Evaluating Neighbor Rank and Distance Measures as Predictors of Semantic Priming

This paper summarizes the results of a large-scale evaluation study of bag-ofwords distributional models on behavioral data from three semantic priming experiments. The tasks at issue are (i) identification of consistent primes based on their semantic relatedness to the target and (ii) correlation of semantic relatedness with latency times. We also provide an evaluation of the impact of specific model parameters on the prediction of priming. To the best of our knowledge, this is the first systematic evaluation of a wide range of DSM parameters in all possible combinations. An important result of the study is that neighbor rank performs better than distance measures in predicting semantic priming.