Memory 4: LTM and implicit learning Flashcards
(40 cards)
What is LTM an outcome of and what does it store?
- Outcome of explicit learning
- Storage of unlimited amounts of semantic information (in contrast to phonological or visuospatial) over long periods of time (minutes to years).
What is explicit memory and what two types of memory does it consist of?
- Explicit memory (declarative) - explicitly acquired material - we are aware of what we have learned and when we have learned it
- Explicit memory consists of episodic and semantic memory (Tulving, 1972)
What is episodic memory?
Specific personal events - remembering - most often affected in amnesia
Allows you to re-experience personal events (time travel) - Tulving
What is semantic memory?
Facts/information about the world - knowing (meaning, societal rules, sensory knowledge) but can be acquired on a single occasion - can be partially spared in amnesia - general knowledge
How have separate stores for semantic and episodic memory been found with neuropsychological double dissociations?
- Patients with brain lesions of the hippocampal area show impaired episodic memory, but only modest problems with semantic memory (Spiers et al., 2001)
- Patients with semantic dementia (loss of concept knowledge due to anterior lesions of fronto-temporal lobes) have unimpaired episodic memory (Adlam et al., 2009)
What evidence supports that while seperate, episodic and semantic memory stores may be interdependent?
- Retrieval from episodic and semantic memory activates the same neuronal networks, including frontal, temporal, and parietal brain areas (Burianova et al., 2010).
- Recall of remote episodic events requires semantic knowledge as retrieval framework (Irish et al., 2011): Patients with semantic dementia have unimpaired recall of recent episodic/autobiographical memory, but impaired recall for remote events.
- Is semantic memory the residue of many episodes?
- They seem to interact
Frederick Bartlett (1932) criticised Ebbinghaus’ approach to investigate memory on the basis of meaningless material (nonsense syllables) - Learning meaningless material only looks at repetition habits, not at how memory works in everyday life.
- Must be measured with material that has meaning to us
- A naturalistic approach must be taken.
- Basic assumption: LTM is based on meaning.
What did he do in his ‘Effort after meaning’ experiment and what were the results?
Material: Complex folk tales from unfamiliar cultures.
Task: Free recall of story events.
Results: Remembered stories are shorter, more coherent, and fit more closely with the participants viewpoint than the original story.
Explanation: (Culturally and socially developed) schemas (long-term representations of knowledge) are used to make sense of (extract meaning) and integrate (store) new material.
e.g. European ppts omit supernatural aspects of Indian folk tales
Supports that meaning may facilitate episodic long term learning
What is Craik and Lockhart’s levels of processing hypothesis?
Information is encoded and processed at varying depths.
Each process leaves a memory record, but deeper processes leave a more durable memory trace.
- The deeper we process information, the longer it stays
- e.g. printed word -> visual characteristics -> spoken sound -> meaning (increasing processing depth)
Craik and Tulving (1975) investigated the Levels of Processing hypothesis:
Material: Processing of visually presented words. e.g. fish
3 task conditions (yes, no answers) (increasing in processing depth):
Case: Visual judgement e.g. is it written in uppercase letters?
Rhyme: Phonological judgement e.g. does it rhyme with dish?
Sentence: Semantic judgement e.g. does it fit into the sentence: ____ live in the sea?
Recall: Unexpected recognition (50% new, 50% seen words).
What was recall like for each condition?
- Recognition performance is best for words that are processed in a semantic context and worst for words that are processed in a perceptual context, both for yes and no answers in each learning context
- Recall better for deeper processes - supports hypothesis about durable memory trace
- However, words given a positive response are associated with higher recognition performance, but only for ‘deeper’ processing conditions (not in the case condition).
- Recognition (memory) improves as a function of processing depth, i.e., more elaborate processing leads to better memory.
- Deep processing (sentence > rhyme > case)
- Meaningful processing (yes > no) - better at recognition when answer to the word was yes - correct semantic background leads to more meaningful processing
What are issues with measuring depth of processing?
It cannot be time spent processing (assumed serial processing - but processing visual, phonological, semantic features does not happen in a serial manner), as different stimulus features may be processed in parallel (Craik & Tulving, 1975).
Time not a good measure
Does deeper processing always lead to better performance?
No
Processing has to be transfer appropriate
e.g., factual knowledge about how to ride a bike is not the skill of riding a bike - deep semantic processing of bike riding techniques would not help you to ride a bike
e.g., best immediate recall performance for ‘shallow’ learning condition (SSSS; Roediger & Karpicke, 2006).
