| English tenses are an inseparable part of the grammatical research.Linguists mainly focus on the descriptive studies of tenses,such as the relationship between tenses and time;the classification,meaning or teaching methodology of tenses.Language has two basic functions: descriptive function and evaluative function(Liu,2006;Wang,2008).This paper holds that the descriptive function of tense is reflected in the indication of time;The evaluative function of tense is the APPRAISAL meaning expressed by the author when the verb tense is inconsistent with the objective time.As a necessary grammatical category,tense has a certain correlation with time,but there is no mandatory indication.It is an important means for people to recognize time and express personal evaluation.The previous studies have mostly focused on the description of tenses,and only a few scholars have analyzed the appraisal meaning of tenses: the mood and modality of tenses change in different situations;local grammar holds that the past tense can reflect the speaker's attitude of liking or disliking;Remoteness expresses the psychological distance between speaker and listeners.However,the previous researches only analyzed part of the appraisal meanings of tenses,and did not systematically analyze it.Meanwhile,with the continuous development of computer algorithms in recent years,natural language processing has become one of the main fields of language research.However,most of the researches focus on the algorithms of character level,and the research on Evaluative tenses is very rare.K-proximity,Decision table and Maximum entropy need large computation space with low recognition efficiency,and fail to take appraisal tense changes into account.As a result,some misrecognized “errors” are automatically “corrected”.Based on the above analysis,this paper focuses on the following three research questions:1.What are the English evaluative tense patterns?2.What is the appraisal meaning of English evaluative tense patterns?3.How to identify English tenses efficiently based on natural language processing technology?This paper holds that there are two conditions for tenses to have appraisal meaning:(1)words or tenses functions as the time reference,which are called time reference points;(2)The tense used in the sentence is different from the time of the reference point.This paper defines the tense with evaluation significance as“evaluative tense”,and the word or tense playing the role of time reference as “time reference point”.The collocation mode of “time reference point + evaluative tense” is a necessary condition for tenses to have appraisal meaning,which is called “the English evaluative tense patterns”.The appraisal meaning produced by the evaluative tense is called "the appraisal meaning of English tenses”.In order to analyze the dynamic meaning of evaluative tenses,this paper proposes a stratified dynamic APPRAISAL model of tenses by combining the APPRAISAL system,stratification,and covert evaluative model of news discourse together.Different from the previous studies,this model can analyze all the appraisal tense patterns,and provide a basis for the analysis of the appraisal meaning of tenses at the lexico-grammar,the discourse-semantics and context(register and genre)strata.Affected by the topics and interpersonal relationships at the register strata,tenses in the lexico-grammar strata express the appraisal meaning,which also realize the meaning of intentionality and acceptability at the discourse-semantics strata;and finally,it generate the attitudinal meaning of emotion,judgment and appreciation.In order to solve the research problems,this paper collects the random samples from the British National Corpus,the NPL vector data and self-built corpus with SSCI and SCI papers.The consistency of temporal annotation is ensured through the automatic code assignment of Clasw 7 Tagset,and then with the manual annotation of UAM,we complete the multi-layer marking.In order to make this research practical,based on natural language processing technology,this paper proposes a new evaluative tense algorithm based on the data statistics and meaning analysis of the English evaluative tense patterns.The deep reinforcement learning algorithm is proposed to classify and recognize tense words at the discourse-semantics strata by combining the text expressions of Allen's relation vector,BCNN and CNN.Then,the random forest partial least squares(PLS),based on Softmax,is further proposed: PLS absorbs the Principal Component Analysis and extracts information from independent variables.It is suitable for data with a smaller sample size and brings great convenience for the evaluation and analysis of complex English tenses.Finally,at the clause level,this paper proposes a new algorithm based on appraisal-driven computing model: attention mechanism and appraisal-driven computing model to improve the performance.The appraisal meaning is divided into M categories with N features,and the degree score set of each appraisal feature is set for clauses,so as to write the computer formula.The artificial intelligence recognition of evaluative tenses is completed through six steps: appraisal meaning computation model,LSTM construction,attention mechanism inclusion,natural language inference instantiation,convergence and feature fusion,and model training and testing.Based on the above research,the following research results are obtained:1.Based on the corpus,this study verifies the actual use of English tenses through data statistics.This paper found that Halliday's 36 secondary tenses were only theoretically idealized,and 15 secondary tenses were counted in the corpus of this paper,among which the past future perfect continuous(T28)only appeared in compound sentences.There are 28 kinds of tense collocation models with inconsistent tense and time.Both news and academic texts have a high proportion of present tense instead of past tense.2.This study proposes a theoretical framework of multi-strata dynamic analysis of the evaluative meaning of English tenses,which complements the analysis of tenses outside the lexical grammatical level in Martin's APPRAISAL theory.The appraisal meaning of tenses is dynamic and changes with different sentences and the discourse environment,which is affected by many factors at all levels.The “time shifting”function of verbs produces appraisal meaning through the change of tense.Attitude is the appraisal meaning of tenses;Engagement is the source of the appraisal meaning,in which the verb is defined as the “shell verb”;Graduation is the degree to which the appraisal meaning is enhanced or weakened by increasing or shortening the psychological distance.3.Through the simulation experiment of the new algorithms for evaluative tenses,it is found that the new algorithms for the lexical phrase layer significantly improve the interpretation degree of the model and the expression of nonlinear structure.Through deep reinforcement learning,the average recognition rate reaches 92.7%.Through the least square method,the highest recognition rate can reach 98.99%.Through comparative analysis,it is found that the new tense-driven appraisal algorithm proposed in this paper has advantages in accuracy,recall rate and regression value.Based on the above results,this study has the following implications: the corpus-based tense profile and evaluative tense pattern statistics provide data support for future tense study;the multi-strata dynamic tense appraisal model provides an analysis framework that can be applied to all tenses.In practice,the new algorithm for evaluative tenses is of great significance to the AI learning of the overall accuracy of English tenses,and also provides a feasible solution for future speech recognition,machine translation and other artificial intelligence research on temporal. |
PDF Full Text Request