Semantic Memory

Semantic Memory

Semantic / Episodic Distinction (Tulving, 1983)

Semantic Memory: Memory for factual information (Centipedes have 100 legs)

Episodic Memory: Memory for events (Remembering your first kiss)

Procedural Memory: Memory for procedures (How to tie your shoe)

Propositional Networks

Proposition: Smallest unit of meaning that can have a truth value. Propositions are not made up of words in any language. Rather they are represented in a kind of abstract metnal language. Propositions are made up of one predicate and one or more arguments.
Arguments: Noun-like things, or other propositions

Predicates: These are things that say things (or predicate things) about arguments.

Attributes: Kind of like adjectives. One place predicates.

Relations: Says how two or more arguments are related. Multiple place predicates.

Propositional Network: Groups of propositions that are connected together in complex ways in memory. Many models of semantic memory propose that semantic memory is simply a propositional network.

Argument Overlap: One common view is that propositions are connected to each other in memory, when they share arguments.

Quillan's Hierachical Network Model

Premises:

Semantic Memory is a propositional network
Arguments are represented as 'nodes" in this network
Relations are "links"
When verifying a statement like "John is in Arkansas" activation spreads from the John node and from the Arkansas node until the two meet up.
Strong Cognitive Economy: Information is stored at the most general level possible and is not redundantly stored.

Evidence for:

Lachman, Lachman & Butterfield (1979): Compare verification time for statements that varied in terms of the number of links needed.

Canary can sing (1 link)
Canary has wings (2 links)
Canary can breathe (3 links)

`Collins & Quillian (1970): Subjects given back to back questions. When questions answered from the same level of the network speed of answering the second question is faster.

Different Levels: Are canaries yellow? Do canaries breathe? SLOW RESPONSE TIME
Same Level: Do canaries eat? Do canaries breathe? FAST RESPONSE TIME

Evidence against:

Rips, Shoben & Smith (1973): Compare verification time for statements that did not vary in terms of number of links needed. Found consistent differences in verification times.

Robin is a bird
Turkey is a bird

Reverse Category Size Effect:

Category Size Effect refers to the general finding that people are slower to identify members of larger categories than of smaller categories.
Reverse Category Size Effect is the opposite finding. For instance, people identify a chicken as an animal more quickly than they identify it as a bird.

Spreading Activation Model Collins and Loftus (1973)

Model is like the Quillian model but tries to deal with typicality effects. The length of each link is a function of the relatedness of the concepts. The model also embraces weak cognitive economy meaning that there can be some redundant information.

Feature Comparison Model (Rips, Shoben & Smith, 1973)

According to the feature comparison model decisions such as "Is a canary a bird?" are not based on activating a semantic network but by looking at the features of the item and comparing them to the features that characterize the category.

Two kinds of features:

Defining Features: All members of the category have this feature

Characteristic Features: Features that members of the category typically have

How a question like "Is a canary a bird?" is answered. Step 1: Look at all the features (both defining and characteristic). If canary has enough features of bird then say "YES" right away. If canary has hardly any features of bird say "NO" right away. If its somewhere in the middle, then go on to a second stage

Step 2: Look ONLY at the defining features. If it has the defining features say "YES"

How this explains the results?

Canary has all the defining features of bird, plus it has a lot of characteristic features. So when asked, "Is a canary a bird" people say "YES" right away in step 1 and never have to go through step 2
Penguin has all the defining features of bird but not many of its characteristic features. So when asked "Is a penguin a bird" in step 1 they aren't sure. It doesn't have enough features to say "YES" right away. It doesn't have so few features that they say "NO" right away. So they go on to step 2 and look only at the defining features. Doing so convinces them that penguins are birds, but it takes longer because people have to go through two steps instead of only one.

ACT-R (Anderson, 1993)

Like the spreading activation model in many ways. Currently the most precise and well worked out Propositional Network models of semantic memory.
Speed of activation depends on how strong the link is compared to other competing links. Each link is competing for the activation. This can produce a Fan Effect. The more links the longer it takes to verify any one proposition.
Anderson (1976): FAN EFFECT

A hippie is in the park

A hippie is in the church

A hippie is in the bank

A captain is in the park

A captain is in the church

A debutante is in the bank

Etc.

Did you see the following sentences?

A hippie is in the park? --Slow

A captain is in the park? --Fast

ACT also makes use of a production system. A production system is simply a set of IF THEN rules that are formally called productions.

These lecture notes were written by Jim Lampinen at the University of Arkansas. Feel free to link to them but please do not copy these notes in whole or in part for use in your own course without my permission.