The rapid growth of machine-learning interatomic potentials (MLIP), particularly the foundational models for materials chemistry, such as MACE, has significantly strengthened our capabilities in studying the complex dynamics of materials through statistical sampling. Here, by performing molecular dynamics with a fine-tuned MACE model, that are further combined with essential dynamics analysis, a tool that is commonly applied in (bio)molecular simulations, we examine the temperature-dependent ion dynamics in a prototypical inorganic solid superionic conductor, namely argyrodite Li₆PS₅Cl. Our analyses show clear localised phonon-like dynamics for Li⁺ ions below 500 K with trajectories lying in a low-dimensional manifold that is highly nonlinear, which is opposite to the case in the high-temperature (∼800 K) end. Between 500-700 K, the system exhibit strong competition in the dynamics between the Li⁺ and framework ions, that are reflected in both the eigenspectrum and the projected low-dimensional embeddings characterising these dynamics not showing a simple monotonic dependency upon rising temperatures, and is further supported by investigations employing Langevin dynamics that systematically examine how damping the framework motions affects the dynamics of the Li⁺ ions, thanks to the computational efficiency provided by MLIP. This work thus contributes towards the broader theoretical and data-driven framework for studying the lattice dynamics of solids.