The paper has two goals. The first is to provide a data-centered, epistemically justified judgment about the status of moralizing high god theory’s contested ability to explain the origins of cooperation in China. The second is to demonstrate the utility of new quantitative methods for the potential disconfirmation or confirmation of close reading interpretations (or at least of those that are operationalizeable and falsifiable). Central to ongoing debate about the role of gods on cooperation is a widening difference of scholarly opinion on what have come to be identified as ‘Moralizing High God Theory’ and ‘Broad Supernatural Punishment Theory.’ We introduce an automated technique for association mining for the humanities and operationalize the moralizing high gods theory in the context of an ancient Chinese corpus. Both theories argue that supernatural agents with certain traits bear a special relationship to cooperation. Those traits include a concern with human morality, monitoring of human behavior, and desire to punish wrongdoers. Moralizing high gods theory predicts that high gods will, ceteris paribus, bear stronger semantic relationships to those traits than will low gods like deities, sprites and ghosts. Broad supernatural punishment theory predicts that semantic relationships with these traits will be spread across different categories of supernatural agents. Sinological interpretations of these texts using close readings tend to be split between the two. We design and apply an association mining technique from corpus linguistics to calculate collocations of token members of agent categories (high gods, deities, sage kings, ancestors) and trait categories (punishment, reward, morality, cognition, religion) in a large corpus. This corpus contains 5.7m characters of classical Chinese and is composed of the most important classical Chinese texts dating from pre-Warring States times through the Han Dynasty.