A glimpse of the time-course of intonation processing in European Portuguese

Isabel Falé, Isabel Hub FariaInterspeech 2005

TL;DR

This study investigates how European Portuguese speakers process sentence-level intonation to distinguish between statements and questions. They found that listeners can identify declarative sentences very early, relying on initial prosodic cues, while interrogative sentences require processing closer to the end of the utterance. This suggests a dual processing strategy for intonation in this language, highlighting the importance of both global and local prosodic information.

What Problem Does This Paper Address?

Accurate and efficient communication relies heavily on prediction, allowing listeners to anticipate upcoming information and reduce cognitive load. In speech processing, this prediction is facilitated by linguistic cues available early in the speech signal. However, for European Portuguese (EP), it was previously believed that key intonational features distinguishing sentence types (like statements vs. questions) occur late in the utterance, near the end. This would imply listeners must wait until the end of a sentence to process its intonation, potentially leading to communication delays and silent gaps. This paper challenges this notion by investigating whether EP speakers can predict sentence type earlier than previously thought, suggesting that relevant prosodic cues might be available sooner.

Methods

The researchers employed the Gating Paradigm, a technique originally developed for word recognition, adapted for sentence-level intonation processing. In this method, auditory stimuli (sentences) are presented in progressively longer segments, starting from a short initial piece and extending to the full sentence. Participants are asked to identify the sentence type (statement or question) at each gate and then rate their confidence in that identification.

A novel aspect of this study is their phonetic criterion for gating. Instead of using fixed temporal intervals (e.g., every 30ms), they gated the sentences at the end of each stressed vowel. This approach is based on the assumption that stressed vowels in EP are significant prosodic units that carry crucial intonational information.

Two key measures were derived:

  • Isolation Point (IP): The point in the sentence where correct identification of the sentence type is achieved and maintained across subsequent gates.
  • Recognition Point (RP): The point where subjects rate their identification as "sure" and maintain that rating until the end of the sentence.

Datasets

The study used 5 pairs of stimuli, totaling 10 sentences. Each pair consisted of two acoustically identical sentences produced by the same speaker, differing only in their intonational contour (one declarative, one interrogative). These stimuli were gated at various locations, primarily defined by stressed vowels. The study involved 20 European Portuguese native speakers. No new benchmark datasets were introduced; rather, existing stimuli were adapted for the gating paradigm.

Key Insights and Ideas

  • Early Prediction of Declaratives: Listeners can identify declarative sentences with high accuracy very early in the utterance, specifically by the first stressed vowel. This indicates that global prosodic cues, likely related to pitch range, are available and informative from the outset.
  • Later Processing of Interrogatives: In contrast, identifying interrogative sentences generally requires processing information closer to the end of the sentence, often at the last or penultimate stressed vowel. This aligns with the common understanding that rising intonation at the end of a sentence is a key differentiator for questions.
  • Dual Processing Strategy: The findings suggest that EP speakers employ a dual strategy for intonation processing: utilizing early global cues for statements and relying on later, more localized cues for questions.
  • Confidence Reflects Processing Time: The recognition point (when subjects feel "sure") consistently occurs later than the isolation point (when they can correctly identify the sentence type), reflecting the time needed to build sufficient confidence in their interpretation.
  • Specific Question Contour: One interrogative sentence in the study (pair C) was identified earlier than others, suggesting that specific acoustic features within its intonational contour (e.g., an initial high and steady pitch plateau) facilitated earlier recognition.

Key Results

  • Declarative Identification: For statements, the Isolation Point (IP) was consistently achieved at the first stressed vowel (FSV) across all stimulus pairs.
  • Interrogative Identification: For questions, the IP was generally achieved later, near the last stressed vowel (LSV) or sentence final (SF) position. An exception was noted for one question (pair C), which was identified earlier.
  • Recognition Point (RP): RP generally occurred later than IP, indicating that subjects needed more information to be confident in their identification. An exception was observed for one question in pair B, where RP appeared sooner than IP, which the authors suggest might be an artifact of their measurement.
  • Table 1 Summary: Table 1 consolidates the IP and RP locations, showing a clear tendency for statements to be recognized earlier than questions.

Limitations and Future Work

  • Artifact in RP Measurement: The authors acknowledge that the earlier-than-expected RP for a question in pair B might be an artifact of their quantification method.
  • Need for More Data for Confidence: The consistent delay between IP and RP highlights that listeners require more data to feel confident about their sentence type identification, even when they can correctly guess it early on.
  • Future Directions: The study suggests that both global and local prosodic cues play a crucial role in intonation processing. Future work could explore the specific acoustic features that contribute to early identification of statements and the precise nature of the local cues used for interrogatives. Further investigation into individual differences in processing strategies and the impact of different sentence structures on intonation processing would also be valuable.

Summary updated

BibTeX
@inproceedings{fale2005glimpse,
  author = {Isabel Falé and Isabel Hub Faria},
  title = {{A glimpse of the time-course of intonation processing in European Portuguese}},
  booktitle = {Interspeech},
  year = {2005},
}
Figure