Sophie Piron, Mixed commentary: Ch 1 Q 9 and Ch 2, Fig 2.2.

Explain the 'U-shape' and 'readiness' phenomena that cnxists claim to explain and model.

Children frequently exhibit various U-shaped behaviors. I wonder if adults exhibit the same type of behaviors when they learn a new language. As we know, adults exhibit a slower learning rate than children when it comes to learning a new language. A
component of the readiness phenomenon must be lost. Speaking in terms of a network, connections have acquired various sets of weights and this slows down the readiness and ability of the network to change the relations built. Connections might have the tendancy to become too tight.

Let's now correlate the U-shaped curves found in children's behaviors with the readiness phenomenon. For children, we can think that the organization of new knowledge takes place during the down phase of the
U-shaped curve. The dramatical improvement occurs because of the readiness to gain new knowledge, i.e. (in terms of network) to set weights between nodes.
Another fact might be taken into account: we know that children are able to differentiate very quickly the phonemes of their own language from the ones of others. This means they posess a broad horizontal distinction between languages (inter lingua level). What is true of phonology could be expanded in some way to syntax and semantics. Bit by bit, a vertical distinction (intra lingua level) takes place: the reinforcement of weights happens. This should make it difficult from now on to "travel" on the horizontal axis. By that time, the readiness phenomenon will have disappeared.
According to this, adults would show a sigmoïd curve like the one on page 53 (figure 2.2.), where the upper part of the curve would always be at a lower level than the upper part of children's U-shaped curves. Meaning that adults don't master the new language as well as children do. We can imagine that the linear part of adults' sigmoïd curve won't exhibit the same degree of sensitivity to inputs than the positive linear part of a U-shaped diagram. The problem remains if adults also start with a negative linear relation.

Comment from Tom:
(1) Horizontal and vertical is a useful way to look at this. (2) Re adults' slower learning rate: my understanding is that this is mainly re acquiring the sound system of the L2, and adults are often faster at other aspects where their greater cognitive resources can come into play. Can anyone clarify this? Anyway, your readiness idea could still be applicable to the phono-level.

It would seem, following the connectionist arguments, that there is little, if any connection between universal outcomes and innateness, except for those outcomes which are invariant within the species. Using seemingly universal outcomes as justification for innateness now seems suspect. Johnson & Morton (1991) call only those things which occur within the organism itself, and occur without any influence from the outside environment, innate.  Language, probably much to Chomsky, Pinker et al's  frustration, doesn't fall into this category. Their position though, is, or perhaps was, logical. If all humans walk, it's innate;  if all humans have language, it's also innate. The problem arises when we look at cases such as those of Genie, or others who have been raised in isolation. Further problems arise in light of current knowledge of brain processes.  Clearly, the environment contains a vast source of information that has perhaps been overlooked/underrated.  Perhaps  the human brain is far more sophisticated and capable than we have allowed until now, and is able to "decode" as it were information/patterns in the environment that it is able to use in learning, storing and organizing information. It would seem that the brain doesn't need the innateness granted it (with relatively little subsequent learning taking place) but rather has the means to  absorb/ process/ store/ organize/ integrate/and learn by using the information available to it in the environment. The claim that universal outcomes are sufficient diagnostics for innateness seems now to be tenuous at best, certainly in the area of language acquisition at least.  

Response from Tom: But why didn't Genie learn grammar if language can be learned? Seems a switch has been turned off, just like the one that says no more height increase after puberty or many others which clearly ARE innate. (Actually I want to keep out of this part of the course except as reader but this is just to demonstrate how responses will appear).

Susan Ayotte, Q 17: Are universal outcomes sufficient diagnostics [= signs] of innate mechanisms?

The authors' use of a particular context to answer this question (all Brits speaking English does not imply their genomes encode for Eng.) illustrates just how definition-dependent this issue is: specifically, how narrowly/broadly one defines "innate" and "universal outcomes". The fact that our genomes do not encode for a particular language does not mean that there are no innate mechanisms at molecular/cellular/archtectural levels which may constitute the primary sine qua non processes leading to the near universal outcome of language development in children by +/- age 4. It's possible that these processes, in interaction with language itself, are responsible for ensuring that language gets there first, and sort of colonizes the brain pathways before other learning has a chance to do so, possibly even permiting other forms of cognitive development to take hold. This phenomenon must have some sort of survival value for the human species.
Since the innate/primal division is difficult to establish for humans (experimental constraints), we need a non-invasive way of modelling human learning which allows us to explore plausible answers to the question of the "how" of the universal language oucome and not just the "what". The question then becomes: what are these minimal innate mechanisms (constraints) which can account for universal outcomes, and how can we formulate models/simulations which will elucidate and/or approximate them? Can we zoom in on the perceptual biases that make language learnable by means of these models ?

