In order to solve the problem of overcorrection in automatic post-editing translations, the authors propose to make advantage of the neural post-editing (NPE) to build two special models: one is used to provide minor edit operations, the other is used to provide single edit operation, and make advantage of machine translation quality estimation to establish a filtering algorithm to integrate the special models with the regular NPE model into a jointed model. Experimental results on the test set of WMT16 APE shared task show that the proposed approach statistically outperforms the baseline. Deep analysis further confirms that proposed approach can bring considerable relief from the over-editing problem in APE.

Because positional language models did not consider semantic relationship between the words in different positions, the authors present an effective model named “positional language models with semantic information”. Firstly, the authors use Gaussian kernel function to measure the position relationship between words. Secondly, the authors present a technology which is named “smoothed mutual information” to measure semantic relationship between the words, and also prove that smoothed mutual information can effectively solve the problem that a large number of two words could not calculate the transition probability between them only by mutual information. Then the authors prove that positional language models are a special case of positional language models with semantic information. Finally, applying this new model to the area of information retrieval can obtain a retrieval model based on the new model. The experiment show that the retrieval model based on the new model performs better than a retrieval model based on positional language models for using in information retrieval.