Source-linked AI summary
Articulatory Entrainment and Coordination Complexity in Spontaneous Autistic and Non-autistic Dialogue
Thanushi Withanage, Carol Espy-Wilson, Elizabeth Redcay, Desi Jones, Noah Sasson
TL;DR
Articulatory entrainment in spontaneous dialogue, especially involving autistic speakers, remains underexplored, and many prior studies use task-based analyses or invasive measurements. The paper applies acoustic-to-articulatory inversion and coordination measures to spontaneous dyads; same-neurotype dyads show increasing coordination complexity, while mixed-neurotype dyads show patterns consistent with sustained or increased effort.
Problem
Articulatory coordination in spontaneous dialogue, particularly autistic and mixed-neurotype interaction, is less understood than alignment studied in task-based paradigms or with invasive measurements.
Source: §1 · p. 1Method
The authors use a speaker-independent framework based on acoustic-to-articulatory inversion and compare speakers’ coordination patterns across conversation halves.
Source: §1 · p. 1, §3.2 · p. 3Results
Same-neurotype dyads show increasing articulatory coordination complexity as conversations progress, whereas mixed-neurotype dyads show patterns consistent with sustained or increased articulatory effort.
Source: §5 · p. 4Takeaways & Limitations
In this study, articulatory coordination patterns provide an interpretable way to examine how speakers adapt over the course of neurodiverse interactions.
Source: §5 · p. 4The entrainment test is specifically whether the speakers’ eigen-spectrum gap decreases from the first half to the second half of conversation.
Source: §3.2 · p. 3
Abstract
from arXiv · showhide
Articulatory entrainment, the adaptation of vocal tract coordination to facilitate interaction remains underexplored in spontaneous dialogue, particularly among autistic speakers. Many prior studies have utilized task-based, phoneme-level analyses with invasive measurement techniques. Here, we introduce a speaker-independent framework to quantify articulatory entrainment in spontaneous dyadic conversations among autistic and non-autistic adults using acoustic-to-articulatory inversion and coordination complexity metrics. We examine temporal changes in articulatory coordination across interaction. Non-autistic dyads exhibit increasing coordination complexity and stronger entrainment over time, autistic dyads show moderate effects, and mixed dyads demonstrate the least alignment. Greater articulatory entrainment correlates with higher self-reported conversational success, indicating increase in coordination complexity as a marker of effective social interaction.
1. Introduction
Articulatory alignment has been linked to rapport, mutual understanding, and conversational success, but its development in spontaneous dialogue—especially across autistic and mixed-neurotype interactions—remains understudied. This study introduces a speaker-independent, non-invasive framework to examine articulatory entrainment in naturalistic dyads.
1. Introduction
- Alignment across acoustics, prosody, facial motion, and language has been associated with rapport, mutual understanding, and conversational success.The authors argue that examining specific communicative dimensions can avoid framing autistic speakers’ differences as deficits.Source: §1 · p. 1
- Articulatory entrainment means voluntary or involuntary adjustments that increase similarity between partners’ speaking styles and articulatory gestures.Prior articulatory kinematics studies have often used electromagnetic articulography or ultrasound, which require specialized setups and may alter natural speech through sensor placement.Source: §1 · p. 1
- Task-based paradigms and invasive articulatory measurements leave spontaneous, naturalistic dialogue—particularly autistic and mixed-neurotype interaction—less understood.Existing approaches include shadowing, repetition, and picture-description tasks; natural speech also poses phoneme-boundary challenges from coarticulation and articulatory undershoot.Source: §1 · p. 1
- The study introduces a non-invasive, speaker-independent framework that uses acoustic-to-articulatory inversion to quantify the degree and direction of entrainment in spontaneous, task-free dyads.It examines how coordination changes over time, varies across neurotype pairings, and relates to perceived conversational success.Source: §1 · p. 1
2. Datasets
The paper analyzes two datasets of previously unacquainted adults in task-free dyadic conversations: a large online corpus of non-autistic dyads and a smaller face-to-face corpus spanning three neurotype pairings.
