Short-Term Memory and Working Memory Lecture Notes

Short-term Memory and Working Memory Mind, Brain, & Behaviour 1 PSYC10003 Learning & Cognition Week 4 Lecture 1 Meredith McKague [email protected] In our first lecture on memory, I introduced the different components and processes of memory in relation to Atkinson and Shiffrin’s (1968) multi-...

Short-term Memory and Working Memory Mind, Brain, & Behaviour 1 PSYC10003 Learning & Cognition Week 4 Lecture 1 Meredith McKague [email protected] In our first lecture on memory, I introduced the different components and processes of memory in relation to Atkinson and Shiffrin’s (1968) multi-store model. We then unpacked the first box in the multi-store model and explored visual sensory memory – using Sperling’s classic studies to define the capacity and duration of the iconic trace. In this lecture, we focus on the second box from the multi-store model - short-term memory or working memory. In the first part of the lecture, we will look at short-term memory as it was originally conceived in the Atkinson and Shiffrin multi store model. We then discuss experiment findings that suggested STM should be thought of as a working-memory system. We discuss the evolution of Atkinson and Shiffrin’s original single box for STM into the multi-component working memory system proposed by Alan Baddeley. 0 Learning Outcomes Describe Atkinson and Shiffrin’s model of STM – including capacity, duration, and role of maintenance rehearsal in transfer to LTM. Describe the tasks and experimental findings used to estimate capacity and duration of short-term memory Explain how serial position effects provide evidence for separate STM and LTM stores, and for transfer between them. Explain how levels of processing”, challenged the original model of STM. Describe the components of Baddeley’s working memory model, including their proposed roles in cognition and neural basis. 1 1 What is STM? • Our conscious representation of ‘the present moment’. Information in STM comes from sensory registers and retrieval from LTM Encoding in STM relates information from sensory stores to long-term memory • A temporary store in which we integrate current sensory experience with long-term memory to achieve current goals • Capacity • Limited • Duration • 15-30 seconds Original model focused on maintenance rehearsal as the primary encoding mechanism, resulting in transfer to LTM 2 The information that comes into STM is selected from the current sensory input and from long-term memory. The content of STM is determined by our current attentional goals. STM enables the conscious representation of our current experience and goals. STM is used to make sense of our current experience by relating current sensory experience to knowledge from long-term memory. As the name implies, the key feature of STM is that its duration is rather brief, although significantly longer than sensory memory - being a matter of seconds, rather than milliseconds. The duration of memory traces in STM, and therefore their chances of being consolidated into long-term memory traces, depends on the extent to which we can maintain attention and engage in encoding processes whilst the information is in STM. Atkinson and Shiffrin originally proposed a mechanism they called maintenance rehearsal (or sometimes just rehearsal) to keep information active in STM. This was conceived of as an ‘inner voice’ (a language-based auditory code) that can be sued to mentally rehears information until it is transferred to LTM. This idea was strongly 2 challenged by the levels of processing experiments we will talk about in this lecture. The capacity of STM is limited to very few items that can be attended at once. However, the definition of what constitutes an ‘item’ in STM can be tricky, as we will see. We look at the question of capacity first. 2 Verbal STM capacity • Assessed using a digit-span task. • Immediate serial recall of verbally presented digits in the order they were presented • Systematically increase length of sequence to determine the “span”. • A participant’s span is reached when they fail on two trials at a given series length. • So, if you were unable to complete both trials without error at series length 8, then your span would be 7. • Average span is 7 (+/- 2) • “seven plus or minus two” items of information. 3 This estimate of STM capacity comes from a task we call the digit span task. In this task, a participant hears a sequence of random digits, spoken at the rate of one digit per second. The task is to repeat back the sequence in the order it was presented as soon as the series has finished being spoken. We call this immediate serial recall – “serial” because memory for the order in which the numbers were said is crucial. Verbal STM is thought to be important for encoding the correct sequence of information as it was presented – of course the order of sensory input is important especially if we think about processing of speech. The digit-span is calculated by the longest sequence you can recall correctly. So, if I got as far as recalling at least one of the two trials of 8 digits correctly but was unable to do either of the two sets of items at series 9, I would have a digit span of 8. 