Academia.eduAcademia.edu

Figure from:Five Notes on Recursive Syntax

See full PDFdownloadDownload figure
Sub-cortical pathways lead to cortex (insular cortex) In addition to a horizontal spreading, say between anterior areas of front left hemispher (FLH- Broca) and posterior areas related to the temporal lobe (TL-Wernicke), we also find sub-cortical portaling where ‘neuro-bundling’ which orginate well below the surface cortext form pathways to surface cortex. This was been suggested with regards to what we find with the insular cortex, whereby repetive practice of a particular task which was associated with Broca processing (Frontal lobe) was seen as migrating both vertically and horizontally to the deeper insular cortex (Temporal lobe).

Figure 13 Sub-cortical pathways lead to cortex (insular cortex) In addition to a horizontal spreading, say between anterior areas of front left hemispher (FLH- Broca) and posterior areas related to the temporal lobe (TL-Wernicke), we also find sub-cortical portaling where ‘neuro-bundling’ which orginate well below the surface cortext form pathways to surface cortex. This was been suggested with regards to what we find with the insular cortex, whereby repetive practice of a particular task which was associated with Broca processing (Frontal lobe) was seen as migrating both vertically and horizontally to the deeper insular cortex (Temporal lobe).

Related Figures (12)

Whenever statistical averages of a competency of a certain Skill <x> are spread across a given demographic <y >, what one finds is a very consistent (probabilistic) average. This average, shown above as 34.1% / 34.1% (= 68.3%) on both sloping sides of the bell, indicates the average skill set of <x>. The largest subset of ‘people studies’ shows an average-level of competency for skill x, and this is consistent across all skills looked at. (Note: The stableness of this bell-shape ratio comes close to what we know of the Fibonacci ‘golden ratio’). The scale here -4 to +4 could be Competency of a Learned Skill <X> / (L2)
Whenever statistical averages of a competency of a certain Skill <x> are spread across a given demographic <y >, what one finds is a very consistent (probabilistic) average. This average, shown above as 34.1% / 34.1% (= 68.3%) on both sloping sides of the bell, indicates the average skill set of <x>. The largest subset of ‘people studies’ shows an average-level of competency for skill x, and this is consistent across all skills looked at. (Note: The stableness of this bell-shape ratio comes close to what we know of the Fibonacci ‘golden ratio’). The scale here -4 to +4 could be Competency of a Learned Skill <X> / (L2)
4 (See Stephen Jay Gould’s Full House: The spread of Excellence from Plato to Darwin (1996) for discussion of the ‘right-wall’). What is so intriguing about the so-called ‘right-wall’ when it comes to a learned skill is that its extreme high-level of mastery mimics what we know of any biological-bases which governs earning. In fact, it’s not overt ‘learning’ at all, but rather a state of biologically determined acquisition. It is in this sense that the terms ‘learned’ versus ‘acquisition’ makes its way into the L1 vs. L2 literature— namely, L1 (pre-critical-period native first language) is biologically determined and so does not suffer the competency spread found of bell-shape learning, while L2 post-critical-period second language learning) shows bell-shape statistics. (post-critical-period second language learning) shows bell-shape statistics.
4 (See Stephen Jay Gould’s Full House: The spread of Excellence from Plato to Darwin (1996) for discussion of the ‘right-wall’). What is so intriguing about the so-called ‘right-wall’ when it comes to a learned skill is that its extreme high-level of mastery mimics what we know of any biological-bases which governs earning. In fact, it’s not overt ‘learning’ at all, but rather a state of biologically determined acquisition. It is in this sense that the terms ‘learned’ versus ‘acquisition’ makes its way into the L1 vs. L2 literature— namely, L1 (pre-critical-period native first language) is biologically determined and so does not suffer the competency spread found of bell-shape learning, while L2 post-critical-period second language learning) shows bell-shape statistics. (post-critical-period second language learning) shows bell-shape statistics.
(For back propagation, neuro-pathways would also go backwards, in so-called error- trained learning).
(For back propagation, neuro-pathways would also go backwards, in so-called error- trained learning).
Within a treelet structure (above), consider the ‘register-sets’ [boxes marked subject, predicate, relation, etc) to be analogous to neurons (cells) as found in the brain. The Hebbian express ‘What fires together wires together’ accurately describes this local level of function as ‘frequency- effects’ plays a critical role at this level (just as with word learning)—any associative model that relies on strength of reinforcement must establish such local wiring & firing domains attributed to one-one associations (behaviorism). In this sense, keen connectionism is a new form of behaviorism. Notice how treelet structures allow nodes to be coded in ways which establish categories, such that (categorical) relations hold between e.g., subject and predicate. This is the main difference between a serial search which would be conducted exclusively via ‘nodes only’ versus searches which could be done in parallel over nodes and categories. Treelets resolve the problem that other low-level processing had (SRNs, ML-Ps) since treelets, with these new categorical relations between nodes can generalize to new concepts. Hence, category, creativity, and the ability to function with variables inters into the processing domain. Such ingredients would be critical for any true human-like language to be mimicked successfully in any Al machine. So, what Al needs is a program (a domain) that uses (abstract) codes/symbols and not just (associative-based) nodes. Recall, that in human language, much of our language intuition goes Fig.2 Register-sets (Copied and reinserted here as found in Fig. 4 of Appendix 1).
