The Interchangeability of the Significant Features to Ditransitive

103 theme prefers prepositional dative and non-given theme prefers double object construction.

4.2.2 The Interchangeability of the Significant Features to Ditransitive

Further steps were done to check the interchangeability of the significant features in particular and of the model in general. The corpus probability model which is built by Bresnan from 2360 instances of dative construction is taken and tested to this research‟s data set of Benefactive. Out of thirteen significant features to dative, six features which are also significant in benefactive cases are in use. The other seven features are left out because whether this research does not occupy the features as predictors or the research does not find the features significant to benefactive construction. The logit probabilistic formula of mixed- effect binary logistic model of Bresnan‟s dative is applied in this research benefactive data set. Both probabilistic formulas are as follow: constant+β1x1+β2x2+β3x3+… +βnxn e In p constant+ β1x1+β2x2+β3x3+… +βnxn 1+e expconstant+ β1x1+β2x2+β3x3+… +βnxn 1+ expconstant+ β1x1+β2x2+β3x3+… +βnxn In p = constant+ β1x1+β2x2+β3x3+… +βnxn 1 - p 104 The following logit formula of benefactive probabilistic model which occupies the following constantintercept and coefficients is replaced with the logit formula of Bresnan‟s dative probabilistic model. p = constant+ β1x1+β2x2+β3x3+… +βnxn 1 - p = -6.091-0.59syntacticcomplexity a+2.27beneficiaryanimacyb+2.57beneficiary givennessc- 2.18themepronominalityd+1.09personofbenefi ciarye-1.34theme definitenessf The model above is replaced by the following model: p = constant+ β1x1+β2x2+β3x3+… +βnxn 1 - p = 0.95-1.21syntacticcomplexity a+2.54beneficiaryanimacyb+0.99beneficiary givennessc- 1.18themepronominalityd+0.48personofbenefi ciarye-1.37theme definitenessf While Bresnan‟s model reaches 95 of model fit accuracy when it is applied in dative data set, surprisingly, the result shows that it can only reach 60.5 of model accuracy when it is applied in benefactive data set. It suggests that the different of size of the six features affecting benefactive and dative is great enough to decrease the model fit accuracy. The accumulated size different of the six features in Bresnan‟s dative and this research benefactive explains the decreasing of the model fit accuracy until almost 35. 105 Similarly, when the corpus probabilistic formula of the benefactive construction is applied in dative cases, the model fit accuracy extremely declines. As the researcher does not have the access to Bresnan‟s dative data set, a new set of dative construction is made. 40 instances are taken from TIME Magazine Corpus, annotated and coded wit h the same system as Bresnan‟s. The following benefactive probabilistic model is applied in newly made dative data set. p = constant+ β1x1+β2x2+β3x3+… +βnxn 1 - p = -6.091-0.59syntacticcomplexity a+2.27beneficiaryanimacyb+2.57beneficiary givennessc- 2.18themepronominalityd+1.09personofbenefi ciarye-1.34theme definitenessf The corpus probabilistic model of benefactive construction, beforehand reaches 91 of model accuracy in predicting benefactive construction. Surprisingly, when the corpus probabilistic model of benefactive is employed to the dative data set, the model only reaches 47.5 accuracy. It can presumably be explained by two major factors. Similar with the result of the first step of model cross-ditransitivity, the first factor is the different size of the six features toward benefactive and dative construction. Whereas, the second factor is most likely the significance level of the seven features which the corpus probabilistic model of benefactive does not possess. As stated above, thirteen features significantly affect the choice of dative construction, while only six features affect the alternation of benefactive. Thus, when benefactive probabilistic model formula with only six features is employed to predict dative construction, the model is inevitable to be 106 less accurate. It shows that the seven features, which are affecting dative but insignificant toward benefactive, have a big role toward the choice of dative construction. Apparently, the research cannot provide equivalently comparable statistical data of the roles of the rest seven features toward dative and benefactive construction. It is because the greatly different number of instances taken in the data set for benefactive and dative. It leads to a question for the future research whether dative is indeed more complicated than benefactive, or simply it is because the data set for this research in benefactive is not large enough to find as many features a s found in Bresnan‟s dative. However, based on the data set used in this current research, the results presumably suggest that dative construction is more complicated than benefactive. It is attested that it possesses more significant features to be regarded when predicting the construction than the benefactive one. On the other hand, the six features significant to the dative and benefactive constructions appear to hold stronger effect toward the choice of benefactive construction rather than to dative. In conclusion, six features are identified relevant to the ditransitive construction, which are directionally similar yet are different in the size of the effect. The six features syntactic complexity, animacy of beneficiary, givenness of beneficiary, pronominality of theme, person of beneficiary, and definiteness of theme are significant in both benefactive and dative. The direction of the features suggest the same way in both construction, yet the size of the effect is greatly differ so that the probabilistic model of benefactive and dative are not interchangeable. When the probabilistic model of dative is applied in benefactive 107 data set, the result shows that it can only predict 60.5 of the instances. Similarly, when the probabilistic model of benefactive is applied in a newly made data set of dative constructions from TIME magazine, it can only correctly predict 47.5 of the constructions. 108

CHAPTER 5 CONCLUSIONS AND RECOMMENDATIONS

Chapter five presents the conclusions and recommendations. The conclusions summarize the findings of this research and the recommendations provide suggestions for further research related to the topic of study.

5.1 Conclusions

In general, the analysis of corpora data in this study resulted in the identification of features which are significant to the choice of benefactive construction along with the probabilistic formula to predict the occurrence of the instances in benefactive construction. The research finds six features to be significant toward the choice of benefactive construction. The six significant features are syntactic complexity, beneficiary animacy, beneficiary givenness, person of beneficiary, syntactic complexity, beneficiary animacy, beneficiary givenness, and person of beneficiary . Four features carry positive coefficient, meaning that directionally the instances with those four features favor benefactive PP construction. The four features with positive coefficient include syntactic complexity, beneficiary animacy, beneficiary givenness, and person of beneficiary . The result shows that when the log scale of length different beneficiary-theme is positive, the construction prefers benefactive PP. Similarly, when the beneficiary is inanimate, non-given, and non-local, the construction seems to appear in benefactive PP. Two other features possess negative coefficient suggesting that the instances with those two features favor double object construction. The two