The impact of attention load on the use of statistical information and coarticulation as speech segmentation cues

Abstract

MaterialAL material. To achieve realistic coarticulation, we used naturally produced utterances rather than synthesized speech, and created concatenated and coarticulated versions of the AL stimuli (for the detailed method, see Fernandes et al., 2007). On the basis of Fernandes et al.'s pretests, we already know that the use of naturally produced rather than synthesized utterances does have an impact on ALL performance.All the natural speech stimuli were recorded in a soundproof room (using WaveLab Lite Program, version 2.52) by a European Portuguese female native speaker using a Shure M58 microphone and sampled at a rate of 22.05 kHz with 16-bit conversion. Adobe Audition 1.5 and Praat 4.2.24 (available at http://fon.hum.uva.nl/ praat/) were used for editing of the digitized versions of the stimuli (i.e., syllables, words, and part-words). These stimuli were closely matched to a synthesized stream in all relevant acoustic parameters, such as speaking rate (approximately 270 syllables per minute), AL stimuli duration, lack of lexical stress cues to word boundaries, and pitch contour (through flattening to monotone 220 Hz).The AL used by Fernandes et al. (2007) was adopted here. It was constituted by six trisyllabic AL words (i.e., / /, / /, / /, / /, / /, / /). TPs between adjacent syllables were always higher within than between AL words. Half of the AL words had higher TPs (high-TP words, ranging from .75 to 1) than the other half (low-TP words, ranging from .50 to .58), but all occurred with the same frequency in the AL stream.

Effects of cognitive load on speech recognition

“…An important difference between their experiments and ours, 3 however, is that the acoustic cues in the Fernandes et al study were far more salient and unambiguous than those in the present ones. This important contrast confirms that CL's effect on sensory analysis is restricted to detailed phonetic processing, perhaps those details for which there are no clear and categorical phonological consequences, e.g., small VOT differences (the present experiments) vs. meaningful allophonic distinctions (Fernandes et al, 2010).…”

“…This result suggests that more processing resources are required for spoken stimuli with low than high levels of lexical involvement. Likewise, Fernandes, Kolinsky, and Ventura (2010) found that the extraction of novel words from a continuous speech stream is more greatly affected by a concurrent visual task when the word boundaries in the stream are cued by statistical regularities than by coarticulatory cues. The authors concluded that CL has a particularly detrimental effect on domain-general processes, e.g., statistical computation, and less effect on phonetic processing.…”

The role of Gestalt grouping principles in visual statistical learning

“…This form of statistical learning is fairly simple, consisting of frequencies of co-occurrence of two shapes that may not involve chunking. A compatible pattern of results was found in modified versions of the Saffran et al (1996) paradigm under dual-task conditions, with learning being compromised when participants were required to perform a secondary task (Toro, Sinnett, & Soto-Faraco, 2005; see also Fernandes, Kolinsky, & Ventura, 2010). Yet all of the above studies did not isolate the effect of perceptual learning on statistical learning.…”

Divided attention disrupts perceptual encoding during speech recognition

“…Within that definition, increased cognitive load can be elicited by contrasting focused attention (one stimulus) with selective attention (ignoring a concurrent stimulus), or focused attention with divided attention (processing both the target and the concurrent stimuli), or levels of divided attention in which the difficulty of a competing task is manipulated. Although speech processing might appear to proceed "automatically," a number of studies have shown that cognitive load has a detrimental effect on tone threshold perception (Macdonald and Lavie, 2011), phoneme perception (Casini et al, 2009;Gordon et al, 1993), word segmentation (Fernandes et al, 2010), and the ability to selectively attend to a single talker in a multi-talker environment (Francis, 2010). Findings such as these indicate that some aspects of speech processing are resource demanding, and hence vulnerable to attentional disruption, but the precise locus of the disruption within the speech recognition system is unclear.…”