The Choice of Plausible Alternatives task serves as a challenging benchmark for commonsense causal reasoning. Existing traditional statistics-based approaches represent and extract commonsense causal knowledge primarily according to the word proximity in text corpus, and use asymmetric PMI metric to judge plausibilities of premise-alternative sentence pairs. Focusing on clause-level proximities in text, this paper takes three kinds of weak causal evidences into consideration, and designs a probabilistic method to aggregate these weak evidences extracted from text corpus. Experimental results have shown that our probabilisitc models on two GB-level text corpora (PersonalStories-1.6M and BookCorpus) even outperforms (or be competitive to) the state-of-the-art statistical models (CausalNet and CausalNet-MWE) learned on terabytes’ text corpora.