Fodor, J. and B. McLaughlin (1991). Connectionism and the problem of Systematicity: Why Smolensky's Solution doesn't work. in Horgan and Tienson (eds.) Connectionism and the Philosophy of Mind, Kluwer Academic Pub. The Netherlands. 45 23.06.99 17:37  Home/Search   Document Not in Database   Summary   Related Articles   Check

....we are to have a substantive connectionist account of systematicity is: rst, it must be made clear what property D is, and second it must be shown that D is a property that connectionist systems can have by law. J. Fodor and B.P. McLaughlin, Connectionism and the Problem of Systematicity [35] 3.1 Iterated Function Systems Consider again the RAAM decoder in Figure 2.3. The behavior of this decoder can be described by the set of equations: T k 6 6 6 6 6 6 6 6 7 7 7 7 7 7 7 7 = f B B B B B B B B 2 6 6 6 6 6 6 6 6 7 7 7 7 7 7 7 7 5 ....

....illustrated a principle by which a neural network may satisfy the traditional cognitive science objections to connectionism. Unfortunately, many earlier attempts at addressing this issue have su ered from overly simplistic assumptions that enabled them to be defeated as straw man arguments [36][35]. Though we cannot claim to have resolved this fundamental debate, we hope to have nudged it in what we perceive as the right direction. 8.2 Future work We can envision at least two directions for In nite RAAM. 8.2.1 Higher dimensional attractor networks The error reduction gained from moving ....

J.A. Fodor and B. McLaughlin. Connectionism and the problem of systematicity: Why Smolensky's solution doesn't work. Cognition, 35:183204, 1990.

....more philosophical criticism was presented by Fodor and Phylyshyn [Fodor and Pylyshyn, 1988] who addressed the capacity of these models to account for sistematicity and compositionality. This critical position, somehow favourable to the symbolic approach, led to a violent debate [Smolensky, 1988, Fodor and McMaughlin, 1990] that showed how the cognitive science community was divided on this topic. Radial Basis Function Networks are a particular kind of Neural Network which until now, are little exploited in Cognitive Modelling. They are characterized by 6 having a transfer function in the hidden units layer, ....

Fodor, J. and McMaughlin, B. P. (1990). Connectionism and the problem of systematicity: why smolensky's solution doen't work. Cognition, 35:183--204.

.... these systematically according to the syntactic structure of the representation (which directly expresses the structure of the represented expression) They pointed out that connectionists neither can exhibit nor supply a novel explanation to this systematic ability of cognitive agents (cf. Fodor and McLaughlin, 1990), since connectionists cannot generate syntactically structured representations or apply processes sensitive to this structure without resorting to symbolic representations. Rather than accepting the implementational role for connectionism, the validity of the premises of these claims were ....

Fodor, J. A. and McLaughlin, B. P. (1990). Connectionism and the problem of systematicity -- why Smolensky's solution doesn't work. Cognition, (35):183--204.

....in between the perfect solution equivalent to the implementation of a symbolic system and a solution consisting in pure rote learning of the training set with disregard for its combinatorial structure. Such a result can be seen as an answer to the arguments raised in (Fodor and Pylyshyn, 1988) and (Fodor and McLaulghlin, 1990) claiming that connectionist systems would have to implement a symbolic system to display generativity and systematicity. 139 8.2 Shortcomings The limitations of this thesis are numerous. While the simple approach taken allowed a clear interpretation, and was thus preferable, in our opinion, to ....

J.A. Fodor and B.P. McLaulghlin. Connectionism and the problem of systematicity: Why Smolensky's solution won't work. Cognition, 35:183--204, 1990.

.... can be met (e.g. Pollack, 1988, 1990; Smolensky, 1990; Chalmers, 1990; Niklasson and Sharkey, 1992; Brousse, 1993) Unfortunately, it has generally been unclear whether these models actually do have this implication (see, for instance, the extensive philosophical debate in Smolensky, 1988; Fodor and McLaughlin, 1990; van Gelder, 1990, 1991; McLaughlin, 1993a, 1993b; Clark, 1993) Indeed, we know of no major supporter of classical orthodoxy who has felt compelled, by connectionist models and arguments, to concede in print that connectionists have in fact delivered a non classical explanation of ....

