INFERENCE GENERATION DURING TEXT COMPREHENSION: AN OVERVIEW OF THEORY AND RESEARCH

Get Complete Project Material File(s) Now! »

CHAPTER3 INFERENCE GENERATION DURING TEXT COMPREHENSION: AN OVERVIEW OF THEORY AND RESEARCH

Introduction

Although, in the previous two chapters, the role ofinferencing during reading comprehension was identified as an important one, the discussion so far has centred mainly around the broader issue of reading and its function in the learning context. The purpose of this chapter is to return to the central topic ofinferencing in this study and to present an overview oftheory and research into the role ofinferencing during text comprehension during the past twenty years or so. The notion ofinferencing and the nature of inferential processes are examined more closely, and the issues and questions that characterise this domain of study are set out in greater depth. The position of this study in relation to the larger domain of inference research will also be identified in the course ofthe discussion.
Since the seventies there has been a proliferation of research into language processing and comprehension in different but related fields such as~ey~~inguistics, cognitive@Cfi9logy and artificial <Ult~iiig~’!ic~ and some of this research has focussed on inferencing during interrelated; in fact, Rickheit, Schnotz & Strohner (1986:5) describe comprehension itself as « an inferential activity », while Sanford(1990:515)wryly states that »theubiquity ofinferences in text comprehension makes the study oftext comprehension look like a subset ofthe study of inference making ». The early recognition in the seventies of the prevalence of inferences in discourse processing caused a shift in research focus to more discriminating questions about the nature of inferences and the conditions that generate them. The questions that have informed and guided inference theory and research concern issues such as:

• What is an inference? This question is concerned with definitional features of inferences and with the nature of inferencing in general. Definitions of inferences inevitably draw one into consideration ofthe different kinds of inferences that occur, since the question What is an inference? easily prompts an answer like It depends on what kind ofinference one is referring to.
• What kinds of inferences are drawn during comprehension? Although this question is closely related to the first one, it also concerns the taxonomic classification of inferences. What kinds of inferences do readers make and how should they be classified? As Sanford (1990: 516) points out, « (t)here is the sheer difficulty of deciding what is the possible set of inferences that could be made over any given period of reading a discourse ».
• When are inferences drawn? Here the controversy revolves around whether inferences are generated <!~-line.. that is, during the comprehension process, or whether they are generated~ that is, later, after comprehension, usually in response to a prompt or a specific task.
• How are inferences drawn and constrained? This question concerns the kind of knowledge and operations or processes involved in generating inferencing, and the factors that constrain them. Not all inferences that are potentially permitted by a text are made, for this would lead to an inferential explosion. It relates t• b•th representational and procedural issues underlying inferencing, and to text-based and reader-based variables that constrain inference generation.
• How are inferences detected and data obtained in order to study them? These are more practical questions that raise methodological concerns. Despite theic~inferences are very elusive processes to pin down, measure and analyse and the field is, understandably, fraught with methodological problems.

The literature review in this chapter will attempt to deal with each of these questions in tum. Much of the work related to these issues is highly theoretical in that it is concerned with establishing what skilled L 1 subjects do when they read and comprehend texts. Afterwards, some of the research related to more applicational matters such as differences between good and poor readers and inference training in the classroom will also be described, and the approach adopted in the present study will be identified.

What is an inference?

