Agent-based Negotiation In Intelligent Pinyin Input Method

As the development of agent-based negotiation,the theoretical study of agent-based negotiation has been applied in daily life.Text,the main information carrier in human-computer interaction,seriously affects human-computer interaction experience.Chinese input method would be the necessary program for Chinese users.In various type of Chinese input methods,Chinese pinyin input method possesses a series of good features.It has been the most popular one in all the Chinese input methods.In the field of machine translation,computer-assisted translation(CAT)has gotten the focus of researchers in recent years.CAT concentrates on how to assist human interpreter to accomplish the translation task efficiently and improve translation quality by computer.Inspired by agent-based negotiation,this thesis abstracted the application scene of Chinese input method in CAT as negotiation domain to accomplish input tasks,and designed an agent-based negotiation intelligent pinyin input method.This input method works in English-Chinese translation environment and concentrates on how to predict the text segment which interpreters want to input.The experiment results proved that the input method can effectively combine Chinese pinyin input method and machine translation features through agent-base negotiation to predict the text segment,and finally,improve the input efficiency for interpreters.The main work of this thesis is as follows:(1)Abstract the Chinese pinyin input application scene as agent-based negotiation domain.The essence of Chinese pinyin input process is decoding the pinyin sequence to Chinese text sequence.This thesis extracts several key features from common Chinese pinyin input method and machine translation as negotiation issues.In negotiation,agents would finish their decoding task independently and generate counter offers according to their preferences.Then the negotiation-based decoder would integrate all offers and generate negotiation outcome,pinyin decoding result.Negotiation ends,the decoding task is finished.The negotiation-based input method would output the negotiation outcome as the decoding result for users.(2)Construct the feature functions for issues.Then use the logarithmic linear model to fuse the features and construct the scorer which is used during the decoding process.The result of feature fusion,scorer,is not only the key for agent to propose offer,but also the most important component for the whole negotiation-based pinyin decoder to decode pinyin sequence.With respect to the different type of features,we applied two main methods to construct corresponding feature functions,including language-model-based evaluation metric and text similarity metric method.(3)Validate the model and determine negotiation agent parameters.The aim of agent-based negotiation pinyin decoder is to fuse all the input features properly,and effectively predict text segments which interpreters want.Scorer model,which is constructed by feature fusion through logarithmic linear model,is the key for the pinyin decoding process.We used a number of sample cases and designed a series of experiments to test the validity of proposed negotiation-based decoder,and finally,used minimize error rate tool(Z-MERT)framework to optimize parameters.