Week 5 - Memory

Question 1

Why do we need memory?

Answer 1

• Food – remember where you put it
• Avoiding danger
• Procreation?

Question 2

Define STM and LTM

Answer 2

• Short-term memory (STM):
  • Information currently held in memory
  • Limited capacity
  • Typically conscious access
  • Fades without refreshing
• Long-term memory (LTM):
  • Stored information
  • Essentially unlimited capacity
  • Works without current or any conscious access
  • Does not necessarily fade

Question 3

Outline the Multi-store Model of Memory (Atkinson & Shiffrin, 1968)

Answer 3

• Serial information processing model (computer inspired)
• Three stores (sensory, short-term, long-term memory)
• Each store is unitary (not modality-specific)
• Unrehearsed information is lost

Question 4

Outline Visual Sensory Memory (iconic memory)

Answer 4

George Sperling‘s whole report (Sperling, 1967):

• Present letters for 50 milliseconds
• Task – report as many letters as possible
• Results - typical number of items that can be reported: 4.5 (=37.5%)
• But participants had impression they initially saw more items than they could report

Question 5

Outline the partial report paradigm

Answer 5

Variation of the task - partial report paradigm:

• Same stimuli but now you only need to report a subset of the letters based on the cue that follows.
  
  • Report letters from one row only
• Results - typical number of items that can be reported: 3.3 out of 4 (=82%)
• Higher capacity than estimated with whole report task

Question 6

How fast does the iconic memory decay?

Answer 6

• When delay between letters and cue is systematically varied - performance drops to whole report performance within a second
• Result - almost all the information is available right after presentation, but very fast decay

Question 7

Outline Auditory Sensory Memory (echoic memory)

Answer 7

Auditory analogue of Sperling‘s partial report (Darwin, Turvey & Crowder (1972):

• Three groups of letters, one on left, one on right, one on both ears
• Cue - report letters from left, right or both ears
• Results - similar to iconic memory, but decay is slower

Question 8

How can impairments of echoic memory be measured?

Answer 8

• With an EEG component, the Mismatch Negativity (MMN). The MMN is an auditory ERP component:
  
  • Peaks between 160-220 ms
  • Reflects comparison of short-lived memory trace with current stimulus – shows automatic processing which can be useful for speech processing.
  • Subjects do not require to attend stimuli for an MMN to be elicited -> automatic and passive
  • Sounds presented to participant with one sound occasionally different from the rest (higher tone), and measure the EEG signal, negativity occurs for the deviant tone

Question 9

How is echoic memory impaired in Alzheimer patients? (Pekkonen et al., 1994)

Answer 9

• Interested in finding out why Alzheimer patients’ memory is impaired.
• Varied time interval between the tones – 1 sec or 3 sec.
• The idea was, if echoic memory has already faded at 3 sec interval you shouldn’t see a MMN because there is nothing to compare the new stimulus against as it has already faded.
• For healthy control group, MMN was observed even at 3 secs intervals → Echoic memory still intact
• For Alzheimer patients, MMN was only observed at 1 sec intervals, but not for 3 secs intervals → Echoic memory decays between 1 and 3 secs

Question 10

The MMM claims there is a distinction between STM and LTM, what is the evidence for this?

Answer 10

1. Primacy and recency effect in STM (Murdock, 1962):

• Task - learn a list of words and recall
• Free recall (order doesn‘t matter)
• Results:
  
  • Primacy effect - first words remembered better than middle words - words in LTM already
  • Recency effect - last items were remembered better - words still in STM
  • Middle - too long to still be in STM, too short to already be in LTM

Murdock noticed participants were moving their lips when trying to remember words. Could rehearsal be to blame?

1. The role of rehearsal in STM (Glanzer & Cunitz, 1966):

• Task - similar to Murdock, learning and free recall
• Conditions - immediate recall or delayed recall (after 10 or 30 secs counting backwards verbally)
• Results:
  
  • Primacy effect was unaffected (because part of LTM)
  • Recency effect varied as a function of delay
  • Rehearsal only affects recency effect (because STM relies on rehearsal)
  • Rehearsal important for STM. STM and LTM seem to be separate.

1. Patient H.M. (Scoville & Milner, 1957):

• Had severe epilepsy, hippocampus removed
• Resulted in anterograde amnesia – LTM affected, couldn’t acquire new LTM but STM, intelligence, language intact
• Supports notion of STM and LTM being independent
• Patient H.M. showed residual learning abilities (e.g., intact ability to learn a mirror-tracing task).
• No ability to acquire new memories of facts and events (e.g., no recall of the training for the mirror-tracing task).

