PROJECT SUMMARY/ABSTRACT
People with Parkinson's disease (PD) experience restricted communicative participation [1,2], which means that
they have difficulty “taking part in life situations where knowledge, information, ideas or feelings are exchanged”
[3]. This restricted participation can have significant consequences for one's well-being, including social isolation,
loss of employment, deterioration of relationships, and difficulty accessing services [3,4]. We currently do not
have behavioral treatment targets that directly and effectively improve participation. The most basic form of
communicative participation is an interaction between two people, a dyad, engaged in dialogue. The small body
of research that has examined interaction behaviors (i.e., interdependent verbal behaviors) in people with PD
(and/or dysarthria) has used descriptive or case study designs to qualitatively describe the dialogues [5-11]. In
a prior R21, we extended this work, leveraging the existing rich literature on dialogue theory that was developed
with neurotypical (NT) adults, coupled with speech signal processing techniques and advanced statistical
modeling, to quantify interaction behaviors in task-controlled dialogue corpora [12-16]. Reasonable conclusions
from this collection of studies are that the interaction behaviors of PD-NT dyads are fundamentally different from
those of NT-NT dyads; and that these differences correlate to restricted participation in the dialogues of people
with PD. The next step is to move from these simple correlations to deriving causal, mechanistic relationships
between interaction behaviors and participation outcomes, allowing us to identify candidate treatment targets for
improving participation in people with PD. We will use two aims to do this. The first aim, using a between-within
design, will compare interaction behavior from the dialogues of PD-NT and NT-NT dyads and quantify their
variation across typical communicative situations. We will audio-record people with PD and NT controls (dyad
condition) engaged in a problem-solving and a rapport-building dialogue (goal condition) with a familiar and an
unfamiliar partner (partner condition). The resulting 480 dialogues will be annotated, and automated metrics of
interaction behaviors, including individual interaction behaviors of each person (articulatory precision, rhythmic
predictability, language complexity) and coordinative interaction behaviors (entrainment, conversational repair,
turn-taking), will be extracted. Linear mixed effects models will be used to assess these behaviors by dyad, goal,
and partner condition. Our second aim looks to assess the impact of each interaction behavior on communicative
participation outcomes in the dialogues of people with PD within a causal inference framework. Whereas SA1
compares the interaction behaviors across dyad types and communicative situations, SA2 uses the interaction
behaviors as input to a causal inference model. This will allow us to quantify how much each interaction behavior
contributed to the associated participation outcome for that dialogue and rigorously simulate the effects of
intervening on each interaction behavior. Assessment of the impact of each behavior will be done using Bayesian
multilevel modeling. Results will inform a subsequent hypothesis-driven clinical trial of interaction targets.
