2.3.2 Research on Evaluation in Academic Writing

Previous descriptions of evaluation provide a reasonably organized framework and a comparative perspective，from which we may examine the distributions of these linguistic resources across several registers（e.g.Biber et al.1999）.Studies have recently focused considerable attention on the multiplicity of ways evaluation is expressed in academic texts.One of the effects of the greater interest in this area is the recognition of the increased variety of linguistic devices of evaluation under examination，including the investigation of reporting verbs（Thompson&Ye 1991），a given lexical item（Lindemann&Mauranen 2001；Oakey 2005；Silver 2003），or a particular word class or type（Charles 2003；Hunston 2005）.In what follows，we will mainly discuss the earliest and clearest attempt made by Hyland at depicting interactive features in science writing.

2.3.2.1 Hedging

Hedging is the expression of tentativeness and possibility.It indicates the writer’s caution in presenting statements and reluctance to make full commitment to propositional information，and in this sense it is closely related with epistemic modality（Lyons 1977）.Since 1990s，this line of research has gained fresh impetus，especially from the innovative studies by Hyland（1998b）.The basic assumption is that all research statements result from careful theorizing or rigorous laboratory procedures，and writers are always conscious of experimental limitations，possible exceptions，and alternative explanations.They recognize that their result can never include every variable to account for all aberrations，but are invariably the product of incomplete information.Writers use hedges to adjust the strength of claims and modify their confidence in their statements.Such devices represent the different degrees of certitude that writers confer on their claims and suggest the adaptation of cautious position.Hyland（1996c）categorizes all scientific hedges into two major types along with a few sub-categories.The categorization is summarized in Figure 2.1.

As shown in Figure 2.1，reader-oriented hedges avoid categorical assertions and invite the reader to participate in a dialogue with the writer.Clear cases of such functions are realized by phrases such as we do not know the reason for，our result，and in our hands.The central device here is the use of personal subject.Hyland argues that an overt acceptance of personal responsibility mitigates the expression of a proposition and signifies a reader-oriented hedge.In addition，personal attribution is also used to soften claims by using the combination of first person pronouns and epistemic verbs such as I/we propose/believe/infer that.

Writer-oriented hedges enable the writer to refer to speculative possibilities while at the same time to guard against possible criticism.The most distinctive signal of such hedges is the absence of writer agency.The writer’s responsibility can be reduced by the use of passive constructions（e.g.is assumed to），clausal subjects（e.g.it might be speculated that）or the construction of abstract rhectors（e.g.these data indicate that）.

Attribute hedges are a finite set of items which are labeled by Ernst（1984）as degree of precision adverbs.A number of these act as downtoners（Quirk，Greenbaum，Leech&Svartvik 1985：509）which weaken the force of an attribute，such as partially，quite and barely.Some of them indicate greater precision in conveying the sense in which a proposition may be held to be true，such as generally and approximately.Still others indicate the precise standpoint from which to judge the truth of a claim，such as viewed in this way and from a practical point of view.

The principal motivation of reliability hedges is to clarify the state of knowledge.Such hedges suggest the writer’s reservations concerning whether the situation actually obtains and are most commonly expressed by epistemic modal verbs（e.g.may），epistemic adjectives（e.g.it appears possible that），nouns（e.g.raises the possibility that），and adverbs（e.g.probably and apparently）.It should be noted that reliability hedges essentially make claims contingent due to knowledge limitations（instead of an additional unwillingness），which is sometimes made explicit by employing devices such as it is not known whether.

2.3.2.2 Metadiscourse

Metadiscourse is commonly known as“discourse about the discourse”which“does not add anything to the propositional content but is intended to help the listener or reader organize，interpret，and evaluate the information given”（Crismore，Markkanen&Steffensen 1993：40）.Metadiscourse is commonly classified into two types：textual and interpersonal，the former relating to the organization of the discourse，and the latter to the writer’s stance towards either the propositional content or the reader.This taxonomy was first developed by Vande Kopple（1985）and was slightly modified in Crismore，Markkanen&Steffensen（1993）.

