A Corrective Training Algorithm for Adaptive Learning in Bag Generation

expressions. However, this approach is still very intuitive and simple because it only adjusts the word frequencies in run-time, and does not revise the statistical information in the long-term memory. Multiple language model is based on several corpora of different fields. Basically, a small amount of similar text are imported and interpolated with the original texts, when the context domain is presented. The extra cost of this approach is the context determination. The similarity measures among test sentences and the pre-defined context domains may introduce additional errors. The sampling problem in training corpus is one of the major sources of errors in corpus-based applications. This paper proposes a corrective training algorithm to best-fit the run-time context domain in the application of bag generation. It shows which objects to be adjusted and how to adjust their probabilities. The resulting techniques are greatly simplified and the experimental results demonstrate the promising effects of the training algorithm from generic domain to specific domain. In general, these techniques can be easily extended to various language models and corpus-based applications. In this paper, we would not like to touch on the power of language models. We focus on the sampling problem in training corpus. A corrective training algorithm, which can be also regarded as a dynamic adaptive learning algorithm, is proposed for bag generation. It exploits the run-time feedback information to best-fit the run-time environment. That is, when error occurs, the error result will be corrected by users. Through the modification, the system learns and adapts. It learns the differences between the correct result and the error result. These form the useful run-time feedback information. In other words, the system learns from the mistakes it makes. Under this way, we first propose a language model to deal with the sentence generation, i.e., bag generation, problem and a generic corpus is used to extract the corresponding statistics information. Then the training algorithm will try to adapt the generic language model into a specific one according to the useful run-time feedback information. At the same time, the probabilities of the related entries in training table are adjusted. In the following sections we first introduce the bag generation algorithm, then describe the adaptive learning model for bag generation. Before concluding we