Connectionist symbol processing : dead or alive ?
Connectionist symbol processing : dead or alive ?
复制标题
联结主义符号处理：死还是活？
DOI：
--
复制
发表时间：
1999
期刊：
影响因子：
--
通讯作者：
B. B. Thompson
中科院分区：
文献类型：
--
作者：
Douglas S. Blank;M. Coltheart;J. Diederich;B. Garner;R. Gayler;C. L. Giles;L. Goldfarb;M. Hadeishi;B. Hazlehurst;M. J. Healy;J. Henderson;N. G. Jani;D. S. Levine;S. Lucas;T. Plate;G. Reeke;D. Roth;L. Shastri;J. Sougné;R. Sun;S. Wermter;W. Tabor;B. B. Thompson
1998年8月，Dave Touretzky在连接主义者的电子邮件列表中问道：“连接主义者的符号处理已经死了吗？”这个问题引发了一场有趣的讨论和思想交流。我们认为把这种交流记录在一篇文章中可能会有用。我们征求了大家的意见，这篇集体文章就是结果。在connectionist的电子邮件列表上公开征集了捐款。收到的所有捐款都经过两到三次非正式审查。几乎所有这些都被接受了，只是进行了不同程度的修改。鉴于贡献的数量和种类，这些文章涵盖了该领域的广泛工作，尽管绝不是完整的。本文中的文章性质各异：立场总结、个人研究总结、历史叙述、争议问题讨论等。我们并没有试图将这些不同的部分联系在一起，或者将它们组织在一个连贯的框架内。尽管如此，我们认为读者会觉得这本合集很有用。在网络中实现符号处理是解决困扰符号系统的许多问题的良好开端。Tony Plate将HRR应用于类比就是一个很好的例子[147]。使用连接主义表示和方法，在模拟生成MAC/FAC系统中消除了昂贵的符号相似性估计过程[60]。不幸的是，整个MAC/FAC混合模型（像许多这样的模型一样）有一个致命的规律，阻止它成为一个自主的、灵活的、创造性的、智能的（类比制造）机器：整个系统组织仍然是严格的“符号”。他们的方法要求类比被编码为符号和结构，在类比过程中没有给感知或语境效应留下空间（关于这个问题的详细描述，见Hofstadter 86]）。由于这些原因，在神经网络（甚至是真正的神经元）中完全实现genner的刚性框架是没有帮助的。与大多数混合系统一样，Plate的混合解决方案解决了MAC/FAC纯符号系统的许多问题。毫无疑问，混合系统比它们的象征性亲戚要好。然而，无论符号和结构在哪里，我们似乎都面临着其他的脆性和刚性问题。
Preface In August 1998 Dave Touretzky asked on the connectionists e-mailing list, \Is connectionist symbol processing dead?" This query lead to an interesting discussion and exchange of ideas. We thought it might be useful to capture this exchange in an article. We solicited contributions, and this collective article is the result. Contributions were solicited by a public call on the connectionists e-mailing list. All contributions received were subjected to two to three informal reviews. Almost all were accepted with varying degrees of revision. Given the number and variety of contributions, the articles cover a wide, though by no means complete, range of the work in the eld. The pieces in this article are of varying nature: position summaries, individual research summaries, historical accounts, discussion of controversial issues, etc. We have not attempted to connect the various pieces together, or to organize them within a coherent framework. Despite this, we think, the reader will nd this collection useful. Implementing symbol processing in networks was a good rst step in solving many problems that plagued symbolic systems. Tony Plate's HRR as applied to analogy is a great example 147]. Using connectionist representations and methodologies, an expensive symbolic similarity estimation process was eliminated in the analogy-making MAC/FAC system 60]. Unfortunately, the entire MAC/FAC hybrid model (like many such models) has a fatal aw that prevents it from leading to an autonomous, exible, creative, intelligent (analogy-making) machine: the overall system organization is still rigidly \symbolic". Their method requires that analogies be encoded as symbols and structures, which leaves no room for perception or context eeects during the analogy making process (for a detailed description of this problem, see Hofstadter 86]). For these reasons, implementing Gentner's rigid framework completely in a neural network (or even real neurons) won't help. Plate's hybrid solution, like most hybrid systems, solved many problems of the MAC/FAC purely-symbolic system. No doubt, hybrid systems are better than their symbolic relatives. However, wherever symbols and structures remain, we seem to be faced with other problems of brittleness and rigidity.