D. S. Blank, L.A. Meeden, and J. B. Marshall. Exploring the symbolic /subsymbolic continuum: A case study of RAAM. In J. Dinsmore, editor, The Symbolic and Connectionist Paradigms: Closing the Gap, pages 113--148. Lawrence Erlbaum Associates, 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... Parallel Parsing system [11] the PARSEC model [8] the CDP approach [9] the SPEC architecture [10] the SCAN architecture [14] and the recurrent network models [4, 5, 13] 2 Confluent Preorder Parser (CPP) In this paper, syntactic parsing is accomplished via a holistic transformation (see [1] and [2] in which the connectionist representation encoding the input sentence is directly mapped to the connectionist representation encoding the target parse tree (Figure 2) Compared with other connectionist parsers, the Confluent Preorder Parser (CPP) using this paradigm, succeed to produce ....

D. S. Blank, L.A. Meeden, and J. B. Marshall. Exploring the symbolic /subsymbolic continuum: A case study of RAAM. In J. Dinsmore, editor, The Symbolic and Connectionist Paradigms: Closing the Gap, pages 113--148. Lawrence Erlbaum Associates, 1992.

.... a transformational holistic computation of f( Delta) The process of computing f (x) in this fashion is called transformational holistic inference and is diagramed in Figure 2(a) This characterizes the techniques employed in [Chalmers, 1990] and the syntactic transformation experiments of [Blank et al. 1992]. For example, Chalmers trained a RAAM to auto associate parse trees of active and passive English language sentences. After this training process had converged, the hidden layer of the RAAM yielded distributed representations for each of the sentences, thus providing the E( Delta) and D( Delta) ....

.... the particular mappings obtained for E( Delta) and D( Delta) The feasibility of the transformational holistic approach hinges upon the learning complexity of g( Delta) When E( Delta) and D( Delta) are learned independently of g( Delta) as they were in the experiments of [Chalmers, 1990] and [Blank et al. 1992], the target inference task has no effect upon the resulting representation scheme. In such a case, a low learning complexity for g( Delta) can only be expected when the ideal representations for auto association happen to be appropriate for the given inference task. 3.2 Confluent Inference One ....

[Article contains additional citation context not shown here]

Douglas S. Blank, Lisa A. Meeden, and James B. Marshall. Exploring the symbolic/subsymbolic continuum: A case study of RAAM. In J. Dinsmore, editor, Closing the Gap: Symbolism vs. Connectionism. Lawrence Erlbaum Associates, 1992. To Appear.

....of linguistic representations. What we roughly distinguish as letters, words, sentences, etc. need to be encoded in a proper and systematic manner, permitting direct, holistic operations over the resultant abstract representations rather than over external sequential forms [Chalmers, 1990] [Blank, 1992] [Hammerton, 1998] Those representations should be static, unique characterization of the original objects, which is necessary for reproducing them back into their sequential form. They should allow holistic transformations and associations to representations from other modalities visual, ....

....compact distributed representations of the static input patterns through an autoassociation. RAAM was further extended to a Sequential RAAM (SRAAM) for sequential processing. Different implementations of the latter model had variable success even when applied to trivial data [Chalmers, 1990] [Blank, 1992], Blair, 1997] Kwasny, 1995] Hammerton, 1998] The development of global memory recurrent neural networks, such as the Jordan Recurrent Networks [Jordan, 1986] and the Simple Recurrent Networks (SRN) by Elman [Elman, 1990] stimulated the development of models that gradually build ....

[Article contains additional citation context not shown here]

Blank, D. S., Meeden, L., Marsgall, J., Exploring the symbolic/subsymbolic continuum: a case study of RAAM, In Dinsmore, Closing the Gap: Symbolism vs. Connectionism, Dinsmore, J., Ed., Lawrence Erlbaum Assocuiates, 1992.

....highly generalizable active passive sentence transformations. Subsequent studies have reported similar findings. The other achievements have included transformations in domains such as the simple manipulation of logical formulae and the simulation of De Morgan s rule [32, 33] plausible inference [5]; English to Spanish translation [14] sequential parsing [47] DRAFT PAPER FOR COMMENTS 6 necessary inference [3] and the manipulation of Horn clauses [11] Much of this research continues as the Connectionists penetrate further and further into Cognitivist territory [34] Nonetheless, the ....

