Rich semantic representation based approach for text generation

Natural Language Generation (NLG) focuses on the generation of written texts in natural language from some underlying semantic representation of information. A new semantic representation called Rich Semantic Graph (RSG) has been proposed to be used as an intermediate representation during recent research for Natural Language processing applications. In this paper, a new model to generate an English text from RSG is proposed. The proposed model can be exploited in Text Summarization, Machine Translation and Information Retrieval applications. In this model, WordNet ontology is used to generate multiple texts according to the word synonyms. Also, the model enables users to determine the output text style by selecting one of two writing styles (Cause-Effect and Description-Narration). Finally, the model evaluates the generated texts to rank them based on two criteria: most frequently used words and discourse sentence relations.