Towards an Integrated Connectionist/Symbolic Theory of Higher Cognition (REU Supplement)

本项目开发了一个研究项目，子符号范式（SSP），它利用连接主义计算作为数值神经计算和符号心理计算之间的桥梁。并形成一种整合，其中后者是前者的近似高级描述。虽然许多连接主义研究已经解决了认知心理学的问题和数据，但SSP关注的是高级认知中的补充数据和问题，这些数据和问题集中在语言学和认知科学的基础上。SSP已经取得了重大进展，包括(a)将符号结构表示为低级分布式活动模式的数学技术，以及(b)对这些表示执行大规模并行的结构敏感处理；(c)将联结主义计算分析为最大化和谐，这是表征的一种格式良好的度量；(d)在句法/语义学和音韵学领域对理论语言学的贡献，这些领域依赖于使用和声来衡量语言结构的良好性；(e)对上述结果的解释对认知科学基础的贡献。这项由本基金支持的新研究将建立在(d)，扩展语法理论作为一个结构良好的优化系统，以及(e)，系统地发展一种基于连接主义的策略类型学，以解释人类高级认知的惊人生产力。SSP正在为认知科学的新方法论的发展做出贡献，这种方法论不仅仅是其姐妹学科的总和，而是一种能够产生所有其他学科单独进行的结果的方法论。通过整合语言、哲学和数学分析、计算机模拟以及离散和连续计算的技术，这种新方法可以开发出更强大的连接主义系统、创新的符号形式、一般的认知原则构成了大脑和心灵的连接主义理论和符号理论统一的真正进展。

ABSTRACT This project develops a research program, the Sub-Symbolic Paradigm (SSP), which exploits connectionist computation as a bridge between numerical neural computation and symbolic mental computation, and forges an integration in which the latter is an approximate higher-level description of the former. While much of connectionist research has addressed problems and data from cognitive psychology, SSP has focused on complementary data and problems in higher cognition that are centrally located in linguistics and the foundations of cognitive science. SSP has already made significant progress, including (a) mathematical techniques for representing symbolic structures as lower-level distributed patterns of activity and for (b) performing massively parallel structure-sensitive processing on such representations; (c) the analysis of connectionist computation as maximizing Harmony, a well-formedness measure on representations; (d) contributions to theoretical linguistics in the areas of syntax/semantics and phonology that rely on using Harmony to measure the well-formedness of linguistic structures; and (e) contributions to the foundations of cognitive science deriving from the interpretation of the preceding results. The new research to be supported by the present grant will build on (d), expanding the theory of grammar as a well-formedness optimizing system, and on (e), systematically developing a typology of strategies for connectionist-grounded explanations of the striking productivity of human higher cognition. SSP is contributing to the development of a new methodology for cognitive science which is more than just the sum of those of its sister disciplines, a methodology capable of yielding results which are beyond the reach of all the other disciplines proceeding individually. By integrating linguistic, philosophical, and mathematical analysis, computer simulation, and techniques from both discrete and continuous computation, this new methodology can develop more powerful connectionist systems, innovative symbolic formalisms, and general cognitive principles that constitute real progress towards the unification of connectionist and symbolic theories of brain and mind.