| With the rapid development of the World Wide Web as well as the “affective turn”in humanities and social sciences,evaluative language has risen to be an active interdisciplinary focus of linguistics and computer science.Evaluative language is nowadays generally studied from two approaches: linguistic approach and computational approach.The linguistic approach is largely theory-driven,focusing on the relations between certain linguistic units and evaluative meaning in texts,while the computational approach,also called sentiment analysis(SA),is largely application-driven and aimed at qualitative analysis and computational processing of evaluative texts.As a sub-branch of NLP,SA plays an important role in public opinion analysis,decision-making in business and government affairs.Though the linguistic approach is different from the computational approach,its emphasis on the in-depth linguistic investigation of evaluative texts is still meaningful for SA.Therefore,with concerns on the requirements of SA,the present study focuses on the attitudinal resources in evaluative texts in an attempt to provide implications to the computational processing of evaluative language.As the core of evaluative meaning,attitudinal meaning is strongly associated with certain linguistic markers.Under constructionist view,as pairings of form and attitudinal meaning,such markers are known as “attitudinal constructions” and expressive of attitudinal meaning in texts.Thus three research questions are introduced:in Chinese evaluative texts,what attitudinal constructions are present? How significant is attitudinal construction in the expression of attitudinal meaning? Should attitudinal constructions be taken into account in the computational process of Chinese SA?Previous studies on Chinese constructions fail to give answers to them.In an attempt to explore the significance of attitudinal constructions for SA,the present study takes a corpus-driven approach to identify Chinese schematic negative attitudinal constructions from 1000 pieces of Microblog comments which are retrieved and collected randomly with a Web crawler from five controversial topics under heat debate in Microblog.Based on manual identification,30 negative attitudinal constructions were found in the 1000 pieces of Microblog comments.10 of them are semi-frozen constructions;17 are phrasal constructions;3 of them are compound-sentential constructions.The distributional pattern of the 30 constructions in the categorization of semi-frozen,phrasal,compound sentential constructions is similar to that in “Modern Chinese Construction Knowledge Base”.The analysis shows that negative attitudinal constructions account for a considerable portion of attitudinal resources in Microblog comments.It is a preliminary verification of the hypothesis that attitudinal constructions play a considerable role in the expression of attitudinal meaning in evaluative texts.Thus,the influence of attitudinal constructions should not be neglected in the computational process of Chinese SA.For the requirements of SA,the present study designs a septet of <polarity,sentiment,assessment,stance,speech act,consequence,order> to present the negative attitudinal meaning of the 30 constructions with an attribute-value matrix.Semantics and syntax of the 30 constructions are analyzed.Most of the constructions are found to be [sentiment +] or [assessment +],but only about half of them are both [sentiment +]and [assessment +].Most identified [sentiment +] constructions are encoded with a fundamental type of negative sentiment,“dissatisfied”.According to the values of“speech act” feature,the 30 constructions are categorized into six types,“blame”,“refute”,“disparage”,“mock”,“dissuade” and “forbid”.Blaming construtions,refuting constructions and disparaging constructions account for 90% of the identified constructions.The above analytical results may reveal some distributional features of Chinese schematic negative attitudinal constructions in general.However,applying the a-priori knowledge of attitudinal constructions to Chinese SA is confronted with two problems,firstly,the reduced SA precision rate caused by the probabilistic polarity of attitudinal constructions and secondly,the difficulty of the accurate detection of the form of some attitudinal constructions.The present study is a genre-specific research which is only concerned with attitudinal constructions in Microblog comments.In other words,the hypothesis that constructions account for a considerable portion of attitudinal resources holds true for the genre of Microblog comments but not for all types of evaluative texts.To confirm the hypothesis,it is necessary to conduct additional analysis of other genres of evaluative texts based on a larger corpus. |