Retrieval and Organizational Strategies in Conceptual Memory (PLE: Memory)

A Computer Model

Janet Kolodner

• Publisher: Psychology Press
• Year: 2014
• Print ISBN: 9781848724112
• eBook ISBN: 9781317685913
• Edition: 1
• Topic: Psychology
• Subtopic: Cognitive Psychology & Cognition
• Index: Psychology

'Someday we expect that computers will be able to keep us informed about the news. People have imagined being able to ask their home computers questions such as "What's going on in the world?"…'. Originally published in 1984, this book is a fascinating look at the world of memory and computers before the internet became the mainstream phenomenon it is today.

It looks at the early development of a computer system that could keep us informed in a way that we now take for granted. Presenting a theory of remembering, based on human information processing, it begins to address many of the hard problems implicated in the quest to make computers remember.

The book had two purposes in presenting this theory of remembering. First, to be used in implementing intelligent computer systems, including fact retrieval systems and intelligent systems in general. Any intelligent program needs to use and store and use a great deal of knowledge. The strategies and structures in the book were designed to be used for that purpose. Second, the theory attempts to explain how people's memories work and makes predictions about the organization of human memory.

I	INTRODUCTION

1	The Nature of a Long-Term Memory for Events

1. GETTING STARTED

There are many manifestations of memory use in our everyday life. When someone asks us a question, and we attempt to answer it, it is quite obvious that we are remembering something. When we are “reminded” of something that happened previously, it is also obvious that memory is involved. But in many of the other tasks we do each day, we are not cognizant of the fact that memory is necessary to the task. In order to perform even the simplest task, such as mailing a letter, for example, it is necessary to remember the procedure for doing the task. When we empathize with another person’s problems because we have had similar problems, we are remembering our own previous experiences. When we understand more of a story than is told to us, we are using our memories to fill in the details. When we walk to the right place at the end of the day to get into the car, we are remembering, whether consciously or unconsciously, where the car was parked in the morning.

One prerequisite to achieving intelligence in computer systems is to give them memories with many of the capabilities people have. For intelligent fact retrieval, memory is especially important. Consider, for example, the following hypothetical situation involving the computer:

You haven’t been keeping up with the news lately, and you are concerned about developments in the Middle East. You know that the secretary of state went on a trip to the Middle East recently. You want to find out what he discussed during that trip and what was accomplished. Instead of leafing through two weeks worth of newspapers, you walk up to the computer terminal at the library and have the following conversation:

> When did Cyrus Vance return from the Middle East?

LAST NIGHT.

> Which heads of state did he talk to there?

MENACHIM BEGIN OF ISRAEL, ANWAR SADAT OF EGYPT, KING HUSSEIN OF JORDAN, AND KING KHALAD OF SAUDI ARABIA.

> Did he talk to President Assad of Syria?

NO.

> …

Although the situation above is hypothetical, the computer program illustrated and the dialog with that program are real. The dialog is with CYRUS (Computerized Yale Retrieval and Updating System), a computer program that stores episodes from the lives of former Secretaries of State Cyrus Vance and Edmund Muskie, and later retrieves those episodes to answer questions posed to it in English. Although CYRUS itself was designed to store information about important people, we can imagine CYRUS embedded in a larger system capable of storing a wide range of current events. Such a system would be useful to people who haven’t been keeping up with the news and want to catch up quickly or learn about some event in detail, to reporters who need to find background information for a new story, or to school children as interactive encyclopedias. On a very practical level, it would be useful to newspapers, which already maintain huge files containing biographies of famous people. One can easily imagine other uses for computer memories capable of natural language interaction.

Any system that keeps track of even a subset of current events will be very large, maintaining data bases of many hundreds of thousands or millions of events. In a system that large, it is important to organize the data so that only relevant records are searched, and so that the number of relevant records to be examined at any time is as small as possible. In a system that keeps track of current events, the memory will constantly be changing. As new events happen, they must be added to the memory. As memory grows, it is imperative that good organization be maintained: otherwise, the initially well-organized system will be useless after many additions.

How can we achieve such results? The algorithms and structures a memory must have are not at all obvious. While we want the computer to perform like a person, it is difficult to discover the way people remember. People effortlessly recall past events and episodes, yet they are largely unaware of the cognitive processes they use in remembering. Since it is usually an unconscious process, it is hard to monitor the human retrieval process to see how it is actually done. We do know, however, that in people memory is a cognitive process that operates on concepts present in memory. We must therefore describe memory in terms of processes that work on conceptual information.

This book reports the results of research aimed at constructing a theory of long-term event memory. A process model of long-term memory is presented as a theory of conceptual information processing. Because people are good conceptual information processors, we turn to people as an available example of the intelligent processes we are trying to explain.

There are three questions that must be answered in addressing this problem:

1. What processes are implicated in retrieving events from memory?

2. What kind of memory organization do the retrieval processes imply?

3. What are the processes for adding new events to memory—how can good memory organization be maintained and how does memory organization change as new events are added?