The initial discussion of metadiscourse was not confined to academic texts.It is Hyland who marshals a range of evidence in support of the view that academic writing is essentially an interactive accomplishment which can only be achieved by using a range of metadiscoursal devices.Hyland incorporates hedging into the broader concept of metadiscourse in the analysis of metadiscoursal features in CEO’s letters（1998a）and in 28 research articles in four academic disciplines（1998c）.He defines metadiscourse as“those aspects of the text which explicitly refer to the organization of the discourse or the writer’s stance towards either its content or the reader”（1998c：438）.This definition is based on the view of seeing writing as a social engagement between the writer and the reader.Hyland states that：

Metadiscourse focuses our attention on the ways writers project themselves into their work to signal their communicative intentions.It is a central pragmatic construct which allows us to see how writers seek to influence readers’understanding of both the text and their attitude towards its content and the audience.（1998c：437）

The schemes of interpersonal metadiscourse developed by Vande Kopple，Crismore，and Hyland are summarized in Table 2.1：

Later Hyland&Tse（2004）re-appraise the issue of metadiscourse，and for the first time propose to blur the distinction between what is textual and what is interpersonal，giving full weight to the interactive nature of metadiscourse.Part of the reason is that there are serious difficulties with the attempt to identify two single，discrete functions of metadiscourse.In fact，most textual resources do not constitute a neatly separate set which can be clearly distinguished from interpersonal aspects.The result is a new modal of metadiscourse incorporating two types of resources：interactive resources which set out an argument to explicitly establish the writer’s preferred interpretations，and interactional resources which alert the reader to the writer’s perspective towards both propositional information and the reader themselves.

2.3.2.3 Model of Interaction in Academic Discourse

While the broad characterization of evaluation in academic texts is interesting，it is still unclear how far these evaluative resources are actually used in particular registers.In order to investigate how academic writers use language to express a stance and relate to their readers，Hyland（2005b）develops a model of interaction which unifies and integrates a diverse array of evaluative features by drawing on interviews and a corpus of 240 research articles.This model is shown in Figure 2.2.

As shown in Figure 2.2，interactions in academic texts are managed by writers in two main ways：stance and engagement.Stance can be seen as an attitudinal dimension and includes features which refer to the ways writers present themselves and convey their judgments，opinions，and commitments.It is comprised of four main elements.Hedges（e.g.possible）indicate the writer’s decision to withhold complete commitment to a proposition.Boosters（e.g.obviously）allow writers to express their certainty in what they say and to mark involvement with the topic and solidarity with their audience.Attitude markers（e.g.hopefully）indicate the writer’s affective attitude to propositions，conveying surprise，agreement，importance，frustration，and so on.Self-mention refers to the use of first person pronouns and possessive adjectives to present propositional，affective and interpersonal information.

Engagement is an alignment dimension where writers acknowledge and connect to others，recognizing the presence of their readers，pulling them along with their argument，focusing their attention，acknowledging their uncertainties，including them as discourse participants，and guiding them to interpretations.There are five main elements to engagement.Reader pronouns（e.g.we）are the most explicit way that readers are brought into a discourse.Personal asides allow writers to address readers directly by briefly interrupting the argument to offer a comment on what has been said.Appeals to shared knowledge（e.g.of course）seek to position readers within apparently naturalized boundaries of disciplinary understandings.Directives（e.g.note）instruct the reader to perform an action or to see things in a way determined by the writer.Questions are the strategy of dialogic involvement par excellence.

It has been found by Hyland that effective academic writing heavily depends on rhetorical decisions about interpersonal intrusion.The model of stance-engagement shows how writers select and deploy community-sensitive linguistic resources to represent themselves.It offers a plausible description of academic interaction and enables writers to anticipate and understand readers’background knowledge，interests and expectations.

In a word，Hyland is probably among the first to seek to increase our understanding of the interpersonal dimension of academic writing.His findings prompt us to re-evaluate most previous studies（e.g.the sociological research）that“see scientific knowledge as less a coherent body of objective truth about the world than a set of justifiable beliefs constructed through interactions among members of the scientific community”（Hyland 1996b：252）.

Hyland’s findings are generally supported by authentic examples selected from a collection of academic texts，but his collection often contains a small number of texts and can hardly be seen as a large electronic corpus.One limitation is therefore that recurrent word combinations and patterns are not easily detectable.