3 Duration of STM • The Brown-Peterson task • Remember 3 consonants “D-P-R” • Memory probed (tested) at 3-second retention intervals. • To prevent rehearsal participants were required to count backwards in 3’s until given a signal to stop • For example: Participants hear “D-P-R – 306” 4 The first attempts to measure the duration of STM were with a task called the Brown-Peterson task, named after the researchers who developed it. In this task you hear 3 random letter names spoken aloud, one per second - e.g, “D, P, R” - and then you hear a number spoken (say “306”). The number cues you to start counting backwards in 3s from 306 until told to stop and recall the letters you heard. The Brown Peterson task was designed to measure the decay of the STM trace over time by filling the retention interval with a task that prevented verbal rehearsal of the letter names. You can see from the graph that the ability to retrieve the names of the 3 letters in the order they were spoken drops off very quickly indeed. After just a 3 second interval only .5 (or 50%) of trials were likely to be recalled correctly with all three letters named in the correct order. By 9 seconds this has dropped to 20% and we are down to zero after 12-18 seconds. 4 Maintenance rehearsal and transfer to LTM • Verbal rehearsal keeps information active in STM and strengthens the trace to increase the chance it will be stored in LTM. • Evidence? 5 5 Serial position effects and transfer to LTM • Immediate free recall of lists of numbers or words is affected by the position of items in the list • Primacy effect provides evidence for transfer to longterm memory for items that receive more rehearsal • Recency effect reflects availability of information still in short-term memory • Recency effect is reduced by introducing a filled retention interval • Primacy effects are eliminated if rehearsal is prevented by introducing a concurrent task (repetition of a word) 6 The primacy effect reflects the finding that we are typically good at retrieving the first items presented. The recency effect is the finding that we are also very good at retrieving the last items. The items in between are harder to retrieve. This pattern of effects was argued to provide evidence for the existence of separate, LTM and STM stores, with transfer to LTM through rehearsal. 6 Towards working memory: Levels of Processing • The purpose of a STM is to encode information meaningfully. • Meaningful processing of information during encoding will produce long-term memory traces • Shallow processing is less effective for long-term retention. • Craik & Tulving (1975) test this hypothesis with their study of Levels of Processing. • Test the idea that LTM for words is influenced by the ‘depth’ (level) of the encoding process used in STM. 7 Pages 265-266 of textbook Chapter 7 on memory briefly covers this work. Participants in this study did not know they were participating in a memory experiment. They thought that the experimenters were interested in measuring their response times to respond to a series of questions about a set of 60 words. The experiment went like this. Participants saw a list of words presented one at a time on a computer screen. Before each word was presented, they heard one of three different kinds of questions (20 words allocated to each question kind). The three different kinds of questions were designed to engage three different ‘levels of processing’ when participants encoded the word that followed. The first kind of question was designed to engage participants in the ‘shallowest’ level of encoding (processing); it focused on the visual features of the word and asked “Is this word written in upper case?” - the word subsequently presented on the screen was written in either written in upper- or lower-case letters and the participant needed to respond either “yes’ or ‘no’ by pressing one of two designated response keys. The second kind of question required participants to focus on the sound of the word (decode it from print to sound) - the question asked “does this word rhyme with X?”, where X is some other word that did or did not rhyme with the word subsequently presented on the computer screen. Again, participants answered ‘yes’ or ‘no’ with a 7 button-press. This ‘rhyme’ condition requires slightly ‘deeper’ processing (encoding) than the ‘case’ task because it requires participants to read the word and mentally generate its sound, rather than focus only on the visual features of case. The third kind of question required the ‘deepest’ level of