联结主义与批评话语分析

符号主义和联结主义是认知科学的两大研究范式,早期基于系统功能语法的批评话语分析在不经意间吸收了符号主义范式的理念。20世纪80年代以来,随着符号主义的日渐式微及更多认知心理学实证研究的发现,联结主义得以卷土重来,其构建的信息加工模型的动态性、整体性和高效性对批评话语分析阐释理想读者对话语的浅层加工及相应的心理表征具有一定的借鉴意义。     

认知心理学两种研究范式的争论与反思

符号加工和联结主义是认知心理学的两种研究范式,二者在认知心理学内部产生了一系列的争论。在一般 认知问题上的争论主要表现在语义理解、心理表征以及一些具体的认知问题上;在方法论问题上的区别主要表现在是 结构模拟、黑箱方法、还原主义还是结构与功能模拟、灰箱方法和整体主义上。两种范式争论的本质问题是究竟哪种 范式更加符合认知过程的本质。联结主义范式似有取代符号加工范式的趋势,但它也不能完全揭示人类的高级认知功 能,因此出现了行为主义的进化论研究范式。     

联结主义是一种生物模型还是功能架构?

一些联结主义者一直将联结主义模型与神经生物学之间的结构和加工相似性,作为其生物合理性的根据。但是这些生物合理性的说明是成问题的：一方面,加工相似性是高度简化的,并且加工相似性对于认知解释并不是充分和必要的;另一方面,联结主义中的一些主要功能及算法示例,目前还不能找到神经生物学基础。所以联结主义和经典主义一样,是一种与实现层面无关的认知功能描述。     

关于符号处理范式在认知科学中的地位和前景

文章在概述符号处理范式（或“ＰＳＳＨ”）的理论框架和基本原理的基础上,就联结主义对该范式的挑战以及符号处理论者的应答与辩护作了客观的述评。论文表明符号处理范式仍具有较强的生命力,并不因为它受到联结主义的挑战就会失去自身存在的价值。从“信息处理”基础假定的合理性、ＰＳＳＨ具有的解释力以及它与联结主义的内在联系来看,断言联结主义即将取代符号处理范式尚为时过早     

从哲学视角看人工智能的发展——对“中文屋论证”的批判性考察

1980年,美国著名哲学家塞尔提出的"中文屋论证",对人工智能领域发起深刻挑战,其引起的争论甚至延续至今。通过对这一论证的逻辑结构及哲学蕴涵的深入考察,发现"中文屋"思想实验实际上缺乏实证性的依据,并且只对符号主义模式人工智能构成威胁,对于联结主义模式人工智能论证无效,甚至塞尔的很多观点实际上支持了联结主义。因此,这实际上为人工智能的发展提供了一条可能的进路。     

联结主义与思想的构成性难题

经典主义通过"思想的构成性论证"指出:联结主义表征要么缺乏构成性,要么只是经典主义的落实形式。落实的联结主义和消除的联结主义成为回应这一主张的两极。构成性质在思想中的表现和地位,是认知架构争论的焦点。     

论认知科学联结主义模型的前景

本文在描述联结主义模型的基本特征的基础上,从理论优势和应用价值两个方面对联结主义的前景作了初步分析。联结主义以其新的计算模式、人脑作为并行处理系统的真实工作模型、解决常识知识问题的可能性等,而明显优于符号处理模型。在神经网络的技术应用方面也有广阔的前景     

ChatGPT能够理解吗?从符号奠基问题的角度看

符号奠基问题来自约翰·塞尔的中文屋论证,它指的是一个基于形式符号系统的计算机程序如何获取和理解语言意义的问题。随着技术的发展,联结主义现已成为人工智能的主流范式,ChatGPT正是联结主义的产物之一。和中文屋一样,ChatGPT也能在一个“窗口”中和人类进行对话,但其不是纯粹的符号系统,而是一种基于神经网络的大型语言模型。和其他联结主义模型类似,ChatGPT也具有“分布式表征”“自我进化”等特性,这些特性赋予了ChatGPT一种内在论式的语言理解能力。同时,一种基于ChatGPT的机器人方案也能为解决符号奠基问题提供新的思路。如果该方案能成功,人们就很难说,面前的机器人没有智能。     

学习过程中的两种认知研究范式分析及启示

学习过程研究不仅在心理科学领域．而且在教育科学领域都处于十分重要的基础地位。目前教育技术领域中．虽然认知取向的学习过程研究成果已经得到广泛认知和应用．但由于学习过程是一个复杂的心理过程，已有的符号信息加工认知研究范式已呈现出一定的局限性．新联结主义认如研究范式的出现培学习研究带来了全新的视角．从符号信息加工研究到新联结主义研究的转换已是一种明显的趋势。升学习过程的符号信息加工认知研究范式进行深入剖析。阐释新联结主义研究范式下的学习研究成果．将大大有助于思考学习机制认知研究范式变换教育技术带来的影响及启示。     

人工智能研究进路的范式转化

在人工智能发展的50年中,人们对人工智能的认识经历了多次变化和范式转化过程。从符号主义、联结主义到行为主义．人工智能研究中智能思想的形成以及认识的变化吸取多元学科理论观点和方法,在假说与实证、改变与重构的经验积累及实践选择中发展。这种以模拟人类思维、心灵复杂现象的人工智能学科．其研究势必遵循多元的研究取向和研究方式,并逐步向实体智能化（逼近于人体智能）发展。     

Introduction to Connectionist Modeling in Education

The basic features of connectionist modeling are reviewed with respect to educational issues. Connectionist simulation is becoming a major theoretical and applied tool for educational research. Examples of connectionist learning are reviewed including simulations of learning for: text to speech, learning verb endings, addition tables, and learning from instruction. The contrast between connectionist and symbolic processing and the use of connectionism in cognitive science research, biological plausibility, and potential connectionist technology are commented upon. Cautions and guidelines are provided regarding the dangers of premature extrapolation of findings from the connectionist simulations to early impact of educational practice.     

