Bowers: Knowledge Flashcards
lecture 1- How do we recognize words according to priming research on the effects of capitalization?
We recognize words not by their shape but by the letters that code for them.
Lecture 1- What are the two main theories of letter recognition?
Context-dependent coding: Different recognition units exist for each letter in every possible position in a word (e.g., “a1,” “a2,” etc.).
Context-independent coding: Letters are recognized independently of their position.
Lecture 1- What evidence supports the context-independent coding theory?
Priming research shows that transposed-letter words (e.g., “calm” vs. “clam”) prime more effectively than words with fully different letters, suggesting we recognize letters irrespective of their position.
Lecture 1- Why is competition important in word recognition?
Similar words compete to be recognized, worsening priming effects and increasing the likelihood of errors.
Lecture 1-What is cascaded processing in word recognition?
It is when the semantic meaning of a word is encoded alongside or before the full recognition of the word.
Lecture 1- How does priming research support the concept of cascaded processing?
Priming evidence aligns more with the cascaded processing approach, where semantics are encoded early.
Lecture 1-
What are frequency effects in word recognition?
High-frequency words are read more quickly than low-frequency words.
Lecture 1- What are regularity effects in word recognition?
Regular words (following standard phonetic rules) are read more quickly than irregular words.
Lecture 1- How do frequency and regularity interact in word recognition?
The regularity effect is only found for low-frequency words
Lecture 1- What are the three types of dyslexia, and how do they differ?
Surface Dyslexia: Difficulty reading irregular words but fine with nonwords and regular words.
Phonological Dyslexia: Difficulty reading nonwords but fine with regular and irregular words.
Deep Dyslexia: Difficulty with nonwords, irregular words, and regular words; better with high-imageable words and often makes semantic errors (e.g., reading “pig” as “elephant”).
Lecture 2-What is the main difference between localist representations and distributed coding?
Localist representations suggest that each unit in the brain codes for a specific concept or item, often referred to as “grandmother cells.” In contrast, distributed coding proposes that each unit is connected to a hidden layer of units with varying strength relationships, and the output is not directly linked to the input, as it is influenced by these complex connections.
Lecture 2- How is distributed coding modeled after computer algorithms?
Distributed coding is modeled after how computer algorithms are trained. Over time, the system learns the correct weighted connections through a process called back-propagation, where it associates the right inputs and outputs, much like how a machine learns from experience.
Lecture 2- What is the main argument against localist coding?
The main argument against localist coding is the skepticism about the existence of “grandmother cells” or neurons that carry out specific tasks for particular stimuli. Studies have investigated this by looking at neuron activation to specific stimuli, but critics argue that this research might have missed other stimuli that could activate the same neurons.
Lecture 2-What is the superposition catastrophe, and how does it challenge distributed coding?
The superposition catastrophe is a problem in distributed coding where two separate inputs, such as two different words, cannot be held in short-term memory (STM) simultaneously. The code generated by the two inputs becomes ambiguous, making it difficult to analyze them correctly within the model.
Lecture 2- How did Botvinick and Plaut (2006) attempt to address the superposition catastrophe?
Botvinick and Plaut (2006) suggested that if the model is given inputs one at a time, the hidden layer can rehearse each input individually and co-activate them, thereby overcoming the superposition catastrophe.
Lecture 2- Does the solution proposed by Botvinick and Plaut (2006) mean that distributed coding has fully overcome the superposition catastrophre
Not necessarily. While the rehearsal and co-activation process can help, research has shown that the neural activation pattern shifts from being distributed to becoming more localized when dealing with multiple syllables. This suggests that the model might be adapting to handle multiple inputs by organizing activation based on the position of letters, which may not be a complete solution to the superposition catastrophe.
Lecture 2- How do neural activation patterns change when dealing with single vs. multiple syllables?
For single syllables, neural activation follows a classic distributed coding pattern. However, when multiple syllables are involved, the activation pattern becomes more organized and localized, which helps the model handle multiple inputs without causing confusion in the coding.
Lecture 3-
How do modular and non-modular systems interact in perception and action?
Perception starts with vertical (modular) systems that process sensory inputs. These feed into more advanced, non-modular (horizontal) higher-level cognitive systems, which then direct motor systems and action.
Lecture 3- What is information encapsulation in modular systems?
Information encapsulation means that higher-level cognitive processes do not influence lower-level perceptual modules. For example, memory does not alter how we initially perceive sensory information—it is only after perception that filtering occurs.
Lecture 3- What is a key feature of modular systems in terms of response speed?
Modular systems operate reflexively and produce very fast responses, independent of higher cognitive processing.
Lecture 3- What is selective adaptation research, and what does it suggest about phoneme perception?
Selective adaptation research involves habituating participants to a certain phoneme (e.g., “g”) and then presenting two similar phonemes (e.g., “g” and “b”). When the “g” receptor becomes fatigued, participants are more likely to perceive the ambiguous sound as “b.” This suggests that phoneme receptors adapt and become less sensitive with repeated exposure.
Lecture 3- How does hallucination research support findings from selective adaptation research?
In hallucination studies, when participants are shown a word with a missing phoneme (e.g., “legi-lature”), they tend to fill in the missing sound and report hearing the full word (“legislature”). When this stimulus is repeated, participants begin to exhibit the same adaptation effects seen in selective adaptation research, reinforcing the idea that phoneme perception is influenced by prior exposure.
Lecture 3- What does pause detection research reveal about word processing?
n pause detection studies, participants hear a word, sentence, or phoneme followed by a pause, and their reaction time to detecting the pause is measured. Findings suggest that:
Pause detection is worse for non-words.
Pause detection is worse for “late-unique” words (words that start similarly to many others and only diverge later, such as “legislation” vs. “legitimate”). This suggests that the brain is actively predicting words based on their early phonemes.
Lecture 3- What is a major criticism of pause detection research?
One criticism is that reaction times can be unreliable. However, the late-unique word effect has been replicated using ERP (event-related potential) research, which strengthens its validity.
