City Research Online

Abstract

This thesis began with the assumption that the clinical goal is to identify articulatory, segmental speech errors. The main objective was to explore the potential of using phonetic-contrast analysis for this purpose in Belgian Dutch speakers with dysarthria. In this approach, the speaker reads a list of monosyllabic words, and the listener either transcribes each word orthographically (open mode) or chooses from among the target and a set of foils (closed mode). The perceived phonemic substitutions are then coded as contrasts in a single phonetic feature, e.g., initial-stop voicing. The secondary objectives of the thesis were to (a) identify vulnerable phonetic contrasts in Belgian Dutch dysarthria and (b) assess the correlation between word-reading accuracy and intelligibility in spontaneous speech.

A phonemically-balanced word list was developed (117 words). It was read by 10 subjects with dysarthria (due to various aetiologies) and 8 neurotypical controls, all from the Antwerp region. The speakers with dysarthria also delivered monologues on topics of their choosing. Online listening sessions were conducted in which the single-word stimuli were identified using both an open and a closed mode. The monologues were assessed using a syllable-accuracy metric derived from orthographic transcription (Lagerberg et al., 2014).

Phonetic-contrast analysis showed significant promise with regard to consonants: more than 78% of substitutions could be coded using 13 contrast categories. Vowel confusions, however, did not typically lend themselves to categorisation based on a single phonetic feature (e.g., height), partly due to the configuration of the Antwerpian vowel space. Contrasts that were no more vulnerable in dysarthria than in the control group included the voicing of word-initial stops and confusions between high, front vowels. Word accuracy for the dysarthric group was significantly higher for the open than the closed mode (mean absolute difference 13.1% ± 6.9%). The two response modes also yielded different error profiles. In each mode, for each dysarthric speaker, vowel and consonant contrasts were separately ranked according to error rate. Pearson’s r between the ranks for the two modes, calculated for each speaker, ranged from 0.34 - 0.72 for consonants and 0.17 - 0.86 for vowels. Prominent consonant confusions included initial-stop devoicing, singletons perceived as clusters (a distortion error), and confusions between fricatives and another manner of articulation. Vowel confusions typically corresponded to either (i) reductions, such as shortening or monophthongisation, or (ii) confusions between phonemes that are relatively close together in F1-F2 space in the Antwerp accent. A correlation of 0.76 (one-tailed p < 0.01) was observed between word intelligibility and intelligibility in spontaneous speech. Overall, the findings suggest a complex set of interactions between speaker characteristics, listener characteristics and the methods used to elicit and analyse speech.

Publication Type:	Thesis (Doctoral)
Subjects:	R Medicine
Departments:	Doctoral Theses School of Health & Psychological Sciences > School of Health & Psychological Sciences Doctoral Theses School of Health & Psychological Sciences School of Health & Psychological Sciences > Language & Communication Science

Export

Downloads