Ambiguity, competition, and blending in spoken word recognition

Abstract

A critical property of the perception of spoken words is the transient ambiguity of the speech signal. In localist models of speech perception this ambiguity is captured by allowing the parallel activation of multiple lexical representations. This paper examines how a distributed model of speech perception can accommodate this property. Statistical analyses of vector spaces show that coactivation of multiple distributed representations is inherently noisy, and depends on parameters such as sparseness and dimensionality. Furthermore, the characteristics of coactivation vary considerably, depending on the organization of distributed representations within the mental lexicon. This view of lexical access is supported by analyses of phonological and semantic word representations, which provide an explanation of a recent set of experiments on coactivation in speech perception (Gaskell & Marslen-Wilson, 1999).

Leading up the lexical garden path: Segmentation and ambiguity in spoken word recognition.

“…Implementations of this theory using recurrent neural networks have been shown to simulate the effect of subcategorical mismatch on the lexical access process (Gaskell & Marslen-Wilson, 1997b;MarslenWilson & Warren, 1994; however, see Norris, McQueen, & Cutler, 2000). They have also been able to simulate effects of lexical competition and bottom-up mismatch without directly implemented inhibitory connections between lexical units (Gaskell & Marslen-Wilson, 1999). However, the networks reported by Gaskell andMarslen-Wilson (1997b, 1999) were unable to track lexical activations across word boundaries and were consequently incapable of identifying short words embedded in longer items.…”

Gradiency and visual context in syntactic garden-paths

“…When the distractor object's name did not share phonetic features with the target object's name (e.g., a raccoon opposite the carriage, or a crayon opposite the tower), there was significantly less curvature in the mousemovement trajectory. These results were interpreted as evidence for parallel partial activation of multiple lexical items competing over time (e.g., Gaskell & Marslen-Wilson, 1999;Luce, Goldinger, Auer, & Vitevitch, 1998;McClelland & Elman, 1986). 1 With a similar visual display, this kind of continuous competition is also observed in computermouse trajectories toward semantic categories for taxonomic classes (e.g., Mammal and Fish), when participants are given atypical animal exemplars to classify (e.g., whale, seal) compared to typical members of those categories (e.g., horse and trout).…”

“…When the distractor object's name did not share phonetic features with the target object's name (e.g., a raccoon opposite the carriage, or a crayon opposite the tower), there was significantly less curvature in the mousemovement trajectory. These results were interpreted as evidence for parallel partial activation of multiple lexical items competing over time (e.g., Gaskell & Marslen-Wilson, 1999; Luce, Goldinger, Auer, & Vitevitch, 1998;McClelland & Elman, 1986). 1…”

Word length and lexical activation: Longer is better.

“…At the lexical level, the combination of excitation and inhibition controls a lexical entry's activation level, which represents how well it matches a spoken word relative to all other words in the lexicon. Distributed connectionist models (e.g., the distributed cohort model [DCM]: Gaskell & Marslen- Wilson, 1997Wilson, , 1999 use the same mechanisms but, through training, form shared representations of words in their hidden units.Activation has proven to be a productive metaphor for contemplating the dynamics of recognition within this modeling tradition. In particular, the gradedness of activation nicely captures the recognition system's sensitivity to the quality of the speech signal and the composition of the lexicon, as a host of studies have demonstrated (Connine, Blasko, & Titone, 1993; Davis, MarslenWilson, & Gaskell, 2002;Vroomen & de Gelder, 1995;Warren & Marslen-Wilson, 1987;Whalen, 1981).…”

“…At the lexical level, the combination of excitation and inhibition controls a lexical entry's activation level, which represents how well it matches a spoken word relative to all other words in the lexicon. Distributed connectionist models (e.g., the distributed cohort model [DCM]: Gaskell & Marslen- Wilson, 1997Wilson, , 1999 use the same mechanisms but, through training, form shared representations of words in their hidden units.…”