From Molecule to Metaphor: A Neural Theory of Language (Bradford Books)
by: Jerome A. Feldman
By
* Publisher: The MIT Press
* Number Of Pages: 377
* Publication Date: 2006-06-02
* Sales Rank: 285571
* ISBN / ASIN: 0262062534
* EAN: 9780262062534
* Binding: Hardcover
* Manufacturer: The MIT Press
* Studio: The MIT Press
* Average Rating: 4.5
* Total Reviews: 5
Book Description:
In From Molecule to Metaphor, Jerome Feldman proposes a theory of language and thought that treats language not as an abstract symbol system but as a human biological ability that can be studied as a function of the brain, as vision and motor control are studied. This theory, he writes, is a "bridging theory" that works from extensive knowledge at two ends of a causal chain to explicate the links between. Although the cognitive sciences are revealing much about how our brains produce language and thought, we do not yet know exactly how words are understood or have any methodology for finding out. Feldman develops his theory in computer simulations--formal models that suggest ways that language and thought may be realized in the brain. Combining key findings and theories from biology, computer science, linguistics, and psychology, Feldman synthesizes a theory by exhibiting programs that demonstrate the required behavior while remaining consistent with the findings from all disciplines.
After presenting the essential results on language, learning, neural computation, the biology of neurons and neural circuits, and the mind/brain, Feldman introduces specific demonstrations and formal models of such topics as how children learn their first words, words for abstract and metaphorical concepts, understanding stories, and grammar (including "hot-button" issues surrounding the innateness of human grammar). With this accessible, comprehensive book Feldman offers readers who want to understand how our brains create thought and language a theory of language that is intuitively plausible and also consistent with existing scientific data at all levels.
Date: 2007-08-17 Rating: 5
Review:
Remarkable, but perhaps too ambitious
This is a remarkable book, albeit possibly too ambitious. Feldman has little use for Chomsky's theory of language (but admires his analysis of language structures). By analogy, he shows the same kinds of arguments could be used to show the ability to dance is genetic, and embodied in its own dedicated brain structure, as to say language is. What Feldman is about, though, is not to engage in polemics, but to attempt to develop a theory of language which draws on current knowledge and perspectives from a variety of fields. In fact, he says tongue in cheek, "the human genome does seem to code for a tendency to engage in bitter (academic) wars that are senseless to an outsider".
The key point is that any theory of language should reflect what we know about neuro-science. He implicitly makes the case that mastering language, while a wonderful achievement, is not much more amazing than mastering visual interpretation. There is a progression: one learns to control one's body, then use many of the same "mirror" neurons to interpret visually movements by others, than use many of the same neurons to give meaning to language describing movement. Grammar is a link between language and meaning, and is first learned by "matching sentences to what the child already knows visually". More abstract uses of language build on schema's for movement and emotional experience by way of metaphor, and Feldman does a wonderful job in giving the reader a great feel for how metaphor works, and how metaphor builds on past mastery of other metaphors organized in "cultural frames", and the role of parameterization.
Feldman may be overly ambitious in trying to communicate too much of the technical underpinnings, so that the reader sometimes gets bogged down. He does a nice job in giving some basics of neuro-science such as Hebb's rule for how neural connections are developed and embody learning. He nicely shows how triangular nodes can be used to represent and retrieve facts, and the neural basis for the well known concept of priming, in which visual or verbal cues impacts subsequent interpretation of language, when the two follow closely in time. However, he is less successful with belief networks, and "DP connectionism" (what is known in other fields as neural network algorithms). Feldman uses the approach of providing toy problems, i.e. very simplified examples, and sometimes this just is not sufficient to get a real feel for things.
Date: 2007-05-07 Rating: 5
Review:
How do we learn, use and construct language?
Jerome Feldman wrote a nice book on the question of how language came to be. It is readable for outsiders (and has some little jokes that make you smile once in a while). By taking you step to step to the concept of a neural theory of language the book can be read chapter by chapter. For me, the theory it develops looks quite satisfactory. Noam Chomsky is one of the guys I would like to comment on this book, maybe sometime in the future he does. Buy this book if you want to know how we construct and communicate information to others and how we use context, experience and grammar to understand. This is a typical American book, providing readers outside the field with a comprehensible overview of recent research. Well done!
