Task adaptation from a set of run-time feedback information has become increasingly crucial for corpus-based natural language applications owing to the variant run-time environment. An order-based adaptive learning algorithm is proposed in this paper for task adaptation to best-fit the run-time environment in the application of Chinese homophone disambiguation. It shows which objects to be adjusted and how to adjust their probabilities. The proposed technique is significantly simplified and robust. Experimental results demonstrate the effects of the learning algorithm from generic domain to specific domain. This technique can be easily extended to varied language models and corpus-based applications.