Development and Validation of Sentences Without Semantic Context to Complement the Basic English Lexicon Sentences

Abstract

Purpose The goal of this study was to develop and validate a new corpus of sentences without semantic context to facilitate research aimed at isolating the effects of semantic context in speech perception. Method The newly developed corpus contains nonsensical sentences but is matched in vocabulary and syntactic structure to the existing Basic English Lexicon (BEL) corpus. It consists of 20 lists, with each list containing 25 sentences and each sentence having four keywords. Each new list contains the same keywords as the respective list in the original BEL corpus, but the keywords within each list are scrambled across sentences to eliminate semantic context within each sentence, while maintaining the original syntactic structure. All sentences in the original and nonsense BEL corpora were recorded by the same two male and two female talkers. Results Mean intelligibility scores for each list were estimated by calculating the mean proportion of correct keywords achieved by 40 normal-hearing listeners for one male and one female talker. Although small but significant differences were found between some pairs of lists, mean performance for all 20 lists fell within the 95% confidence intervals of the mean. Conclusions Lists in the newly developed nonsense corpus are reasonably well equated for difficulty and can be used interchangeably in a randomized experimental design. Both the original and nonsense BEL sentences, all recorded by the same four talkers, are publicly available. Supplemental Material https://doi.org/10.23641/asha.13022900.

Figure 1.

Mean proportion of correctly “filled in” words, averaged across all 20 lists of original Basic English Lexicon (BEL) sentences and nonsense BEL sentences, scored using three methods. The first set of bars represent the mean proportion of reported keywords that were exactly correct (Exact), the second set of bars show the proportion of keywords that were either exactly correctly or contained the right word but wrong tense or plurality (Exact + tense), and the third set of bars show the proportion of reported keywords that were either exactly correct, correct except for an error in tense or plurality, or had the same semantic meaning as the missing keyword (Exact + tense + meaning). Results for the original BEL sentences are shown by black bars and those for the nonsense BEL sentences are shown by white bars. Error bars represent 1 standard error of the mean between listeners.

Figure 2.

Speech perception for nonsense Basic English Lexicon sentences recorded by an older female talker and a younger male talker. Red circles and blue squares show individual data for the male and female talker, respectively. The red and blue bars indicate mean performance for each list for the male and female talkers, and black bars show list averages across talker. The horizontal red, blue, and black dotted lines show the overall performance average across lists for the male talker, the female talker, and across talkers, respectively.

Figure 3.

Scores for nonsense Basic English Lexicon sentences recorded by an older female talker (blue) and a young male talker (red), plotted relative to each participant's mean across lists. The red and blue bars indicate mean relative scores for each list for the male and female talkers, respectively, and black bars show averages across the two talkers. The horizontal dotted lines show the 95% confidence interval for performance on all lists across talkers.