Disentangling lexical, morphological, syntactic and semantic influences on German prominence - evidence from a production study

Barbara Samlowski, Petra Wagner, Bernd MöbiusInterspeech 2012

TL;DR

This study investigates how different linguistic factors (word boundaries, lexical stress, sentence focus) interact to influence syllable prominence in German. By analyzing spoken sentences with carefully controlled verb prefixes, the researchers found that lexical stress plays a crucial role, even influencing unstressed syllables, and that unbound prefixes at phrase boundaries receive the highest prominence.

What Problem Does This Paper Address?

Understanding what makes a syllable "prominent" in speech is key to many areas of linguistics and speech technology, from speech recognition to prosody generation. Prominence isn't just about loudness; it's a complex acoustic phenomenon influenced by various linguistic levels. Previous research often conflated word stress (inherent to a word's meaning) and sentence stress (highlighting specific words for information). This paper tackles the challenge of disentangling these influences, specifically in German, by examining how word boundaries, lexical stress, and sentence focus (semantic emphasis) interact to shape syllable prominence. They focus on German verb prefixes, which can have minimal pairs differing only in stress and meaning, making them ideal for this investigation.

Methods

The researchers conducted a production study where 30 native German speakers read sentences containing specific verb prefixes. These sentences were meticulously designed to manipulate four key factors:

  • Lexical stress: Whether the stress falls on the verb prefix or the verb stem.
  • Sentence stress: Whether the verb is in a broad sentence focus (accented) or not.
  • Morpheme boundary (mb): The prefix is attached to the verb stem.
  • Phrase boundary (pb): The prefix is split off and appears at the end of the phrase.
  • Word boundary (wb): The prefix appears as a standalone function word (like a preposition).

Acoustic analyses were performed using a prominence tagger that considered duration, pitch movement, intensity, and spectral emphasis. Syllable durations were also calculated. Statistical analysis (Wilcoxon tests) was used to compare prominence and duration across different sentence conditions. Additionally, acoustic similarity between syllables was measured using cross-correlation of spectral envelopes.

Datasets

The study used a newly created dataset of 56 German carrier sentences, each featuring one of four specific verb prefixes ("um", "unter", "über", "durch") and their corresponding verbs. These sentences were illustrated with pictures to aid comprehension and ensure correct reading. The sentences were designed to cover 7 distinct conditions, systematically varying lexical stress, sentence stress, and prefix boundedness. The test items were mixed with 48 filler sentences and presented in a quasi-random order.

Key Insights and Ideas

  • Lexical stress is king: The study strongly confirms that lexical stress is a primary driver of syllable prominence, even influencing unstressed syllables within a prefix.
  • Unbound prefixes get the spotlight: Prefixes that are split off and appear at the end of a phrase (unbound) receive the highest prominence, suggesting phrase boundaries play a significant role.
  • Interaction matters: The interplay between lexical stress and sentence focus is complex. While lexical stress consistently affects prominence, sentence stress's impact is more pronounced on lexically stressed prefixes.
  • Acoustic similarity reveals stress effects: Syllables in lexically stressed prefixes are acoustically more similar to unbound prefixes than their unstressed counterparts, highlighting a shared prosodic characteristic.
  • Robustness varies: Some syllables (like [be] and [te]) are more acoustically robust across different conditions than others, showing less variation.

Key Results

  • Prominence: Lexically stressed prefixes generally showed higher prominence than unstressed ones. Unbound prefixes (condition 5) consistently received the highest prominence values. Sentence stress had a significant effect on prominence for lexically stressed prefixes but not for unstressed ones.
  • Duration: Similar trends were observed for syllable duration, though the correlation between prominence and duration was moderate (Pearson correlation: 0.31).
  • Acoustic Similarity: Syllables in lexically stressed prefixes were significantly more similar to unbound prefixes (condition 5) than syllables in unstressed prefixes. This similarity was particularly strong for the syllable [?y:].

Limitations and Future Work

  • Limited scope: The study focused on a small set of German verb prefixes and a specific set of sentence structures.
  • Production study: While valuable, production studies capture idealized speech. Future work could explore spontaneous speech.
  • Automatic tagger: While sophisticated, the automatic prominence tagger's performance might have limitations, and manual verification could be beneficial.

The authors suggest that future research could explore a wider range of German prefixes and sentence types, as well as investigate the perceptual correlates of these acoustic differences. The findings also lay the groundwork for improving prosody generation in speech synthesis systems.

Summary updated

BibTeX
@inproceedings{samlowski2012disentangling,
  author = {Barbara Samlowski and Petra Wagner and Bernd Möbius},
  title = {{Disentangling lexical, morphological, syntactic and semantic influences on German prominence - evidence from a production study}},
  booktitle = {Interspeech},
  year = {2012},
}
Figure