Phoneme restoration and empirical coverage of interactive activation and adaptive resonance models of human speech processing

Abstract

Grossberg and Kazerounian [(2011). J. Acoust. Soc. Am. 130, 440-460] present a model of sequence representation for spoken word recognition, the cARTWORD model, which simulates essential aspects of phoneme restoration. Grossberg and Kazerounian also include simulations with the TRACE model presented by McClelland and Elman [(1986). Cognit. Psychol. 18, 1-86] that seem to indicate that TRACE cannot simulate phoneme restoration. Grossberg and Kazerounian also claim cARTWORD should be preferred to TRACE because of TRACE's implausible approach to sequence representation (reduplication of time-specific units) and use of non-modulatory feedback (i.e., without position-specific bottom-up support). This paper responds to Grossberg and Kazerounian first with TRACE simulations that account for phoneme restoration when appropriately constructed noise is used (and with minor changes to TRACE phoneme definitions), then reviews the case for reduplicated units and feedback as implemented in TRACE, as well as TRACE's broad and deep coverage of empirical data. Finally, it is argued that cARTWORD is not comparable to TRACE because cARTWORD cannot represent sequences with repeated elements, has only been implemented with small phoneme and lexical inventories, and has been applied to only one phenomenon (phoneme restoration). Without evidence that cARTWORD captures a similar range and detail of human spoken language processing as alternative models, it is premature to prefer cARTWORD to TRACE.

TISK 1.0: An easy-to-use Python implementation of the time-invariant string kernel model of spoken word recognition

“…It can be useful to reduce the lexicon size in order to better understand such Bpure^effects. Magnuson (2015), for example, reduced the TRACE lexicon to a single word in order to isolate the effects of feedback. 2 In these simulations, we did not try to modify the parameters to promote accuracy.…”

“…The reduplication strategy has often been critiqued, beginning with McClelland and Elman themselves (1986, p. 77). Hannagan et al (2013) review some of the arguments, and Magnuson (2015) makes a case for TRACE as a model building (McClelland & Elman, 1986). The black squares at bottom indicate the input patterns (which would be features over time in the actual model), presented in sequence (corresponding to CAT).…”

Lexical feedback in the Time-Invariant String Kernel (TISK) model of spoken word recognition

“…Some have argued that this scheme is simply implausible (e.g., Grossberg & Kazerounian, 2011;Norris, 1994), largely because of the numbers of nodes and connections it would take to implement a realistic phoneme inventory and lexicon. Magnuson (2015) presents a case for the TRACE architecture as a model of echoic memory. Hannagan et al (2013) connections.…”

“…This means that the critical pattern a model must be able to simulate is (a) robust activation of a lexically-consistent phoneme when it is replaced with noise, but (b) weak or absent activation when it is replaced with silence (see Grossberg & Kazerounian, 2011and Magnuson, 2015, for a debate about how phoneme restoration should be modeled).…”

“…To test TISK's ability to simulate phoneme restoration with and without feedback, we used a single word, luxury (/l^kS^ri/), that was the key example in a relevant debate (Grossberg & Kazerounian, 2011;Magnuson, 2015). We examined the activations of the "expected" phonemes in initial, medial, and final positions (/l/, /∫/, and /i/, respectively) when they were intact versus when they were replaced with silence or increasingly strong noise.…”