The Language Instinct is very well written. The arguments and examples presented are interesting and persuasive. Many of the examples caught my interest.
In his example of BEV vs SAE, he shows and states that certain auxiliaries are used in their full forms(I have seen) while others are normally contracted. Where is the logic or rule that keeps some auxiliaries intact, contracts most, and yet deletes others?
Another topic I found interesting was the stuff on pidgen and creole languages. Pinker states that pidgen "did not offer the speakers the ordinary grammatical resources... no consistent word order... and no consistent way to indicate who did what to whom." Is it possible that the word order and some grammatical features of the pidgen came from each individual speaker's mother language?
A thought came up when I was studying the list of sentences on page 45 obeying the agreement rule. In one sentence, Sarah says the word "hisself". This word seems to bolster Pinker's proposition that language acquisition is not picked up through imitation. I would come to the conclusion that Sarah upon hearing the word "herself" deduced that the correct form for the male reflexive would be "hisself" by adding "self" to the end of the possessive pronoun(his and her books). This would show that Sarah then used pattern association and not imitation as a means of learning language. Then again, maybe my logic is a bit screwy.
I guess my final comment about the Pinker reading is that I'm very impressed with how he put a whole course of syntax into one chapter.
I don't have much of a reaction to the Modelling text. It was pretty straightforward and we covered all the stuff in class. However, I find it odd that the threshold on page 56 was set to two. How was that number chosen? At random? It works with the data, but I don't quite see how. I also find the diagram in figure 3.19a a bit disturbing. It's not how I'd imagine the interaction between axons and dentrites. C'est tout.
Some thoughts of how the connectionist model might handle some characteristics of the language 'instinct' described by Pinker:
(1) Pinker talks about subconscious grammar rules which people with SLI cannot perform(i.e. wug test), and suggests that this may lead back to a 'grammar gene' that is not fully functional in these people. This approach seems to point more towards the idea that language ability is localized rather than distributed. I have a feeling that the SLI samples in Pinker could be explained well in terms of a distributed model since the general language is present, but only details, as use of proper morphology, are missing. Also, from the idea I got about SLI, it is not like the ability to use correct endings has completely been wiped out, but rather the LIKELYHOOD that the correct morpheme will be used is reduced. This also seems to point towards a distributed model.
(2)The idea of prototypes, or the typicality effect. If synaptic connections are strenthened every time a pattern of stimuli is presented together with an association pattern, and learning occurs through several presentations, then what is learned in the end must resemble the average presentation, so recall works best when a pattern most likely to occur -- a prototype -- is presented. That makes sense, and works well with the Rosch color experiments. However, is it really always the case that the most common input from our environments becomes the one that we react best to? Do kids learn mostly nouns in the beginning because they are presented with mostly nouns, or is there something else going on that is more internal and less reliant on the outside world? In other words, the idea of typicality is based on the nature of input, but a lot of mental activity, certainly language, is very 'self-driven', thus that effect does not seem to apply so much.
The thing that kept coming back to my mind as I was reading was a brief mention in the second chapter of Pinker's book. He is discussing the formation of Creoles by children whose parents speak rough Pidgins. While it is facinating that these children are able to form languages in a single generation and that these languages contain full and complex grammars, it is even more facinating that the grammars of the Creoles are alike in so many aspects, and that, as Bickerton points out, this structure is so similar to the grammatical errors that children learning to speak English will often make. Bickerton says that their errors make it seem like there is a basic, underlying grammar "bleeding through a veneer of whitewash." This seems to suggest that there is a more basic and instinctive grammar than the one that we use in speaking English. If this is the case, it is surprising that languages like English were formed in the first place; why do creoles so consistently form with a common grammar that is so unlike the accepted grammar of English?
An argument based on children whose parents spoke a creole seemed, to me, to have some holes in it. Even though the adults were not able to speak to each other in a fully developed and grammatically complex language, each certainly had their own language that they spoke and that they likely used to communicate with their children. So the children were not completely without contact with languages that would enable them to learn the general concepts of a complex grammar; they might simply have applied this learned attribute of language to the creole that they were developing to communicate with one another. The obvious counter-example here, though, is the child whose parents were both deaf but spoke broken ASL; despite not having any other contact with speakers of ASL (and not being able to hear his parents' communication in spoken English) the child was able to master more complex grammatical structures in ASL.
