基于条件随机场的藏语自动分词方法研究与实现

藏语自动分词是藏语信息处理的基础性关键问题,而紧缩词识别是藏语分词中的重点和难点。目前公开的紧缩词识别方法都是基于规则的方法,需要词库支持。该文提出了一种基于条件随机场的紧缩词识别方法,并在此基础上实现了基于条件随机场的藏语自动分词系统。实验结果表明,基于条件随机场的紧缩词识别方法快速、有效,而且可以方便地与分词模块相结合,显著提高了藏语分词的效果。     

Estimates of soil phosphorus for different soil series

Over a range of soil series, the responses to applied P were measured with Swedish turnips and potatoes in field experiments and with oats in a pot experiment. Phosphorus was extracted from soil samples, taken before the experiments were started, by acetic acid, Morgan's, bicarbonate, resin and lactate methods. The relationships between the amounts of soil P extracted and the responses to applied P showed that curved models, using reciprocals or logarithms of soil P, were generally superior to linear and quadratic models especially for the field experiment data. Including loss-on-ignition in the model significantly improved the relationships for some soil series.Reciprocal and logarithmic models were used to determine the ranges for classifying soil P into three classes and to calculate responses to soil P at each site. For the experiments with Swedish turnips, soil P and calculated responses were allocated to three classes and compared with the corresponding classification of the actual field responses. Using different ranges of critical values for classifying soil P for a specific series is shown to be at least as important as using the best method of extraction. The results illustrate how the reliability of a three-class system for predicting the P status of these acid, Scottish soils can be increased by using different class intervals for soil P for different series.