A methodology for the study of rhythm in drummed forms of languages: application to Bora Manguare of Amazon

Julien Meyer, Laure Dentel, Frank SeifartInterspeech 2012

TL;DR

This paper introduces a novel semi-automatic methodology to analyze the rhythmic structure of drummed languages, specifically applied to the Bora Manguaré of Amazon. By correlating drum beats with speech phonemes, they reveal that drummed rhythms primarily encode vowel-to-vowel intervals, offering a new perspective on how these complex communication systems convey linguistic information beyond just pitch.

What Problem Does This Paper Address?

Drummed languages, where drum beats mimic spoken language features, are fascinating but challenging to study. While it's known they can encode pitch and basic prosody, the precise mechanisms by which they convey rhythmic information, particularly at the word level, remain largely unexplored. Previous research often characterized rhythm in broad strokes, asserting a general mimicry of speech rhythm or a beat-per-syllable correspondence. However, a detailed, data-driven analysis of inter-beat durations (IBDs) and their relationship to specific linguistic units within drummed forms of languages was lacking. This paper aims to bridge this gap by developing and applying a methodology to quantify and analyze these rhythmic cues in the Bora Manguaré, a drummed language directly linked to a spoken Amerindian language.

Methods

The authors developed a semi-automatic methodology for analyzing drummed language rhythm. The core steps involve:

  1. Data Collection and Transcription: Eliciting and recording 95 Manguaré drummed messages from expert drummers, followed by transcription in ELAN, aligning drum beats with spoken language phonemes.
  2. Phoneme Clustering: Two strategies were explored to associate phonemes with drum beats:
  3. Syllable Clustering: Grouping phonemes based on syllable boundaries.
  4. Vowel-to-Vowel (V-to-V) Clustering: Associating drum beats with phonemes occurring between two consecutive vowels. This strategy was ultimately favored.
  5. Automatic Beat Detection: Employing a beat synchronization algorithm to detect drum beat timings, followed by manual verification. This involved filtering based on energy peaks, inter-beat duration limits, and energy rate between beats to ensure accurate detection of drum transients.
  6. Frequency Estimation: Calculating the power spectral density around each beat to estimate peak frequency, which was then correlated with the high/low pitch of the drum to determine the tone of the drummed beat.
  7. Inter-Beat Duration (IBD) Analysis: Measuring the time intervals between consecutive drum beats and analyzing these durations in relation to the phoneme clusters identified by the V-to-V strategy, as well as tonal configurations.

Datasets

The study utilized a newly collected dataset of 95 Manguaré drummed messages, elicited from five expert Bora Manguaré drummers. These messages were recorded and transcribed in ELAN. After pre-analysis, the corpus was refined to 1806 word utterances and 4452 pairs of drum beats/intervals, representing 197 unique words. This dataset is described as being archived and accessible online.

Key Insights and Ideas

  • Drummed Rhythm Encodes Vowel-to-Vowel Intervals: The most significant finding is that inter-beat durations in Bora Manguaré primarily correspond to the intervals between vowels in the spoken utterance, rather than syllable durations. This suggests a more nuanced rhythmic encoding than previously assumed.
  • Phoneme Influence on IBDs: Within the V-to-V framework, the presence of consonants influences IBDs. Specifically, IBDs increase with the complexity of the V-to-V cluster (e.g., V < VC < VVC ~ VCC).
  • Tonal Configurations Impact IBDs: The tonal sequence of consecutive beats (LL, LH, HL, HH) significantly affects IBDs, with LH patterns generally showing the shortest durations.
  • Vowel Nucleus as the Drum Beat Anchor: The results strongly support the hypothesis that the drum peak corresponds to the vowel nucleus, rather than encoding the onset consonant of a syllable.

Key Results

  • The V-to-V phoneme clustering strategy was found to be more informative than syllable clustering for analyzing IBDs.
  • IBDs in the V-to-V clustering case showed a clear ordering: IBD(V) < IBD(VC) < IBD(VVC) ~ IBD(VCC). The difference between VVC and VCC was borderline significant (p=.0324).
  • The effect of tonal patterns on IBDs was statistically significant across various cluster types (V, VC, VVC, VCC). For example, for V clusters, F(2,229)=77.16 (p<.0001).
  • The LH tonal pattern consistently resulted in the shortest IBDs across different V-to-V structures.

Limitations and Future Work

  • Limited Data for Individual Consonants: The study acknowledges that the corpus of measured IBDs was not extensive enough to analyze the influence of individual consonant types separately. This is identified as a promising area for future research.
  • Comparison with Spoken Bora: The authors suggest a systematic comparison of drummed IBDs with V-to-V spoken IBDs in Bora as a crucial next step to further validate their findings and understand the mapping more deeply. They note preliminary measures on spoken Bora V-clusters suggest a similar dependency on tonal duplets.
  • Extending the Corpus: To fully analyze the impact of specific consonants and other factors, the corpus of measured IBDs needs to be extended.

Summary updated

BibTeX
@inproceedings{meyer2012methodology,
  author = {Julien Meyer and Laure Dentel and Frank Seifart},
  title = {{A methodology for the study of rhythm in drummed forms of languages: application to Bora Manguare of Amazon}},
  booktitle = {Interspeech},
  year = {2012},
}

Loading related content...

Figure