On Being Systematically Connectionist

Niklasson, Lars; Gelder, Tim van

doi:10.1111/j.1468-0017.1994.tb00227.x

Cited by 49 publications

(37 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, an extension of the RAAM networks of Pollack (1990) that was proposed by Chalmers (1990) and Niklasson and Gelder (1994) uses two separate networks, one to encode (and decode) all possible instances of variable in a fixed input bank, and another to transform the encoded representations in a systematic way; the structure of these models precisely parallels the division between encoding and computation in standard symbolic models. Such networks are best seen as implementations of those models, not as genuine, nonsymbolic alternatives.…”

Section: Alternative Architecturesmentioning

confidence: 99%

Rethinking Eliminative Connectionism

Marcus

1998

Cognitive Psychology

271

180

View full text Add to dashboard Cite

Humans routinely generalize universal relationships to unfamiliar instances. If we are told ''if glork then frum,'' and ''glork,'' we can infer ''frum''; any name that serves as the subject of a sentence can appear as the object of a sentence. These universals are pervasive in language and reasoning. One account of how they are generalized holds that humans possess mechanisms that manipulate symbols and variables; an alternative account holds that symbol-manipulation can be eliminated from scientific theories in favor of descriptions couched in terms of networks of interconnected nodes. Can these ''eliminative'' connectionist models offer a genuine alternative? This article shows that eliminative connectionist models cannot account for how we extend universals to arbitrary items. The argument runs as follows. First, if these models, as currently conceived, were to extend universals to arbitrary instances, they would have to generalize outside the space of training examples. Next, it is shown that the class of eliminative connectionist models that is currently popular cannot learn to extend universals outside the training space. This limitation might be avoided through the use of an architecture that implements symbol manipulation.

show abstract

Section: Alternative Architecturesmentioning

confidence: 99%

Rethinking Eliminative Connectionism

Marcus

1998

Cognitive Psychology

271

180

View full text Add to dashboard Cite

show abstract

“…For example, Frank (2006) noted that the models by Bodén (2004) and Hadley et al (2001) only managed to behave systematically because they were specifically tailored for that purpose. Likewise, the systematic connectionist model proposed by Niklasson and van Gelder (1994) was criticized by Hadley (1994a) and Phillips (1998) for depending on hand-crafted input representations that explicitly encoded syntactic class information. In addition, Haselager and van Rappard (1998) remarked that the model required a very extensive and carefully arranged training regime.…”

Section: Demonstration Versus Explanationmentioning

confidence: 99%

“…A connectionist model that performs such operations could therefore be considered an implementation of a symbol system. For instance, the Recursive Auto-Associative Memory (RAAM) models proposed by Chalmers (1990) and Niklasson and van Gelder (1994), consist of two networks: The first learns to encode the possible instantiations of a variable, and the second performs transformations over the resulting representations. This architecture, so Marcus (1998b) argues, 'precisely parallels the division between encoding and computation in standard symbolic models ' (p. 270) and is therefore not a relevant counterexample to Fodor and Pylyshyn's (1988) claims.…”

Section: Implementation Of a Symbol Systemmentioning

confidence: 99%

Connectionist semantic systematicity

2009

View full text Add to dashboard Cite

Fodor and Pylyshyn (1988) argue that connectionist models are not able to display systematicity other than by implementing a classical symbol system. This claim entails that connectionism cannot compete with the classical approach as an alternative architectural framework for human cognition.We present a connectionist model of sentence comprehension that does not implement a symbol system yet behaves systematically. It consists in a recurrent neural network that maps sentences describing situations in a microworld, onto representations of these situations. After being trained on particular sentences-situation pairs, the model can comprehend new sentences, even if these describe new situations. We argue that this systematicity arises robustly and in a psychologically plausible manner because it depends on structure inherent in the world.

show abstract

“…In the classical RAAM and its variants, systematicity is based on the properties of the neural code; similar concepts are represented by topologically close vectors, which enables relevant generalization (Pollack, 1990;Niklasson and van Gelder, 1994). Bodén and Niklasson (2000) recently proposed an extension of RAAM that learns the properties of objects, based on their relations within the training set.…”

Section: Resultsmentioning

confidence: 99%

“…For example, the RAAM has been used to perform holistic computations, e.g. active-passive transformations of grammatical phrases (Chalmers, 1990;Chrisman, 1991), or logical transformations of symbolic expressions (Niklasson and van Gelder, 1994). Niklasson and van Gelder noted that holistic processes in RAAM are sensitive to structure, and therefore address Fodor and Pylyshyn's second argument.…”

Section: Introductionmentioning

confidence: 99%

Linear recursive distributed representations

Voegtlin

Dominey

2005

Neural Networks

View full text Add to dashboard Cite

Connectionist networks have been criticized for their inability to represent complex structures with systematicity. That is, while they can be trained to represent and manipulate complex objects made of several constituents, they generally fail to generalize to novel combinations of the same constituents. This paper presents a modification of Pollack's Recursive Auto-Associative Memory (RAAM), that addresses this criticism. The network uses linear units and is trained with Oja's rule, in which it generalizes PCA to tree-structured data. Learned representations may be linearly combined, in order to represent new complex structures. This results in unprecedented generalization capabilities. Capacity is orders of magnitude higher than that of a RAAM trained with back-propagation. Moreover, regularities of the training set are preserved in the new formed objects. The formation of new structures displays developmental effects similar to those observed in children when learning to generalize about the argument structure of verbs.

show abstract

On Being Systematically Connectionist

Cited by 49 publications

References 17 publications

Rethinking Eliminative Connectionism

Rethinking Eliminative Connectionism

Connectionist semantic systematicity

Linear recursive distributed representations

Contact Info

Product

Resources

About