Studying non-locomotor uses of locomotor appendages can provide insight into the role of functional trade-offs in the evolution of animal anatomy. Bird tails have obvious aerodynamic functions, but they can also serve as potentially conflicting signals in territory defense, courtship, or predatory avoidance behaviors. To look for evidence of non-aerodynamic tail function in flying birds, we used multiple high speed cameras to record wild American Goldfinches (Spinus tristis) during individual and agonistic flights at established bird feeders. We reconstructed their three-dimensional movements based on a wand calibration in Argus (which uses direct linear transformation, DLT). Quantifying such variable behavior requires larger sample sizes than traditional lab-based biomechanical studies. To automate markerless digitizing, we trained two levels of DeepLabCut models, and developed custom code modules to convert data between DLT and DeepLabCut formats. The first model tracks each bird in each camera view, finds bounding boxes around each bird, and uses the output to create cropped videos for each bird. The second model performs detailed tracking of important kinematic landmarks on each bird in the cropped videos. Using cropped videos allows for faster and more accurate landmark tracking than running DeepLabCut on full resolution videos. Another module then rescales the coordinates from the cropped videos to full dimensions, converts the data format from DeepLabCut to DLT, and uses Argus to perform the 3D reconstruction. All custom code is available on GitHub. We found evidence of both aerodynamic and behavioral signaling functions of bird tails; for example, goldfinches fan and depress their tails during deceleration, and exaggerate those motions during agonistic interactions achieving greater decelerations.