From Molecule to Metaphor : A Neural Theory of Language
The continuing development of the Neural Theory of Language can be tracked at the ECGweb wiki .
M2M Errors and some Updates, By Section
Almost all of the corrections appear in the paperback edition, March 2008.
Front Matter p.xi, line 2: ", said, " -> "said, "
p. xii, line -5: "principle concern" -> "principal concern"
Acknowledgements Mary -> Margy Avery, add Colleen Lanick
Section 1. Embodied Information Processing
p.5, line -5, "appearing to closer" -> "appearing closer"
Section 2. How the Brain Computes p. 38, line -10; add two paragraphs:
Here is an additional important fact about neural computation and learning. Current evidence suggests there is no erasing in the adult brain. As we will see in Chapter 6, long term memory of facts, skills, or situations is captured by structural changes in the connections between neurons. There is no process for selectively reversing these changes. This is why it is so hard to alter your behavior patterns or to change the beliefs of others. The only known neural mechanisms for changing behavior involve inhibiting a pathway or bypassing it with a more active alternative.
This does not imply that repressed memories are always accurate. My memory of last week is spotty and research has shown that people misremember even important events like their wedding. Some neurons die naturally and trauma or disease can cause massive destruction. But the fact that there is no mechanism for selectively eliminating unwanted neural connections has a fundamental role in our mental life and institutions.
p. 46, pg 4: "...generated in
the cell nucleus"
p. 51, pg 2: "... a nucleus, which (carries out) -> (is involved in) the manufacture of new molecules..."
p52, Caption for 4.4 should be "Drawing of a receptive channel, shown
protruding through a cell wall. On the right, the channel is open, allowing
neurotransmitters to enter the cell.
p. 53, line "whether or not
p. 69, color version of Figure 5.2
Greenspan, R.J. An Introduction to Nervous Systems, Cold Spring Harbor Press, 2007.
A very well written and informative book on the information processing abilities of simple animals, including
paramecium, which is a single cell.
Spitzer, M. The Mind in the Net. MIT Press, 1999.
This is an excellent book relating aspects of brain function to a variety of neural net models. It goes well beyond M2M
on biochemistry and on clinical applications. I should have known about it years ago.
Section 3. How the Mind Computes p.88, There are even stronger timing constraints known for visual tasks. A series of papers from Simon Thorpe and colleagues shows that many visual decisions are so fast that the temporally first spikes must carry the crucial information.
Van Rullen R, Thorpe S (2001) Rate Coding versus temporal order coding: What the Retinal ganglion cells tell the visual cortex. Neural computation, 13, pp1255-1283.
p. 90, Caption for Figure 7.2 should be:
Priming Model. The noun and verb meaning of "rose" compete; semantic activation of flower provides a priming advantage to the noun sense.
p. 94, A paper by Richardson and Matlock, forthcoming in Cognition shows that fictive motion elicits eye movements along trajectories in consistent and predictable ways.
p. 97, Gallese and Lakoff provide a detailed examination of the philosophical notion of concept from the embodied perspective.
Gallese, V., & Lakoff, G. (2005). The Brain's Concepts: The Role of the sensory-motor system in reason and language. Cognitive Neuropsychology, 22, 455-479.
p. 112, line 17 "pitch" -> "tone"
p.116, Caption for 9.3 should be " Layered PDP connectionist network. Circles depict nodes and solid arrows are forward connections. The boxes depict training inputs, see text."
p120, line -14 "his point" -> "this point"
Pecher, D. and R. Zwaan (2005) Grounding Cognition . Cambridge, Cambridge University Press.
This is an excellent collection of articles on the behavioral and linguistic evidence for the embodied nature of concepts. It is consistent with the story of this book, but features a variety of different perspectives. I should have included it.
Section 4. Learning Concrete Words
p. 134: the section header "Syntactic Cues ..." is misplaced; it belongs before the previous paragraph on p. 133, line 14.
p. 137, line -2 " located in" -> " identified with"
p. 158, line -1 "if" -> "of"
p.168, Figure 13.2 . The arrow heads should point into the "apply hand" ellipse
Section 5. Learning Words for Actions
p176 line -8 double hyphen
Section 6. Abstract and Metaphorical Words
p190, Recent work shows that language interference with color grouping is restricted to the left (language) hemisphere.
"Whorf hypothesis is supported in the right visual field but not the left." , A. L. Gilbert, T. Regier, P. Kay and R. B. Ivry, Proceedings of the National Academy of Sciences. 103, 489-494. 2006
p. 216, color version of Figure 17.1
Section 7. Understanding Stories
p229, Fig 18.1 The label above the hexagon in the center bottom should be "step", not "walk"
p229, line -6, "depends" -> "depending"
p230, line -7, "previously" -> "above"
p232, line 1, "interrupted" -> "suspended"
p236, line 4 "figure 13.2" -> "figure 11.2"
p240, line 8 "autarchy" -> "autarky"
p246, line -2, "Economic state" -> "Current Economic state"
Section 8. Combining Form and Meaning p261 line -14 "Turkish" -> "Turkic"
p262 chapter 11 -> chapter 15 p266 chapter 11 -> chapter 15
p268 line 22, "cases" -> "features"
p288 "table 10.1" -> "table 11.1"
"table 10.2" -> "table 11.2"
"figure 10.1" -> "figure 11.1"
Section 9. Embodied Language
p308 ref -> Narayanan et al. 1998
p312 add CHILDES reference
p316, line -1 "black" -> "dark"
p318, line -11 "figure 10.2" -> "table 10.2"
p331, People interested in embodied philosophy should also track the work of Jesse Prinz - philosophy.unc.edu/prinz.html
Bloom, P. should be 127
add: event structure 207-211, 245-247,
executing schema, see X-schema
References and Further Reading
Additional References are listed or linked in the appropriate section.
Chang, N. thesis date should be 2007
Ramachandran should be "V.S."
There are many books, movies etc. about AI and robots that talk. These three are especially recommended because they are intrinsically good and also have significant scientific input. Each of them gives a good feel for the issues involved in trying to build a system that could understand language.
For David Lodge's novel. Thinks, Aaron Sloman was the scientific consultant.
For Richard Power's Galatea 2.2, it was Gary Dell, UIUC.
Christos Papadimitriou, author of Turing, is himself a distinguished computer scientist. He got further input from Berkeley AI colleagues, but did not take my suggestions about neural theories of language.