....render Fodor and Pylyshyn s vague, amorphous notion rigorous and precise, he finds it fractures into a number of kinds and levels. Thus, while the initial notion of systematicity was not tied to learning in any particular way it asserted merely that cognitive capacities always come in clumps (Fodor and McLaughlin, 1990) Hadley found that to turn the systematicity challenge into a specific modeling problem for connectionists, it had to be reformulated as a learning problem: given that a system can acquire some capacities in a clump, does it automatically acquire other capacities in that clump The precise ....

Fodor J. A. and McLaughlin B. P. 1990: Connectionism and the problem of systematicity: Why Smolensky's solution did not work, Cognition, 35, 183 - 204.

....the challenge can be met in principle (for the models, see Pollack 1988, 1990; Smolensky 1990; Chalmers 1990; Niklasson Sharkey 1993; Brousse 1993; etc. Whether these models can or do meet the challenge has been the subject of much philosophical debate (Smolensky 1988; van Gelder 1990, 1991; Fodor McLaughlin 1990; Sharkey Jackson 1992; McLaughlin 1993a, 1993b; Clark 1993; Matthews 1994; etc. A consensus has emerged that the only way to deliver a non classical explanation of systematicity is to construct models that utilize representations that are compositionally structured, and hence can be the basis ....

....introduced into cognitive science for the first time in Fodor Pylyshyn s 1988 paper. In view of this, it is a surprising fact that Fodor Pylyshyn characterize systematicity only very vaguely. For example, Fodor has described systematicity as the idea that cognitive capacities come in clumps (Fodor McLaughlin 1990, p 184) In this paper, we focus on just one component of the general phenomenon, the systematicity of inference. The 1988 paper gestured at this phenomenon as follows: organisms should exhibit similar cognitive capacities in respect of logically similar inferences. p 47) You don t, for ....

Fodor J. A. & McLaughlin B. P., (1990), Connectionism and the problem of systematicity: Why Smolensky's solution did not work, Cognition, 35, pp 183 - 204.

....1991) 5 Unfortunately, in some major debates on this matter attention has been directed primarily towards some of the weaker notions. There has in particular been a tendency to equate distributed representations with microfeatural representations (as in the dreary cup of coffee debate e.g. (Fodor McLaughlin, 1990; Smolensky, 1991) 6 In this equation, f(s) specifies the input pattern, where s is an index for input units; g(t) specifies the hidden layer pattern, where t is an index for hidden units; w ts is the weight from unit s to unit t; and q t is a bias on unit t. 7 More precisely: suppose we have ....

Fodor, J. A., & McLaughlin, B. P. (1990) Connectionism and the Problem of Systematicity: Why Smolensky's Solution Doesn't Work. Cognition, 35, 183-204.

....it can be regarded as an implementation of a classical (relational) system. Whether, in fact, connectionism does provide an alternative theoretical basis is still debated (Fodor, 1997; Smolensky, 1995) and given the extensive literature on this issue (see Fodor Pylyshyn, 1988; Smolensky, 1988; Fodor McLaughlin, 1990; van Gelder, 1990, among others) it would be inappropriate to address it here. Nevertheless, alternative or not, the problem of determining what connectionist properties support systematicity remains. Acknowledgements We thank Yoshiro Miyata for the PlaNet simulator. The first author was ....

Fodor, J. A., & McLaughlin, B. P. (1990). Connectionism and the problem of systematicity: Why Smolensky's solution doesn't work. Cognition, 35, 183--204.

.... and process these systematically according to the syntactic structure of the representation (which directly expresses the structure of the represented expression) They pointed out that connectionists neither can exhibit this ability of cognitive agents, nor supply a novel explanation to it (cf. Fodor and McLaughlin, 1990), since connectionists cannot generate syntactically structured representations or apply processes sensitive to this structure without resorting to symbolic representations. Rather than accepting the implementational role for connectionism, suggested by Fodor and Pylsyhyn, the validity of the ....

Fodor, J. A. and McLaughlin, B. P. (1990). Connectionism and the problem of systematicity { why Smolensky's solution doesn't work. Cognition, 35:183-204.

