Word meaning-in-context can vary in a very fine-grained manner, raising the question of how to predict this variation from context. I present a Magnitude Estimation (ME) task in which participants judged similarity in meaning between pairs of verb occurrences consisting of a motion verb and a singular definite NP subject (e.g. ``The kid runs'' vs. ``The rabbit runs''). Differences between the subject nouns in these pairs are hypothesized to predict intuitions about verbal semantic similarity. I explore four measures of noun similarity, two based on noun animacy, as well as conceptual and distributional similarity measures. I find that all four factors are significant predictors of verbal semantic similarity judgments, but that the best model combines all four measures. This is taken as an indication that a proper use of converging sources of evidence enables a much finer-grained study of word meaning in context than is generally undertaken.