Question 11

What are the challenges for the MMM?

Answer 11

1. Evidence that STM is not just passive store
   * Levels of pressing effects (Craik & Lockhart, 1972; Craik & Tulving, 1975):
   • Task - list of 60 words. Each word came with question:
   • visual quality: Is the word in uppercase?
   • phonemic quality: Does the word rhyme with “bar“?
   • semantic quality: Does the word fit in this sentence?
     * Then you do a surprise test and see how many of the words they remember
     * Results - with equal exposure duration, the deeper the processing (semantic judgment), the better the memory performance
     * Speaks against multi-store model’s assumption that LTM quality merely depends on length of rehearsal – also affected by deeper processing
2. LTM can be intact with impaired STM

• Patient K.F. (Shallice & Warrington, 1970):
  
  • patient with motorcycle accident
  • Impaired STM (telephone number) but normal LTM - how can information be transferred to LTM without STM?

1. STM seems to have subsystems

• Patient K.F:
  
  • Visual STM intact, only verbal STM had problems - two subsystems?
• Proactive interference in STM (Wickens et al., 1976):
  
  • Retroactive interference - input of new stuff makes representation of old stuff worse
  • Proactive interference - old stuff interferes with learning new stuff
  • Task - learn these words (always three at once)
  • Same category in the first three trials
  • Then 1) same category 2) related category 3) different category
• Results:
  
  • Results in first trial similar
  • Drop in performance for trials 2 and 3 – performance gets worse if same category in previous trial due to proactive interference
  • Trial 4 – no/less proactive interference when category switches
  • Suggests that STM is not unitary, may be semantically sorted

Question 12

The MMM spawned what new model?

Answer 12

Working Memory Model (WMM) (Baddeley & Hitch, 1974)

Question 13

What does the WMM assume?

Answer 13

• Central executive - supervisory system; controls flow of information; attentional system without store. The most important component is the central executive. It has limited capacity, resembles attention and deals with any cognitively demanding task. The phonological loop and the visuo-spatial sketchpad are slave systems used by the central executive for specific purposes.
• Two subsystems - visuo-spatial sketchpad and phonological loop for short-term storage
• Episodic buffer - limited capacity passive system for integration of information between visuo-spatial sketchpad and phonological information

Two assumptions of the model:

1. If two tasks use the same component, they cannot be performed successfully together.
2. If two tasks use different components, it should be possible to perform them as well together as separately.

• Rather than seeing STM as a passive store fed by active processes putting information in or taking information out, they thought of STM as fundamentally active and dynamic – seen by all the processes involved.

Question 14

What evidence is there for the WMM subsytems? and outline fMRI data

Answer 14

1. Dual task method (Hitch & Baddeley, 1976):

• Assumptions:
  
  • WMM has limited capacity, gradually filled up
  • If WMM is filled due to a task, fewer resources available in second task
• Dual Tasks:
  
  • Digit span task (0-8 digits) – remember digits
  • Reasoning task that requires verbalisation
• Results:
  
  • Some impairment, but not very much
  • Reasoning task takes longer if you have to remember more digits – suggests there is some interaction, but impairment is not very much (error rate doesn’t change)
  • Central executive – believed this was doing the reasoning task
  • Phonological loop – believed this was doing the digit task

1. Other dual task method (Duncan, Martens & Ward, 1997):

• Participants were looking at word streams and had to detect target words.
• Two visual streams – horizontal or vertical pair (either attending both streams or one - two words or four)
• At the same time, had to do an auditory task:
  
  • Listening to words and had to respond to target words
  • High and low voice (two auditory streams)
• Results:
  
  • If you have to monitor a visual and auditory channel, you are at the same advantage as if you have to monitor just one channel.
  • But if they are within the same modality, e.g. two visual streams or two auditory streams, then your performance is impaired. Because you have to separate your resources within the one modality you have to do two streams. Whereas before, you can use your full auditory resources to do the auditory task and the full visual resources for the visual task.

fMRI data - how much do brain regions overlap than represent single categories?

• The more they overlap, the worse your performance.
• Mixed-benefit was strongly negatively correlated with overlap in occipitotemporal cortex – more resources available.
• Interpretation - the larger the neural population at hand, the more efficient processing

Question 15

What evidence is there for the phonological loop?

