Influence of lexical tones on intonation in kammu

Anastasia Karlsson, David House, Jan-Olof Svantesson, Damrong TayaninInterspeech 2010

TL;DR

This paper investigates how lexical tones in Kammu, a language with tonal and non-tonal dialects, influence sentence-level intonation and focal accent. The findings reveal a surprising hierarchy where lexical tone exerts a stronger influence than sentence accent, and even focal accent, in the tonal dialect, suggesting that lexical tones can override typical intonational patterns.

What Problem Does This Paper Address?

Understanding the interplay between lexical tone and intonation in tone languages is a long-standing challenge. It's difficult to disentangle the contribution of lexical tones (inherent pitch on syllables) from sentence-level intonation (pitch contours conveying grammatical or pragmatic information like focus). Previous research often relied on constrained speech materials or models that assumed intonation was superimposed on tones. This paper addresses this by studying Kammu, a language with recently developed lexical tones and distinct tonal and non-tonal dialects, allowing for a direct comparison of how tone influences intonation. Specifically, it examines how focal accent is realized in both dialects and how lexical tones affect this realization.

Methods

The researchers used a novel approach involving controlled speech material. They designed sentences with specific lexical tone combinations and varied the position of focal accent (on the noun or adjective within a noun phrase). Four pictures were used to elicit responses to questions that systematically varied the oppositions (e.g., eagle vs. lizard, black vs. red). Speakers were asked to produce eight specific sentences, each containing a final noun phrase with all four possible combinations of lexical tones. Each sentence was repeated three times by each speaker. Acoustic analysis was performed using the Praat software to measure fundamental frequency (F0) contours, specifically focusing on F0 maxima in the last three words of the utterance. F0 values were converted to semitones (St) and normalized to allow for inter-speaker comparison. The analysis focused on maximal F0 values, the final rise (difference between F0 maxima on the adjective and noun), and the "focus effect" (difference in the final rise when the adjective is focused versus when the noun is focused).

Datasets

The study utilized a custom-designed speech corpus. The material consisted of elicited utterances from 5 non-tonal speakers and 10 tonal speakers of Kammu. The speakers ranged in age from 17 to 57. The utterances were recorded in Laos and Thailand. No pre-existing benchmark datasets were used; instead, a controlled experimental design was employed to generate the necessary data.

Key Insights and Ideas

  • Lexical Tone Hierarchy: The most striking finding is that lexical tones in the tonal dialect of Kammu appear to be the dominant prosodic feature, influencing sentence intonation and focal accent more strongly than these features influence the tones themselves. This challenges the common assumption of intonation being superimposed on tones.
  • Tone-Lexical Conflict: When lexical tones conflict with the typical rising intonation pattern (e.g., a high-low tone combination), the lexical tone can neutralize or even reverse the expected rise, demonstrating its powerful influence.
  • Focus Subordination: Focal accent, while present, is subordinate to sentence accent and lexical tone. The sentence-final rising gesture is a robust feature in the non-tonal dialect and is significantly modified by lexical tones in the tonal dialect.
  • Dialectal Comparison: Comparing the tonal and non-tonal dialects clearly highlights the impact of lexical tones. The non-tonal dialect exhibits a consistent final rising gesture, while the tonal dialect shows considerable variation influenced by the specific lexical tones present.
  • Preservation of Features: Despite the strong influence of lexical tones, all three prosodic features (lexical tone, sentence accent, and focal accent) are still preserved by speakers of the tonal dialect, indicating a complex but organized prosodic system.

Key Results

  • Non-tonal dialect: Speakers consistently exhibit a pronounced high gesture on the final word, with a larger final rise (4.5 St) when focus is on the final adjective compared to when focus is on the pre-final noun (3.5 St). The "focus effect" is 1.0 St.
  • Tonal dialect:
    • A final rise is generally present, except for the high-low tone combination where a fall can occur when the pre-final noun is focused.
  • The "focus effect" is always positive, meaning the final rise is higher when the final adjective is focused than when the pre-final noun is focused. This effect ranges from 0.2 to 0.9 St across different tone combinations.
  • Lexical tones are clearly distinguished in F0 values, with high tones consistently having higher F0 than low tones, regardless of focus position.
  • The low-low tone combination is compatible with a rising gesture, similar to the non-tonal dialect.
  • The high-high tone combination shows a limited focus effect (0.2 St), possibly because the high lexical tone on the pre-final noun already raises its F0, limiting the range for the final adjective.

Limitations and Future Work

  • Limitations: The study acknowledges that the controlled material might not fully capture the nuances of spontaneous speech. The analysis focused on F0 maxima, and other aspects of the F0 contour were not as deeply investigated. The paper also notes that the tonal system in Kammu has developed relatively recently, which might influence its interaction with intonation.
  • Future Work: The authors suggest that these findings have implications for how tone and intonation are analyzed in other tone languages. Further research could explore the production and perception of these interactions in more detail, potentially using the PENTA model or similar frameworks to better model the integration of lexical and postlexical tonal features. Investigating the role of other prosodic features and the impact of different types of focus (e.g., broad vs. narrow) would also be valuable.

Summary updated

BibTeX
@inproceedings{karlsson2010influence,
  author = {Anastasia Karlsson and David House and Jan-Olof Svantesson and Damrong Tayanin},
  title = {{Influence of lexical tones on intonation in kammu}},
  booktitle = {Interspeech},
  year = {2010},
}
Figure