Lecture 13: The future Flashcards
What are Marr’s 3 levels? Explain shortly.
Computational theory: what and why?
- What is language and why does it have certain cognitive properties?
Representation & algorithm: how?
- What is the representation for the input/output can be transformed (algorithm)
Hardware & implementation: where?
- Hardware explanation of a cognitive process
What is the “Ontological incommensurability problem (OIP)” ? On which of Marr’s 3 levels does it show up?
It shows up in the second level, the “Representation & Algorithm” level.
- Ontological = relation between concepts/categories
- Incommensurability = having no common measure
-> OIP states that linguistics components do not have an equivalence to neuroscience components
Example: a neuron is not the neurobiological equivalent of a linguistic unit.
But both types of components are valid, and do we really need causational links between computational-representational (CR) theories and neurological theories? Not really, it’s enough to know how the concepts are correlated.
CORRELATION DOES NOT MEAN CAUSATION
Example: activity in BA44 (Broca’s area) is correlated with processing phonology, syntax and semantics
On which levels of Marr’s levels model do computational-representational theory focus on, and on which did neurobiology focus on? What did scientist recently suggest concerning those two theories?
Most of the computational-representational theories have been occupied with the first 2 levels (computational theory, representation & algorithm), defining what language is and why language is the way it is, and describing how it operates. The neurobiology has been doing the hardware part.
Scientists see an issue in separating those theories; they should rather work together on all three levels. This is meant not only for language, but any kind of cognition.
What is the “Granularity mismatch problem (GMP)” ? What could we do to solve this problem, and why don’t we do this?
- Granularity = scale / level of detail
- Mismatch = fail to correspond
GMP is the problem of CR theories having a lot of detail, while neurobiology is quite broad/non-specific.
Example: the magnet effect (see lecture 8) is very specific, while the processing of phonology (broad) is all tied to the same area (BA 44).
In order to solve this problem, we would need to record neurobiological processes in more detail, but we don’t have the technology to do that yet.
What are the three approaches to combining CR (computational-representational theories) and NB (neurobiology) successfully?
Correlational approach
- Find correlations between brain representations and computation of language
Integrated approach
- Use NB information to select and interpret CR
Explanatory approach
- NB explains and is explained by representation and computation. (In the explanatory approach, brain structures are specialized in specific processing and representations and these explain the brain structure.)
In a study, participants listened to different phonemes and their brain activity was recorded using MEG (which has high temporal and spatial resolution). The phonemes were [i], [y], [u].
- [i] & [y] have a similar lipshape while pronouncing, [u] is different
- [y] & [u] have similar tongue-position while pronouning, [i] is different
All of these sounds had a spike at N100, but were not all the same spatially.
Which approaches to combining CR (computational-representational theories) and NB (neurobiology) does this study fit with? Explain.
This study fits with both the correlation and integrated approach
- Correlation: there is a correlation between the activity of specific areas of the brain, and different phonemes
- Integrated: use the result of this study to conclude that different phonemic features are of importance in language processing
Which approaches to combining CR (computational-representational theories) and NB (neurobiology) has been used to explain the hearing of barn owls?
The explanatory approach.
Has the explanatory approach been used in studies on human language? Why or why not?
No.
Levels of CR are smaller than the levels of NB; sometimes CR’s levels are smaller than neurons. These levels are not anatomically defined areas.