To explain what an inference is – or is not – obviously calls for a definition of the concept. Although definitions are useful because they help to delimit the concept under discussion, one must be mindful that they are determined to a large extent by underlying theoretical assumptions and methodological purposes.
they are different sides ofthe same coin. As Singer points out (1994:480), these two terms « refer to activities on the part of the speaker and the understander, respectively », a distinction generally observed in the formal study oflanguage processing. When a speaker/writer implies something, the listener/reader is expected to make an inference.
Let us start examining the question of what an inference is from the long-standing formal and traditional perspective. Formal approaches to the definition of inference within the domain of language comprehension have used the criterion of deducibility. In other words, does the process entail a valid deduction? If so, then it is an inference. This formal definition defines a specific type of inference, the logical or deductive inference, and the criterion ofdeducibility gave rise to a distinction between logical/necessary and possible/probable inferences. As was pointed out in Chapter 1, logical inferences are based on formal rules. In logic, a specific conclusion [4c] that is derived from specific premises is termed a deductive inference [the primary premise 4a and the secondary premise 4b].
[4a]  If an iron rod is heated, it will expand.
[4b]  Hector is heating an iron rod.
[4c]  The iron rod will expand.
The argument in [4] is a conditional argument, a type of causal inference, called Modus Ponens, where the secondary premise confirms the antecedent clause in the primary premise. There are four kinds of conditional inferences, which are classified in terms of whether the secondary premise confirms or denies a clause of the primary premise (Franks, Mulhern & Schillinger 1997 :293). Other common types oflogical inference include class instantiation (Great Danes are large dogs. Hedda is a Great Dane. Therefore Hedda is a large dog) and disjunction (Either the fridge has tripped or the washing machine has tripped. The washing machine has not tripped.
Therefore the fridge has tripped).
Logical inferences are context-free because the premises entail the conclusion, and the conclusion is valid by virtue of the fact that it is not possible to assert the premises and deny the conclusion without contradicting oneself. In other words, logical infere!_lces are 100% cer.tain. Overton (1990, in Franks et al.1997 :287) defines a deductive inference as  » a process whereby a proposition (the conclusion) is arrived at and accepted on the basis of other propositions (the premises) that were originally accepted ». Definitions of an inference, from a logician’s point of view, do not involve gap-filling operations. Instead, they involve reasoning that reaches conclusions that are not simply plausible or possible, but logically necessary, and they therefore involve different cognitive processes from those used in probable inferences (Franks et al. 1997:286).
Because everyday discourse is embedded in context and is far messier than the discourse of formal logic, looser definitions of inference involving possible/probable inferences (also called invited or pragmatic inferences) are typically adopted by researchers working in language processing. Probable inferences are possibly true implications of sentences in the text, given the state of affairs in the world. These inferences are typically based on people’s general background knowledge: « although they are often probable, they are not certain » (Singer 1994:481 ). This was illustrated in Chapter 1 with the example of Hector burning his mouth. Negation of a sentence from which a possible inference is drawn does not lead to an inconsistency.
Vonk & Noordman (1990) state that the necessary/deducibility criterion for defining an inference is problematic because in terms of which criteria can an inference be said to be necessary? Comprehension is not a monolithic concept, but a graded one. In similar vein, Sanford (1990) argues that the necessary-possible distinction, though intuitively appealing, does not have a readily defined boundary, which weakens its usefulness. Although the logical/necessary-probable distinction serves to differentiate inference categories, « its utility as a processing distinction is somewhat limited » (Singer 1994:481), since detecting all the logical inferences of a message would lead to an inferential explosion. Likewise, some possible inferences are more likely to occur than others due to their high probability and « computational simplicity » (Singer 1994:481 ). Due to these kinds of problems, several researchers working within text comprehension avoided formal definitions of inference and adopted more procedural or computational approaches to the question of what constitutes an inference during text processing.
Let us briefly consider one such approach where, in the early stages of inferential research, the criterion of automaticity or processing time was used to determine whether or not a cognitive process involved inferential activity. In other words, processing time had implications for definitions of what constitutes an inference. The early work by Haviland & Clark (1974) suggested that a lot of inferential activity occurs when readers identify referents for definite noun phrases. In the by now oft -cited example, the researchers found that readers exposed to texts such as [10] took longer to compute the referent for the beer in [lOb] than readers did in [9b].
[9a]  Mary got some beer out of the car.
[9b] The beer was warm.
[lOa] [ 1Ob]
Mary got some picnic supplies out of the car.
The beer was warm.
In example [9], a linking operation is performed via repetition of the noun beer, while in [ 10] the linking operation that the reader computes is that picnic supplies include beer. The differences in comprehension times in short samples oftext such as these were explained in terms of the extra processing time that is needed to form a bridging assumption or backward inference as in [ 10]. The implication, as Brown & Yule (1983:257) point out, is that « inferences take time ». Linking operations that required extra processing time were regarded as non-automatic processes, and hence inferences, whereas linking operations that occurred rapidly and automatically were regarded as not involving inferential activity. However, such an approach is somewhat impractical, since one can only establish post facto whether a cognitive process is or is not an inference.
Empirical support for such an approach was contradictory. Sanford & Garrod (1981) used similar two-sentence texts involving missing links to those used by Haviland & Clark (1974), but they found that some missing links did not require extra processing time, e.g.:
[lla] [llb] Mary dressed the baby.
The clothes were made of pink wool.
[12a]
[12b]
Mary put the baby’s clothes on.
The clothes were made of pink wool.
The researchers found that no extra processing time was required in the implicit text in [11] compared to the explicit text in [12]. The explanation they proffered moved beyond processing time to include representational issues. In [11] the missing link is that dressing a baby involves putting clothes on the baby. The researchers suggested that clothes is part of a reader’s representation of the verb dress, and that when this verb is activated, then clothes is also automatically activated, whether it is explicitly mentioned or not. The conclusion based on this kind of psycho linguistic data was that when the missing link activated by a particular text is part of a reader’s knowledge representation, then no extra processing time is needed, but when the missing link does not form part of the knowledge representation, then extra processing time is required to make a bridging inference. Such an approach incorporates both procedural and representational aspects of text processing into definitions ofwhat constitutes an inference.
This approach obviously has implications for how one defines inferences. Sanford (1990), for example, argues that it is misleading to say that in a sentence such as He unlocked a door the inference He used a key is necessarily made. Rather, such knowledge becomes accessible (Sanford 1990:519), but doesn’t necessarily translate into the actual drawing of an inference. This gives rise to a position where inferences are defined in terms of a distinction between activation (or priming)¹ and inferencing. It has been argued that the activation of a concept in semantic memory leads to the activation of related concepts, such that a spread of activation occurs along neural pathways (Singer 1988: 178). Corbett & Dosher (1978) distinguished between activation and encoding. They argued that the former involve the transient activation of implied ideas – although concepts are often briefly activated, they do not necessarily result in an inference being made, i.e. they are not always necessarily encoded. There has been a lot of research about priming of concepts at different levels, such as at the level of word meaning (Meyer & Schvaneveldt 1971, in Singer 1988), the factual level (Anderson 1976), and the level oftext meaning (McKoon & Ratcliff 1990), but a discussion of these issues will take us beyond the scope of this chapter.

