Cognition and the Brain Study Guide: Semantic Representation and Auditory Perception

Semantic Representation

Main Points: Semantic Dementia and ATL

• Semantic dementia is a temporal lobe variant of frontotemporal dementia. It's a selective neuropsychological impairment that involves anomia (inability to name object/sound) and multimodal comprehension impairment (all domains such as pictures, words, and smells, are affected)
• Patients with semantic dementia have atrophy in the Anterior Temporal Lobe (ATL); semantic dementia is a graded (by familiarity and specificity), asymmetrical, bilateral disease.
• The ATL is the basis for modality-invariant contribution to coherent semantic representations
• Hub-and-spoke model of conceptualisation: the hub (ATL) coordinates all of the input from the different spokes (areas in the cortex) to represent the semantic conceptualisation of the concept.
• The ATL is locally connected (little evidence of large jumps in white matter tract)
• The ATL has graded representation --- certain parts are more sensitive to certain modalities (e.g. audition).
• The ventral ATL is responsive to stimuli of all modalities
• The ventrolateral ATL is the region within the ATL that has the most atrophies
• Repeated Similarity Analysis: compared similarity of brain responses to similarity of stimuli

Keywords/Concepts: Semantic Memory and Aphasia

• Semantic memory
• Agnosia: disorder where an individual can't transform sensation into meaning; the meaning is intact but access to the meaning (i.e. door) is impaired.
• Semantic Aphasia (completely different from semantic dementia): lack of semantic control; problem with executive control of semantics (meaning is there but can't control it); this is different from Semantic Dementia

Brain Mechanisms of Flexible Cognitive Control

Main Points: Cognitive Control and Attention

• How does our brain focus/selective on specific tasks while also easily shifting/flexible between tasks?
• Spotlight Effect: we focus on individual parts of a stimulus (we can zoom in and out, move from one part to another, etc.), but since we are moving our eyes so fast, we create the illusion that we are seeing everything in one glance.
• Attention in space, to an object or feature, and in time
• Attentional Blink (Rapid Series Visual Presentation): when two relevant stimuli are presented close to each other, it's harder to attend to the second stimulus because you are still processing the first stimulus.
• Exogenous (bottom up) vs. Endogenous (top down) cues
• Change blindness: unaware of the change; under normal circumstances, the change is transient but if there is a blank between the images, this causes a transient everywhere, making it harder to detect the change.

Multiple Demand (MD) System

• Multiple Demand (MD) system: parts of the brain (usually frontal and parietal cortices) that allow us to focus on certain stimuli while also flexibly switching between stimuli.
• Pre-supplementary motor area, intraparietal sulcus, inferior frontal gyrus, middle frontal gyrus, and anterior insula/frontal operculum
• Active for a diverse range of tasks
• Represents a variety of information
• Strongly shaped by task demands
• Prioritizes difficult and attended information
• Fast time course

Adaptive Coding Hypothesis and Neural Activity

• Adaptive Coding Hypothesis: single neurons dynamically adjust their response profiles to encode information that is currently relevant for behavior
• Frontoparietal emphasis on task relevant information supports dominance of that information throughout the brain
• Multivariate Analysis -> measuring neural activity in voxels
• Linear Pattern Classifier: a machine learning algorithm in which you feed data (about different categories) into the model and the model determines the boundary between the categories. Use the linear pattern classifier to measure how well the neurons are coding certain stimuli/features.
• The strength of the object coding varies with explicit allocation of attention (e.g. the more you attend to a feature, the higher the decoding accuracy will be)
• Attentional Episodes: how quickly you can change your attention from one feature to another; in easy tasks, you don't attend to any one feature in particular
• Is the activity of the MD network (e.g. responding to specific stimuli) relevant to behavior?
• Analysis of Errors -> are errors tied to errors in the MD network's coding of the stimuli and the rules? The MD coding predicts error type.
• When you make a rule error, the neural coding corresponds to the opposite response in the MD system
• Particular error is predicted by the information coded
• The relationship between patterns of coding and behavior is stronger in the MD network compared to the visual cortex
• For the visual cortex, rule errors correspond to accurate stimulus coding and at chance unspecified error coding
• Strong link to behavior for codes in MD regions and at later timepoints
• Causal role in information coding -> if you perturb parts of the MD network (DLPFC), classification accuracy decreases and impacts the rest of the MD network; the MD networks supports prioritization of relevant information, it doesn't suppress irrelevant information
• By stimulating the brain with TMS and concurrently using MEG + fMRI, we can discover causal pathways
• Brain damage -> can provide causal insight; can investigate neural processing/decoding at an individual basis; each individual is very different (some show attentional prioritization deficits and others do not)

Basic Mechanisms of Auditory Perception

Main Points: Sound Creation and Tonotopic Mapping

• Sound is created when an object vibrates, transmitting vibrational energy to the surrounding medium (usually air). This vibration travels into the ear where it moves fluid in the Cochlea, which vibrates the basilar membrane and the stereocilia (hair cells), consequently, activates the cochlear/auditory nerve.
• Tonotopic mapping: the basilar membrane is stiffer on the base side (high frequencies) and less stiff on the apex (low frequencies).
• Sounds are largely characterized by frequencies and loudness (loudness helps us differentiate between vowels and consonants)
• The frequency that corresponds to the lowest threshold is the characteristic frequency of that group of neurons
• Spectrogram: heatmap that illustrates frequency across time and the intensity/loudness of the signal
• The absolute threshold of a sound (the loudest it can get) depends on the frequency of the sound
• Spread of excitation: the louder the sound is, the more neurons that are recruited (even if the sound doesn't correspond to their specific frequency)

