The F0 fall delay of lexical pitch accent in Japanese infant-directed speech

Yoko Saikachi, Mafuyu Kitahara, Ken'ya Nishikawa, Ai Kanato, Reiko MazukaInterspeech 2012

TL;DR

This study investigates how Japanese mothers modify the timing of lexical pitch accents when speaking to their infants (infant-directed speech, IDS). They found that pitch accent timing is delayed more often in IDS, especially when certain prosodic cues like boundary pitch movements or non-lexical vowel lengthening are absent. This suggests that IDS might exaggerate certain prosodic features to enhance infant attention, potentially as a trade-off with other prosodic exaggerations.

What Problem Does This Paper Address?

When adults speak to infants, they often alter their speech patterns, a phenomenon known as infant-directed speech (IDS). While it's well-established that IDS features exaggerated intonation (higher pitch, wider pitch range) and slower speech rates compared to adult-directed speech (ADS), less is understood about how these modifications affect the fine-grained acoustic characteristics of speech at the lexical level. Specifically, this paper focuses on Japanese lexical pitch accents, which are marked by a sharp drop in fundamental frequency (F0). In Japanese, this F0 drop can sometimes be delayed beyond the accented syllable, a phenomenon called "F0 fall delay" or "ososagari." Previous research indicated this delay is more common in IDS, but it wasn't clear why or under what specific conditions this delay occurs. This paper aims to fill that gap by examining the factors influencing F0 fall delay in Japanese IDS, considering syllable structure and prosodic context.

Methods

The researchers analyzed recordings from the RIKEN Japanese Mother-Infant Conversation Corpus (R-JMICC). They focused on bisyllabic words with the pitch accent on the first syllable. To pinpoint the F0 fall delay, they used a two-piece linear regression method to identify the "turning point" of the F0 contour – the start of the sharp F0 fall. This turning point was then compared to the end of the accented mora.

The analysis involved several steps:

  1. Token Selection: They selected specific tokens based on syllable structure (e.g., CV' and CV'N) and excluded those with problematic F0 contours or unclear segmentations.
  2. Turning Point Determination: Using Praat software, they extracted F0 contours and applied the two-piece linear regression to locate turning points. Manual adjustments were made where necessary.
  3. F0 Fall Delay Calculation: They calculated the frequency of "ososagari" (F0 fall delay) by comparing the turning point's timing relative to the end of the accented mora.
  4. Factor Analysis: The frequency of F0 fall delay was then analyzed based on:
  5. Syllable structure of the accented syllable.
  6. Presence of Boundary Pitch Movements (BPMs) on the following syllable.
  7. Presence of non-lexical lengthening (<H>) of the accented vowel.
  8. Combinations of these factors (e.g., excluding tokens with BPMs or <H>).

Statistical analyses, including paired-sample t-tests, were used to determine the significance of observed differences between ADS and IDS.

Datasets

The study utilized the RIKEN Japanese Mother-Infant Conversation Corpus (R-JMICC). This corpus contains conversations between 21 mothers and their 17-24 month old children (Infant-Directed Speech, IDS) and also conversations between the mothers and a female experimenter (Adult-Directed Speech, ADS). The speech signals were accompanied by detailed linguistic annotations, including intonational labels based on the X-JTOBI scheme. The data used for analysis was limited to the "book part" of the IDS sessions.

Key Insights and Ideas

  • IDS Exaggerates F0 Fall Delay: The core finding is that F0 fall delay is significantly more frequent in IDS than in ADS, particularly when certain prosodic cues are absent. This suggests a deliberate modification in IDS.
  • Prosodic Context Matters: The occurrence of F0 fall delay is not uniform. It's strongly influenced by the prosodic environment, specifically the presence or absence of Boundary Pitch Movements (BPMs) and non-lexical lengthening (<H>).
  • Potential Trade-off in Exaggeration: The paper proposes that IDS might involve a trade-off between different types of prosodic exaggeration. When BPMs or non-lexical lengthening are present (which are themselves forms of prosodic exaggeration), the F0 fall delay might be less pronounced. Conversely, in their absence, the F0 fall delay is more likely to occur, perhaps to compensate or to highlight the accent in other ways.
  • Feminine Connotation: The authors briefly mention that F0 fall delay has been linked to conveying femininity in Japanese, suggesting a potential link between IDS modifications and the perception of maternal speech.

Key Results

  • Overall Frequency: While overall F0 fall delay was observed in both ADS and IDS, the average frequency was slightly higher in IDS. However, this difference was not statistically significant when all data was considered.
  • Effect of BPMs: The presence of BPMs on the syllable following the accented syllable significantly reduced the frequency of F0 fall delay in both ADS and IDS.
  • Effect of Non-lexical Lengthening (<H>): Similarly, non-lexical lengthening of the accented vowel (<H>) also significantly reduced the frequency of F0 fall delay in both ADS and IDS.
  • IDS vs. ADS Differences Emerge When Controlling for Prosody: When tokens with BPMs or <H> were excluded, the difference in F0 fall delay frequency between IDS and ADS became statistically significant. Specifically, excluding <H> tokens, excluding BPM tokens, or excluding both led to significantly higher F0 fall delay in IDS compared to ADS.
  • Syllable Structure: While CV' syllables occurred more frequently than CV'N syllables in both conditions, the effect of syllable structure on F0 fall delay was less pronounced and not always statistically significant in IDS.

Limitations and Future Work

  • Limited Syllable Structures: The study focused only on bisyllabic tokens with the accent on the first syllable and excluded certain syllable structures (CV'H, CV'V) due to segmentation difficulties. This limits the generalizability to other syllable types or accentual patterns.
  • Manual Annotation Reliance: While efforts were made to ensure accuracy, the determination of turning points and the identification of BPMs and <H> relied on manual annotation and subjective judgment to some extent.
  • Future Work: The authors suggest that future research should consider the role of these prosodic modifications in infant language acquisition. They also highlight the importance of using databases with detailed linguistic annotations to fully understand the characteristics of lexical pitch accents in IDS. Further investigation into the "trade-off" mechanism between different prosodic exaggerations in IDS would also be valuable.

Summary updated

BibTeX
@inproceedings{saikachi2012f,
  author = {Yoko Saikachi and Mafuyu Kitahara and Ken'ya Nishikawa and Ai Kanato and Reiko Mazuka},
  title = {{The F0 fall delay of lexical pitch accent in Japanese infant-directed speech}},
  booktitle = {Interspeech},
  year = {2012},
}
Figure