Research On Chinaese Voice Instruction Parser For Home Service Robot

In the robot research fields, such as robotics navigation, rescue robots and robot assistants, the problem of controlling the robot by voice has become one of the most important research topics. How to parse the voice instructions to a sequence of actions which robot can perform is important for the robot to complete the tasks. Based on a comprehensive analysis of domestic and foreign research, this thesis has comprehensively used many aspects of theories knowledge and done studies on the approach for the problem of parsing Chinese voice instructions.Firstly, given the related knowledge as well as the method used to create the knowledge base, the knowledge base is required by the robot, and then the thesis also given the statement parsing mechanism for parsing and semantic analysis of forms of expression.Secondly, the thesis has given the method to create an entity attribute library, according to the acquired surface information, by inquiring about the entity attribute library, extracted the semantic categories of each entity, obtaining the required semantic categories information to complete the task of taking delivery of items, extracting the deep information of the Chinese instructions. And then mode of processed the extracted deep information.Thirdly, based on in-depth analysis and studies of the characteristics of model instructions of the taking delivery items tasks, an approach for creating a parser based on the combinatory categorial grammar is proposed in this thesis. The problem of mapping the model instructions to action sequences of instructions is converted to find the the maximizd conditional likelihood of training data by using the command parser learning algorithm. Training is performed on model instructions annotated with the corresponding action commands sequences. Then obtains the optimal parse of Chinese voice instructions by adopting stochastic gradient descent method to maximize the likelihood values.Finally, experimental results show that the proposed method is feasible and effective, and the analysis of experimental results are made.