| The rapid development of information technology has provided more convenience for the study of changes in natural language.In recent years,with the rise of artificial intelligence and the high-speed advancement of deep learning technology,continuously enriched language corpora and improved data processing capabilities have made largescale language computation possible.This provides new theory,methodology,resource and technological support for automatically discovering the diachronic changes in word meanings.This dissertation focuses on the monitoring and discovery of diachronic changes in word meanings.Large-scale diachronic corpus with a long-time span is constructed.Based on a combination of pre-training models and morphological analysis,dynamic word embedding representation technology is used to generate semantic vector,and then the semantic similarity of word is measured by vector distance.By comparing the degree of semantic similarity and distribution characteristics of word meanings in different periods,the detection of changes in word meanings over time is carried out.A multi-dimensional and multi-strategy fusion method for discovering diachronic changes in word meanings is proposed,which is based primarily on conceptual meanings,with grammatical meanings and coloring meanings as auxiliary factors,and incorporates examination of the use domains of lexical and formal markers.The main work of this dissertation includes: firstly,constructing a large-scale longterm corpus with a time span of 73 years and approximately 2.75 billion words,mainly employing newspaper data and social media data,and conducting targeted annotation processing on it;secondly,starting from three dimensions of conceptual meaning,coloring meaning and grammatical meaning,designing and proposing a method for automatically discovering diachronic lexical semantic changes;thirdly,using words whose dictionary definitions have changed as references,the effectiveness of the method is verified through experiments and analysis;fourthly,automatically mining words with semantic changes in a large-scale corpus,analyzing and summarizing the characteristics of their semantic changes from linguistic perspective.At the same time,this dissertation verifies some patterns and rules of lexical semantic changes over time based on experiments and corpus data analysis: firstly,conceptual meaning,coloring meaning and grammatical meaning are the main parts of lexical semantic change;secondly,grammatical meaning changes are more likely to occur in monosyllabic words and verbs;thirdly,the transfer of meanings presents a multi-directional and multi-dimensional characteristic.The characteristics and innovations of this dissertation are mainly reflected in three aspects.First,a new method for automatically discovering diachronic lexical semantic changes with multiple dimensions and strategies is proposed,which detects the diachronic changes of word meanings from three dimensions of conceptual meaning,coloring meaning and grammatical meaning.It uses changes in the word use domain,as well as changes in form markers and collocations of words,as clues to assist in change detection,providing a new perspective for subsequent research.Second,the proposed dynamic word embedding representation method that integrates morphological information enhances the accuracy of embedding representation of different senses of polysemous words,provides more accurate input for lexical semantic change computation,and provides new ideas for applying dynamic word embedding representation to semantic computation research.Third,a new method for data-driven screening of candidate words for semantic change is proposed.Unlike manually selecting candidate words or simply choosing content words as candidates,it can better combine linguistic features such as morphology,part-of-speech and syllables,as well as statistical features such as word frequency,distribution and usage stability to comprehensively screen candidate words with a higher probability of undergoing semantic changes,providing a new optimization direction to improve the efficiency of lexical semantic change detection. |