encoding because participants had to focus on the meaning of the word - This kind of question asked “Could this word complete the sentence “the X ate the grass”, where X was a word that either made sense in the sentence or did not (e.g., cow or car ). The 20 words in each question condition were matched to each other on factors that affect memory such as how long they are, how frequently they occur in print, and their part of speech (noun or verb, etc), and each set of 20 words occurred in each question condition across participants. So, the questions that preceded each of the words served to control the “level of processing” used to encode each of the words. After the question/answer encoding task, participants were given another task to do for 5 minutes. Then, they were given a surprise recognition memory test. The recognition memory task consisted of 180 words presented in a different random order to each participant. 60 of these words were those that had been in the question/answer task; 120 were other words that were of similar length and frequency and part of speech that had not been seen in the experiment. The task was to indicate for each word whether it was ‘old’ or ‘new’. An ‘old’ response was used to indicate that the word had been seen in the question/answer task. ‘New’ responses indicated that the word had not been seen previously in the question/answer task. The graph in the slide shows the results. Participants were much more accurate at recognising words that had been in the ‘sentence’ condition than they were at recognizing words from the ‘rhyme’ condition, and words in the ‘case’ condition were least well recognised. This study formed the basis for the idea that ‘deeper’ the encoding during study will lead to richer connections with existing long term memory representations which increases the likelihood of being able to retrieve the information later. The important implication of this work for the multi-store model was that levels of processing shifted memory researchers to think about STM as a system that supports learning and reasoning rather than focusing on mere rehearsal of information with the primary goal being immediate recall. This leads us very naturally on to the concept of a working memory system as a development from the original STM system of the multi-store model/ Image adapted from page 274 of Craik, F. I., & Tulving, E. (1975). Depth of processing and the retention of words in episodic memory. Journal of 7 experimental Psychology: general, 104(3), 268. 7 A shift from STM to Working Memory • Studies like Craik and Tulving’s suggested the need for a more detailed account of STM as multi-component system that supports meaningful encoding and active reasoning and problem-solving. • Rather than focusing on maintaining information for immediate recall, the focus shifts to thinking about STM as providing a mental work-space that helps us to achieve our current goals and update our understanding of the world. • Alan Baddeley introduces his model of Working Memory 8 8 The PL and VSS are independent but interacting sub-systems, one for visualspatial information and one for auditory-verbal information PL and VSS access and update languagebased and visual representation s in long-term memory Central executive directs attention to and retrieves information from PL and VSS for integration in the episodic buffer. Integrate, multi-modal memory traces formed in the episodic buffer and stored in episodic longterm memory 9 Baddeley’s multicomponent approach to working memory represented a development of earlier models of short-term memory, such as the Atkinson and Shiffrin model, but differed in two ways. First it abandoned the concept of a unitary short-term store in favour of a multi-component system, and second it emphasised the function of such a system in problem-solving and complex cognition, rather than in pure maintenance of memory. The initial three-component model of working memory proposed by Baddeley and Hitch consisted of an attentional controller, the central executive, aided by two subsidiary systems, the phonological loop, capable of holding speech-based information, and the visuospatial sketchpad, which performs a similar function for visual information. The episodic buffer (EB) was added in a later version of the model. It provides a storage system that binds together the inputs from the visual/spatial and auditory systems and integrates them in multimodal (i.e., multisensory) representation of the current contents of awareness. The buffer is ‘episodic’ in the sense that it holds integrated episodes or scenes, and it is a ‘buffer’ in the sense that it provides a temporary holding place to represent the current contents of consciousness. The EB allows multiple sources of information to be considered simultaneously, creating a model of the environment that may be manipulated to solve