Integrating knowledge and memory

Integrating knowledge and memory

Integrating new knowledge in traditional data bases is accomplished by simply adding an item to its key location, deleting an item from a key directed location, or modifying fields of an existing item with specific input information.

When an item in inventory is replaced with a new one, its description is changed accordingly. When an item is added to memory, its index is computed and it is stored at the corresponding address

More sophisticated memory systems will continuously monitor a knowledge base and make inferred changes as appropriate

A more comprehensive management system will perform other functions as well, including the formation of new conceptual structures, the computation and association of casual linkages between related concepts, generalization of items having common features and the formation of specialized conceptual categories and specialization of concepts that have been over generalized

Hypertext

Hypertext systems are examples of information organized through associative links, like associative networks

These systems are interactive window systems connected to a database through associative links

Unlike normal text which is read in linear fashion, hypertext can be browsed in a nonlinear way by moving through a network of information nodes which are linked bidirectionally through associative

Users of hypertext systems can wander through the database scanning text and graphics, creating new information nodes and linkages or modify existing ones

This approach to documentation use is said to more closely match the cognitive process

It provides a new approach to information access and organization for authors, researchers and other users of large bodies of information

Memory organization system

HAM, a model of memory

One of the earliest computer models of memory was the Human Associative memory (HAM) system developed by John Anderson and Gordon Bower

This memory is organized as a network of prepositional binary trees

An example of a simple tree which represents the statement “In a park s hippie touched a debutante” is illustrated in Fig 9.5

When an informant asserts this statement to HAM, the system parses the sentence and builds a binary tree representation

Node in the tree are assigned unique numbers, while links are labeled with the

following functions:

C: context for tree fact        P: predicate

e: set membership     R: relation

F: a fact  S: subject

L: a location    T: time

object

As HAM is informed of new sentences, they are parsed and formed into new tree-like memory structures or integrated with existing ones

For example, to add the fact that the hippie was tall, the following subtree is attached to the tree structure of Fig below by merging the common node hippie (node 3) into a single node

When HAM is posed with a query, it is formed into a tree structure called a probe. This structure is then matched against existing memory structures for the best match

The structure with the closest match is used to formulate an answer to the query

Matching is accomplished by first locating the leaf nodes in memory that match leaf nodes in the probe

The corresponding links are then checked to see if they have the same labels and in the same order

The search process is constrained by searching only node groups that have the same relation links, based on recency of usage

The search is not exhaustive and nodes accessed infrequently may be forgotten

Access to nodes in HAM is accomplished through word indexing in LISP

Memory Organization with E-MOPs

One system was developed by Janet Kolodner to study problems associated with the retrieval and organization of reconstructive memory, called CYRUS (Computerized Yale Retrieval and Updating System) stores episodes from the lives of former secretaries of state Cyrus Vance and Edmund Muskie

The episodes are indexed and stored in long term memory for subsequent use in answering queries posed in English

The basic memory model in CYRUS is a network consisting of Episodic Memory Organization Packets (E-MOPs)

Each such E-MOP is a frame-like node structure which contains conceptual information related to different categories of episodic events

E-MOP are indexed in memory by one or more distinguishing features. For example, there are basic E-MOPs for diplomatic meetings with foreign dignitaries, specialized political conferences, traveling, state dinners as well as other basic events related to diplomatic state functions

This diplomatic meeting E-MOP called $MEET, contains information which is common to all diplomatic meeting events

The common information which characterizes such as E-MOP is called its content

For example, $MEET might contain the following information:

A second type of information contained in E-MOPs are the indices which index either individual episodes or other E-MOPs which have become specializations of their parent E-MOPs

A typical $MEET E-MOP which has indices to two particular event meetings EV1 and EV2, is illustrated in Fig 9.6

For example, one of the meetings indexed was between Vance and Gromyko of the USSR in which they discussed SALT. This is labeled as event EV1 in the figure. The second meeting was between Vance and Begin of Israel in which they discussed Arab-Israeli peace. This is labeled as event EV2

Note that each of these events can be accessed through more than one feature (index). For example, EV1 can be located from the $MEET event through a topic value of “Arab-Israel peace,” through a participants’ nationality value of “Israel,” through a participants’ occupation value of “head of state,” and so on

As new diplomatic meetings are entered into the system, they are either integrated with the $MEET E-MOP as a separately indexed event or merged with another event to form a new specialized meeting E-MOP.

When several events belonging to the same MOP category are entered, common event features are used to generalize the E-MOP. This information is collected in the frame contents. Specialization may also be required when over-generalization has occurred. Thus, memory is continually being reorganized as new facts are entered.

This process prevents the addition of excessive memory entries and much redundancy which would result if every event entered resulted in the addition of a separate event

Reorganization can also cause forgetting, since originally assigned indices may be changed when new structures are formed

When this occurs, an item cannot be located, so the system attempts to derive new indices from the context and through other indices by reconstructing related events

The key issues in this type of the organizations are:

The selection and computation of good indices for new events so that similar events can be located in memory for new event integration

Monitoring and reorganization of memory to accommodate new events as they occur

Access of the correct event information when provided clues for retrieval