The concept of a universal grammar seems like it might be very important with respect to getting computers to understand and use natural language. Instead of programming or training them to understand English or some other specific language, maybe efforts should be focused more on developing a program or network that has some sort of universal grammar, and can learn any language. This is obiously a harder proposition, but it might be more likely to bear fruit.
There seem to be huge advantages for distributed representations over localist representations. Other than space requirements and facility of adding new information, are there any advantages to localist representations over distributed ones? And in what situations would these advantages outweigh the many disadvantages?
I must preface these reaction notes with the sheepish confession that I have not yet read Ch. 4 in Pinker, and I'm not quite through the assignment in the other book--but I've done all I can do at present, and I'll try to do some more before class.
One question that struck me while reading the Connectionist Modeling book is probably related to the fact that I'm taking Computer Architecture this semester as well. I'm wondering what kind of sychronization issues arise in neural networks. Obviously, in the human brain timing of distributed processing is governed by biochemical factors--the speed at which the signals can be transmitted. But in synthetic models, do timing or sychronization issues arise? I haven't thoroughly thought this out, but looking at figure 2.2 on p. 40 made me wonder--couldn't some confusion arise when a node (such as a name of /another/ Jet) is excited and then inhibited? I would assume a situation could arise in a complex network where the wrong name is excited initially and only later inhibited. Does human cognition show artifacts of such difficulty? How do simulated connectionist models resolve (or choose not to resolve) this issue?
I'm also very curious to learn how/if one could build a neural network that models various cognitive features described by Pinker. Specifically, can a computer-simulated neural network make the pidgin-to-creole jump described by Pinker (pp. 24-29), or is that jump the result of a uniquely human language instinct which is more complex than a connectionist network? What else would one need to add to a neural network in order for it to exhibit a language instinct? Similarly--though perhaps more likely--how does one build a network that mimics Broca's aphasia? Is it possible to design a simplified network which models the multiple intellegences of the brain--i.e., a network which has both linguistic and mathematical capacity--and then damage the network in such a way that the mathematical intellegence remains but the linguistic capacity is impaired?
Also, I think I need a very brief review of Turing machines...
After reading the first few chapters of the Pinker text, I am confused as to the scope of his core argument. The focus of his argumentation appears to be that the ability to communicate via language is inherent in all homo-sapiens, a skill hard-wired into our brains via some form of evolution. What confuses me is what scope this theory appears to have based on the argumentation in the first few chapters. What constrains this skill-specific brain specialization to just language? For example, in citing the Wynn experiment, he mentions that babies ``updat[e] their counts as dolls were added or subtracted'' (59). Does he believe that there's a counting instinct or an addition instinct specific to our brains that even babies have? What about music? Is there a singing instinct? Just how much of the brain is specific and how much is all-purpose? Is Rain Man an arithmetical ``chatterbox?''
I am also curious to hear arguments refuting the ``from scratch'' (20) examples Pinker uses to support his argument. Most interesting are the following claims:
Pinker discusses several lines of evidence that seem to converge on one conclusion: language is a built-in function of humans. I don’t wish to try to refute each one, but I think it is worthwhile to examine his most convincing evidence, which can be summarized as "little kids are really good at learning language."
First, one should be skeptical whenever anyone mentions, as Pinker does, a "critical period" for language learning – not because such a period does not exist, but rather because we have no reason to believe that the period exists for the purpose of learning language. A critical period is nothing more than the time in one’s development of maximal neural plasticity and brain metabolic rate. After the age of five, we lose some 50% of synapses in the brain, and learning becomes more difficult as a result. However, this difficulty extends to behaviors other than language, such as learning to play the piano, which clearly no one would argue is innate.
Pinker also discusses the "poverty of input" argument, which states that certain grammatical structures are rarely, if ever, used in the presence of an infant, yet children figure out how to construct sentences containing such structures. Therefore, the "basic design of language is innate."
First, I wonder if cognitive scientists have not already built neural networks capable of producing the poverty of input result Pinker discusses, that is, deciding which auxiliary to move to the front of a sentence during question formation (a final project for this class, perhaps?). Such a network would be an existence proof for a system without innate specifications solving a problem argued by many linguists to be intractable.
