Psy 416: Reasoning and
Problem Solving
Notes
for Chapter 9 Semantic Memory
Go
to Chapter 8 notes
-
Let us consider the words canary, robin, penguin,
and piano. As speakers of English we know many things about canary.
We know that it refers to a kind of bird. We know that the thing it refers
to is often yellow, has wings, has a heart, and is warm-blooded. We know
that canaries fly, sing, and lay eggs. We know that a robin is a kind of
bird, is warm-blooded, has a red breast, flies, has a heart, eats worms
and lays eggs. We know that a penguin is a kind of bird, is warm-blooded,
swims, eats fish, and lays eggs. We know that the kind of object piano
refers to, is quite different from all of these.
-
How do we represent the knowledge that we all have about
words? There are many redundancies, and we know of many relationships among
them. Does that mean that words with similar meanings are in some sense
share parts of their meaning representation? By what process or algorithm
do we identify appropriate relationships? In order to demonstrate our knowledge,
either directly by answering questions about it, or indirectly, by using
the meaningful information appropriately, we must have access to and be
able to operate on the right information.
-
#1 and #2 lead us to consider how lexical information is
structured in memory and what the processes are that we use to interrogate
that memory. We certainly feel that some of the things that we know about
these concepts we did not learn directly, but we could figure it out. From
a reasoning or problem solving, or computational perspective, we have to
be able to identify some procedure (recursive process) which can effectively
do the task.
-
Lexical items are of various sorts. The parts of speech in
grammar help somewhat, but does not designate all of the differences that
seem to exist. Content words such as nouns, verbs, and adjectives function
differently from function words like prepositions, pronouns, and conjunctions.
Each of these categories have words which function differently from one
another. I am particularly interested in the fact that some nouns are Basic
Level words. These seem to have a special status, probably due to their
means of acquisition.
-
Many researchers use some variant of a semantic network to
represent the lexicon in memory. Basically this model has the lexical entries
connected to one another by labeled links or edges. These links identify
the kind of relation that exists between the items connected with one another.
-
Answers are derived by traversing the network to find them.
Some inferences that may be made are defeasible, which allows for normal
or usual inferences to proceed. This works by including the concept of
default reasoning which can be overruled by specific information.
-
SnePS a semantic network system developed in Buffalo by S.
Shapiro and his colleagues minimizes the kinds of links by making the links
themselves accessible to the processor.
-
Some issues:
-
Cognitive economy versus processing efficiency. Performance
in solving problems using semantic networks is a function of both the structured
relations among the items and the processing mechanisms that use the data
in the network.
-
What kinds of links are there?
-
Why do the same arcs have different response latencies?
-
Why do some links, which logically should be closer, take
more time than those further away?
-
Some tasks do not seem to fit the structured models that
work for other tasks.
-
There are nuances or different meanings of words which are
easily understood by people, but which are very difficult to include in
the models.
-
Where do these networks come from? What is the process or
mechanism of concept acquisition that would construct such networks?
Schema theory--Notes
for Chapter 8
The lexicon seems to relate to what has been called lexical
or conceptual meaning. This gives what is usually called the meaning of
the words. When, however, words are used to talk about stuff or to refer
to stuff we seem to enter a different ballgame. Certain phrases often refer
to things other than themselves, and sentences can be used to make assertions,
ask questions, request activities, promise something, create a mood, tell
stories, etc. We will not deal with many of these issues in this class,
but one aspect of the more global properties of language will be used,
either as a metaphor, or directly, for other reasons. Much language is
contextually interpreted. That is, it relates to some integrated and (usually)
coherent domain, and only makes sense within that domain. Our current term
for this domain to which the discourse relates is schema.
Some properties of schemata or issues relating to schema
theory
1. Conceptual structures that help us understand, interpret,
and remember incoming information
2. Related to Gestalt theory in that one develops, structures
(and restructures) the information.
3. Sometimes called constructivist theory because rather
than simply reproduce the information that is received the information
is restructured into something that the agent can call his own. Information
is often added, subtracted, ignored, or transformed depending on how the
agent views the schema and the relation of new information to the schema.
4. Schema theory is useful for reasoning, categorization,
story interpretation, evaluation, inferences, and much more.
5. We have problems agreeing as to the properties of
schemata. Bartlett--schemata, Minsky--frames; Schank and Abelson--scripts;
6. There are lots of data that have been interpreted
to support a schema analysis.
-
Bartlett: Schemas: War of the ghosts
-
Dooling and Lachman: Columbus story: Understand sentences
better when they are accompanied with a contextual frame
-
Bransford et al; and Johnson: washing clothes, serenading,
sentential images, serenade,
-
context sensitivity for interpretations
-
Bransford and Franks: larger units; One doesn't recognize
particular sentences, but compares them to "full schema"
-
Other Bransford: remember particular sentences that fit into
the frame
-
Beth Loftus: remembering as a function of the questions asked
or the context. Childhood sexual abuse repressed and then remembered
-
Greenspan and Segal: Using linear arrays; Understanding input
as a function of the context
-
Story grammars--Rumelhart, Mandler and Johnson
-
Fillmore; B. Tversky; Segal et al: Contextualization, perspective
taking and deixis.
-
Pragmatic errors
-
Barwise and Perry: situation semantics
-
Bruner: Concepts of sharpening and leveling, modification,
bias in memory, errors of memory, intrusions, reversals, etc. --
-
Lawson: pictures and sentences
-
SNePS and CASSIE: evaluating sentences in context
Go to Chapter 9 notes