Gradient sensitivity to within-category variation in words and syllables.

Abstract

Five experiments monitored eye movements in phoneme and lexical identification tasks to examine the effect of within-category sub-phonetic variation on the perception of stop consonants. Experiment 1 demonstrated gradient effects along VOT continua made from natural speech, replicating results with synthetic speech (McMurray, Tanenhaus & Aslin, Cognition, 2002). Experiments 2-5 used synthetic VOT continua to examine effects of response alternatives (2 vs. 4), task (lexical vs. phoneme decision), and type of token (word vs. CV). A gradient effect of VOT in at least one half of the continuum was observed in all conditions. These results suggest that during on-line spoken word recognition lexical competitors are activated in proportion to their continuous distance from a category boundary. This gradient processing may allow listeners to anticipate upcoming acoustic/phonetic information in the speech signal and dynamically compensate for acoustic variability. KeywordsSpeech Perception; Categorical Perception; Word Recognition; Subphonemic Sensitivity; Visual World ParadigmThe remarkable ability of human listeners to understand spoken language stands in stark contrast to the highly variable acoustic cues that signal phonemic distinctions. Attempts to find context-independent or invariant cues to simple phonetic features such as place of Corresponding Author: Bob McMurray, Department of Psychology, E11 SSH, University of Iowa, Iowa City, IA 52242, bobmcmurray@uiowa.edu, 319-335-2408 (phone), 319-335-0191 (fax). 9 We were surprised that Experiment 2 shows gradient effects despite the steep labeling function, whereas Experiment 3, with a more continuous labeling function does not. We conducted detailed analyses of the timecourse in order to understand these conflicting results. Two factors appear to conspire to drive these effects. First, Experiment 2 has extremely low variability, making it possible to detect even small differences. Second, our analyses suggest that the, likelihood of detecting gradiency depends on the overall proportion of fixations to competitors. There were relatively more fixations to competitors in Experiment 2 than in Experiment 3, perhaps driven by the fact that there were no other objects on the screen. Details of these analyses are available from the first author. articulation have met with little success, suggesting that simple invariants do not exist and that speech perception must employ at least partial context-dependent sensitivity. At its heart, the search for invariant acoustic cues in the speech signal is based on the assumption that perception is fundamentally a problem of transforming a continuous and contextually varying signal into discrete and contextually invariant units such as phonemes and words. Continuous aspects of the signal are thought to represent noise that must be discarded during perception, and so-called invariant cues represent nuggets of discreteness buried in this continuous signal. NIH Public AccessIt is now clear that the variance reduction assumption is ...

Continuous dynamics of color categorization

“…These results parallel the results from speech perception (McMurray et al, 2008;McMurray, Tanenhaus, & Aslin, 2002), and suggest that parallel activation and gradiency, not discrete categories, are the norm across perceptual categorization. Such processes can be modeled by timevarying dynamical systems (e.g., McClelland & Elman, 1986).…”

“…If one participant had six steps on the green side, but another had only four, we could analyze only the first four from both (otherwise, there would be missing cells). Thus, from McMurray, Aslin, Tanenhaus, Spivey, and Subik (2008).…”

Role of labeling mediation in speech perception: Evidence from a voiced stop continuum perceived in different surrounding sound contexts

“…Again, a peak in discrimination accuracy was predicted for pair 5-7, which crossed the categorical boundary indicated by the identification results, that is, NS-Step 6. The stimulus pairs were divided into three groups, and analyzed: (1) pairs frequently heard as /ba/, that is, pairs 1-3, 2-4, and 3-5 (within-/ba/ pairs), (2) those frequently heard as /da/, that is, the pair 7-9 (within-/da/ pair), and (3) the pair that crosses the categorical boundary, that is, the pair 5-7 (cross-boundary pair). As in the Sentence Condition, the pairs that have the boundary step, i.e., NS-Step 6, were excluded from the analysis, although they are shown in the figure.…”

“…Concretely, categorical perception in an original sense predicts (1) discrimination peak at a pair that strides a categorical boundary and (2) close to the chance level of discrimination within the same phonemic category.…”

“…Although listeners may be aware of acoustic variation of allophones under some experimental conditions [1,2], traditional studies have often discussed human's selective sensitivity to phonemic contrast during speech perception in terms of categorical perception [3][4][5].…”