....debate and its implications for cognitive architecture have had little if not no impact on the day to day practices of the connectionists version of cognitive science. Despite the seminal papers of Fodor and Pylyshyn (1988) and Smolensky (1988) and numerous rebuttals from both sides (see Fodor McLaughlin, 1990; Fodor, 1997; Smolensky, 1995b, among many others) architectures that predate the debate still appear to be the connectionists first choice (e.g. Plunkett Elman, 1997) In fact, enthusiasm for the recent Rethinking Innateness series (see the cover note to Elman, Bates, Johnson, ....

....the feedforward and recurrent networks. Such networks permit states that have no counterpart in higher cognition (specifically, states of weak systematicity) Yet, the negative results do not imply that all of connectionism is unsuitable for cognitive modeling 18 . What is does tell us, to echo Fodor and McLaughlin (1990), is that there must be additional principles above those of units and weights that organise network connectivity in such as way as to enforce systematicity. One possibility is tensor networks (Smolensky, 1990) 19 , whose connectivity is arranged in such a way as to implement the inner (fi) and ....

[Article contains additional citation context not shown here]

Fodor, J. A., & McLaughlin, B. P. (1990). Connectionism and the problem of systematicity: Why Smolensky's solution doesn't work. Cognition, 35, 183--204.

.... the English word John is a classical constituent of the English sentence John loves the girl and every tokening of the latter implies a tokening of the former (specifically, every token of the latter contains a token of the former; you can t say John loves the girl without saying John ) (Fodor and McLaughlin, 1990, p. 186) If related expressions are constructed by using the same constituents, according to Fodor and Pylyshyn, this could explain why humans possess a systematicity of thought. What does it mean to say that thought is systematic Well, just as you don t find people who can understand the ....

....connectionist representations are context sensitive, the question is, what is to be considered the representations for a constituent. Do we have to use the stable symbolic constituents or is it possible to use the more approximate context dependent connectionist representations According to Fodor and McLaughlin (1990), it could turn out that, complex superpositional vectors will have it in common with Classical complex symbols that they have a unique decoding into semantically significant parts. Of course, it could also turn out that they don t, and no ground for optimism on this point has so far been ....

[Article contains additional citation context not shown here]

Fodor, J. A. and McLaughlin, B. P. (1990). Connectionism and the problem of systematicity -- why Smolensky's solution doesn't work. Cognition, (35).

....theoretical force of methodological solipsism means that the computational psychologist rejects, utterly and without reservation, any conception of the mind as a semantic engine. For quite some time in the literature, there have been advocates of the syntactic engine as a model of the mind (cf. Fodor, 1980; Stitch 1983; Churchland, 1990), and advocates of the semantic engine as a model of the mind (cf. Dennett, 1981; Haugeland, 1981) with both schools remaining cooly contemptuous of the other. But a possible taxonomy of Engines does not stop at a black and white choice of either syntactic or semantic, because of the possibility ....

Fodor, J.A. & McLaughlin, B.P. (1990) `Connectionism and the Problem of Systematicity: Why Smolensky's Solution doesn't work.' Cognition, 35, 183-204.

....for representational theories. This was mainly as a reaction against the explicit representationalism of the earlier deliberative Artificial Intelligence (AI) with its concomitant need for a central executive. There was a similar reaction by the Connectionists [52, 53, 45] against the Cognitivists [20, 19]. An idea shared by both the behavior based and the Connectionist approaches in the 1980s, was that deliberative AI and, indeed, the Classical theory of mind, with its symbolic veneer could account for only a small part of what we call intelligence. The rest is shared with our animal forbears and ....

J.A. Fodor and B.P. McLaughlin. Connectionism and the problem of systematicity: Why Smolensky's solution doesn't work. Cognition, 35:183--204, 1990.

.... and process these systematically according to the syntactic structure of the representation (which directly expresses the structure of the represented expression) They pointed out that connectionists neither can exhibit this ability of cognitive agents, nor supply a novel explanation to it (cf. Fodor and McLaughlin, 1990), since connectionists cannot generate syntactically structured representations or apply processes sensitive to this structure without resorting to symbolic representations. Rather than accepting the implementational role for connectionism, suggested by Fodor and Pylsyhyn, the validity of the ....

