| Argument structure alternation refers to different yet related argument realization patterns of some verbs in semantically similar constructions(Perek,2015,p.146).And the dative alternation in English captures the notion that two constructions,namely the double object construction and the prepositional dative construction,can be used interchangeably(Rappaport & Beth,2008,p.129;Beth,2015;Perek,2015,p.146).Previous studies have investigated this issue in various aspects,such as the mechanism of dative alternation,dative variants,dative verbs,the usage features,etc.However,there are still some limitations.For starters,existing research approaches are inadequate or one-sided in explaining the relationship between the two variants and how they alternate.Meanwhile,dative alternation is influenced by various informationstructure-related factors of the recipient and the theme,like semantic and pragmatic information(Perek,2015).Most studies have not taken into account the semantic differences of dative verbs,while verbs play a crucial role in options of argument structure realization patterns.As a matter of fact,there do exist some differences in givetype,send-type dative verbs,in that give-type verbs are associated with only one event schema(a caused possession meaning)while another type has two event schemas(both caused motion and caused possession meanings)(Rappaport & Beth,2008).But a few studies are conducted to explore whether there are differences between dative alternations containing the two types of dative verbs.Moreover,corpus-based methods having been widely used to investigate the usage features of dative alternation,mono-factorial analysis is often adopted.But multifactorial analysis is less mentioned,in which many factors,as well as their interactions,can be analyzed more systematically.What’s more,most studies adopt only one single corpus-based model,and fewer studies combine different statistical models to provide a more comprehensive analysis of dative alternation.In light of that,the study,introducing the verb-sensitive approach into the theoretical framework,establishes an integrated approach called the verb-sensitive allostruction model to investigate the general features of the send-type dative alternation while observing whether there are differences between dative alternations with send and give.With a corpus-based quantitative method,the current study,on the basis of previous studies,establishes three corpus-based probabilistic models in order to provide plausible explanations for those corpus data using the dative verb send as a case.Moreover,their predictive power and accuracy of speakers’ choice on two allostructions are discussed as well.In this way,the feasibility of the verb-sensitive allostruction model can be proved.The research questions are what follows:(1)What are the general features of send-type dative alternation? What are the similarities and differences compared with previous research conclusions on the dative alternation containing the prototypical dative verb give?(2)Can the corpus data support the choice of the two variants(the double-object construction and the prepositional dative construction)of send-type dative alternation? If so,how? How are the predictive ability and the accuracy of the statistical models?(3)What are the reasons for the choice of the two construction variants of send-type dative alternation? Are the results of the research able to provide further evidence for the verb-sensitive allostruction model?To answer these questions,the current study adopts the corpus-based approach,the corpus data of which are mainly extracted from Corpus of Contemporary American English(COCA).For the word send and its infections,a query is run in each section,and2001 hits are retrieved and filtered manually from each returned data,which are then annotated in terms of a total of 12 variables about the recipient and the theme that are based on previous studies(Bresnan & Ford,2010;R?thlisberger et al.,2016;Zhang,2020)—accessibility,definiteness,pronominality,animacy,concreteness,length difference,and style.Later on,three statistical models are established in R software.Through these statistical models,the quantitative structure of the dataset can be understood well.The results of the models are analyzed and compared to draw a conclusion about the general features of send-type dative alternation.The research has found that:(1)Variables that have significant effects on send-type dative alternation in English include pronominality of recipient,length difference,definiteness of recipient,pronominality of theme,animacy of theme,definiteness of theme,animacy of recipient,accessibility of recipient,accessibility of theme as well as style,among which pronominality of recipient is the most significant predictor while concreteness of theme is insignificant.And multiple interactions exist between these variables,which are then visualized through the tree & forest model.For example,pronominality,discourse accessibility,and animacy interact with each other in many different ways.The interactive effects being taken into account,the relative importance of these variables is ranked as follows: pronominality of recipient > accessibility of recipient > pronominality of theme > animacy of recipient > length difference > definiteness of theme >accessibility of theme > animacy of theme > definiteness of recipient > style.Moreover,the current research compares the send-type dative alternation and the give-type dative alternation to summarize the general features of send-type dative alternation.It has been found that there are both similarities and differences between the two kinds of dative verbs.As for the similarities,significant variables are almost the same,among which concreteness of theme is an insignificant predictor.When the theme is definite,the prepositional dative construction is preferred for both types.And interactions between different predictors exist in usage features of both the give-type and the send-type dative alternation.When it comes to the differences,the most important predictor,for the two types,is not the same: accessibility and length difference have the most significant effects in the give-type dative alternation,while pronominality is the most important one for the send-type dative alternation.Besides,the relative importance of these significant variables is ranked in different ways for two types of dative alternation.(2)There are three statistical models adopted in this study.For the logistic regression model,C value(index of concordance)is 1,and accuracy is 89%;for the tree& forest model,C value is 1,and accuracy= 88%,89%;as to the na?ve discriminative learning model,C=0.94 and accuracy= 87%.Thus,all of them can offer very similar results across the corpus datasets,but it is also clear that the former two models outperform the NDL model.To be more specific,firstly,all three models have their ways of assessing the importance of variables.The tree & forest model can be used to detect interactions between the variables and their relative importance in a visual way.Also,each variable’s coefficient and weight are two kinds of indicators of the importance for the other two models.Although differences do exist in each model’s results owing to their distinct algorithms,it is without a doubt that there is an overlap and convergence between x them.In the second place,the logistic model and the tree & forest model can both provide reasonable outcomes for the significance test,which can be indicated by the p-value,while it is not available to the NDL model for now.Given the advantages as well as the weaknesses,the study draws a conclusion that the tree & forest model can be utilized as an instrumental method to complement the logistic regression model,whereas the NDL model can still be offered as a helpful alternative with its higher cognitive plausibility and psychological reality.(3)Through the observations on the results demonstrated by the statistical models,it is pretty clear that the use of dative alternation is indeed influenced by factors as regards information packaging.Some principles,such as the end-focus principle,the end-weight principle(Quirk et al.,1972),harmonic alignment effects(Bresnan & Ford,2010),are provided further evidence.Meanwhile,it shows that the prepositional dative construction and the double-object construction are different syntactic realizations in specific contexts,supporting the allostruction model.What’s more,the differences between the send-type and the give-type of dative alternations suggested by the results of the study prove the reasonability of the verb-sensitive approach put forward by Rappaport & Beth(2008),in that different characteristics of dative verbs set constraints to the situations in which they are used.And the links between a dative verb and the dative allostructions are far more complex.In a nutshell,the proposed theoretical framework,i.e.,the verb-sensitive allostruction model,is plausible.The present study has both theoretical and methodological implications.First,it provides an alternative view on the argument structure alternation and further proves the allostruction model theory.Meanwhile,with the verb-sensitive approach taken into consideration,the study compares two types of dative alternation so as to summarize the general features of the send-type dative alternation.Additionally,it sheds some light on the nature of language grammar and language use and supports the probabilistic and dynamic nature of grammar.With human brains having a complex cognitive mechanism,language use is affected by various variables,including arguments,categorization and environmental contexts,and there exist interactions between these factors.Hence,speakers’ choice of variants does not happen randomly but can be described as probabilistic and systematic(Tagliamonte,2015,p.313).As for methodological implications,three statistical models are established and analyzed in the current study.The logistic regression model,the tree & forest model and the na?ve discriminative learning model have been combined in an attempt to summarize the general features of the send-type dative alternation.Last but not least,this paper ends by discussing future perspectives of dative alternation studies from the aspects of research data,annotation,and research tools.It calls for more converging evidence from other disciplines or methods,such as psycholinguistic or neurolinguistic experiments,cross-constructional analysis,and so forth. |
