Where to associate stressed additive particles? evidence from speech prosody

Bettina BraunInterspeech 2012

TL;DR

This paper investigates how the German additive particle "auch" (also) is prosodically realized and how listeners associate it with other parts of a sentence. The study finds that while stressed "auch" isn't always linked to contrastive topics, its presence influences sentence completion, suggesting a connection between prosodic prominence and information structure.

What Problem Does This Paper Address?

Linguists have long debated how "additive particles" like "auch" in German connect to the rest of a sentence, particularly when the particle itself is stressed. Traditional theories often link stressed "auch" to "contrastive topics" – elements that are explicitly contrasted with something else. However, empirical evidence has been mixed. While studies show that the constituent associated with stressed "auch" (the "associated constituent" or AC) is produced with more prominence, it's not always clear if this prominence equates to a full contrastive topic. This paper aims to clarify this by comparing the prosodic features of ACs with those of known contrastive and non-contrastive topics, and by testing how listeners complete sentences based on these prosodic cues.

Methods

The researchers conducted two main experiments:

  • Production Experiment: Native German speakers read sentences designed to elicit different information structures: contrastive topics, non-contrastive topics, and sentences with stressed additive particles. The prosody of their speech was analyzed, focusing on accent types (like L+H and L+H) and tonal alignment (how pitch contours rise and fall relative to stressed syllables). A key methodological choice was using a between-subjects design, meaning each participant only heard one condition. This was intended to make participants less aware of the experimental manipulation compared to within-subjects designs used in prior work.
  • Sentence Completion Experiment: Participants listened to sentence fragments and were asked to type how they would naturally complete them. The fragments were presented with three different intonational patterns: contrastive topic, non-contrastive topic, and a "corrective focus" accent (which has a distinct falling pitch). The goal was to see if the intonation of the fragment influenced whether participants used an additive particle in their completion.

Datasets

The paper does not introduce new benchmark datasets. Instead, it uses carefully constructed German utterance triplets for its experiments. These triplets were designed to control for phonetic variables and elicit specific information structures.

Key Insights and Ideas

  • Prosodic Prominence is Key, Not Just Contrast: While stressed "auch" might not always be associated with a full-blown contrastive topic, its associated constituent is marked by prosodic prominence. This prominence seems to be a crucial factor in how listeners process and associate the particle.
  • "Auch" Sits in the Middle: The prosodic realization of constituents associated with stressed "auch" falls acoustically between that of contrastive and non-contrastive topics. This suggests a nuanced information structure role, not a strict binary contrast.
  • Listeners "Hear" the Connection: Even when not explicitly prompted, listeners are more likely to use additive particles in their sentence completions when the preceding fragment has a contrastive topic accent. This highlights the role of prosody in guiding pragmatic inference.
  • Beyond Contrast: The Role of Nuclear Fall: Additive particles also appeared more frequently in completions following a "corrective focus" accent (characterized by a nuclear fall). This suggests that the relative prominence of an accent, regardless of its specific type (prenuclear rise vs. nuclear fall), might be more important for associating additive particles than previously thought.

Key Results

  • Production: In the production experiment, contrastive topics showed significantly later alignment of both low (L) and high (H) tonal targets compared to non-contrastive topics. The prosody of constituents associated with stressed "auch" fell between these two conditions, showing no significant differences from either.
  • Completion: In the sentence completion task, additive particles were used significantly more often (over twice as often) in continuations following a contrastive topic accent compared to a non-contrastive topic accent. The corrective focus condition also saw a notable occurrence of additive particles, though not significantly different from the other two.

Limitations and Future Work

  • Speaker Awareness: The authors acknowledge that differences between their study and prior work (like Sudhoff & Lenertová, 2006) might be due to the between-subjects design used here, which potentially made participants less aware of the experimental manipulation. Future work could explore how speaker awareness impacts these prosodic realizations.
  • Dialectal Variation: The authors suggest that dialectal differences might also play a role in the realization of prenuclear accents and their association with additive particles.
  • Disentangling Prominence and Accent Type: A key future direction is to further investigate the role of accent prominence versus specific accent types (like prenuclear rises vs. nuclear falls) in the association of additive particles. This would require more controlled experiments to isolate these factors.
  • Ambiguity and Perception: The paper hints that in ambiguous sentences, the prominence of the associated constituent might be even more critical for listener perception. Further research could explore this in more detail.

Summary updated

BibTeX
@inproceedings{braun2012where,
  author = {Bettina Braun},
  title = {{Where to associate stressed additive particles? evidence from speech prosody}},
  booktitle = {Interspeech},
  year = {2012},
}
Figure