Answer 15

1. Most people can remember seven numbers/digits (Miller, 1956):
   
   • A variety of tasks indicate that we can keep 7 ± 2 items in mind (e.g. digit span task, judgment task, subitising)

New - Idea that STM has slots

What can be stored in a slot?

• Cowan (2001): Chunking, supported by LTM; real capacity closer to 4
• Limitations of the phonological loop – does not just depend on the number of items, but the certain qualities of a word:
  • Similarity of words affects span of phonological loop
  • Similarity of sound
  • Length of sound relevant?
• The more syllables a word has, the fewer words can be remembered - phonological loop is not about the number of items, but the length of an auditory signal
  
  • Phonological loop is correlated with time it takes to say words out loud
  • Time to say words counts (span = 2 secs)

Question 16

What evidence is there for the central executive?

Answer 16

1. Executive WM distinct from WM maintenance (Owen, Evans, & Petrides, 1996):

• Spatial monitoring task:
  
  • Trial1: Search through red circles until one turns blue
  • Trial2: Search again, but avoid previous target location
  • 8 Trials per round
  • Requires updating of WM
• Passive WM task:
  
  • Sequence is shown that needs to be reproduced
  • No manipulation of the stored sequence is required
• Results of PET scan revealed:
  
  • More activity in DLPFC for spatial monitoring task – involves more executive control
  • More activity in VLPFC for passive task
  • Indicates there are different processes involved – executive control more related to the DLPFC

Question 17

How can you measure the working memory capacity?

Answer 17

1. The change detection task - Phillips (1974); Luck and Vogel (1997):

• Measures visual working memory (WM)
• Stimuli are presented long enough to encode them into WM
• Retention interval follows
• Probe Display - has something changed?
• Idea - the better WM, the more likely a change can be detected because the item that has changed is in working memory
• Results (Luck and Vogel, 1977):
  
  • Subjects remembered ~3 items without problems
  • Verbal load (2 digits in mind) did not affect performance – shows there is a limit in visual working memory
  • Load – additional verbal load (phonological loop)

1. Different condition in the change detection task – conjunctions:

• Conjunction - detect change in either orientation or colour (8 features – 4 colours and 4 orientations)
• What will happen?
  
  • Will performance drop because each feature needs a slot?
  • Or will performance stay the same because the entire item is stored?
  • Conjunctions should be harder to remember if each feature is stored separately
• Results:
  • Participants either had to remember just orientation, just colour, or both
  • Similar WM performance for orientation and colour – suggests they were stored as whole objects
  • Conjunction - detect change in either orientation or colour (= up to 12 features in WM)
  • Performance for single features or conjunctions was identical

1. Did another experiment with more features – gap, size, orientation, colour

• Even with four features, no impairment in performance
• If you can store an object, you can store all features that come with it

Maybe we have different stores for different features? (e.g. one for orientation, one for colour, etc.)

• So did another experiment with features from same dimension:
  
  • Participants had to either attend the colour of the large squares or small squares
  • Even with features from same dimension, no impairment in performance
  • Conclusion - items stored in working memory as integrated objects, not individual features → Slot model is correct

Question 18

How can the capacity K be measured? (how many items can we hold in working memory?)

Answer 18

• If we are presented with 6 items and only remember 3, does that mean our WM is at 50%?
  • No because we could guess
• Single detection theory takes this into account. You can get a more precise K by using this formula:
  • Cowan’s K = N × (H – FA)
  • Assumption: WM has slots that can be filled
  • Set size (items presented): N
  • Capacity (number of slots): K

Question 19

What are the neural measures of WM?

Answer 19

The contralateral delay activity (CDA) (Vogel & Machizawa, 2004) (Vogel, McCollough, & Machizawa, 2005):

• Variation on the change detection task - lateralised change detection task
• Items presented bilaterally
• A cue indicated from which side items had to be encoded
• Items from the uncued side would never be probed in the end of a trial
• The manipulation to this task allows us to measure the lateralised EEG signal
  
  • Similar spatial setup to N2pc experiments (but note that items need to be encoded to WM here and that set size varies)

What does this mean?

