This chapter reviews two research areas that are closely relevant to the research topic in this book：discourse coherence and multimodal discourse analysis.The underlying assumption for the review is that theories and frameworks of discourse coherence in linguistic texts can be hopefully extended to multimodal texts.From the review conclusions can be drawn in the following aspects：

First of all，the review of discourse coherence shows that there are various factors and aspects responsible for the construction of discourse coherence，including the external socio-communicative factors，the reader's cognitive and psychological factors，and the text internal factors.The relations between these different aspects of coherence construction have been inadequately studied（Zhang，1999）.The relationships between the multiple layers of coherence construction，including socio-communicative，mental，text-internal factors such as global coherence，local coherence，and cohesion，need to be made clear.

Since this book takes the perspective of discourse analysis，the emphasis will be placed on the text-internal factors for coherence instead of the external ones.This emphasis on text leads to two important tasks.One is how the internal factors influence external ones，and the other is the relations between the multiple internal factors.It is not enough to have an intuitive recognition that these factors don't have the same status in the construction of discourse coherence.More efforts should be made to shed light on the specific roles and functions of each major factor in the weaving of texture.For this end，these distinct layers of factors should be selectively re-organized within a single，coherent and operational model with which systematic，clear，empirically sound and effective analysis results can be achieved.

Secondly，even within the text-internal aspect of discourse coherence，the situation is complex.Global coherence，local coherence and cohesion are the three layers that are widely recognized and discussed.The biggest problem in these studies on the three levels of factors is that the opinions on them are highly diverse yet interrelated.This situation is especially serious for global coherence compared to other two aspects.For example，there has not been a clear definition or criterion of global coherence.Although discourse topic and macro-structure are widely recognized as closely related to global coherence，there are still problems in the analysis，such as the difficulty to define discourse topic.Moreover，the way discourse topic and macro-structure operate to exert the global control on the overall discourse is still to be further clarified.To solve the problem，clear definitions and criteria for global coherence，local coherence and cohesion should be made，at least for the specific genre（s）of the data being analyzed.

Thirdly，if the frameworks of discourse coherence in purely linguistic sense are to be extended to account for the coherence construction in multimodal texts，the features of multimodal texts must be taken into account，and the definitions and criteria of these levels of coherence should be adapted accordingly.Therefore，the definitions and criteria of these levels of coherence should be flexible and compatible enough to accommodate the complexity of means of expression，such as written language，images，charts，and so on.Although the works on visual-verbal relations as reviewed in Sections 2.2.2 and 2.2.3 could be regarded as contributing to the research question of coherence in multimodal texts，they belong to the category of local coherence and cannot exhaust the potential of integration of resources from different semiotic modes into a coherent discourse，such as the global connectivity and cohesive devices in the expression level.The implication is that for each of the level of coherence，namely global coherence，local coherence，and cohesion，a working definition and criterion should be made in a broad semiotic sense to account for the coherence construction across multiple semiotic modes.

Fourthly，despite the current explosion in semiotically-based multimodal studies，the methods employed are often not up to the complexity of the artefacts analyzed，and as a consequence，the results obtained are still too often presented in idiosyncratic ways，with little constraint on the style of analysis carried out and the kinds of conclusions drawn（Bateman，2009）.This limitation in methodology is most obviously manifested in the data based on which the results are obtained.The data used in the existing researches of visual-verbal relations are not systematically collected and cannot represent features of a specific genre.Since semiotic configurations，and also relations between semiotic modes vary greatly with discourse genres，it is impossible to draw a general pattern that applies to all genres.The implication is that in analysis from a semiotic perspective the data must be representative of one specific genre instead of a mixed collection.

Finally，another limitation in methodology in existing studies in the field of multimodality is its subjectivity.Multimodality researches can seem rather impressionistic in its analysis（Jewitt，2009：26）.The meaning of images，gestures，and other nonverbal modes，depends to a large extent on the researcher's interpretation.This is in part an issue of the heritage of multimodality from the social semiotic perspective of linguistics.It is beyond doubt that multimodal research needs to be increasingly reliable，evaluable and free of pre-structured conceptions about the nature of multimodal phenomena（Hiippala，2012）.For this end，it will be very beneficial if the results of the textual analysis of coherence can be triangulated by psycholinguistic research methods which can reveal the mental mechanisms and activities in the reader's comprehension process.

【注释】

[1]Samet and Schank（1984）did not use the terms“local coherence”or“global coherence”，but in Unger's（2006）review the expository connections between sentences are called local coherence and the story line is called global coherence.

[2]MOPs is the short term for Memory Organization Packets.It represents procedural knowledge，and constitutes an extension of Schank and Abelson's scripts（Schank&Abelson，1977）.A MOP represents a chain of events that are temporally ordered and causally connected to reach a goal，and includes planning knowledge.（van Geenen and Witteman，2006）