These questions cannot be answered independently of each other. The organization of memory places constraints on the types of retrieval and updating processes memory must use. On the other hand, memory organization, and therefore procedures for adding information to memory, must be designed with retrieval requirements in mind.

The remainder of this chapter presents an overview of some of the problems involved in retrieving events from memory and some of the requirements the retrieval process places on memory’s organization. We do that by looking at examples. We also discover some of the capabilities and limitations of human memory. In so doing, we discover the requirements of a computer program that “remembers” and motivate many of our design decisions. An overview of CYRUS, a computer program that implements the theory, is also presented.

2. THE RECONSTRUCTIVE NATURE OF HUMAN MEMORY

For the most part, people are expert at recalling information about their pasts. When people try to retrieve information about a particular episode in their lives, however, they often find they don’t have the entire episode immediately available. In that case, they attempt to reconstruct the rest of the episode. When asked to recall a recent trip, one person started his reconstruction as follows:

Let me see. When we got there, we must have had to find a hotel. Yes, I remember, we had a guidebook with a lot of hotels in it, and we called a few of them until we found one with a vacancy. It was late afternoon, so we must have gone out to eat soon after that.¹

Clearly, the trip was not stored in memory in one large chunk. Instead, it must have been stored in bits and pieces requiring reconstruction to put them back together. Such reconstruction can happen through application of generalized knowledge. In the protocol above, the knowledge that “one must have a hotel to stay in during a trip” and “usually after arrival, the next step of a trip is to check in at the hotel” allowed “finding a hotel” to be recalled.

The process seems to involve building a description of “what must have happened” and then filling it out with “actual” details. We can thus conclude the following about human memory for events:

Memory Principle #1

Human remembering is often a process of reconstructing what must have happened rather than directly retrieving what did happen.

This process of reconstruction is almost always used by people trying to remember episodes, as evidenced by the fact that people tend to “remember” details incorrectly so often. In fact, retrieval confusions and false recall must be inherent to a process that produces probable rather than actual explanations. We see later why we might want to give this capability to the computer despite the fact that it often produces errors in details.

3. MEMORY ORGANIZATION FOR RECONSTRUCTIVE RETRIEVAL

In discovering a memory organization for reconstructive memory, we must find one that not only supports such retrieval, but also requires it. To begin addressing that issue, consider the following hypothetical dialog:

Q: Have you been to Saudi Arabia recently?

A: Yes, most recently last month, to discuss the Camp David Accords with King Khaled and Prince Fahd.

Q: Where did you go afterwards?

A: To Syria. I was touring the Middle East talking to each of the Arab leaders about the Accords.

The answers given in this dialog would have been reasonable ones for Cyrus Vance to give while he was secretary of state. During the 4 years he was in office, however, he went on hundreds of trips, among many other memorable activities. How could he choose the correct ones from memory to answer specific questions? If memory were arranged in lists, then we could imagine a process that searched down an appropriate list until the correct episode was found. This seems unreasonable, however, for a memory with hundreds or thousands of events of the same type: Searching down a long list is a slow process that becomes slower as additional items are added to the list. People’s memories do not slow down as they learn more.

Furthermore, if people searched down long lists in their memories to retrieve events, then it would be easy for them to enumerate experiences of particular types. For instance, a question such as the following would be easy to answer:

(Q1): Recall all the times you have been to museums.

This question, however, is not easy for people to answer, as suggested by the following protocol:

I know I’ve been to a lot of museums in Europe. I’ve been to England, and I went to a number of museums there—some in London—the British Museum, the National Gallery, and a few smaller galleries. … I was at a museum in Brighton—the Royal Pavilion. I’ve been to museums in Paris—the Louvre and some smaller ones. In Rome, I’ve been to…. In Naples, to…. In Florence, to….

The lists of experiences people enumerate are constructed on the fly as they are answering a question. In the protocol above, the person tried to recall “experiences in museums,” “experiences in museums in Europe,” “experiences in museums in England,” “experiences in museums in London,” and so on, filling in (or reconstructing) additional details with each iteration. Generalizing from that, we can make the following hypothesis:

Memory Principle #2

Remembering requires progressive narrowing in on a description of the event to be remembered.

We can describe that process as a process of specifying details that can differentiate the targeted event from other events in memory.

The protocol above and others like it suggest that narrowing the description of a sought event, or focusing in on an event by providing more detail, is a necessary part of remembering (Williams, 1978). Thus, we can state the following principle of memory retrieval:

Memory Principle #3

In a reconstructive memory, memories are not directly enumerable. Instead, the features that describe individu...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Original Title Page
• Original Copyright Page
• Dedication
• Table of Contents
• Preface
• Part I: Introduction
• Part II: Retrieval Strategies
• Part III: Maintaining Memory Organization
• Part IV: State of the Program, State of the Art
• References
• Author Index
• Subject Index