• The CDA increases with the number of stimuli
• The CDA increases up to 3 items, and then it asymptotes – asymptotes right at the point of participants WM capacity (closely related)
• No difference between red and green line because you’ve reached your working memory capacity
• Subjects with higher working memory showmore CDA increase from 2 to 4 items

Question 20

Does suppression contribute to WM? (Feldmann-Wüstefeld & Vogel, 2018)

Answer 20

• Standard change detection task with distractors – remember squares, ignore circles
  • Either the targets or the distractors would be lateral
  • Top example would allow us to measure the CDA
  • Lower example would allow us to measure the PD as a measure of suppression
• Four conditions:

1. 2 distractors with targets lateral
2. 2 distractors with distractors lateral
3. 4 distractors with targets lateral
4. 4 distractors with distractors lateral
   * Results:
   
   • No significant difference with CDA
   • Difference in the PD component with different number of items – larger PD component for four distractors than two distractors, suggesting there is more suppression required for four items than two
   • Red line is more positive than blue line suggesting sustained positivity – CDAp
   • PD correlated with working memory capacity (K) suggesting those with a larger PD (more suppression) also have a large working memory – so suppression does contribute to working memory performance

Question 21

How do we represent items in working memory?

Answer 21

There is an ongoing debate whether WM capacity is represented as slots (discrete) (remember an item entirely or forget it entirely) or a resource (continuous) (might get an idea of some memories):

• Evidence for resource - when items change drastically, performance can be extremely good even for 8 items. This can be explained by 8 items being represented very vaguely - which is however enough to detect dramatic changes (Bays & Husain, 2008)
• Evidence against resources - in line with slot models, when some items are cued (indicating they would be more likely probed), there is no difference between neutral and invalid trials. Resource models would predict medium performance for neutral and bad performance for invalid trials (slot models: all or nothing). (Carlisle et al., 2011)

Question 22

How do we measure transfer from working memory to LTM?

Answer 22

• Can use CDA component to measure transfer from WM to LTM (Carlisle et al., 2011):
• Task - visual search task, target can change from trial to trials
• Memory component – targets had to remember what target they had to look for
• They were not interested in the search array results, but the CDA elicited from the target template that was kept in mind during the waiting period.
• Results:
  
  • If a target was presented for the first time, you would see a large CDA component.
  • In trials 3-4 where the target is repeated, the CDA shrinks slightly.
  • In trials 5-7, the CDA shrinks even more.
• The results suggest a transition from working memory to LTM:
  
  • CDA is a marker for working memory – it gets smaller but they can still do the task so must be somewhere else, indicating LTM.

Question 23

What are the different analysis’ for Working Memory Representations in the Brain

Answer 23

1. Univariate analysis - undifferentiated sum of activity:

• E.g. we present a stimulus and then measure the brain activity during the retention interval and compare this to the same brain area during stimulus presentation.
• Blind to more subtle differences in activity patterns (e.g., single voxels in fMRI or single electrodes in EEG)

1. Multivariate pattern analysis (MVPA) can identify a more fine-grained pattern of activity by weighting variables – more precise picture of how WM works in the brain

• MVPA involves training an algorithm to find a liner combination that is most active in one condition compared to another condition. Example:
  
  • Weighted sum = 3×A +B –C –2×D
  • During stimulus presentation: 3 × 6 + 2 – 0.5 – 2 × 0.5 = 18.5
  • During retention: 3 × 4 + 0.5 – (-1) – 2 × (-2) = 18.5
• Study example - Decoding feature and task type in frontal and sensory areas of the brain (Riggall & Postle, 2012):
  
  • Participants were presented with 2 tasks - either remember direction or remember speed of stimuli
    
    • 2 features - direction and speed
  • Results:
    
    • Total brain activity during retention interval reduced in occipital cortex (D) but not frontal cortex (A)
    • MVPA classifier could decode in occipital cortex (H) but not frontal (E) which feature was held in WM
    • MVPA classifier could recode in frontal cortex which task (keep speed or direction in mind) was done (I)

We can learn from this that the prefrontal cortex is responsible for the executive control of tasks – what task am I supposed to do. Sensory regions are important to reflect actual features:

• Occipital regions - sensory recruitment “actual content“
• Frontal regions - top-down processes, executive control
• WM = delayed activation of sensory regions, mediated by frontal region control

Question 24

Outline non-declarative (implicit) LTM

Answer 24

• Forms of long-term memory that influence behaviour but do not involve conscious recollection/you can’t voice what you know, only shown through behaviour. Knowledge that you have that changes the way you behave. Types of implicit memory:
  1. Procedural memory – memory that involves knowing how to perform certain actions/gradually acquired. Memory that enables highly skilled behaviours or actions such as touch typing or walking. As you acquire the skill, the information is very much conscious. Once acquired, the information is more implicit – unconscious. Putting conscious control on this process slows you down. Skill learning can generalise to numerous stimuli:
• Basal ganglia involved in motor learning
• Motor cortex involved in retention of movements
• Both coordinate by the cerebellum