From Tom: (a) So true that connectionism is an extremely definitions-sensitive zone. Then, when you consider that in the light of the fuzziness that we now attribute to definitions (under Prototype Theory), supported largely by the connectionists... (b) The "how" questions - that's what the connectionists are trying to answer - right? I couldn't tell if you were criticizing or articulating the cnxist program.

Tim Dougherty, Q 21: Summarize in your own words three important benefits of simulations in cognitive research.

The first chapter of Elman argues for three benefits of the simulation model in cognitive research. First,  the computer itself forces the researcher to follow a rigorous research path, because of the necessary precision that the  program demands. ie. no imprecision is allowed if the ideas of the researcher are to be tested using the computer as a research tool.

Second, because it is a computer model that is being used, the process is available on an ongoing basis for inspection, analysis, manipulation,  adjustment to the input, and corrections to the output, as needed. There are few human learners that are willing to subject themselves to this type of scientific inquiry!

However it's the third reason cited, that the models possess nonlinear characteristics, and "that the models exhibit emergent behaviors"(p.44) that I think provide the greatest benefit. This nonlinear behavior, which is remarkably similar to the learning pattern found in humans, at least as far as language learning goes, (and I suspect all learning), plus the emergent behaviors that the model can demonstrate, that also mimic the course of human learning, are  the aspects of the connectionist model that have the possibility of providing the learning/research/teaching community with information that will have great practical benefit. If the nonlinear connectionist model allows us to better understand the reasons for the way human learning takes place, or allows us to better understand the process of the learning taking place, it will advance the present state of knowledge about learning and teaching  significantly.

No. If the visual cortex becomes specialised in vision by being exposed to visual stimuli, then it is not specified in the genome and therefore is not innate. Any other cortex patch might have done the same job if it had been stimulated in the same way. In this sense, visual representation is not innate but rather the result of a more global innate structure which guarantees that the brain will develop in such a way that some patch of cortex, the one stimulated by visual input, will develop into visual cortex.

The distinction might seem less relevant when applied to basic physiological development, like that of the visual cortex. One easilly imagines that no matter what the visual stimuli are, the area of the cortex which receives them will specialise in vision all the same. However, the distinction becomes more interesting when applied to learning and the development of full-fledged cognitive processes. Certain areas of the cortex end up specialising in specific tasks, the ones that are presentedto the brain, and then growing subsequently more rigid in their specialisation as a result of training. Exposure would then be directly responsible for the biological shaping of the mind, which is suggested to be irreversible (Ho! My God...).

It is very interesesting to know just how much information is latent in the environment if we believe that the environement and the information it contains ( the external stimulus) shapes the very core of our developing biology. If our cortex gets specialised in such or such a way, according to the stimulus it receives while developping, then it is crucial to know how much information lies in the environment if we wish to have any understanding of our very nature or any influence on its development.

From Tom: Strong reasoning. I found your initial 1 & 2 dichotomy very clarifying.

What strikes me upon reading the first chapter is that the title of the book is very well chosen; it really is about Rethinking Innateness, about setting different parameters to define innateness. When comparing the Chomskian model and the connectionist model, I am reminded of the parallel between Newtonian physics and quantum physics. The latter proved there was no precomposite building blocks to matter. The authors of Rethinking Innateness put forward that there is no precomposite building blocks to grey matter either; cells of the cortex specialise according to the external stimuli they receive ( in a relative way, if you may); the microcircuitry is unspecified. Yet, they suggest that the macrocicuitry may be specified :

"In this form of nativism, knowledge is not innate but the overall structure of the network constrains or determines the kinds of information that can be received, and presentations that can subsequently be stored. In other words, the macrocircuitry [...] may be specified even if the microcircuitry is not."
(Rethinking Innateness, page 30)

Couldn't this prespecified macrocircuitry be compared to the innate structure refered to by Chomsky?

It seems like we're making the definition of innateness jump from one architectural level to the next. So connectionism is not necessarily anti-nativist in that it does not negate the existence of an innate structure. What it negates is that representation be built within this innate architecture.

***