Date: 2006-09-17 Rating: 2
Review:
Wordy, convoluted, unsatisfying
I read this book immediately after reading Hawkins' _On Intelligence_. Whereas OI was written in very layman's language, FMTMANTOL was written in a very academic style. Its style seemed to contradict the content. On the one hand, it gave some very simplified and hand-waving view of molecules, but went over the same simple points again and again in too much detail. Specifically, the author seemed fascinated with metaphor. Obviously that's the title of the book, but it seemed rather than just explaining how metaphor works, he as obsessed with pointing out the metaphors he was using to try to describe neurons and such. I don't have the book in front of me at the moment so I can't give any specific examples. Then (on the other hand), the meat of the book covers some theories on schemas and other logical/symbolic processing the brain allegedly does. This too is done in excruciating and difficult-to-read detail. The same points are repeated over and over in numerous ways, like one of those online personality tests that asks you the same question with slightly different wording. It seems that if he's covering things as simply as he does in the molecular section, then he doesn't need to go into too much detail, and certainly doesn't need the formal tone he takes. And if he's covering the more advanced stuff, then he also doesn't need to go into so much repetition that it seems he's writing for an unsophisticated audience.
I actually described the experience of reading it as "The most excruciating displeasure I've ever had".
Anyway, I think the book could have been 1/2 to 1/3rd the length it was.
Also, especially after reading Hawkins' book, I was especially disappointed that it didn't discuss the role of anatomy and brain structure. Certainly in a book that alleges to explain "from molecule to metaphor", there should be some discussion about brain regions, anatomy, cortex, etc. There wasn't even any relevant mention of how strong a role feedback plays.
He talks about molecules leading to synaptic transmission, but doesn't even show a spike train or time-voltage plot, as would be expected in even the most rudimentary exposition on neural function. A Hodgkins-Huxley model would also have gone well here as one level in between purely molecular function and neural transmission. He also could have/should have covered something about how memories are formed, at a molecular level (given the title of the book), rather than at the neural networks level, where he treats memories simply as synaptic weightings.
He seems to motivate his writing with how wonderful it'll be when we have intelligent machines, but he doesn't discuss anything other than some cheesy software models of "schemas", which seem to be able to parse natural languages, and says nothing of what people normally consider "real" intelligence. I think he should have made mention of the vast differences in processing power, power consumption (watts), operations/joule, memory, and associative power of the brain vs. real computers. I'm not a linguist or cognitive researcher, but my sense is that although schemas might be a theory of how it all works, it's not really what's going on. My analogy is that Newton had a theory of physics, but then Relativity adds more detail and classical physics "falls out" of more modern physics. I think schemas are like this: once we have a better (and artificially reproducible) theory of intelligence and language, then schemas as they are represented in this book will "fall out" of the theory.
Also, I'm puzzled about his insistence on equating intelligence with language. Although he is very clear about embodiment being a requirement for language, he makes little mention of how to get an artificial system to interact with the environment. Personally, I don't think you can begin to tackle language until you have a robot that can physically deal with the real world 1/10th as well as a real animal does -- something that can run on uneven terrain, use stereoscopic vision, and fumble for its keys in the dark. THAT to me is true embodiment, and until we have that figured out, I think theories about how language works with so-called schemas are too far out there to believe what's "really" going on in the brain. The schemas viewpoint of things seems waaay too much like computer science and trying to hack some way to parse a language without a lot of input.
The other stuff he should have mentioned, which, being an EE like me I think he should have, and being interested in artificial embodiment, is the research happening in neuromorphic engineering. People are using silicon to build neurally plausible *actual* systems, that work in hardware, without resorting to schemas, etc. They are mimicking cortex, visual areas, auditory function, etc., in real circuitry with similar orders of power, speed (faster than, even), and parallelism as real wetware. The advantage is of course that these circuits don't just simulate function like a digital computer, they *do* the function; he even says in his philosophical chapter at the end that the simulation of water, no matter how realistic, still isn't wet. Anyway, I'm getting off course here. The point is I think he gets way too far away from the physicality of the brain in trying to support his schemas idea.
Also, I don't think he did any of the research in his book himself. Not that that's such a bad thing, but given the level of wordiness, redundancy, and trying-to-explain-ness of it all, it almost seemed as though he were explaining his own theories. The book is a summary of other people's theories on language and not a whole lot on other things related to intelligence. One notably missing item in his explanation of language is the ability of humans to *hear* language. No serious mention of auditory processing or phonemic understanding is made. All the experiments on the computer are made with text.