My second objection to the poverty of input argument makes use of Pinker's discussion of children learning the –s agreement suffix (despite parental misuse). Is this not a case of learning a superfluous (at least in the deepest grammatical sense) language rule despite insufficient information? If it is, then the argument that language rules (such as auxiliary displacement in question formation) learned despite a poverty of input must be a part of the innate "basic design of language" is just plain false. Children can pick up on many rules, sometimes cleaning up noise from the environment, but it does not mean that they must have been born knowing the rules in any sense.
I didn't really have big questions or find things unclear in these early chapters, so this week I'll comment more on how this relates to previous reading and experiences I've had.
What an eclectic book! I thought I'd be reading about linguistics, but instead I've learned more biology and psych from these first chapters by far. Granted, as a linguistics major, I've already read Whorf's Language, Thought, & Reality and Berlin & Kay's big paper on color names and hierarchies of inclusion in different languages, and discussed Al Bloom's work as well.
Pinker's doing an awful lot of generalizing. "Whorf was out of his league. Here's my theory. Here're a few summaries of case studies to back it up. There, now don't you agree with me?" Granted, this book is for a lay readership, and granted, there isn't much more than case studies to go on -- these are very people-oriented issues, hardly cut-and-dried -- but I'll need a bit more convincing. Crazy as Whorf was, I liked some of his ideas!
But don't take that to mean that I dislike Pinker. Sure, I'm not sold on instinctual language just yet, but I'm hooked on the idea, and most of the book's still yet to be read....
Where does Pinker come up with such a disparate selection of case histories, random episodes in history, and the like? I'm amazed at his knowledge and/or research abilities, but remain vaguely skeptical that he's ooh-ing and aah-ing us with obscure and fascinating references, but secretly selecting just the case studies that complement his theories....
I'm also in the enjoyable position of seeing his theories at work several times a week in the person of Michael Iain McNary (age 8 1/2 months). Everyone talks to him, and he babbles back a stream of ever-more-various phonemes. By the time this semester is over, he might be speaking his first meaningful sentences. His mom's a linguist, and does speak "motherese" to him, but (and I asked her last night) only because it feels comfortable for her to do so, not because she thinks it will help Michael learn to talk. I look forward to trying to understand what's going on in his head these days.
He does an awful lot of imitating, though, which leads me to question Pinker's discounting of imitation. When we're at a restaurant reading the menus, he tries to grab one and hold it in front of him. When we're playing a game on a table, he'll pull himself up and try to see what's going on, and move the pieces around -- quite disastrous at times! Sure, maybe he's just curious, but he's displaying his curiousity by doing what the adults are doing, and what's to say he doesn't or won't try to do the same with language in the coming months? Well, I'll find out.
Pinker's book is an entertaining read. His review of psycholinguistic experiments is the most interesting thing to me, although I'd love to see data or foot/endnotes directing me to the data of the experiments that he sums up for us. While he does occasionally throw around some percentages, his summations of experiments are for the most part quantitative. His arguments would be stronger if they directed us to some data. His introduction to Chomsky's theory on language and the very brief introduction to X'-theory are impressively concise, but not new to a linguist.
Chapter 3 on "mentalese" was the most curious to me. While I can see that thought can happen independant of language and vice versa, I'm not convinced that it always does. What I mean is that while i agree that people can think outside of the bounds of language, I think that a person's language does inform how she may think, insofar as much as it a marker of culture. For example, while Chinese students do understand counterfactuals, would the language's lack of such a construction indicate that its speakers would not normally think in such a way. I think this line of thinking also applies to the question of the gender-neutral singular pronoun in English: he. Many children and adults struggle with or naturally resist the use of he as gender-neutral, as indicated by the common use of they in its place (as in (1)).
(1) Everyone put their book under their desk.
But, once someone learns from some prescriptive grammar teacher that they should use he, and start to do so, would they not start to think primarily of men in supposedly gender-neutral situations.
Do the idiosynchrosies of a language limit a person's ability to think certain thoughts? I don't think so, but I wonder if said idiosynchrosies do influence how a native speaker does think.