To Örebro University

PURPOSE: The purpose of the current study was to create a set of video-recorded real-sign-pseudosign pairs for Swedish Sign Language (Svenskt teckenspråk, STS). The pseudosigns should be based on the overall phonotactic structures of STS; each real-sign-pseudosign pair should be matched in terms of neighborhood density (ND) and phonotactic probability (PP); the real signs and pseudosigns should have similar distributions of ND, PP, and duration; and the set as a whole should have distributions of ND, PP, and (for the real signs) sign frequency (SF) similar to that of the STS. To achieve this, a secondary purpose was to develop algorithms to calculate ND and PP for STS and to investigate how the metrics correlate.

METHOD: Based on publicly available data sources for STS, an initial data set was formed, which was utilized in an automatic algorithm to generate phonotactically feasible pseudosign candidates, as well as to calculate ND and PP values for both the real signs and the generated pseudosign candidates. The use of an automatic matching algorithm was followed by manual evaluation of the selected pseudosign candidates. The selected matching pairs of signs and pseudosigns were video recorded by four actors (two males and two females). RESULTS: Four hundred sixty-six sign-pseudosign pairs, matching in ND and PP, were video-recorded. The real signs and pseudosigns had similar distributions of ND, PP, and duration, and the distributions of ND, PP, and SF values in the recorded set were similar to those of the STS as a whole.

CONCLUSIONS: The resulting data set has been made publicly available and is suitable for use in psycholinguistic experiments involving STS. The present work presents a new standard for estimation of lexical metrics in sign language that can be applied to other sign languages than STS in future studies.

SUPPLEMENTAL MATERIAL: https://doi.org/10.23641/asha.29318822.

OBJECTIVE: To introduce the Phoneme Discriminability Level (PDL) and compare its ability to predict scores on the Situated Phoneme (SiP) test with those of a phoneme-level version of the Speech Intelligibility Index (SIIP) and the original Speech Intelligibility Index (SII).

DESIGN: SII, SIIP, and PDL values were calculated for 24,444 SiP-test trials and used to train and evaluate logistic regression models. Models were trained on half of the study sample and statistically evaluated on the other half.

STUDY SAMPLE: Seventy-four adult native speakers of Swedish (mean age = 62 years), with hearing ranging from normal to severe loss. All had symmetric hearing without conductive components.

RESULTS: Models based on PDL and SIIP showed similar overall predictive power, and both outperformed the SII-based model. SIIP and SII were better at predicting average session scores, whereas PDL was better at predicting scores for individual test words. Adding random intercepts for test words improved the predictive accuracy of the SIIP and PDL models, but only at the level of individual words, not whole sessions.

CONCLUSIONS: All investigated models are useful for predicting SiP-test scores. SIIP and PDL appear suitable for optimising SiP-test settings and estimating critical differences.

OBJECTIVE: To introduce the urban outdoor version of the Situated Phoneme (SiP) test and investigate its test-retest reliability.

DESIGN: Phonemic discrimination scores in matched-spectrum real-world (MSRW) maskers from an urban outdoor environment were measured using a three-alternative forced choice test paradigm at different phoneme-to-noise ratios (PNR). Each measurement was repeated twice. Test-retest scores for the full 84-trial SiP-test, as well as for four types of contrasting phonemes, were analysed and compared to critical difference scores based on binomial confidence intervals.

STUDY SAMPLE: Seventy-two adult native speakers of Swedish (26-83 years) with symmetric hearing threshold levels ranging from normal hearing to severe sensorineural hearing loss.

RESULTS: Test-retest scores did not differ significantly for the whole test, or for the subtests analysed. A lower amount of test-retest score difference than expected exceeded the bounds of the corresponding critical difference intervals.

CONCLUSIONS: The urban outdoor SiP-test has high test-retest reliability. This information can help audiologists to interpret test scores attained with the urban outdoor SiP-test.

The speech perception ability of people with hearing loss can be efficiently measured using phonemic-level scoring. We aimed to develop linguistic stimuli suitable for a closed-set phonemic discrimination test in the Swedish language called the Situated Phoneme (SiP) test. The SiP test stimuli that we developed consisted of real monosyllabic words with minimal phonemic contrast, realised by phonetically similar phones. The lexical and sublexical factors of word frequency, phonological neighbourhood density, phonotactic probability, and orthographic transparency were similar between all contrasting words. Each test word was recorded five times by two different speakers, including one male and one female. The accuracy of the test-word recordings was evaluated by 28 normal-hearing subjects in a listening experiment with a silent background using a closed-set design. With a few exceptions, all test words could be correctly discriminated. We discuss the results in terms of content- and construct-validity implications for the Swedish SiP test.

In the current thesis, a Swedish phoneme-level speech-audiometric test in natural background noise was developed. The test is called the Situated Phoneme (SiP) test. In the first study, different types of word metrics thought to influence lexical access were developed and calculated for more than 800 000 phonetically transcribed Swedish words, which were then assembled in a database called the AFC-list. In the second study, groups of monosyllabic AFC-list words with minimal phonemic contrast were selected as linguistic stimuli for the SiP-test, using a method by which the influence of word frequency, neighborhood density, phonotactic probability and orthographic transparency was controlled. All test words were recorded to sound files, of which the accuracy was validated in a listening experiment with 28 normal-hearing adult native speakers of Swedish. In the third study, a method was developed by which realistic masker sounds, spectrally matched to each set of test phonemes in the SiP-test material, were generated for the SiP-test based on a database of urban outdoor sound events. In the fourth study, the validity of six statistical methods for significance testing of observed score differences applicable to the SiP-test were investigated. Analyses were based both on computer simulated test sessions and on SiP-test sessions with human participants. In the latter, the SiP-test speech material was presented against the urban outdoor masker sounds at different difficulty levels to 74 people with normal hearing to severe hearing loss in a listening experiment using a multiple-alternative forced choice paradigm. Based on the results, a computational prediction model for the SiP test was developed, by which the underlying success probability of specific SiP-test trials could be estimated. In turn, this enabled the use of significance-test methods based on the Poisson’s binomial distribution, resulting in improved significance-test validity. In addition, the human SiP-test results were analyzed in terms of test-retest reliability, learning effects, content-, construct- and criterion validity within the remains of the thesis.

Purpose: As factors influencing human word perception are important in the construction of speech perception tests used within the speech and hearing sciences, the purposes of this study were as follows: first, to develop algorithms that can be used to calculate different types of word metrics that influence the speed and accuracy of word perception and, second, to create a database in which those word metrics were calculated for a large set of Swedish words.

Method: Based on a revision of a large Swedish phonetic dictionary, data and algorithms were developed by which various frequency metrics, word length metrics, semantic metrics, neighborhood metrics, phonotactic metrics, and orthographic transparency metrics were calculated for each word in the dictionary. Of the various word metric algorithms used, some were Swedish language reimplementations of previously published algorithms, and some were developed in this study.

Results: The results of this study have been gathered in a Swedish word metric database called the AFC-list. The AFC-list consists of 816,404 phonetically transcribed Swedish words, all supplied with the word metric data calculated. The full AFC-list has been made publicly available under the Creative Commons Attribution 4.0 International license.

Conclusion: The results of this study constitute an extensive linguistic resource for the process of selecting test items in new well-controlled speech perception tests in the Swedish language.

Supplemental Material: https://doi.org/10.23641/asha.8330009.