Handbook of Learning and Cognitive Processes (Volume 6)

Linguistic Functions in Cognitive Theory

William Estes

• Publisher: Psychology Press
• Year: 2014
• ISBN:
  9781317671985
• Edition: 1
• Topic: Psicologia
• Subtopic: Storia e teoria della psicologia

Originally published in 1978, Volume 6 concludes the survey of research and theory on learning and cognitive processes that was envisaged when the plan for this Handbook was sketched.

The primary orientation in the planning the Handbook was to concentrate on research and models aimed toward the development of general cognitive theory.

The first five chapters of this volume are organized in relation to one of the research areas that had expanded most vigorously during the period of planning and writing of the Handbook. These chapters treat aspects of psycholinguistics most closely related to research and theory covered in the other volumes. Perhaps the most fertile source of new concepts and models closely related to other branches of cognitive theory has been research on semantic memory. This work is given a critical review and interpretation by Smith in the first chapter of this volume, following which some lines of theoretical developmental leading "upward" into problems of comprehension of meaningful material are reviewed by Kintsch, then connections "downward" into more elementary problems of coding in memory by Johnson. Also, Johnson's chapter shades into the very active current body of work on perceptual and memorial processes in reading, carried further by Baron's examination of perceptual learning in relation to letter and word recognition. Finally, we consider inputs to the psycholinguistic system via speech and speech perception. The strong emphasis of Pisoni's chapter on speech perception rather than production simply reflects both the predominance of research on perceptual aspects of speech in the current cognitive literature and the close relationships of this research to other lines of investigation of perception and short-term memory.

Some knowledge of the history of the subject and some understanding of the way some of the more persuasive concepts and principles have evolved may serve present-day investigators better than boosting their reading rates. The final chapter of the present volume provides some documentation for this last suggestion.

1

Theories of Semantic Memory

Edward E. Smith

Standford University

I. INTRODUCTION

A. The Domain of Semantic Memory

The phrase semantic memory has been used to denote three views about meaning and memory that vary in how inclusive they are. At the broadest level, semantic memory is assumed to be our store for meaningful material, be it our permanent knowledge of the meanings of words or our transient memory of a particular sentence that was presented to us in a laboratory experiment. Under this definition, theories of semantic memory include, among others, the large-scale memory models developed by Anderson and Bower (1973), Kintsch (1974), and Norman and Rumelhart (1975). A second definition restricts the domain of semantic memory to a subset of the above. This is Tulving’s (1972) distinction between semantic and episodic memory, where the former refers to our permanent knowledge about the language and basic facts of the world, while the latter essentially includes everything else. Under Tulving’s definition, semantic memory would contain the meaning of cat as well as the fact that Columbus is commonly credited with the discovery of America, but not a particular sentence that was just presented to us in a psychological experiment. The third definition, that semantic memory contains only knowledge about the language, specifically, the meanings of words and rules for operating them, is the most restrictive of all.

It would be advantageous in several respects to adopt the most restrictive definition. This would permit us to focus our review on a well-defined domain. Also, it might allow us to relate psychological studies of semantic memory to some well-known linguistic and philosophical approaches to semantics, like that of Katz (1972), though perhaps at the cost of less contact with rival formulations in linguistic semantics. There are, however, two problems with adopting the most restrictive definition. One is that even if we limit our attention to the psychological literature that is explicitly referred to as semantic-memory re-search, we find studies dealing with knowledge other than that contained in word meanings. The second problem is that accepting the most restricted definition seems to entail taking sides on an unresolved empirical issue, namely, whether a separable knowledge system exists that deals only with linguistic (as opposed to real-world) information. In view of these considerations, we will have to remain somewhat open on the question of a definition. In the following discussion, we will take semantic memory to be a system that certainly contains word meanings and rules for operating on them, and possibly includes some general world knowledge as well. Though this is an awfully fuzzy stance to take on a definitional matter, it will allow us to be restrictive in our coverage of meaning and memory, and at the same time treat the distinction of linguistic versus real-world knowledge as an open question.

