Listening to every other word: Examining the strength of linkage variables in forming streams of speech

Abstract

In a variation on a procedure originally developed by Broadbent ͓͑1952͒. "Failures of attention in selective listening," J. Exp. Psychol. 44, 428-433͔ listeners were presented with two sentences spoken in a sequential, interleaved-word format. Sentence one ͑target͒ comprised the odd-numbered words in the sequence and sentence two ͑masker͒ comprised the even-numbered words in the sequence. The task was to report the words in sentence one. The goal was to determine the effectiveness of cues linking the words of the target ͑or masker͒ over time. Three such "linkage variables" were examined: ͑1͒ fixed talker, ͑2͒ fixed perceived interaural location, and ͑3͒ correct syntactic structure. All of the linkage variables provided a significant advantage when applied to the target compared to the baseline condition in which the linkage variables were randomized. However, these linkage variables were not effective when applied to the masker. Word position effects were found such that performance in the baseline condition declined, and the advantages of the linkage variables increased, for the words near the end of the sentence. Overall, this approach appears to be useful for examining interference in speech recognition that has little or no peripheral component. The results suggest that variables that link target words together improve their resiliency to interference and/or their recall.

The role of syntax in maintaining the integrity of streams of speech

“…Because the words were spoken individually with neutral inflection when recorded (cf. Kidd et al, 2008b), it also is possible that less of a benefit of voice was found here than would occur for natural speech where coarticulation and intonation patterns may provide cues linking words together. Furthermore, the familiarity of the talkers was not assessed or explicitly manipulated and that factor has been shown to exert an influence on the ability to selectively attend to one talker among competing talkers (e.g., Newman and Evers, 2007;Johnsrude et al, 2013).…”

“…Significant differences were found for identification performance when the target word syntax varied (as here, better performance for correct syntax) but masker syntax was not a significant factor. There were many differences between the present study, the Kidd et al (2008b) study, and the Brouwer et al (2012) study with respect to the methods that were used. Perhaps most importantly, though, Brouwer et al employed semantically more meaningful target utterances than the other two studies.…”

“…The better performance for the last word is likely a recency effect-a well-known phenomenon in serial recall-and has been reported previously using similar stimuli but somewhat different methods (Kidd et al, 2008b; see also Ruggles and Shinn-Cunningham, 2011). Also, because the sequences were time-aligned at the beginning and had variable lengths depending on the words chosen, the last target word occasionally extended beyond the two maskers and thus may have been easier to identify.…”

“…Earlier work from our laboratory (Kidd et al, 2008b) using this corpus and similarly constructed five-word target strings of speech masked by equal-length strings of maskers drawn from the same corpus, found that variations in syntactic similarity between target and masker did not significantly affect performance. In the Kidd et al (2008b) study, the target and masker words were interleaved in time so that the target occupied the odd-numbered words in the sequence while the masker occupied the even-numbered words (cf. Broadbent, 1952).…”

“…Except for one single-word condition in experiment 1B (described below), the target was always a string of five words drawn from a corpus of forty monosyllabic words ("BU Corpus," Kidd et al, 2008b). The corpus was subdivided into the five categories with eight items per category of: <name> <verb> <number> <adjective> and <object>.…”