2.1. CANDOR Dataset
- CANDOR contributes 1,381 retained non-autistic dyadic interactions from 30-minute naturalistic Zoom conversations, with post-session surveys used to derive perceived conversational success.Participants were previously unacquainted; dyads with elevated autistic traits on either Autism Spectrum Quotient social-skill or communication subscale were excluded.Source: §2.1 · p. 2
2.2. UTD Dataset
- UTD contributes 62 previously unacquainted dyads in five-minute unstructured face-to-face conversations: 18 autistic–autistic, 31 autistic–non-autistic, and 13 non-autistic–non-autistic.Autistic participants met ADOS-2 diagnostic criteria, and all participants had an IQ above 80.Source: §2.2 · p. 2
3. Method
The method converts conversational audio into vocal-tract coordination representations and uses changes in those representations to characterize complexity and dyadic entrainment over time.
3.1.1. Audio prepossessing
- CANDOR audio was converted to mono 16 kHz, enhanced, silence-trimmed, and segmented into 40-second chunks; UTD audio was diarized before enhancement and segmentation into 3-second chunks.UTD used a central microphone without separate speaker channels, whereas CANDOR provided separate speaker channels.Source: §3.1.1 · p. 2
3.1.2. TV eigen spectra generation
- The inversion system extracts six time-varying vocal-tract variables, whose pairwise cross-correlations at a fixed 7-second delay form coordination matrices summarized by 90-dimensional eigenvalue spectra.The variables include lip aperture and protrusion, plus tongue-tip and tongue-body constriction degree and location.Source: §3.1.2 · p. 2
- Later-quarter spectra flatten, indicating increasing coordination complexity; the first-quarter-referenced difference spectra likewise show complexity increasing as conversation progresses.The simulation comparison explains that simpler phasing concentrates magnitude in a few lower-ranked eigenvalues, while more complex speech has relatively larger higher-rank eigenvalues.Source: §3.1.2 · pp. 2–3, Figure 2 · p. 2, Figure 3 · p. 3
3.2. Entrainment calculation
- Entrainment is operationalized as global articulatory convergence: each speaker’s average spectrum is reduced to a weighted scalar, and a decreasing inter-speaker gap from the first to second half indicates entrainment.The weighting emphasizes prominent lower-ranked eigenvalues; the selected decay parameter was α = 0.51.Source: §3.2 · p. 3
- A signed dyadic entrainment score combines the two speakers’ directional shifts and normalizes them by their initial separation, representing convergence or divergence relative to that starting gap.The method computes each speaker’s directional adjustment separately; a positive deviation indicates movement toward the partner, and a negative one indicates divergence.Source: §3.2 · p. 3
4. Results
Coordination complexity and entrainment varied over time and across dyad types. In CANDOR, coordination complexity increased early, while positive entrainment was associated with higher perceived conversational success among non-autistic dyads.
4.1. TV coordination complexity
- In CANDOR, mean ω decreased from 0.3552 in the first half to 0.3352 in the second, a shift shared across speakers.The main effect of half was significant (F(1, 1380) = 170.54, p < .001), while speaker and half × speaker effects were not significant.Source: §4.1 · p. 3
- Mean ω decreased from 0.3904 to 0.3560 across CANDOR conversation quarters, with the largest decline for both speakers occurring within about seven minutes.Quarter had a significant main effect (F(3, 4134) = 136.81, p < .001); within-quarter speaker differences were largely non-significant.Source: Figure 4 · p. 4, §4.1 · p. 4
- In UTD, NA–NA dyads showed more complex coordination toward conversation end, AT–AT dyads maintained similar complexity, and mixed dyads became more simply coordinated.The authors suggest mixed-dyad patterns may reflect effort to make speech clearer to a partner of a different neurotype.Source: §4.1 · p. 4
4.2. Entrainment
- In CANDOR non-autistic dyads, positive entrainment (e > 0) was more common among highly successful conversations and was associated with higher perceived conversational success.Positive e includes trajectories that reduce speaker separation or cross during the interaction.Source: §4.2 · p. 4
5. Conclusion
The study presents articulatory coordination and entrainment as markers of interactional alignment and communicative efficacy, with patterns differing across neurotype pairings and datasets.
5. Conclusion
- Across two datasets, same-neurotype dyads showed increasing, more complex articulatory coordination as conversations progressed, while mixed-neurotype dyads showed patterns consistent with sustained or increased articulatory effort.Articulatory entrainment was more prominent and positively associated with perceived conversational success among non-autistic interactions.Source: §5 · p. 4