problems and plan future behavior. The buffer serves as a modelling space that is separate from LTM, but which forms an important stage in long-term episodic 9 learning. The need for the episodic buffer arose from the need to account for interactions between the visual and phonological STM buffers and LTM. The working memory components are represented in the clear boxes at the top of the figure. The blue box at the bottom represents the long-term memory systems. There’s many readable accounts of the development of Baddeley’s model of working memory. The references below are ones I have used in the preparation of this lecture. You are not required to read them. Baddeley, A. (2012). Working memory: Theories, models, and controversies. Annual review of psychology, 63, 1-29. https://www.annualreviews.org/doi/pdf/10.1146/annurev-psych-120710-100422 Baddeley, A. D. (2017). The concept of working memory: A view of its current state and probable future development. In Exploring working memory (pp. 99-106). Routledge. Baddeley, A. (2010). Working memory. Current biology, 20(4), R136-R140. https://www.sciencedirect.com/science/article/pii/S0960982209021332 Baddeley, A. D., & Logie, R. H. (1999). Working memory: The multiple-component model. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive co. ntrol (p. 28–61). Cambridge University Press. https://doi.org/10.1017/CBO9781139174909.005 https://psycnet.apa.org/record/1999-02490-001 Baddeley, A. (2003). Working memory and language: An overview. Journal of communication disorders, 36(3), 189-208. http://www.linguisticsnetwork.com/wp-content/uploads/working-memory-andlanguage-.compressed.pdf Repovš, G., & Baddeley, A. (2006). The multi-component model of working memory: Explorations in experimental cognitive psychology. Neuroscience, 139(1), 5-21. https://cpb-usw2.wpmucdn.com/sites.wustl.edu/dist/1/1008/files/2017/10/repovs20062ht9p2y.pdf Cocchini, G., Logie, R. H., Della Sala, S., MacPherson, S. E., & Baddeley, A. D. (2002). Concurrent performance of two memory tasks: Evidence for domain-specific working memory systems. Memory & Cognition, 30(7), 1086-1095. https://link.springer.com/content/pdf/10.3758/BF03194326.pdf 9 Gray, S., Green, S., Alt, M., Hogan, T., Kuo, T., Brinkley, S., & Cowan, N. (2017). The structure of working memory in young children and its relation to intelligence. Journal of Memory and Language, 92, 183-201. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5157932/ 9 The phonological loop • A mental workspace for manipulating auditory and verbal information. • Digit-span backwards is considered a test of phonological/verbal working memory because you must actively manipulate the information in memory, rather than just maintain the sequence. • Important in language development and verbal reasoning tasks. 10 10 The visuo-spatial sketchpad • A temporary store for representations of visual and spatial information such as faces, objects written words and cognitive maps • Enables the mental manipulation of visually and spatially represented information. • Mental rotation of objects • Visual and spatial mnemonics • Mental arithmetic • “Cognitive maps” for navigation • You can try an online version of the corsi-block tapping task here https://www.psytoolkit.org/experiment-library/corsi.html • And the backwards version here https://www.psytoolkit.org/experimentlibrary/backward_corsi.html 11 The Corsi Block Tapping task (at top) is an example of a spatial working memory task. It works on the same principles as digit-span, but using locations tapped by the experimenter. The experimenter has a view with the blocks labelled so they can repeat sequences given in the test. The participant view has no numbers displayed of course. The sequence of stimuli in the lower image shows another test of spatial working memory, similar in structure to the visual STM capacity tasks on slide 6. A spatial array is presented at encoding and is followed by a brief retention interval of 2500 ms, then you must decide whether the test array that is presented is the same or different from the memory representation you have for the test array – that is, are the black squares all in the same position in the test array as they were in the encoding array? The Central Executive • Executive processes are used in planning and coordinating complex behavior: • Goal orientation • Focus attention • Control of social behaviour • Switching between tasks, updating memory, inhibition of distracting information • Planning and problem solving • Executive processes are governed by circuitry in the pre-frontal cortex, especially dorsalateral prefrontal cortex and the anterior cingulate cortex (ACC) 12 The executive functions of the mind refer to a set of processes that allow us to focus our attention on our goals and control and regulate our behaviours. Working memory functions include a wide range of cognitive