Holographic String Encoding

Abstract

In this article, we apply a special case of holographic representations to letter position coding. We translate different well-known schemes into this format, which uses distributed representations and supports constituent structure. We show that in addition to these brain-like characteristics, performances on a standard benchmark of behavioral effects are improved in the holographic format relative to the standard localist one. This notably occurs because of emerging properties in holographic codes, like transposition and edge effects, for which we give formal demonstrations. Finally, we outline the limits of the approach as well as its possible future extensions.

Toward a scalable holographic word-form representation

“…A variety of word-form representations have been proposed to account for phenomena arising from orthographic similarity, many of which were reviewed and compared in Davis and Bowers (2006) and Hannagan, Dupoux, and Christophe (2011). Our primary concern in this article is not to develop a complete theory of visual word processing but, rather, to demonstrate how the form of the "output" of visual word recognition-a word-form representationmakes predictions about the perceived similarity between verbal stimuli.…”

“…When evaluating slot coding against the constraints from Hannagan, Dupoux, and Christophe (2011; see Table 2), we find that it fails in the ways discussed above-namely, to capture the fact that transposed letter primes produce more facilitation than do replaced letters, that replacement of terminal letters produces less facilitation than does replacement of internal letters, and that preservation of relative (but not absolute) position of letters in a prime still produces facilitatory priming. More in-depth discussion of the merits and demerits of slot coding may be found in Davis and Bowers (2006) and Hannagan, Dupoux, and Christophe (2011).…”

“…Indeed, although we do not model lexical competition here, many encoding schemes-including all of the ones that we investigate-predict positive priming in this condition. However, because such priming requires different procedures to be apparent, we retain the constraint from Hannagan, Dupoux and Christophe (2011) that this condition simply produces the weakest facilitation in the cases considered, rather than no facilitation at all.…”

“…This is much in the spirit of Hannagan, Dupoux, and Christophe (2011), who compared localist implementations of word-form encodings with ones implemented as binary spatter codes (BSCs, a different type of HRR; Kanerva, 1994) and evaluated the benefits of the various theories and their implementations with respect to the similarity constraints imposed by masked priming studies (summarized above and in Table 1). Our primary focus will be on the construction of HRRs via circular convolution, rather than BSCs.…”

“…A related effect is that a target that contains a repeated letter receives just as much facilitation if the prime is missing as it does when it is missing a nonrepeated internal letter. Hannagan, Dupoux, and Christophe (2011) codified the findings above into 20 criteria for assessing word-form encoding schemes, which are given in Table 1.…”