Explaining Learning: From analysis to paralysis to hippocampus

This paper seeks to explain learning by examining five theories of learning—conceptual analysis, behavioural, constructivist, computational and connectionist. The first two are found wanting and rejected. Piaget's constructivist theory offers a general explanatory framework (assimilation and accommodation) but fails to provide an adequate account of the empirical mechanisms of learning. Two theories from cognitive science offering rival explanations of learning are finally considered; it is argued that the brain is not like a computer so the computational model is rejected in favour of a neurally‐based connectionist model of learning.     

CONNECTIONISM AND THE EMERGENCE OF KNOWLEDGE

Procedures for creating used by connectionist networks with hidden units are analyzed. With these networks it is possible to create by modification of the inputs, by modification of collector properties, by modification of connections between units and variation of thresholds of activation, by modification of constraints imposed on the network, by creation of new hidden units or by creation of new outputs. The interest of connectionist networks is in simulating various processes to allow analysis of modifications that are efficient and capable of concretizing the consequences of choices. Networks must be designed as tools capable of helping formalize problems, and not as replicas of cognitive processes.     

Reading: three to thirteen

Askew and his colleagues (1997), put forward a range of important notions as to what might constitute effective approaches to the teaching of numeracy. In particular, they place great emphasis on the way in which teachers who make links between different areas of mathematics are seen to be effective teachers of numeracy. (Askew et al term this approach ‘connectionist’)

The work reported here suggests that the notion of classifying a teacher as, say, connectionist is too simplistic. We report observations which suggest that effective teachers of numeracy incorporate a range of approaches into their teaching.     

Learning the norm of internality: NetNorm,a connectionist model

The objective of the present article is to show that connectionist simulations can be used to model some of the socio-cognitive processes underlying the learning of the norm of internality. For our simulations, we developed a connectionist model which we called NetNorm (based on Dual-Network formalism). This model is capable of simulating the implicit learning of this social norm through a simple process of situation/reinforcement association. In a second simulation, we simulated a non-linear learning in order to reproduce the triple inflexion in the internality learning curve that has been observed for children entering high school. The results support those previously reported in the literature. Moreover, our simulations could be used to develop new experimental predictions.     

Connectionism and learning to read: Steps towards a psychologically plausible model

We consider the prospects and need for a psychologically plausible connectionist model of the development of word recognition skills. We present an outline of the Seidenberg and McClelland (1989) distributed, developmental model of word recognition and naming. We emphasize certain incompatibilities between this model and the assumptions underlying it and psychological evidence from studies of children learning to read. In particular, we emphasize the importance of phonological skills as precursors and facilitators of learning to read. We argue that it may be possible to develop a connectionist model of the development of word recognition skills which, by having a built in knowledge of phonology, will be more consistent with evidence from studies of both normal and dyslexic children.     

解读建筑漫游动画——以广东轻工职业技术学院为例

通过广东轻工职业技术学院建筑漫游动画的制作流程去解读建筑漫游动画的制作方式、创作流程,使人们对建筑漫游动画的认知不仅仅停留在表层,能完整的、更深层次的了解建筑漫游动画。     

Parental beliefs about child development and parental inferences about actions during child-rearing episodes

The way parents’ beliefs on child development support the elaboration of practical inferences during everyday child-rearing episodes was examined. We contrasted two models based respectively on the classical and the connectionist view of schema approaches. According to the classical view, parents activate preformed packages of beliefs in order to produce inferences whereas under the connectionist view, they activate the network of interconnected episodic traces that better fits the information provided by the situation. In the former case, the quality of the inferences depends on the activation of the proper schema whereas in the latter case it depends on the structure of the information given. Two experiments were designed in which parents holding a particular global belief about child development (either constructivism or environmentalism), were presented with a target couple with similar or different views with respect to them. In Experiment 1, constructivist parents bearing in mind the couple’s belief had to judge a set of words describing the couple’s image as parents, their educational goals as well as a number of sentences describing the couple’s child-rearing practices during hypothetical episodes. In Experiment 2, environmentalist parents had to judge the couple’s practices and the amount of information presented about the couple’s ideas as well as its plausibility were manipulated. The results indicate that the accuracy and speed in the production of inferences depends on the information presented in the task. When the information is embedded in episodes and a full, plausible and distinctive account of the couple’s belief are provided, then the production of inferences is performed faster and with more accuracy. The results are discussed in terms of the classical and connectionist views of schema approaches.     

试析认知心理学的三种研究取向及其未来发展

认知心理学是现代西方心理学的一个重要的理论流派。它共有三种研究取向,分别是符号加工取向、联结主义取向和生态学取向。这三种研究取向都不能单独说明认知的本质。认知心理学未来的发展必然会出现实验室研究与现场的生态学研究相融合的趋势。     

Applying a connectionist description of feedback timing

This investigation considers the effects of feedback on memory with an emphasis on retention of initial error responses. Based on a connectionist model (Clariana, 1999a), this study hypothesized that delayed-retention memory of initial lesson responses would be greater for delayed feedback compared to immediate feedback, that feedback effects will be greatest with difficult items, and that there would be a disordinal interaction of feedback timing and item difficulty. High school students (n = 52) completed a computer-based lesson with either delayed feedback, single-try immediate feedback, or multiple-try immediate feedback. There was a significant difference for type of feedback, with retention test memory of initial lesson responses greater under delayed feedback than under immediate feedback. Also, instructional feedback effects varied depending on lesson item difficulty. The findings indicate that a connectionist model can explain instructional feedback effects.