A Study On Metaphorical Calculation Of Natural Language

In our daily communication,one in three sentences is a metaphor.Metaphor is not only a rhetorical device,but also a necessary way of thinking and cognition.It plays an irreplaceable role in our daily life,scientific research and social management.The ubiquity of metaphor and its importance in thinking and cognition determine the urgency of metaphor research.At present,the study of metaphorical computing has become a hot topic in the field of natural research.The study of metaphor has been carried out earlier in foreign countries,and the study of metaphor in China,especially the calculation of Chinese metaphor has been carried out relatively late.There are not many representative theories and the proposed metaphor calculation model has some shortcomings.First of all,Chinese metaphor studies are mostly based on semantic knowledge bases manually constructed by linguists or psychologists and there is a black box operation in the processing process,and there is a lack of metaphor quantitative calculation step-by-step processing procedures and stepby-step processing experimental data results.Secondly,the study of metaphor recognition is the basis of metaphor understanding and metaphor generation.In recent years,some good natural language processing models have emerged,but they are rarely used in metaphor recognition tasks.At present,the accuracy of metaphor recognition is not high.In addition,most of the current related metaphor computing research is based on the theoretical research of the knowledge base.How to construct a tool that can assist the metaphor computing research is also a very urgent task in the current metaphor research.In view of the above-mentioned shortcomings,this project relies on the large-scale Chinese corpus construction project of the Language Engineering Center of the School of Foreign Languages of Donghua University.First studied the metaphor quantitative calculation processing flow then proposed a metaphor recognition method based on BERT pre-training language model and Stacking integrated learning and finally developed for non-computer professional researchers based on the research process A metaphorical auxiliary calculation tool.The main research contents of this paper are as follows:First of all,studied the process of metaphor quantitative calculation,and designed a metaphor quantitative calculation method based on Latent Semantic Analysis.Through in-depth exploration of the underlying semantic relationship between words and words,words and text,and text and text,a step-by-step processing flow in the study of metaphor quantitative computing is constructed.Moreover,calculation methods such as latent semantic analysis,singular value decomposition,and Pearson correlation coefficient are used in this step-by-step process.After the text corpus is processed,the calculation results of each step in the step-by-step process can be obtained.For noncomputing researchers,especially related researchers from the School of Foreign Languages in the context of this project,we provide a step-by-step processing flow and quantitative and evaluable data results to provide support for downstream tasks.In addition,this paper extracts the correlation between content words and content words,and visually analyzes the correlation between content words,further verifying that the metaphor quantitative calculation research in this paper has important research significance.Secondly,studied the method of automatic metaphor recognition.This paper proposes a metaphor recognition method based on BERT and Stacking integrated learning,and used simple classifiers such as support vector machines,decision trees,random forests and logistic regression to simultaneously perform metaphor recognition.These single classifiers are used as control experiments.Finally,a comparative analysis of multiple sets of experimental results proves that the metaphor recognition method proposed in this paper has higher accuracy and more stable metaphor recognition performance.Finally,designed and implemented a natural language metaphor computing system.The system relies on the content of the first two parts of metaphor quantitative calculation research and metaphor recognition method research,and provides a metaphor calculation research including five functions including word segmentation and part-of-speech tagging,extracting content words,generating content word document matrix,LSA content word similarity calculation and Stacking classification prediction Tools to provide relevant researchers with auxiliary tools for metaphor computing research,especially for the large-scale Chinese metaphor corpus being established by the Language Engineering Center of the School of Foreign Languages.