1. Perceptual representation system

• Knowledge about the perceptual world
• How does an object look? (abstract perceptual definition)
• Perceptual learning (Schacter et al., 1990):
  
  • Participants looked at possible and non-possible objects
  • Found possible objects can be more represented in LTM

Question 25

Outline declarative (non-implicit) LTM

Answer 25

* Declarative Memory (explicit) – a form of long-term memory that involves knowing something is the case; it involves conscious recollection and includes memory for facts (semantic memory) and events (episodic memory)/refers to memories that can be declared/voiced. There are two types: 1. Semantic memory – our general knowledge of the world, e.g. grass is green, what grass is, what green is. It is shared amongst each other – we all know certain concepts which allows us to communicate with each other. Starts early in life. 2. Episodic memory – past memory for events/moments in your life. Not shared, unique to the individual. You might share the experience of an event, but YOUR perception of the event will be unique to you. Starts later in life. * Patients can learn a new language (semantic memory) with anterograde amnesia (Hirst et al., 1988) → Semantic memory and Episodic memory are distinct * Patient H.M.: * both semantic and episodic memory impaired * implicit memory was intact * structures within medial temporal lobe (e.g., hippocampus) important for declarative memory (but not for implicit memory)

Question 26

What is memory consolidation?

Answer 26

* Memory consolidation - it is a distinct process that serves to maintain, strengthen and modify memories that are already stored in the long-term memory

Question 27

What are the different brain regions involved in remembering?

Answer 27

* Task – remember these words * Three types of recall: 1. Recall - please say as many words from the list as you remember 2. Recognition - was lemon in the list? 3. Cued recognition - key\_\_\_\_\_ * Results: * Recognition is easier than recall Why is this the case? * There are different brain regions involved in these two tasks: 1. Perirhinal cortex is related to familiarity – this is context free * Involved in the recognition test and slightly in the cued recall 1. Parahippocampal cortex is related to recollection mechanism – context-dependent, specific information from episode: * Involved in recall test * Recall test does not receive information from the perirhinal cortex. Recognition test receives information from perirhinal cortex AND parahippocampal cortex

Question 28

What is easier forgotten?

Answer 28

* Information that is not semantically processed * levels-of-processing account * Information that is retrieved in different context * Contextual encoding account * Information is better retrieved in the place it was learnt * Information that we “want” to forget * Directed forgetting * Frontal lobe lesion stops directed forgetting from working * Example of context retrieval (Godden & Baddeley, 1975). Predicting what will be remembered: * fMRI signal in prefrontal cortex (a) and parahippocampal region (b) during encoding predicts whether word will be remembered or not (Wagner at al., 1998) * measured activity during learning * Lower activity during learning in words that are forgotten

Question 29

What is the standard model of memory consolidation?

Answer 29

* Standard model of memory consolidation: * Memory initially depends on the hippocampus, subsequently on the neocortex * Information is slowly transferred from the hippocampus to the cortex (over years) * Based on long-term potentiation. Found in rabbits by Lomo (1966) * Damage to hippocampus: * Memories from earlier life easier to recall than those in later life (Butters and Cermak, 1986) * The older the event, the more consolidated it is in the cortex and the less dependent on the hippocampus it is. * Dementia patients: * Sementic dementia (SD) patients have intact hippocampus but impaired anterior temporal lobe * SD patients remember relatively recent events * Memories still in hippocampus? * Problems with standard model: * Standard model does not make distinctions between episodic and semantic memories * fMRI studies show no difference in medial temporal lobe activity for old and new memories (Fink et al., 1996)

Question 30

Define the Multiple Trace Theory (Nadel & Moscovitch, 1997)

Answer 30

* Developed in response to the standard model of memory consolidation * Hippocampus’ support of LTM is not time-limited * Retrieved events create multiple traces that make forgetting less likely * Information is more and more schematic (entorhinal cortex) and less contextualized (hippocampus) * Consolidation is transforming, not transferring memories * Different regions of medial temporal lobe with distinct roles: * Hippocampus: concrete contextual memories * Entorhinal cortex: schematic, semantic memories