Cochlear Implants (CI)

• Cochlear Implants:
• Directly stimulates the electrodes along the basilar membrane (the electrode array is directly inserted into the cochlea)
• The CI processes the acoustic wave, using the envelope detection method to separate the different frequencies into bins. Then, it converts these frequencies to electrical pulses that activate the appropriate part of the basilar membrane.
• The signal is hugely degraded in CI because there aren't as many electrodes (around 20) as there are hair cells.
• Localisation cues are impaired in CIs; Interaural Level Differences (ILDs) is the only cue that isn't impaired in CI

Measuring Hearing and Auditory System Health

• Approaches to measuring level of hearing/health of auditory system
• Electrically-Evoked Compound Action-Potential (eCAP): the auditory nerve response is recorded by the CI itself; measuring the response of the auditory nerve from electrode activation
• Electrically-Evoked Auditory Brainstem Response (eABR): the CI activates the electrode and measures the activity in the brainstem
• Cortical Auditory Evoked Potential (CAEP): measuring the response of an auditory stimulus in the auditory cortex
• There is a lot of variability between CI users. Two factors influence the electrode to neuron interface: (1) health of the auditory nerve and (2) the spread of electrical current

Measurements and Electrical Auditory System

• Measurements:
• Forward-Masking Artefact-Reduction Technique: a method for extracting neural responses from noise/stimulus artifacts
• Spread of Excitation Curves (SOE): an estimation of the spread of neural excitation to surrounding neural tissue in response to one electrode being activated
• Panoramic ECAP Method: an improvement of the SOE by recording neural responses to all combinations of masker and probe; goal of this method is to finetune the electrical stimulation and spread (e.g. identifying where the dead neurons are, how each electrode impacts nearby electrodes, etc.) so that sound perception is clearer in CI; it's a way to individualize the CI to the individual (electrodes respond differently in each individual)
• With the electrical auditory system, it's easy to see place pitch, as each electrode corresponds to a different frequency band. The CI is really good at phase locking (temporal dimension); can separate/distinguish between different speeds.

Keywords/Concepts: Hearing Theories and Localization

• How we hear: two theories
• Place Pitch: pitch is encoded based on the location of the basilar membrane that responds to the vibrations. The base of the membrane codes for higher pitched/frequency sounds whereas the apex codes for lower pitch/frequency sounds.
• Phase Locking/Temporal Pitch: pitch is encoded based on the speed or rate at which the neurons are firing (the different speeds of neural firing represent the different frequencies/pitches); works in conjunction with place pitch theory
• Interaural Timing Differences (ITDs): differences in timing between when the sound reaches one ear vs. the other ear; a localisation technique
• Interaural Level Differences (ILDs): differences in the volumes of sound reaching one ear vs the other ear (the ear that is closer to the sound will hear the stimulus louder, indicating that the sound is closer to that ear)
• Difference liemenes: just noticeable differences

Sense of Hearing and Technology

Main Points: Hearing Loss and Devices

• Hearing loss is due to loss of sensory hair cells in the cochlea and it affects around 1 in 3 people
• Pure-tone audiometry (PTA): primary test to determine hearing threshold levels
• Three types of hearing devices: hearing aids, sensory implants/prosthesis (e.g. cochlear implants + implants into the auditory stem), and hearables

Limitations of Hearing Devices and Spectral Blurring

• Limitations of hearing devices: speech-in-noise difficulties (challenging to decipher speech from noise) and current spread/spectral blurring (signals from nearby electrodes are blurred)
• No spectral blurring when the electrodes are spaced or clustered in the middle and basal regions. There is individual variability when it comes to spectral blurring.
• Apical electrodes cause major deterioration of sound quality when the electrodes are blurring. Clinical importance -> it might be useful to deactivate certain electrodes in the apical side of the basilar membrane.

Speech-in-Noise Tests and Confounding Effects

• Requirements for speech-in-noise tests: S(sensitivity), E (efficiency), R (reliability), V (validity), I ( inclusivity), C (comparability), E (efficacy)
• Fixed or adaptive SNR testing
• Confounding effects in speech testing + ways to avoid them:
• Learning effect: provide practice/new stimuli
• Order effect: randomise or counterbalance
• Bias effect: double-blinded testing
• Ceiling and flooring effects: adaptive SNR

Future of Hearing Devices

• Hearing devices of the future will:
• Effectively mitigate noise and reverberation via improving signal quality and cleaning up noisy speech
• Use machine learning (ML) to parse out noise from speech (e.g. training DNNs for speech-in-noise detection)
• Multi-microphone setup: adding more than one microphone to the cochlear implant improves signal to noise ratio
• Choosing between target speakers
• Speaker-informed DNN -> teach the model to pick out a certain voice