TL;DR
This paper investigates the intonation of Czech question-word questions (QWQs), specifically whether the melodic peak (F0 peak) should occur early or late in the sentence. Through a perceptual experiment, they found that native Czech speakers significantly preferred QWQs with an early F0 peak, suggesting this is the more natural and unmarked intonation pattern, aligning with production data.
What Problem Does This Paper Address?
For decades, linguists have debated the "correct" intonation for question-word questions (QWQs) in Czech. These questions, like "Where is my snack?" or "How many points do you have?", are lexically marked as interrogative by starting with a question word (e.g., "kde," "kolik"). This has led some to believe that their intonation should mirror that of statements, with the main melodic peak occurring late in the sentence, near the end. Others argue that the question word itself should carry the peak, resulting in an early F0 peak. This debate has largely been based on introspection rather than empirical evidence. This paper aims to resolve this by empirically testing how native speakers perceive these two intonation patterns.
Methods
The researchers designed a perceptual experiment to test listener preferences. They created two versions of Czech QWQs: one with an early F0 peak and one with a late F0 peak. To ensure a fair comparison, these versions were not simply recorded but meticulously synthesized. They started with natural productions of QWQs, then created a monotonous F0 contour, and finally used the PSOLA (pitch-synchronous overlap and add) method to transplant the original early and late peak contours onto this monotonous base. This ensured that the only significant difference between the two conditions was the peak position.
To avoid listeners noticing the systematic manipulation, they included distractor items: statements and yes-no questions, also synthesized. Thirty-two native Czech speakers, who were not phonetically trained, participated. They were asked to rate the naturalness of the recorded utterances on a 7-point scale, imagining they were judging actors in a screenplay audition. The experiment was administered using the Praat software, and statistical analysis was performed using linear mixed-effects models to account for individual listener and phrase variations.
Datasets
The study utilized synthesized speech stimuli. The core material consisted of eight Czech question-word questions (QWQs), each presented in two versions: one with an early F0 peak and one with a late F0 peak. These synthesized versions were based on real productions from a previous study by the same authors. Additionally, 16 statements and 6 yes-no questions were used as distractor items, extracted from corpora of the Czech language. Some distractor items were also resynthesized with slight F0 contour changes. A small set of training items (one QWQ with an early peak, one QWQ with a late peak, three statements, and one yes-no question) was also created.
Key Insights and Ideas
- Early Peak is Preferred: The most significant finding is the clear perceptual preference for early F0 peaks in Czech QWQs. This suggests that, for native speakers, this is the more natural and unmarked intonation pattern.
- Production and Perception Align: This perceptual preference for early peaks aligns with the authors' previous production data, where most speakers naturally produced QWQs with early peaks. This convergence strengthens the argument for the early peak being the default.
- Late Peak is Not Unnatural: While less preferred, the late peak was not universally perceived as unnatural. It might serve specific communicative functions like contrastive emphasis or implicational cues, making it acceptable in certain contexts.
- Individual Variation Matters: The study highlights the importance of accounting for individual differences in perception. Random effects analysis revealed variations among listeners in their sensitivity to the peak position, and even a few listeners showed a minor effect in the opposite direction.
- Adaptation and Response Time: Listeners became less strict and their response times decreased as the experiment progressed, indicating adaptation to the task. Interestingly, response times were slightly longer for late peak conditions, suggesting they might require a bit more processing.
Key Results
The perceptual experiment revealed a statistically significant preference for the early peak condition. The linear mixed-effects model showed that PEAK POSITION was a significant predictor of the naturalness score, with the late peak condition lowering the evaluation score by 8% of the scale. While the late peak was less preferred, it was still considered acceptable by many listeners. The study also found that the order of presentation of blocks (early vs. late peak blocks) influenced the results, with later trials being evaluated more positively.
Limitations and Future Work
The authors acknowledge that while they found a general preference for early peaks, the induced context in their experiment allowed for the late peak to be acceptable, sometimes even to a greater extent. They also note that individual phrases and listeners behaved somewhat differently, underscoring the role of context and individual experience.
Future research directions suggested include:
- Experimenting with more varied contexts to explore the contrastive and affective effects of peak positions.
- Investigating the impact of peak pitch ranges and peak alignments (synchronizations with segmental features) in addition to peak position.
- Further exploring how speaker experience and cultural norms influence perception.
Summary updated