Fodor, J. A. and McLaughlin, B. P. (1990). Connectionism and the problem of systematicity -- why Smolensky's solution doesn't work. Cognition, 35:183--204.

....constituent cup is different when decomposed from cup with coffee 1 than from cup with rocks 2 , as shown in fig. 6. Figure 6: Context dependent decomposition Smolensky defined the type of compositionality possible for this type of context dependent representations, as weak compositionality. Fodor McLaughlin (1990) argue that this entails that the decomposed representation for a constituent (e.g. cup) has neither the necessary nor the sufficient conditions for being that constituent. F M further argue that this inability to form stable representations for the constituents, is the reason why this type of ....

....not a good model of the real world, since we are not generally in the habit of assigning mental capacities to plants) We understand the thought, just by looking at the representation for it, as one in which there is a relation between an agent (in this case with mental capacities) and an object. Fodor McLaughlin (1990) argue that, the question if constituents can be used for structure sensitive systematic behaviour is highly dependent on the mode of combination, We argue that weak compositional structure is irrelevant to the systematicity problem and of dubious internal coherence. We then argue that strong ....

Fodor J. A. & McLaughlin B. P.,Connectionism and the problem of systematicity: Why Smolensky's solution did not work, Cognition, 35, pp 183 - 204, 1990.

....and thus that connectionism could not be an adequate cognitive architecture until it accounted for this phenomena. Many attempts have been made to meet this challenge (e.g. Smolensky, 1990; Christiansen and Chater, 1994; Niklasson and van Gelder, 1994) but the critics have not been satisfied (Fodor and McLaughlin, 1990; Hadley, 1994a; Hadley, 1994b) 1 Such difficulties have lead some connectionists investigating higher level cognitive activities to propose extensions to standard connectionist architectures. One such investigation developed a technique called temporal synchrony variable binding for use in ....

....by its very nature makes learned parameters independent of variable bindings. By showing this we show that temporal synchrony variable binding is a connectionist architecture that can account for systematicity. The Definition of Systematicity Although neither (Fodor and Pylyshyn, 1988) nor (Fodor and McLaughlin, 1990) provides a precise definition of systematicity, the concept is meant to embody the pervasive regularities in language (and thought) that are traditionally captured using compositional generative grammars. Such grammars express generalizations about the sentences of a language in terms of their ....

Fodor, J.A. and McLaughlin, B. (1990). Connectionism and the problem of systematicity: Why smolensky's solution doesn't work. Cognition, 35:183--204.

....[T]here is another important reason for emphasising non symbolic representations. Their examination represents a topic that is unique to connectionism. Distributed symbolic representations have been applied in the Classical tradition in areas as speech recognition (c.f. Fodor McLaughlin 1990 [3]) However . the study of non symbolic representations is a new departure. 2.1 Compositionality F P claim that compositionality, i.e. constituent structure, is unique to Classical models. You have constituent structure when (and only when) the parts of semantically evaluable entities are ....

Fodor J. A. & McLaughlin B. P.,Connectionism and the problem of systematicity: Why Smolensky's solution did not work, Cognition, 35, pp 183 - 204, 1990.

.... for complex expressions and process these systematically according to the structure of the representation (which directly express the structure of the represented expression) They point out that connectionists can neither exhibit nor supply a novel explanation to these principles (cf. Fodor and McLaughlin, 1990). The validity of the premises of these claims has been questioned (Goschke and Koppelberg, 1991; van Gelder and Niklasson, 1994) However, it is generally accepted also by connectionists that human thought obeys the so called compositionality and systematicity principles. 1 Thus, much research ....

Fodor, J. A. and McLaughlin, B. P. (1990). Connectionism and the problem of systematicity -- why Smolensky's solution doesn't work. Cognition, (35):183--204.

.... with arbitrary accuracy [9] 15] the important question, is: do the models have sufficient bias to account for systematicity Previous answers to this question have been an emphatic No, based on philosophical arguments on the nature of internal representations and processes [5] [4], and statistical analyses of training sets [8] This paper examines this question from a computational perspective. 1 Systematicity Systematicity is the ability to represent (systematic ity of representation) and infer (systematicity of inference) structurally related objects [5] Consider, ....