CHAPTER 1: READING, INFERENCING AND ACADEMIC PERFORMANCE
1.0 Introduction
1.1 The focus of enquiry
1.2 Reading in the learning context.
1.3 Aims and objectives ofthe study
1.4 Overview of methodological framework
1.5 Structureofthethesis
‘-CHAPTER 2: READING: A THEORETICAL AND DEVELOPMENTAL PERSPECTIVE
2.0 Introduction
2.1 Assumptions about reading comprehension: a historical perspective
2.2 ‘Bo<.’k language’ -the language of written discourse
2.3 The development of reading and the causes of reading problems
2.4 Theoretical and developmental perspectives: a summary
+CHAPTER 3: INFERENCE GENERATION DURING TEXT COMPREHENSION: AN OVERVIEW OF THEORY AND RESEARCH
3.0 Introduction
3.1 What is an inference?
3.2 What kinds of inferences are made during comprehension?
3.3 When are inferences drawn?
3.4 How are inferences drawn and constrained?
3.5 How are inferences detected and studied?
3. 7 Inference development and application issues
3.8 Summary
CHAPTER 4: ANAPHORIC INFERENCING
4.0 Introduction
4.1 Anaphora
4.2 Research into anaphoric resolution
4.3 Analytic tools: Categorisation of anaphoric relations
4.4 Research questions and hypotheses
4.5 Research methodology
4.6 Results
4. 7 Errors in anaphoric resolution
4.8 Discussion
5.1 Vocabulary development: an overview of some issues
5 .2 Vocabulary and reading
5.3 Aimsofpresentstudy
5.4. Methodology
5.5 Results
5.6 Discussion
t CHAPTER 6: SEMANTIC RELATIONS AND THEMATIC INFERENCES
6.0 Introduction
6.1 Coherence, texts and reading
6.2 Semantic relations
6.3 Thematic inferences
6.4 Methodology
6.5 Results
6.6 Discussion
CHAPTER 7: GENERAL INFERENCING SKILLS
7.0 Introduction
7.1 Research context
7.2 Methodological issues
7.3 Results
7.4 Discussion
CHAPTER 8: CASE STUDIES
8.0 Introduction
8.1. Background
8.2 Interview procedures and methodological limitations
8.3. Reading profiles of the students
8 .4 C onstruct1. ng meani. ng du n.n g read 1′ ng
8.5 Interest and motivation
8.6 Summary
CHAPTER 9: CONCLUSION
9.0 Introduction
9.1 Revisiting the aims ofthe study
9.2 Overview of chapters
9.3 Limitations ofthe study
9.4 Educational and instructional implications
9.5 Conclusion
APPENDICES
BIBLIOGRAPHY
GET THE COMPLETE PROJECT
INFERENCE GENERATION IN THE READING OF EXPOSITORY TEXTS BY UNIVERSITY STUDENTS