Spoken word recognition without a TRACE

Abstract

How do we map the rapid input of spoken language onto phonological and lexical representations over time? Attempts at psychologically-tractable computational models of spoken word recognition tend either to ignore time or to transform the temporal input into a spatial representation. TRACE, a connectionist model with broad and deep coverage of speech perception and spoken word recognition phenomena, takes the latter approach, using exclusively time-specific units at every level of representation. TRACE reduplicates featural, phonemic, and lexical inputs at every time step in a large memory trace, with rich interconnections (excitatory forward and backward connections between levels and inhibitory links within levels). As the length of the memory trace is increased, or as the phoneme and lexical inventory of the model is increased to a realistic size, this reduplication of time- (temporal position) specific units leads to a dramatic proliferation of units and connections, begging the question of whether a more efficient approach is possible. Our starting point is the observation that models of visual object recognition—including visual word recognition—have grappled with the problem of spatial invariance, and arrived at solutions other than a fully-reduplicative strategy like that of TRACE. This inspires a new model of spoken word recognition that combines time-specific phoneme representations similar to those in TRACE with higher-level representations based on string kernels: temporally independent (time invariant) diphone and lexical units. This reduces the number of necessary units and connections by several orders of magnitude relative to TRACE. Critically, we compare the new model to TRACE on a set of key phenomena, demonstrating that the new model inherits much of the behavior of TRACE and that the drastic computational savings do not come at the cost of explanatory power.

An ERP investigation of dichotic repetition priming with temporally overlapping stimuli

“…Treisman (1971) had already nicely summarized the key issue here in reference to research described in Treisman (1970; 1971) and earlier work on dichotic listening: “If switching between the ears with alternating stimuli is difficult, as the present experiments suggest, how can it lead to so many spontaneous, apparently unavoidable errors in experiments with simultaneous presentation?” Today, many would agree that the answer lies in the quite extensive processing that is performed on information arriving in the unattended ear, and the way this information can be integrated with simultaneously processed information in the attended ear. Here we tentatively propose a straightforward extension of a recent model of spoken word recognition, the TISK model (Hannagan et al, 2013), as a possible solution to Treisman’s paradox, and as an account of the present findings. 2 …”

“…However, it may be less suited for the study of processing linked to lower-level time-specific phonological representations – that is, phonemes coded for their position in time rather than their position in a word. Given that one of the key questions for current research on spoken word recognition is precisely just how such time-specific phonological information makes contact with time-invariant, word-centered phonological representations (Hannagan, Magnuson, & Grainger, 2013), the need for more appropriate methodology becomes obvious.…”

Sentential influences on acoustic-phonetic processing: a Granger causality analysis of multimodal imaging data

“…In contrast, models that embrace top-down or interactive mechanisms (c.f. Gagnepain, Henson, & Davis, 2012; Hannagan, Magnuson, & Grainger, 2013; Marslen-Wilson & Tyler, 1980; McClelland & Elman, 1986) allow higher level representations to directly influence speech processing.…”

Rapid modulation of spoken word recognition by visual primes

“…It is these representations that can be mapped onto the equivalent word-centered sublexical orthographic code via the sublexical O⇔P interface in the BIAM. Here it is important to note that a new model of spoken word recognition, incorporating this kind of sublexical word-centered phonological code, has recently been developed and successfully tested against the standard benchmarks in this field (Hannagan, Magnuson, & Grainger, 2013). In sum, within the framework of the BIAM (Figure 1), the early priming effect would be driven by activation in word-centered sublexical phonological representations (P-units), and processing of Katakana print would be optimized for this.…”