LINGUIST List 14.1823

Tue Jul 1 2003

Review: Neurolinguistics: Arbib, ed. (2002)

Editor for this issue: Madhavi Jammalamadaka <madhulinguistlist.org>


What follows is a review or discussion note contributed to our Book Discussion Forum. We expect discussions to be informal and interactive; and the author of the book discussed is cordially invited to join in.

If you are interested in leading a book discussion, look for books announced on LINGUIST as "available for review." Then contact Simin Karimi at siminlinguistlist.org.


Directory

  • terry, The Handbook of Brain Theory and Neural Networks.

    Message 1: The Handbook of Brain Theory and Neural Networks.

    Date: Mon, 30 Jun 2003 11:34:49 +0000
    From: terry <terrylinguistlist.org>
    Subject: The Handbook of Brain Theory and Neural Networks.


    The Handbook of Brain Theory and Neural Networks, 2nd ed. Arbib, Michael A., ed. (2002) The Handbook of Brain Theory and Neural Networks, 2nd ed., MIT Press.

    Announced at http://linguistlist.org/issues/14/14-864.html

    Bilge Say, Middle East Technical University, Turkey

    DESCRIPTION OF THE BOOK

    This comprehensive reference volume is a collection of around 300 articles by various experts on mostly the fundamentals of neuroscientific brain theory, neural modeling for neuroscience as well as for higher level cognitive processes and selected applications of artificial neural networks. These articles are complemented by articles on the mathematics of the architectures and dynamics of artificial neural networks. The Handbook consists of three parts. Part I and Part II, written by Arbib himself, form an introductory passage into the bulk of 285 articles in Part III. Part I is a brief and condensed introduction to the elements of brain theory and neural networks. The biological neuron and its artificial models are introduced followed by a general introduction to the history and styles of neural modeling and a concluding section on artificial neural networks as dynamic and adaptive systems. Part II states 22 roadmaps organized under 8 major roadmaps, each roadmap consisting of a subset of articles in Part III selected within that interest area. For each roadmap, Arbib has written an introductory guided tour, giving some background on the theme of the roadmap and previewing how each article contributes to that interest area. Additionally, there is a meta-map introducing the 8 major roadmaps: Grounding Models of Neurons and Networks; Brain, Behaviour, and Cognition; Psychology, Linguistics, and Artificial Intelligence; Biological Neurons and Networks; Dynamics and Learning in Artificial Networks; Sensory Systems; Motor Systems; and Applications, Implementations, and Analysis. All articles in Part III (each of 3-6 pages of length, double column) have an Introduction section where the subject is introduced in an accessible manner and basic research pointers are given, and a Discussion section at the end, where a critical evaluation of the research in the subject area of the article is presented. In between, major contributions to research in the area of the article are reviewed, sometimes concentrating on the author(s)' own work, sometimes presented in a more comparative framework. Each article's main text is followed by three intra-Handbook pieces of information: Roadmap, the roadmaps this article is related with; Background, the articles in the Handbook that might serve as background reading for the article and Related Readings, related articles in the Handbook. These are further followed by a list of references, where major references are marked by a black diamond. The sheer size of the Handbook makes it impossible to review each article so I have decided to review only the 16 articles in the Linguistics roadmap. Of course, there will be other articles of interest for each reader; the roadmaps in Part II make it easier to form one's own roadmap. Language Processing (Shillcock) is a review of connectionist language processing models for both spoken and written language. Written with a cognitive science perspective, especially the strengths of such models in modeling change, namely, language acquisition, language breakdown and language evolution are emphasized. Emergence of complex linguistic behaviour from basic building blocks and gradedness of linguistic phenomena are demonstrated by example models. In Optimality Theory in Linguistics, Zuraw reviews the basic mechanisms of Optimality Theory and extensions such as stochastic constraint rankings. Advantages and disadvantages are presented alongside a brief review of how Optimality Theory has been inspired by neural networks. In Constituency and Recursion in Language, Christiansen and Chater reevaluate constituency and recursion in language from a connectionist perspective. They review various ways that these two concepts can be modeled in connectionist models and the implications of such models for language processing. Neurolinguistics (Gordon) introduces neurolinguistics as the study of all aspects of brain and language: basic methods and models, and how those models are extended by data obtained from normal humans as well as those with language deficits. In Reading, Holden and Van Orden review psycholinguistic experimental research and connectionist modeling on reading. They concentrate on the dichotomy as to whether word recognition is analytic or synthetic (or holistic) and show how nonlinear dynamic view of language processing can help resolve this issue. In Imaging the Grammatical Brain, Grodzinsky gives an account of how studies on lesioning data and functional neuroimaging can help shape hypotheses on the basic units of analysis in language and their neural plausibility. In Speech Processing: Psycholinguistics, Chater and Christiansen show how connectionist models can bridge the gap between experimental studies of psycholinguistics within brain theory and traditional linguistic accounts of language processing on issues such as modularity, concentrating on speech segmentation and aural word recognition. In Speech Production, Byrd and Saltzman summarize current research on speech production and show how a dynamic perspective brought about by connectionist models can blur the distinction between phonology and phonetics. Fowler, in Motor Theories of Perception, evaluate the hypothesis of motor systems being used in perception using motor theory of speech perception as a principle example. The studies presented in Language Evolution: The Mirror System Hypothesis (Arbib) can be taken as evidence for a motor theory of speech perception as well as for a theory of language evolution: mirror neurons found in the premotor cortex of the monkey (akin to language area in human brain) respond both when the monkey performs an action or when it perceives someone else doing the same action. In Language Evolution and Change, Christiansen and Dale review the role of connectionist modeling as tests for hypotheses about language evolution, language variation and change. Language Acquisition (Mac Whinney) is an exploration of how connectionist models have challenged (and have been challenged by) Chomskyan assumptions about a language acquisition device. Almor, in Past Tense Learning, reveals the past tense debate, various theoretical issues raised as a result of connectionist modeling of linguistic processes and structures for the acquisition of past tense inflectional forms. In Speech Recognition Technology, Beaufays et al. review some neural network models that are used in large vocabulary, continous speech recognition. Convolutional Networks for Images, Speech and Time Series (LeCun and Bengio) is a presentation of a specific, biologically inspired neural network architecture successfully used in applications such as handwritten character or speech recognition. Likewise, Bourlard and Bengio in Hidden Markov Models, introduce a specific architecture, namely hidden Markov models, as a special case of stochastic finite state automata and outline the benefits of their hybrid usage with artificial neural networks. It is stated in the Handbook that the second edition is an extensive update of the first: of the 266 articles in the first edition, only 9 have remained unchanged; all others have been updated or completely rewritten. About one third of the articles in this edition are on new topics; the emphasis of the shift in coverage has been from applications of artificial neural networks to cognitive modeling, including language and especially modeling for computational neuroscience. It is announced by the publisher that an online version of the Handbook will become available in August on CogNet (http://cognet.mit.edu/), MIT Press' electronic resources and community tools site for researchers in cognitive and brain sciences.

