Social interactions across animal species are governed by the interplay between multimodal sensory information and the internal states of each animal. Here, we investigate this interplay in female Drosophila (fruit fly) as she engages with the male during courtship, a highly dynamic social behavior. Previous work has shown that the male fly's courtship song is shaped by the subtle feedback cues from the female and is influenced by his internal states1,2. However, in the case of females, prior work has focused only on predicting her walking speed averaged over minutes3,4; her fast timescale responses to male cues have remained unexplored. In this study, we construct an internal state model of the female's behavior, aiming to understand the extent to which her locomotion is predictable based on the male cues, and how much it is driven by her own volition. We employ a Hidden Markov Model (HMM) with discrete latent states that is able to predict short-timescale variations in female locomotion. These states correspond to different sensorimotor integration modes, each of which is characterized by a distinct linear mapping from sensory cues to female velocities. Our model accounts for approximately 9\% of the variance in female locomotion, compared to only 6\% using a standard linear regression without internal states. By examining the identified latent states, we found that the contribution of sensory cues to a female's locomotion varies by state, with each state integrating cues across different timescales. By modeling complete sensorimotor transformations and hidden internal state dynamics in the context of natural behavior, and with the ability to leverage the female fruit fly whole-brain connectome6 to connect circuitry with behavior, this work is a necessary step towards a mechanistic understanding of a social interaction.