A number of statistical approaches have been developed in recent years to detect and quantify the strength of natural selection using modern genomic data. However, these approaches all have poor temporal resolution and limited power to detect selection acting on standing genetic variation. Ancient DNA allele frequency data provides the most direct and sensitive alternative for detecting selection at specific loci, and offers the possibility of resolving temporal variation in selection strength. However, ancient DNA sample sizes are typically small, and samples tend to be sparsely and unevenly distributed in space and time. In addition, all approaches are sensitive to confounding effects of demography. Here we present a Bayesian framework for reconstruction allele frequency trajectories through time from ancient allele frequency data that can explicitly accommodate the confounding effects of gene flow between populations and uncertainty in sample ages. We applied this method to ancient European domestic chicken genotype data from TSHR locus, which has been argued to be under strong and recent selection in domestic chickens. We find that the hypothesized sweep allele shows a pattern of strong selection starting 1100 years ago, coinciding with a European-wide known shift in poultry management between the mid-ninth to mid-eleventh century. This shift is associated with religious but also a legal rule that required people to abstain from meat, brought in as part of the Benedictine reform. This directly highlights the importance of ancient DNA and statistical modeling for understanding how cultural practices in the past have shaped modern domesticated species.