Given our quasi-definition, where exactly does semantic memory fit into the broader scheme of cognitive psychology? While it is possible to view this topic in the general context of concept attainment and utilization (Rosch, 1974; Smith, Rips, & Shoben, 1974a), in this chapter semantic memory will be treated as a component of language comprehension. Consider, then, the process involved when an attentive listener comprehends a single sentence. At a minimum, three types of procedures must come into play: pattern-recognition devices that encode the words of the sentence, syntactic procedures that parse the sentence and make apparent the grammatical role of each word, and semantic procedures that interpret the meaning of the sentence. There is good reason to believe that these three types of procedures are mutually interactive rather than sequentially ordered and independent. Thus, pattern-recognition devices can use the information available from syntactic and semantic analyses to facilitate the perception of words (for example, Morton, 1969), while parsing procedures may consider semantic and pragmatic facts as well as syntactic information (for example, Winograd, 1972). But despite the complexities resulting from these interactions, researchers have considered it a legitimate endeavor to focus on one of these procedures for intensive study, while keeping in mind how the procedure of interest could be interfaced with the other processes. Our particular concern is with the set of processes that are involved in semantic interpretation. These include: (a) retrieving the semantic representations of individual words, (b) combining these individual meanings into an overall meaning of the entire sentence (where such combinatorial processes may be partly directed by context), (c) possibly relating this overall meaning to a real-world situation so that the truth or falsity of the sentence can be established, e.g., determining whether a sentence is a true description of a particular scene, and (d) using the meaning of the entire sentence to draw some permissible inferences (where again context may play a decisive role). These four aspects of semantic interpretation are precisely what the study of semantic memory is about. Under this view, then, semantic memory is a critical ingredient in the study of language comprehension.

B. Goals of a Theory of Semantic Memory

Each of the above-mentioned processes—retrieving word meanings, combining meanings, interfacing meanings with real-world situations, and drawing inferences—relates to a particular goal that any theory of semantic memory should aspire to.¹

The first goal is to stipulate the psychological representations of word mean-ings. In principle, this will be accomplished when we have a theory that: (a) offers a general description of meaning components and their structure and (b) tells us how we can proceed to determine detailed aspects of meaning for individual cases. Given both (a) and (b), we could then account for why people judge some word pairs to be synonyms (for example, lad and boy), others contradictories (for example, dead and alive), others coordinates (for example, apple and peach), and so on. These are some of the standard phenomena that linguists have traditionally been concerned with (for example, Katz, 1972; Lyons, 1968). To see how far we are from being able to do this, one need only consider that even if we included linguistic and philosophical research in conjunction with psychological work, we could not offer such lexical specifications for more than a handful of words, and even these might prove problematic. (See, for example, Katz’s 1972 treatment of buy and sell for an indication of the detailed analysis needed to stipulate word meanings and the problems inherent in doing so.) It seems that, for now, we must settle for a more limited goal and accept as a contribution any research that specifies some general structural properties of semantic representations. Such properties might, for example, be that the semantic representation of a word contains only those meaning components needed to define it literally or that the components of such a representation must be hierarchically organized. Most semantic memory research has been primarily concerned with establishing such general properties.

The second goal involves a description of the combinatorial processes that operate on individual semantic representations to yield larger meaning units. These processes must have the cast of productive rules, since they operate on a finite set of lexical entries, yet produce an indefinite number of larger meaning units, like sentence meanings. A list of such processes and their exact input and output conditions would, in principle, constitute a means of fulfilling our second goal. Obviously, progress on this line of research will be critically dependent on progress on our first goal, though it is possible to work on this combinatorial problem without firm foundational knowledge about the nature of semantic representations (see, for example, Rips, Shoben, & Smith, 1975; Rips, Smith & Shoben, 1978).

