Toward a scalable holographic word-form representation

Abstract

Phenomena in a variety of verbal tasks-for example, masked priming, lexical decision, and word naming-are typically explained in terms of similarity between word-forms. Despite the apparent commonalities between these sets of phenomena, the representations and similarity measures used to account for them are not often related. To show how this gap might be bridged, we build on the work of Hannagan, Dupoux, and Christophe, Cognitive Science 35:79-118, (2011) to explore several methods of representing visual word-forms using holographic reduced representations and to evaluate them on their ability to account for a wide range of effects in masked form priming, as well as data from lexical decision and word naming. A representation that assumes that wordinternal letter groups are encoded relative to word-terminal letter groups is found to predict qualitative patterns in masked priming, as well as lexical decision and naming latencies. We then show how this representation can be integrated with the BEAGLE model of lexical semantics (Jones & Mewhort, Psychological Review 114:1-37, 2007) to enable the model to encompass a wider range of verbal tasks.

OrBEAGLE: Integrating Orthography into a Holographic Model of the Lexicon

“…While all but one of these word-form encoding methods were found unable to account for the entirety of the empirical constraints, [3] introduced a word-form encoding that satisfied the desiderata. Our solution, called "terminal-relative" (TR) encoding, is related somewhat to the simplified word recognition model of [1] and to the SERIOL model [16].…”

“…Empirical studies of word recognition (see, for a review, [9,7]) show that humans are indeed sensitive to structure on both those levels, and that such sensitivity is required to account for human word recognition capabilities. In addition, TR encoding is capable not just of capturing the relative similarity between isolated pairs of words, but scales to account for orthographic similarity effects within the entire lexicon, as evidenced in lexical decision and speeded pronunciation tasks [3]. In general, TR encoding is a good balance between simplicity (it is parameter free), veracity (it accounts for many word recognition effects), and scalability (orthographic similarity effects across the entire lexicon).…”

“…In this way, we may capture orthographic similarity (e.g., cat and catch), and perhaps additional semantic relationships (e.g., catch and catcher ). Because OrBEAGLE builds its knowledge of semantics and word order on the basis of a principled orthographic representation, it may better explain a variety of human data, and can be applied to tasks such as fragment completion that no semantic model has previously been suited to model [3]. In the present paper, we examine how well OrBEAGLE accounts for human free-association data, and compare it to BEAGLE.…”

“…[3] proposes an alternative orthographic encoding scheme that better captures a set of empirical constraints enumerated by [9]. This encoding scheme uses Holographic Reduced Representations (HRRs) [15], which are neurally-plausible, distributed, and which result in analytically similar representations for items with similar content or structure.…”

“…This encoding scheme uses Holographic Reduced Representations (HRRs) [15], which are neurally-plausible, distributed, and which result in analytically similar representations for items with similar content or structure. [3] integrates orthographic vectors into BEAGLE, an HRR-based model of lexical semantics and word order [12], and also successfully accounts for lexical decision data. In the present work, after briefly describing HRRs, the Cox et al orthographic encoding scheme, and BEAGLE, we apply this model to human free-association data [4].…”