Fig 1.
Schematic representation of three intervals, left edge to anchor (LE-A), center to anchor (CC-A) and right edge to anchor (RE-A), delineated by points in an initial single consonant, /r/ (top row), or consonant cluster, /kr/ (bottom row), and a common anchor (A). The alignment schema on the left shows simplex onset organization. The schema on the right shows complex onset organization. They key difference between simplex and complex alignment is in the patterns of change in the intervals across /r-/ (top) and /kr-/ (bottom) initial words. On the left (simplex), looking across the top and bottom schemes, RE-A remains stable while LE-A and CC-A increase. On the right (complex), looking across the top and bottom schemes, CC-A remains stable while LE-A increases and RE-A decreases.
Fig 2.
Illustration of temporal alignment in Arabic.
Positional signals in the y-dimension for 3 different receivers, tongue tip, lower lip and tongue back, for 10 repetitions each of bulha, sbulha, ksbulha. The leftmost vertical line (grey) demarcates the center of the initial consonant cluster (or single consonant as in bulha). The middle vertical line demarcates the release of the prevocalic consonant [b]. The rightmost vertical line demarcates the point of inferred maximum constriction in the post-vocalic consonant [l]. Reprinted from [13] under a CC BY license, with permission from Cambridge University Press (S2 File), original copyright 2011.
Fig 3.
Given any sequence of consonants and vowels, here “C C V X”, we exemplify our modelling paradigm by asking: is the sequence parsed in terms of syllables of the simplex or the complex onset type? To evaluate the two hypotheses, H1 vs. H2, the model projects coordination topologies from hypothesized syllable parses. The topology on the top/bottom embodies temporal relations of the simplex/complex onset parse. Absolute time (ms) predictions can be derived from these topologies, and their match to experimental data can be rigorously evaluated.
Fig 4.
Summary of word simulation algorithm.
Consonant landmarks are generated from the release of the immediately prevocalic consonant. The alignment of the vowel is determined by the syllable parse (simplex or complex). All landmarks are associated with a noise term, ɛ.
Fig 5.
Articulatory recordings of Moroccan Arabic.
The top panel shows the movement of the tongue tip during the production of lan ‘to become soft’ by four speakers. The bottom two panels show the tongue tip and lower lip movement during the production of flan ‘someone’ by the same four speakers. The colors of the lines indicate the different speakers. The thick black line shows the average trajectory across speakers. Dotted vertical lines indicate the landmarks that left-delimit the temporal intervals of interest: left edge, center, and right edge. The movement trajectory of the tongue tip is relatively consistent across lan and flan tokens. In particular, the right edge landmark is stable while the center and left edge landmarks shift to the left.
Fig 6.
Duration of measured intervals in Arabic.
Each box corresponds to 567 data points (collapsing over data reported in [13,63]). Left box: LE-A (left edge to anchor interval), middle box: CC-A (center to anchor interval), right box: RE-A (right edge to anchor interval). Intervals shown here were right-delimited by the CMax anchor.
Table 1.
The relative standard deviation (RSD) of three intervals, left edge to anchor (LE-A), center to anchor (CC-A), and right edge to anchor (RE-A), for different Moroccan Arabic word sets and model hit rates for each syllabic organization, simplex onsets and complex onsets.
Table 2.
The relative standard deviation (RSD) of three intervals, left edge to anchor (LE-A), center to anchor (CC-A), and right edge to anchor (RE-A), calculated over Moroccan Arabic word sets using different landmarks, VEnd and CMax, to right-delimit the intervals.
Table 3.
Hit rates for each syllabic organization, the simplex onset model and the complex onset model, for sets of Moroccan Arabic words that show stability reversals.
Fig 7.
Simulation results for the simplex onset model (left) and the complex onset model (right).
The y-axis shows the RSD of the LE-A, RE-A and CC-A intervals. The x-axis shows anchors from lowest to highest variability (1 to 20). For anchors of low variability, anchors 1–6, the RE-A interval has the lowest RSD for the simplex onset model (left) and the CC-A interval has the lowest RSD for the complex onset model (right). Beyond anchor 7, however, stability patterns, expressed in terms of inequalities, change. For the simplex onset model (left), the RE-A interval becomes more variable than the CC-A interval; for the complex onset model (right), the CC-A interval becomes more variable than the LE-A interval. These changes in patterns of RSD inequalities obscure the expected phonetic consequences of the underlying syllabic structure. The main point illustrated is that the mapping between abstract syllabic organization and phonetic stability patterns cannot be expressed in terms of canonical or invariant stability patterns. The same symbolic organization, e.g. that of simplex onsets, surfaces with the expected phonetics of simplex onsets for one range of anchor values (1–6) but also with the expected phonetics of complex onsets for another range of parameter values (anchor value 7 and beyond).
Table 4.
The RSD of three intervals, left edge to anchor (LE-A), center to anchor (CC-A), right edge to anchor (RE-A) calculated across multiple (10–18) repetitions by four speakers of Moroccan Arabic.
Fig 8.
Articulatory recordings of English.
The top panel shows the movement of the tongue tip during the production of ‘row’ by five speakers. The bottom two panels show the tongue tip and tongue back movement during the production of ‘grow’ by the same five speakers. The shading of the lines indicates the different speakers. The thick black line shows the average trajectory across speakers. Dotted vertical lines indicate the landmarks that left-delimit the temporal intervals of interest: left edge, center, and right edge. The movement trajectory of the tongue tip is shifted (to the right) in row relative to grow while the center landmark remains constant across words.
Fig 9.
Duration of measured intervals in English.
Boxes summarize mean duration calculated over 255 data points from 21–40 speakers (depending on the word) of each interval. Left bar: LE-A (left edge to anchor interval), middle bar: CC-A (center to anchor interval), right bar: RE-A (right edge to anchor interval).
Table 5.
The mean, standard deviation, and relative standard deviation of three intervals, left edge to anchor (LE-A), center to anchor (CC-A), right edge to anchor (RE-A), calculated across four English word sets with varying numbers of speakers.
Table 6.
Mean, standard deviation, and relative standard deviation of three intervals, left edge to anchor (LE-A), center to anchor (CC-A), right edge to anchor (RE-A), for 6 groups of 8 productions of row and grows by speakers of American English drawn from the X-Ray microbeam speech production database.
Fig 10.
Interval stability dynamics for Arabic and English.
Regression lines fit to RSDs of LE-A, CC-A, and RE-A intervals, y-axis, plotted against the standard deviation of the right edge to anchor interval for Arabic (left) and English (right) data. All RSDs reported in the paper for both languages are shown in the figure. The patterns in the regression lines for Arabic correspond to the simplex onset dynamic (Fig 7, left); the patterns for English correspond to the complex onset dynamic (Fig 7, right). Regression fits are significant at the p < .01 criterion for all intervals shown.
Fig 11.
Interval stability dynamics for English acoustic data.
Regression lines fit to RSDs of LE-A, CC-A, and RE-A intervals, y-axis, plotted against the standard deviation of the right edge to anchor interval for 96 speakers productions of the raw~draw dyad. The patterns in the regression lines correspond to the complex onset dynamic (Fig 7, right). Regression fits are significant at the p < .01 criterion for all intervals.