The third goal is to determine how to relate semantic representations, whether for individual words or for larger units, to real-world situations. That is, we want to provide an interface between semantic representations and perceptual processes so we can account for how one recognizes objects from descriptions, or answers questions about visual objects, or calls another’s attention to aspects of the physical environment. At this point in time, it is not even clear what form a solution to these problems would take. It might consist of showing that semantic representations include the type of information needed to identify their perceptual referents. This implies that both linguistic descriptions and perceptual representations of events can be reduced to the same semantic representation, a position argued by Clark and Chase (1972) and Pylyshyn (1973). Alternatively, a solution to this goal might consist of a set of rules that interrelate separate semantic and perceptual representations, a proposal similar to one noted by Kosslyn and Pomerantz (1977). In any event, we can see that again progress on this endeavor will be intimately related to progress on the prior goals, though once more it is possible to make progress on the current problem without firm solutions to the prior ones (see, for example, Clark & Chase, 1972).

The final goal is to be able to specify the kinds of inferences that one makes on the basis of sentence meaning, particularly those based on the meanings of the constituent words, for example, Dan is a bachelor implies Dan is male. A conceivable solution to this problem would involve some sort of formal system that generates only acceptable inferences, and possibly a set of psychological principles that accounts for our preferences for drawing some semantic inferences over others. The kinds of formal systems I have in mind are of two types. In one (for example, Katz, 1972), words are specified in terms of primitive meaning components (for example, one sense of bachelor includes the components of being male and unmarried), and permissible inferences are based on these components (for example, Dan is male can be inferred from Dan is a bachelor). In the other kind of system, permissible inferences are determined by a set of general inference rules or meaning postulates (for example, Fodor, Bever, & Garrett, 1974; Stillings, 1975). Under either type of system, there is again a dependence of the present goal on the previous ones since, for example, questions about inferences are inextricably bound to questions about semantic representations.

While I have repeatedly emphasized how success on a particular goal depends on progress on previously mentioned goals, the situation is not nearly as sequential as my discussion might suggest, for all of these goals are thoroughly interdependent, and how we approach any one of them will have implications for the others. As one example of this, Fodor et al., (1974) noted that there is a tradeoff between issues about semantic representation and issues about inferences, such that many aspects of semantic knowledge may either be incorporated directly into semantic representations or handled indirectly via semantic inference rules. As another illustration of this interdependence, consider how our conceptualization of semantic representations can be affected by our notions about combinatorial rules, semantic/perception interfaces, and inferences. In the preceding discussion of my first goal involving semantic representations, it was mentioned that such representations may be explicated in terms of meaning components, but no constraints on potential components were suggested. Having discussed the other goals, it is now apparent how partially to constrain potential meaning components. Specifically, these components should be capable of: (a) being readily employed by compositional rules, (b) making contact with perceptual representations (though this is too strong a constraint for all meaning components to meet), and (c) fostering the semantic inferences that people in fact draw. That is, the general constraint on meaning components is that they be compatible with solutions to the other goals.

These goals provide a natural way of structuring the field of semantic memory and all its offshoots. However, in this chapter I will concentrate primarily on research related to the first goal—specifying the general properties of semantic representations. The reason for doing so is simple: the bulk of semantic-memory research has been primarily concerned with this issue of lexical representations and with the related questions of how these representations are accessed and utilized in comprehending sentences, particularly those whose meanings can be determined without extensive use of combinatorial processes. In the next, or second, section some critical distinctions that will serve to elucidate and organize semantic-memory theories in general will be considered. One of these distinctions is drawn from linguistics and philosophy, while others derive from psychology proper. Armed with these, I will, in Section III, present the major models of semantic memory, all of which are fundamentally concerned with the nature of semantic representations. (For reasons that are not entirely clear, these models deal primarily with nouns and generally neglect the semantics of other syntactic categories.) In addition to describing these models, I will offer critiques of their ability to account for the most important empirical findings. Finally, in Section IV, I will evaluate semantic-memory research against three rather abstract criteria: generally, extendability, and sufficiency. I believe that this kind of appraisal will indicate the major virtues and failings of semantic-memory research as it has...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Original Title Page
• Original Copyright Page
• Table of Contents
• Foreword
• 1 Theories of Semantic Memory
• 2 Comprehension and Memory of Text
• 3 Coding Processes in Memory
• 4 The Word-Superiority Effect: Perceptual Learning from Reading
• 5 Speech Perception
• 6 On the Organization and Core Concepts of Learning Theory and Cognitive Psychology
• Author Index
• Subject Index