitations and Future Work

The authors acknowledge that while pitch accent is a strong cue, it's not always sufficient for unambiguous RQ/ISQ classification. They suggest that future work should explore the additive effects of less powerful cues like voice quality and duration more deeply. Additionally, they plan to investigate the cue-weighting process during RQ/ISQ categorization and conduct further analyses on the whole sentence to better understand error detection responses and prosodic expectancy positivity. The EEG experiment pooled prosodic conditions (e.g., breathy and modal L*+H were both treated as RQ prosody), and future work could benefit from separating these to potentially reveal clearer patterns.

Summary updated

BibTeX
@inproceedings{kharaman2019processing,
  author = {Mariya Kharaman and Manluolan Xu and Carsten Eulitz and Bettina Braun},
  title = {{The Processing of Prosodic Cues to Rhetorical Question Interpretation: Psycholinguistic and Neurolinguistics Evidence}},
  booktitle = {Interspeech},
  year = {2019},
}
Figure