    CRITICAL EVALUATION

    The inspiration to this comprehensive Handbook is stated by its editor as the two great questions: ''How does the brain work?'' and ''How can we build intelligent machines?''. The collection of articles answer the second question more indirectly, mostly in the form of selected applications and mechanisms of neural network modeling. To me, the main attraction the Handbook brings is the shareability and mutual understandability of resources and knowledge among three, sometimes disparate approaches that take neural function and form as their starting points: computational neuroscience-''systematic use of mathematical analysis and computer simulation to provide better models of the structure and function of living brains''-, connectionist computational modeling for cognition-studies modeling human cognitive processes in terms of artificial neural networks, usually using more abstract and simple models than computational neuroscience- and neural computing-''design of machines (software and hardware) with circuitry inspired by but which need not faithfully emulate the neural networks of brains''. The articles are concise, well-edited and comprehensible given the right background. For the interested linguist though, some more technical background presented in a more gentle and accessible manner than the Handbook might be necessary to understand some of the articles in full. Thus, complementary resources such as Lytton (2002) and McLeod et al. (1998) might be advisable. The roadmaps and the various ways of cross-referencing used make the Handbook suitable for exploratory as well as reference reading. The only thing I found missing was a glossary, or at least some highlighting of definitional references in the Subject Index, given the vast terminology covered by the Handbook.

    REFERENCES

    Lytton, W. W. (2002). From Computer to Brain: Foundations of Computational Neuroscience. Springer.

    McLeod,P., K. Plunkett and E. T. Rolls (1998). Introduction to Connectionist Modelling of Cognitive Processes. Oxford University Press.

    ABOUT THE REVIEWER

    Bilge Say is a faculty member at the Cognitive Science Program of Middle East Technical University, Turkey. Her research interests lie mainly in Computational modeling for cognitive science and computational linguistics.