D.S. Blank, L.A. Meeden, and J.B. Marshall. Exploring the symbolic/subsymbolic continuum: A case study of RAAM. In J.Dinsmore, editor, The Symbolic and Connectionist Paradigms: Closing the Gap. Lawrence Erlbaum Associates, New Jersey, USA, 1992.

....was introduced by Pollack [Pol90, Pol89] to allow traditional symbolic data structures, such as trees with labeled leaves, to be represented subsymbolically as distributed patterns of activation. In the last years, different papers have discussed or used the RAAM model with interesting results [Chr91, BMM92, Rei92, SW92]. The basic RAAM can encode arbitrary tree structures of variable depth but fixed branching factor (valence) The idea is to map a symbolic tree into a numeric vector and then to reconstruct a very close approximation of the symbolic tree starting from the numeric vector. The mapping from trees to ....

....is the decoding function, i.e. given a structure does it exist a reduced representation from which the recursive application of the decoding function can extract all the components of the structure The approach we use in performing this exploration is very close to the one used by Blank and al. [BMM92], but adjusted to the LRAAM model and extended in scope. In the next section we will give a geometric interpretation of the binary LRAAM model with respect to the decoding process. 5.1 Geometric interpretation of the decoding process As a first approximation let us consider an LRAAM with binary ....

D.S. Blank, L.A. Medeen, and J.B. Marshall. Exploring the symbolic /subsymbolic continuum: a case study of raam. In J. Dinsmore, editor, The Symbolic and Connectionist Paradigms: Closing the Gap, volume 1, pages 113--148. Lawrence Erlbaum, 1992.

....distributed representations for variable sized recursive data structures. There have been several publications showing the appropriateness of representations produced by the RAAM for subsequent classification tasks [Cad94, GSS95b] and also for more complex tasks even with structured output [DSB92, Chr91, Nic94]. Our approach, however, is different. We present a novel learning scheme that we call backpropagation through structure (BPTS) in analogy to backpropagation through time for recurrent networks [RGM86, Wer90] It allows us to devise distributed representations exclusively with respect to the task ....

....distributed reduced representations w.r.t. a given inference task: transformational and confluent inference. The former could be achieved by first computing unique distributed representations with the RAAM, freezing them and then using them as input for subsequent modules doing the inference task [DSB92, Cad94]. Transformational inference does not imply a very tight coupling. RAAM and inference module may consist of separately trained networks based on different models. Confluent inference takes the intended inference task into account while learning distributed representations for structures. RAAM and ....

J. B. Marshall D. S. Blank, L. A. Meeden. Exploring the symbolic/subsymbolic continuum: A case study of raam. In J. Dinsmore, editor, The Symbolic and Connectionist Paradigms: Closing the Gap. LEA Publishers, 1992.

....this paper, syntactic parsing is discussed in the context of pure connectionism. The Confluent Preorder Parser (CPP) will be presented which exemplifies the Holistic Parsing Paradigm. Instead of a step by step algorithmic approach, syntactic parsing is accomplished via a Holistic Transformation [4, 5]. This transformation maps the connectionist representation encoding the input sentence to the connectionist representation encoding the target parse tree (see Figure 3) Effectively, the parsing mechanism is encapsulated in a black box . With the knowledge of the ultimate input (the sentence) ....

D. S. Blank, L.A. Meeden, and J. B. Marshall. Exploring the symbolic/subsymbolic continuum: A case study of RAAM. In J. Dinsmore, editor, The Symbolic and Connectionist Paradigms: Closing the Gap, pages 113--148. Lawrence Erlbaum Associates, Hillsdale, NJ, 1992.

No context found.

D. S. Blank, L.A. Meeden, and J.B. Marshall. Exploring the symbolic/subsymbolic continuum: A case study of raam. In J. Dinsmore, editor, Closing the Gap: Symbolism vs. Con nectionism, pages 113--147. Lawrence Erlbaum Associates, Hillsdale, N.J., 1992.

No context found.

Blank, D. S., Meeden, L. A., & Marshall, J. B. (this volume). Exploring the symbolic/subsymbolic continuum: A case study of RAAM. Block, N. (1986). Advertisement for a semantics for psychology. In P. French (Ed.), Midwest Studies in Philosophy, Vol. 10 (pp. 615--78). Minneapolis: University of Minnesota Press.

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