....components of a complex representation arises because the explicitness of a component within a representation is relative to the deconstructive process [11] The representation 12, in itself, does not restrict the possible decompositions. This illustration is effectively Fodor and McLauglin s [4] point as to why Smolensky s [16] tensor representational system, in itself, does not guarantee structure sensitivity. The deconstruction process must know about the mode of construction. However, in addition to this point, a constructive function (such as concat and multi) must map each ....

J A Fodor and B P McLaughlin. Connectionism and the problem of systematicity: Why Smolensky 's solution doesn't work. Cognition, 35:183-- 204, 1990.

....as the Compositionality argument, and it implies that combined symbols should be concatenated and preserved in the representation of complex expressions, as tokened constituents. An example of this is the complex logical expression (A B) C . These symbols are what Fodor McLaughlin (F M) [2] define as Classical constituents. 1.1.2 Classical mental processes The mental processes that operate on the complex mental representations should be sensitive to the syntactic structure of the representation, stated by F P as, Structure sensitivity of process. Classical mental ....

....This in fact also applies to the Classical model of the mind, since the need for resources for representing complex expressions grows, with increased complexity of the expression. There has been some debate whether the second type of representation can have constituents, e.g. Fodor McLaughlin s [2] response to Smolensky [4] The problem, according to F M, is that these representations of concepts are context sensitive. An example of this is that the combination of the microfeatured representations of the constituents cup without coffee and coffee , would result in the, rather ....

Fodor J. A. & McLaughlin B. P.,Connectionism and the problem of systematicity: Why Smolensky's solution did not work, Cognition, 35, pp 183 - 204, 1990.

....claimed to hold for contents should also be mapped, modulo any relevant changes due to switching from one technology to the other. For example, whereas a symbolic representation must be explicitly compositional (and so have syntactic and semantic structure that plays a causal role in processing: Fodor McLaughlin, 1990), its connectionist counterpart might characterise such structure in a non causal form (e.g. Smolensky, 1987) or in a more distinctly functional form (e.g. Sharkey, 1992) However, it is our contention that a technology non specific mapping of content alone is insufficient to confer true ....

Fodor, J. A. & McLaughlin, B. (1990). Connectionism and the problems of systematicity: why Smolensky 's solution doesn't work. Cognition, 35, 183--204.

.... the English word John is a classical constituent of the English sentence John loves the girl and every tokening of the latter implies a tokening of the former (specifically, every token of the latter contains a token of the former; you can t say John loves the girl without saying John ) (Fodor McLaughlin 1990, p. 186) If related expressions are constructed by using the same constituents, according to Fodor and Pylyshyn, this could explain why humans possess a systematicity of thought. What does it mean to say that thought is systematic Well, just as you don t find people who can understand the ....

....connectionist representations are context sensitive, the question is, what is to be considered the representations for a constituent. Do we have to use the stable symbolic constituents or is it possible to use the more approximate context dependent connectionist representations According to Fodor McLaughlin (1990), it could turn out that, complex superpositional vectors will have it in common with Classical complex symbols that they have a unique decoding into semantically significant parts. Of course, it could also turn out that they don t, and no ground for optimism on this point has so far been ....

[Article contains additional citation context not shown here]

Fodor, J. A. and McLaughlin, B. P. (1990). Connectionism and the problem of systematicity -- why Smolensky's solution doesn't work. Cognition, (35):183--204.

No context found.

Fodor, J. and B. McLaughlin (1991). Connectionism and the problem of Systematicity: Why Smolensky's Solution doesn't work. in Horgan and Tienson (eds.) Connectionism and the Philosophy of Mind, Kluwer Academic Pub. The Netherlands. 45 23.06.99 17:37

No context found.

Fodor, J. A., & McLaughlin, B. P. (1990). Connectionism and the problem of systematicity: Why Smolensky's solution doesn't work. Cognition, 35, 183-- 204.

No context found.

Fodor, J. and McLaughlin, B., Connectionism and the problem of systematicity: why Smolensky's solution doesn't work, in Connectionism and the Philosophy of Mind, Horgan, T. and Tienson, J., Eds., Kluwer Academic Publishers, The Netherlands, 1991.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC