Semantic Parsing Of Action Text For Text-to-Scene Conversion

With the continuous development of human society,word is no longer the only way to communicate between people.The visual way of expressing people's thought is gradually being pursued and yearned by more and more people.It is time-consuming and needs professional knowledge to make 3D scene manually.So,how to generate the visual scene automatically is the key to realize this hope.The text-to-scene conversion was born under such demand.It can dependent on computer to realize the creation of 3D scenes from natural language description automatically.It is of great scientific significance and social value to free hands and improve understanding of computer.Besides,it also has a broad application in the fields of design,animation,education,military,artificial intelligence and so on.Rendering the action correctly is one of the key points of generating dynamic scene.However,the current research about text-to-scene conversion are focus on the static scene,it is still lack of basic theories and effective solutions about dynamic scene and action semantic.Facing all this,action text is studied.Firstly,in order to meet the needs of text-to-scene conversion,the related verb classification idea is proposed,the concept of action text,the feature of various action text,and the formal expression of action chain structure are proposed.In addition,on the premise of analysis of the factors affecting,the semantic dictionary of action text is established,and a method of semantic analysis of action text is proposed based on the dictionary.Considering the different expressions between dictionary and actual text,the method of extracting the action text is improved.Finally,the threshold is used to distinguish action text and non-action text,extracting action text automatically from the text is realized.And context similarity is as a feature to optimize the similarity calculation method.The experimental results show that the method proposed in this thesis can extract and parse action text from the input under certain conditions,which prove the accuracy of the theoretical analysis and the effectiveness of the algorithm,and provide not only the theoretical support for the exploration of dynamic scene in text-to-scene conversion,but also a reference for computer to understand the semantic of action text.