Dependency tree contains rich syntactic structure information, which is very effective for semantic role labeling. Semantic role labeling is a task with strong syntactic dependence. How to effectively use subgraphs to predict surrounding graph structures for context of dependency tree and encode the dependency relationships between predicates and arguments in the sentence which to be annotated is still a challenging problem. In this paper, we propose a bidirectional graph convolution model based on distribution of graph structure and the attention mechanism, which is used to make better use of all syntactic information in the sentence dependency tree and the dependencies between predicates and arguments, so as to improve the accuracy of semantic role labeling. Without heavy feature engineering and extra knowledge of a particular language, our model obtained 88.9% F1 value on the WSJ test set of public dataset CoNLL2005, and 0.7% improvement compared with the existing best model.

The emergence of strong convective weather has caused serious threats to people’s lives and production activities, and also caused serious damage to the ecological environment. Because this kind of weather has a short duration and a small impact area, it is difficult to be accurate predicted. Doppler weather radar is one of the devices that can sample strong convective weather continuously for a long time, and plays an extremely important role in the detection and prediction of severe weather. The prediction of radar echo intensity by using deep learning methods has largely improved the prediction accuracy of strong convective weather, but the prediction of radar echo shape change is not yet ideal, so the trend of strong convective weather cannot be accurately judged. This paper proposes a weather radar echo prediction method based on cyclic convolutional neural network, which is called the EDD model, its design was inspired by the RU_Net families of deep learning models. The feature fusion layer was added to the neural network, thus ensuring that the feature matrix contains more feature parameters and improving the problem of inaccurate prediction of radar echo shape change due to insufficient information during upsampling. The EDD model is trained using Doppler weather radar echo maps, and the experimental results show that the EDD model used in this paper can improve the prediction accuracy of radar echo intensity and make good predictions of radar echo shape changes, thus helping meteorologists to make more accurate judgments on the occurrence, development and change trends of strong convective weather.

Hidden markov model (HMM) is frequently used for Pinyin-to-Chinese conversion. But it only captures the dependency with the preceding character. Higher order markov models can bring higher accuracy, but are computationally unaffordable to average PC settings. We propose a segment-based hidden markov model (SHMM), which has the same magnitude of complexity as first-order HMM, but generates higher decoding accuracy. SHMM tells a word from a bigram connecting two words, and assigns a reasonable probability to words as a whole. It is more powerful than HMM to decode words containing over two characters. We conduct a comprehensive Pinyin-to-Chinese conversion evaluation on Lancaster corpus. The experiment shows the perfect sentence accuracy is improved from 34.7% (HMM) to 43.3% (SHMM). The one-error sentence accuracy is increased from 72.7% to 78.3%. Furthermore, SHMM can seamlessly integrate with pinyin typing correction, acronym pinyin input, user-defined words, and self-adaptive learning all of which are a must for a commercial Pinyin-to-Chinese conversion product in order to improve the efficiency of pinyin input.