Words are labels generalized over instances of the "same" object. Problem: how do words for things arise? What are the categories they label? In short, what is the mental content for object-words?
Proposals: mental image or template (has all the problems of any template theory (see Hoffman on vision)); rule-based representation.
Solution: the meaning of an object-word is its access to the thing in the world (truth) via a visual representation. Responsibilities of semantic (meaning) theories under this picture: refer under varying conditions, account for words in combinations, account for relations between sentences, acquisition, and processing time.
If words stand for mental classes of objects, what are mental classes? Two broad approaches:
Definitional-- necessary and sufficient features
Prototype -- family resemblances, fuzzy membership and goodness of fit
Problems in all such proposals:
How do you determine features? Are they sensory? high-level visual?
How do you build a category out of the features? Knowing something as a thing is not the same as recognizing it as a class member.
Is there an even more general solution?
Armstrong, Gleitman and Gleitman argue that class membership is not the same as exemplariness. So definitional category inclusion, which is all-or-nothing, is not incompatible with goodness of fit and prototypes. They may operate at different levels or emerge in different kinds of tasks.
"Memory" is a cover term for both storage and access, neither of which is unitary. There is evidence for multiple subsystems of storage and different kinds of retrieval and interaction across subsystems.
The most influential recent theory of storage is Atkinson and Shiffrin's, based on the computer analogy: memory storage (hardware) and mental processes (software). More particularly, A&S proposed three stores: sensory, short-term and long-term. These are accessed and coordinated by a variety of information management procedures sensitive to instruction, stimuli, task, etc. This model, which remains influential in its general outline, accounts for some standard experimental memory effects: primacy and recency, for example. First in and last in are best recalled because the first in has been rehearsed in a clear buffer and sent to long-term store as the experiment proceeds, and the last in is still active in short-term when measurement is applied. This leads to a prediction: it should be possible to dissociate short-term and long-term by manipulating speed of stimuli and nature of tasks. Experiments support this, and so primacy and recency are accounted for by how they fare in the storage-system of memory.
Problem: how many storage systems are there?
Neuropsychological work suggests that multiple storage is the right idea, though it perhaps needs to be reinterpreted. Work on certain kinds of amnesia and other memory disorders suggests that short-term can be in tact while long-term is not and vice versa. For example, H.M., the famous amnesia case, was essentially unable to transfer new information into long-term storage (because of bilateral lesions to hippocampus and amygdala).
Still, an even closer look at patients' performance suggests something more subtle. H.M. could in fact learn if he stayed on task. When interrupted, he would then exhibit his extreme behavior. Cohen's explanation: H.M. had preserved procedural memory but disrupted declarative memory. This suggests that memory might be more accurately investigated as divided between a kind of operational store and an explicit data-structure store.