Broken Symmetries in a Location-Invariant Word Recognition Network

We studied the feedforward network proposed by Dandurand et al. (2010), which maps location-specific letter inputs to location-invariant word outputs, probing the hidden layer to determine the nature of the code. Hidden patterns for words were densely distributed, and K-means clustering on single letter patterns produced evidence that the network had formed semi-location-invariant letter representations during training. The possible confound with superseding bigram representations was ruled out, and linear regressions showed that any word pattern was well approximated by a linear combination of its constituent letter patterns. Emulating this code using overlapping holographic representations (Plate, 1995) uncovered a surprisingly acute and useful correspondence with the network, stemming from a broken symmetry in the connection weight matrix and related to the group-invariance theorem (Minsky & Papert, 1969). These results also explain how the network can reproduce relative and transposition priming effects found in humans.

[1]  Yann LeCun,et al.  Tangent Prop - A Formalism for Specifying Selected Invariances in an Adaptive Network , 1991, NIPS.

[2]  John Shawe-Taylor,et al.  Symmetries and discriminability in feedforward network architectures , 1993, IEEE Trans. Neural Networks.

[3]  E. Rolls,et al.  INVARIANT FACE AND OBJECT RECOGNITION IN THE VISUAL SYSTEM , 1997, Progress in Neurobiology.

[4]  C. Whitney.  How the brain encodes the order of letters in a printed word: The SERIOL model and selective literature review , 2001, Psychonomic bulletin & review.

[5]  M. Sigman,et al.  Opinion TRENDS in Cognitive Sciences Vol.9 No.7 July 2005 The neural code for written words: a proposal , 2022 .

[6]  Carol Whitney,et al.  Supporting the serial in the SERIOL model , 2008 .

[7]  Jonathan Grainger,et al.  Letter visibility and the viewing position effect in visual word recognition , 2003, Perception & psychophysics.

[8]  James L. McClelland,et al.  Locating object knowledge in the brain: comment on Bowers's (2009) attempt to revive the grandmother cell hypothesis. , 2010, Psychological review.

[9]  Geoffrey E. Hinton.  Learning Translation Invariant Recognition in Massively Parallel Networks , 1987, PARLE.

[10]  Sachiko Kinoshita,et al.  Masked priming : the state of the art , 2003 .

[11]  George Kachergis,et al.  Toward a scalable holographic word-form representation , 2011, Behavior research methods.

[12]  Padraic Monaghan,et al.  The Computational Exploration of Visual Word Recognition in a Split Model , 2001, Neural Computation.

[13]  Manuel Perea,et al.  Transposed-Letter Confusability Effects in Masked Form Priming , 2003 .

[14]  Tony A. Plate,et al.  Holographic reduced representations , 1995, IEEE Trans. Neural Networks.

[15]  Ali S. Hadi,et al.  Finding Groups in Data: An Introduction to Chster Analysis , 1991 .

[16]  James L. McClelland,et al.  An interactive activation model of context effects in letter perception: I. An account of basic findings. , 1981 .

[17]  Jonathan Grainger,et al.  A study of relative-position priming with superset primes. , 2006, Journal of experimental psychology. Learning, memory, and cognition.

[18]  Garrison W. Cottrell,et al.  The Early Word Catches the Weights , 2000, NIPS.

[19]  Jordan B. Pollack,et al.  Recursive Distributed Representations , 1990, Artif. Intell..

[20]  Kenneth I. Forster,et al.  Masked form priming with extreme transposition , 2008 .

[21]  D. Plaut,et al.  Locating Object Knowledge in the Brain , 2022 .

[22]  J. Grainger,et al.  Letter position coding in printed word perception: Effects of repeated and transposed letters , 2004 .

[23]  Colin J Davis,et al.  The spatial coding model of visual word identification. , 2010, Psychological review.

[24]  R S Berndt,et al.  A new model of letter string encoding: simulating right neglect dyslexia. , 1999, Progress in brain research.

[25]  Jonathan Grainger,et al.  Protein Analysis Meets Visual Word Recognition: A Case for String Kernels in the Brain , 2012, Cogn. Sci..

[26]  Emmanuel Dupoux,et al.  Holographic String Encoding , 2011, Cogn. Sci..

[27]  Jonathan Grainger,et al.  Cracking the orthographic code: An introduction , 2008 .

[28]  Carol Whitney,et al.  Comparison of the SERIOL and SOLAR theories of letter-position encoding , 2008, Brain and Language.

[29]  S. Lupker,et al.  Does jugde activate COURT? Transposed-letter similarity effects in masked associative priming , 2003, Memory & cognition.

[30]  Jeffrey L. Elman,et al.  Analyzing Cross-Connected Networks , 1993, NIPS.

[31]  M. L. Lambon Ralph,et al.  Age of acquisition effects in adult lexical processing reflect loss of plasticity in maturing systems: insights from connectionist networks. , 2000, Journal of experimental psychology. Learning, memory, and cognition.

[32]  Mariano Sigman,et al.  Hierarchical Coding of Letter Strings in the Ventral Stream: Dissecting the Inner Organization of the Visual Word-Form System , 2007, Neuron.

[33]  J Grainger,et al.  The role of letter identity and letter position in orthographic priming , 1999, Perception & psychophysics.

[34]  Carol Whitney,et al.  SERIOL Reading , 2008 .

[35]  Jeffrey R. Binder,et al.  Tuning of the human left fusiform gyrus to sublexical orthographic structure , 2006, NeuroImage.

[36]  James L. McClelland,et al.  Why there are complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory. , 1995, Psychological review.

[37]  S Lehéricy,et al.  The visual word form area: spatial and temporal characterization of an initial stage of reading in normal subjects and posterior split-brain patients. , 2000, Brain : a journal of neurology.

[38]  K. Plunkett,et al.  A neurocomputational account of taxonomic responding and fast mapping in early word learning. , 2010, Psychological review.

[39]  Tony Plate,et al.  Visualizing the Function Computed by a Feedforward Neural Network , 2000, Neural Computation.

[40]  Jonathan Grainger,et al.  Learning location-invariant orthographic representations for printed words , 2010, Connect. Sci..

[41]  Manuel Perea,et al.  The overlap model: a model of letter position coding. , 2008, Psychological review.

[42]  Xiaohui Xie,et al.  Equivalence of Backpropagation and Contrastive Hebbian Learning in a Layered Network , 2003, Neural Computation.

[43]  Jonathan Grainger,et al.  Modeling letter position coding in printed word perception , 2004 .

[44]  Francis Crick,et al.  The recent excitement about neural networks , 1989, Nature.

[45]  Jonathan Grainger,et al.  Letter position information and printed word perception: the relative-position priming constraint. , 2006, Journal of experimental psychology. Human perception and performance.

[46]  Jeffrey S Bowers,et al.  Contrasting five different theories of letter position coding: evidence from orthographic similarity effects. , 2006, Journal of experimental psychology. Human perception and performance.

[47]  Kurt Hornik,et al.  Multilayer feedforward networks are universal approximators , 1989, Neural Networks.