Within a treelet structure (above), consider the ‘register-sets’ [boxes marked subject, predicate, relation, etc) to be analogous to neurons (cells) as found in the brain. The Hebbian express ‘What fires together wires together’ accurately describes this local level of function as ‘frequency- effects’ plays a critical role at this level (just as with word learning)—any associative model that relies on strength of reinforcement must establish such local wiring & firing domains attributed to one-one associations (behaviorism). In this sense, keen connectionism is a new form of behaviorism. Notice how treelet structures allow nodes to be coded in ways which establish categories, such that (categorical) relations hold between e.g., subject and predicate. This is the main difference between a serial search which would be conducted exclusively via ‘nodes only’ versus searches which could be done in parallel over nodes and categories. Treelets resolve the problem that other low-level processing had (SRNs, ML-Ps) since treelets, with these new categorical relations between nodes can generalize to new concepts. Hence, category, creativity, and the ability to function with variables inters into the processing domain. Such ingredients would be critical for any true human-like language to be mimicked successfully in any Al machine. So, what Al needs is a program (a domain) that uses (abstract) codes/symbols and not just (associative-based) nodes. Recall, that in human language, much of our language intuition goes Fig.2 Register-sets (Copied and reinserted here as found in Fig. 4 of Appendix 1).
usage as arecursive means would looks as follows Clearly, this is very similar to what we saw above showing how the ‘flat recurrent’ structures typical of SRNs and/or ML-Ps, have a difficult time handling such embedded dual processing: fo! instance, just as there is no way to capture ‘boat-house’ vs. ‘house-boat’ as discussed above by simply coding for <boat> and <house>, so too would there be no way to handle ‘box inside pot from ‘pot inside box’, or the kind HORSE vs the individual ‘John’s horse’, etc. A flat structure would ambiguously code both instances as [box/pot inside box/pot] as well as not being able tc capture the dynamics that ‘John’s horse’ and ‘Mary’s horse’ (or ‘Mary’s sandwich’ for tha’ matter), is not one in the same item. simply coding for <boat> and <house>, so too would there be no way to handle ‘box inside pot’ from ‘pot inside box’, or the kind HORSE vs the individual ‘John’s horse’, etc. A flat structure
usage as arecursive means would looks as follows Clearly, this is very similar to what we saw above showing how the ‘flat recurrent’ structures typical of SRNs and/or ML-Ps, have a difficult time handling such embedded dual processing: fo! instance, just as there is no way to capture ‘boat-house’ vs. ‘house-boat’ as discussed above by simply coding for <boat> and <house>, so too would there be no way to handle ‘box inside pot from ‘pot inside box’, or the kind HORSE vs the individual ‘John’s horse’, etc. A flat structure would ambiguously code both instances as [box/pot inside box/pot] as well as not being able tc capture the dynamics that ‘John’s horse’ and ‘Mary’s horse’ (or ‘Mary’s sandwich’ for tha’ matter), is not one in the same item. simply coding for <boat> and <house>, so too would there be no way to handle ‘box inside pot’ from ‘pot inside box’, or the kind HORSE vs the individual ‘John’s horse’, etc. A flat structure
Showing a linguistics syntactic tree, the CST-1 found in [7i] would be represented in [8] below The above notion of ‘locality vs. distance’ as bound by computational time-steps (CTS) has antecedence to levels of computational processing found in the brain—with ‘Low-level’ computations being assigned to exact one-to-one neuro firing (say, having to do with the triggering of a specific node within a connectionists model), while ‘High-Level’ processing establishes distant relationships, say, between nodes. The expression that the brain bootstraps establishes distant relationships, say, between nodes. The expression that the brain bootstraps
Showing a linguistics syntactic tree, the CST-1 found in [7i] would be represented in [8] below The above notion of ‘locality vs. distance’ as bound by computational time-steps (CTS) has antecedence to levels of computational processing found in the brain—with ‘Low-level’ computations being assigned to exact one-to-one neuro firing (say, having to do with the triggering of a specific node within a connectionists model), while ‘High-Level’ processing establishes distant relationships, say, between nodes. The expression that the brain bootstraps establishes distant relationships, say, between nodes. The expression that the brain bootstraps
Do, Be, Have’” (do cook!, is cook-ing, has visit-ed), Inflectional morphology V + ing = N (gerund)
Do, Be, Have’” (do cook!, is cook-ing, has visit-ed), Inflectional morphology V + ing = N (gerund)
Fig. 2. The Lateral spreading of grammar based on frequency-effects: from Broca to Wernicke (walked vs. stalked). ‘Frequency of item
Fig. 2. The Lateral spreading of grammar based on frequency-effects: from Broca to Wernicke (walked vs. stalked). ‘Frequency of item