This paper examines the significance of tasting notes in determining wine scores. Using a dataset of over 6,700 tasting notes and scores by Robert Parker, we develop a deep learning model based on ’Bidirectional Encoder Representations from Transformer’ (BERT) to evaluate the consistency between textual tasting notes and numerical ratings. Our model predicts Parker’s scores with high accuracy, affirming the consistency of his evaluations. We then analyze auction data to quantify the price premium associated with tasting notes and scores. We disentangle intrinsic wine quality from perceptual bias. Our findings suggest that both objective and subjective features of tasting note significantly influence wine prices. Our study highlights the impact of expert judgment (and mis-judgment) on pricing and offers a replicable framework for machine-learning-based wine valuation.