Anyway, despite the shortcomings, some of the ideas are a little interesting. After all, I did finish the book. I think his ideas for language processing would be well suited for expert systems or other domains of limited scope. It was interseting to read about how non-english languages allow for variations in color understanding and word endings.
Overall I'm giving this book a crappy grade for style, readability, and comprehensibility. The subject matter really wasn't that difficult, but his writing styile made it so. The only reason it gets a 2 is because there was some interesting content. I thought it would extend my knowledge after reading Hawkins' book, but instead just made me more scornful of computer science oriented people trying to recreate "intelligence".
Also, some of my biases coming into this were having read Pinker's Blank Slate, Language Instinct, and How the Mind Works, and of course Hawkins' books. Also, I have done some graduate level research in neuromorphic engineering, which further biases my opinions towards that of feeling the need to embody physical stuff before you can think about tackling the language problem.
Date: 2006-08-23 Rating: 5
Review:
Deep Thoughts on the Relation between the Brain and the Mind
"How do our brains compute our minds?" If you have thought about this question, and even if you haven't, you will likely find this book fascinating. Dr. Feldman's answer is a theory developed after years of research by his group at U. C. Berkeley and workers at other places that holds that the embodiment of all thought, and of language in particular, is central to how this works.
While I certainly wouldn't classify this book as bedtime reading, these ideas are presented with great care so that the non-specialist reader can grasp the big picture, even if he doesn't get every detail. The details are there, so the book would also be useful to researchers and students, but Feldman charts a course through them so the layman (like me) can focus on the central notions and not get lost in nor be intimidated by the wealth of information provided.
And what a picture it is - the Neural Theory of Language (NTL) is shown to be consistent with all the experimental findings from relevant disciplines, and to provide a framework that allows for further work where the outstanding scientific questions can be posed with precision. Computer programs are described based on the NTL that have achieved remarkable results, and the ongoing work to address their shortcomings is exhilarating in its promise.
The writing is clear and straightforward, and Dr. Feldman displays flashes of wry humor and a becoming sense of humility regarding what they are about. This is meaty stuff, but if you are interested in doing some thinking about thinking, I definitely recommend this book.
Date: 2006-07-31 Rating: 5
Review:
Great Presentation of the Brain, the Mind, Language, and Thought
If you read "On Intelligence" [...] and it left you hanging, wanting more, this book is perfect for you. But, be prepared to spend a lot more time thinking.
As the book's title promises, Dr. Feldman walks the reader through an explanation of how language and thought have their roots in chemical reactions at the molecular level, a neuron firing. Through twenty-seven masterfully staged chapters, one is exposed to all of the mind-boggling complexity that leads to communication and understanding. I'm sure that each chapter is deserving of a book of its own but the author has managed to give you just the right exposure in a dozen or so pages.
I read the book over a period of a few weeks as I could seldom dedicate more time in a sitting than what it took me to process one chapter. But, this always left me looking forward to getting back to where I left off. It turned out to be a good approach for me; the book is full of backward references for review, and forward references to keep things in context.
Dr. Feldman takes a low-key approach (little sensationalism, mostly matter-of-fact descriptions) to concepts presented. I enjoyed the journey the book took me on; the highlight for me (background in computer science) was the computer models, the simulations. Computer programs can only do what they are told. So, constructing and running computer models is sure to point out any existing weakness in the author's understanding. But, when they exhibit the proposed behavior, how satisfying!
From the preface:
"This book proposes to begin integrating current insights from many disciplines into a coherent neural theory of language... Understanding language and thought requires combining findings from biology, computer science, linguistics, and psychology... If you want to understand how our brains create thought and language, there is a fair chance that this book can help."
There are nine sections to the book:
I. Embodied Information Processing
II. How the Brain Computes
III. How the Mind Computes
IV. Learning Concrete Words
V. Learning Words for Actions
VI. Abstract and Metaphorical Words
VII. Understanding Stories
VIII. Combining Form and Meaning
IX. Embodied Language
Dr. Feldman has spent twenty-five years working in the area and you can tell from the story he has put together. A very interesting one.
guy |