Universal principles underlying segmental structures in parrot song and human speech

Abstract

Despite the diversity of human languages, certain linguistic patterns are remarkably consistent across human populations. While syntactic universals receive more attention, there is stronger evidence for universal patterns in the inventory and organization of segments: units that are separated by rapid acoustic transitions which are used to build syllables, words, and phrases. Crucially, if an alien researcher investigated spoken human language how we analyze non-human communication systems, many of the phonological regularities would be overlooked, as the majority of analyses in non-humans treat breath groups, or "syllables" (units divided by silent inhalations), as the smallest unit. Here, we introduce a novel segment-based analysis that reveals patterns in the acoustic output of budgerigars, a vocal learning parrot species, that match universal phonological patterns well-documented in humans. We show that song in four independent budgerigar populations is comprised of consonant- and vowel-like segments. Furthermore, the organization of segments within syllables is not random. As in spoken human language, segments at the start of a vocalization are more likely to be consonant-like and segments at the end are more likely to be longer, quieter, and lower in fundamental frequency. These results provide a new foundation for empirical investigation of language-like abilities in other species.

Figure 1

Complex vocalizations of the human and the budgerigar. In human language, silence is not a reliable cue for word, syllable, or segment boundaries. For instance, the complex and novel phrase “the zealous sailors sail all seven seas and all four oceans” can be spoken without any intervening silence, as shown in the spectrogram (A). Budgerigar song seems to share this property. Budgerigar song is comprised of complex (B: syllables 3 and 5) and simple (B: syllables 1, 2, 4, 6, 7) syllable types. The complex syllables (C), however, seem to be composed of segments which are similar to the simple syllables. We created the image using R and the packages cowplot, ggplot2, ggpubr, seewave, and viridis^–.

Figure 2

Segmentation and segment validation. (A), an example of an output of our segmentation algorithm on human speech, in this case, an English vocal breath group, “four oceans”. (B), an example the segmentation algorithm applied to budgerigar song. We scaled the algorithm to the different species by making the algorithm dependent on a minimum fundamental frequency setting. For each cluster size of human speech (C), silhouette widths were higher for segments than for syllables ($\overline{x}$ = 0.42–0.25). For the budgerigar units, we included simple syllables and random snippets of syllables that were equally long as the segments as controls (D). Silhouette width scores for segments were statistically different from complex syllables ($\overline{x}$ = 0.32–0.13, V = 105, p < 0.001) and the random snippets ($\overline{x}$ = 0.32–0.22, V = 105, p < 0.001). We created the image using Praat, R, and the R packages ggplot2 and magick^,,,.

Figure 3

Budgerigar vowels and consonants. (A) A hierarchical clustering of budgerigar segments reveals two clear clusters of segments. (B) Using the clustering, we found that one cluster is comprised of periodic, vowel-like units while the other is made up of aperiodic, consonant-like units. The former is, on average, longer, louder, and more periodic than the latter. We created the image using R and the packages factoextra and ggplot2^,,.

Figure 4

Edge effects in budgerigar song. (A) Canonical syllable with aperiodic burst onset, long and low final segment, and a rise-and-fall intensity contour. All groups shared similar positional biases. For all individuals, mean F0 measurements were lower for segments in syllable-final position when compared to medial segments (mean of individual means: n = 14, Medial: $\overline{x}$ = 2522 Hz, s =  ± 181 Hz ~ Final: 2069 ± 165); final segments were, on average, longer in duration than medial segments (n = 14, Medial: $\overline{x}$ = 6.5 ms, s = 0.84 ms ~ Final: 12, ± 4.84); intensity was lowest in initial position (Initial: $\overline{x}$ = 46.7 dB s = 3.71 dB ~ Medial: 56.4 ± 4.03 ~ Final: 52.8 ± 4.46); and periodicity was lowest in initial position with slightly more than a third of segments having periodic vibration (n = 14; $\overline{x}$ = 34.3%, s = 10.5% ~ Medial: 74%, ± 4.9 ~ Final: 54.9%, ± 10.5%). For all four acoustic measurements, a mixed-effect model with position as a fixed effect performed better than a model without position (F0: X² = 2353.4, df = 2, p < 0.001; duration: X² = 3209.6, df = 2, p < 0.001; intensity: X² = 22,186, df = 2, p < 0.001; periodicity: X² = 10,158, df = 2, p < 0.001). (B) The prevalence of aperiodicity in syllable-initial positions and periodicity in syllable-medial positions suggests that budgerigars share the human segment organizational preference (C) for aperiodic onsets followed by periodic signals. We created the figure using R and the packages cowplot, ggpubr, seewave and viridis^–.