Depending on when information is recalled, shallow or deeper processing can each be advantageous
What has to match in order for a test to reveal prior learning?
For a test to reveal prior learning, the processing requirements of the test have to match the processing conditions at encoding (test bias).
If the test asks for semantic and meaningful processing (e.g., delayed free recall), semantic learning leads to the best performance.
Deep semantic learning appropriate for semantic recall and vice versa
What is Fisher and Craik’s (1977) Transfer appropriate processing experiment?
2 types of word-pairs (the capitalised word has to be learned): Rhyme (CAT – bat) vs. associate (CAT – dog).
i.e., 2 learning conditions: Rhyme vs. associative context.
Cued recall: Rhyme vs. associative retrieval cues.
Did matched or unmatched learning and recall conditions do better? (Fisher and Craik, 1977)
Matched
- People learnt best when encoding and retrieval context was the same If words are learned in an associative (semantic) context, recall is facilitated by associative cues. If words are learned in a (shallow) rhyme context, recall is facilitated by rhyme cues. - Meaning that retrieval performance is modulated by encoding context - Overall advantage for deep coding Semantic (meaningful) coding is advantageous i.e., leads to richer and more elaborate memory code which is more readily retrievable. This is in line with the levels of processing hypothesis, but additionally, memory performance in a particular test, depends on the match between encoding and test mode, which is in line with the idea of transfer-appropriate processing.
What is maintenance rehearsal vs elaborate rehearsal?
Which one enhances LTM?
Maintenance rehearsal = continued processing of an item at one level e.g., the code for the loo at Starbucks.
Elaborate rehearsal = linking the rehearsed item with other material in memory e.g., your debit card pin code.
Only elaborate rehearsal enhances long-term learning.
What are the building blocks of LTM?
Semantic memories - semantic coding is advantageous
What is the hierarchical network model? (Collins and Quillian, 1969)
Semantic memory = many hierarchical networks.
Nodes = major concepts
Associations = properties and features
Basic rule = feature information is stored as high up as possible to minimise storage space needed
e.g. ‘has skin’ is in the animal node because all animals have it - would make no sense to have this basic feature under every type of animal
How did Collins and Quillian (1969) provide support for their hierarchical network model?
Task: Decide whether a sentence is true of false
3 conditions:
Same level link - a canary is a canary and a canary can sing
One level link - a canary is a bird and a canary can fly
Two level link - a canary is an animal and a canary has skin
Reaction times are faster for sentences with links closer together in the model hierarchy, both on the concept and properties (features) level
This means Information is stored hierarchically, with faster connections between closer hierarchy levels.
What was Conrad’s (1972) critique of the hierarchical network model?
When sentences controlled for familiarity, hierarchical distance has little effect on reaction times
e.g. a canary can sing is a familiar sentence, a canary has skin is an unfamiliar sentence
Hierarchical distance and familiarity are confounded
What was Rosch’s (1973) critique of the hierarchical network model?
Typical category members possess more concept characteristics than less typical ones - typicality effect
e.g. a canary is a bird and an ostrich is a bird are both 1 level links, however RT is longer for ostrich as it is untypical
The model predictions contradict an observable effect
What are criticisms of the hierarchical network model?
- Too inflexible
- Confounded by familiarity
- Fails to explain observable typicality effects
However, served as a basis for the spreading activation model
What is the spreading activation model? (Collins and Loftus, 1975)
- Semantic memory must be organised on semantic relatedness
- Measuring semantic relatedness:
- Semantic rating of word pairs - cat and dog more closely related than cat and salad
- Creating category member lists (members produced most often must be more closely related to a category) - cat and dog more closely related to animal than axolotl
What is the structure of the spreading activation model?
- Concepts
- Links - length of the links indicates the degree of semantic relatedness
- Basic rule: If a concept is activated, the activation spreads most strongly to semantically closely related concepts and less strongly to semantically more distant concepts.
Meyer & Schvaneveldt (1976) conducted an experiment supporting the spreading activation model:
- Material - strings of letters
- Task - decide whether letters form a word
2 conditions: subsequent words are semantically related e.g. N U R S E D O C T O R, or subsequent words are semantically unrelated e.g. F I S H D O C T O R
What were the results?
- Reaction times to semantically related words are faster compared to unrelated words
- Semantic priming facilitates word recognition by pre-activating semantically related concepts
- Memory activation is faster for semantically (more) related than semantically unrelated (less related) concepts