processes and behavioral competencies, such as verbal reasoning, problem-solving, planning, sequencing, the ability to sustain attention, resistance to interference, utilization of feedback, multitasking, cognitive flexibility, and the ability to deal with novelty. These functions are associated with the very front most part of the brain, called the pre-frontal cortex. This is the most recently evolved part of the human brain and the structure that is most developed in humans compared to other mammals. It is late to develop in children, and it is often first to decline with the diseases of ageing. Researchers generally, characterize executive functions as a specific set of attention-regulation skills involved in conscious goal-directed problem solving. These skills include cognitive flexibility (e.g., switching between tasks, multi-tasking, alternative perspective-taking), working memory (performing mental operations), and inhibitory control (suppression of interfering information). Cognitive flexibility involves thinking about something in multiple ways—for example, considering someone else’s perspective on a situation or solving a mathematics problem in multiple ways, and things like switching attention between two tasks, or multi-tasking. Working memory involves both keeping information in mind and, usually, manipulating it in some way, such as in passage comprehension when a reader must integrate several pieces of information or ideas into a coherent whole, or mental arithmetic. Inhibitory control is the process of deliberately suppressing attention (and subsequent responding) to something, such as ignoring a distraction, stopping an impulsive utterance, or overcoming a highly learned response. 12 Miyake, A. et al. The unity and diversity of executive functions and their contributions to complex “frontal lobe” tasks: A latent variable analysis. Cognitive psychology 41, 49–100 (2000). Zelazo, P. D., Blair, C. B., & Willoughby, M. T. (2016). Executive Function: Implications for Education. NCER 2017-2000. National Center for Education Research. 12 Neural Basis of Working Memory • Executive processes are based within networks in the pre-frontal cortex • The phonological loop (PL) is a lefthemisphere fronto-temporal lobe network. • The visuo-spatial sketchpad (VSS) is a right occipital-parietal network. • The episodic buffer integrates multi-modal information in an integrated ‘episodic trace’ within the parietal cortex (association cortex) 13 The figure shows a simplified depiction of the working memory system mapped to regions of the cortex. The central executive system is in the pre-frontal cortex, including the anterior cingulate (ACC). It controls attention to incoming information from sensation and from LTM. The phonological loop is shown as a (left hemisphere) fronto-temporal network. The visual-spatial sketchpad is a (right hemisphere) occipital-parietal network. The episodic buffer is centered on a parietal area of ‘association cortex’ that enables integration of information across modalities. Figure is from Chai, W. J., Abd Hamid, A. I., & Abdullah, J. M. (2018). Working memory from the psychological and neurosciences perspectives: a review. Frontiers in psychology, 9, 401. Note left and right hemispheres of the brain are not separated in this image. It is the spatial separation of the phonological loop in left hemisphere frontal-temporal circuits from the right hemisphere occipital location of the visual spatial sketch pad that helps to keep the two systems from interfering with each other. The darker pink circle that sits over the parietal lobe, located at the top of the cortex on each hemisphere towards the back of the brain represents the episodic buffer. This is an area also referred to as an “association cortex” because of its ability to integrate inputs from different sensory processes into a unified representation. 13 • Early accounts of STM focused on verbal maintenance of information for immediate recall and transfer to LTM. • Verbal STM capacity approximately 7 +/- 2 items (digit-span) • Duration is very brief, with rapid decay if rehearsal is prevented (BrownPeterson task) • Rehearsal in STM leads to LTM transfer • Primacy reflects LTM, recency reflects STM. Summary • Levels of processing affects transfer to LTM and suggests a working-memory rather than a system for shallow maintenance of information • Craik and Tulving (1975) • Baddeley’s model of working memory expands the concept of STM to a multicomponent, multi-modal system governed by executive processes. • Active workspace for reasoning and problem-solving, not mere maintenance of information. • The WM system is located across an integrated cortical network. • Left and right hemisphere systems for PL and VSS, respectively, governed by pre-frontal executive system (DLPF and ACC), producing an integrated episodic trace in parietal cortex. 14 14