The transfer of connected speech behaviours: <br/>external sandhi and glottalization in English-accented German

Projektinformationen

TRASANGLOTT

ID Finanzhilfevereinbarung: 623394

Projekt abgeschlossen

Startdatum 1 Oktober 2014

Enddatum 30 September 2016

Finanziert unter

• Specific programme "People" implementing the Seventh Framework Programme of the European Community for research, technological development and demonstration activities (2007 to 2013)

Gesamtkosten € 221 606,40

EU-Beitrag € 221 606,40

Koordiniert durch

QUEEN MARGARET UNIVERSITY, EDINBURGH
United Kingdom

Final Report Summary - TRASANGLOTT (The transfer of connected speech behaviours: external sandhi and glottalization in English-accented German)

The aims of the present project are to analyze:
1. the extent and nature of external sandhi and of glottalization of word-initial vowels in English-accented German as compared to native English productions, in relationship to prosodic boundaries,
2. the influence of external sandhi and of glottalization on word recognition by English and German listeners,
3. the influence of external sandhi and of glottalization on the perception of phrase boundaries by English and German listeners.
External-sandhi consists in the modification of sounds at word boundaries, especially in connected speech, for example the assimilation of final /s/ and initial /j/ in miss you producing as output /∫/. Glottalizations consist in the closing and subsequent abrupt opening of the vocal folds – glottal stops – or in perceptually equivalent irregular and low-frequency vocal fold vibrations – creaky voice –. The traditional view is that glottalisation blocks external sandhi, however external sandhi has been previously found to co-occur with glottalisation.
As a first step in order to meet the first project objective, an intensive study of the literature regarding external sandhi and glottalization in English and German was carried out. Secondly the experimental design for carrying out the experiments with ultrasound tongue imaging was developed. In order to create the text materials, which participants would read, it was necessary to search the WebCelex, and several Wikipedia and Google Books databases. The selection of the materials was challenging since the sentences should vary phrasing and accent of the target words, and also trigger external sandhi and glottalization. Moreover, the sentence materials needed to be comparable in German and in English, to allow cross-linguistic analysis.
Dr. Bissiri was introduced to the use of the ultrasound machine and to the analysis of ultrasound data by her project supervisor, Prof. James M. Scobbie, and several colleagues at the Clinical Audiology, Speech and Language (CASL) Research Centre. The training was practical because it involved mastering the different tasks required by the project: setting up the software for the ultrasound recordings, carrying out the recordings and analysing the ultrasound data. Initially, pilot recordings were carried out in order to adjust the materials, the length of the recordings and the pauses, so that they were comfortable for the speakers, and to make sure that the text material chosen was suitable for the purpose. Then recordings with ten Southern British English speakers with German as second language and eight German native speakers were carried out, each recording lasting about two hours.
A method was developed to first carry out automatic acoustic labelling of the data and of the glottalizations. This automatic acoustic labelling was then corrected manually. The manual labelling of tongue contours, which is an extensive task requiring a large time span, was subsequently carried out. Investigations were carried out in the R environment, in order to find a suitable method to analyse the dynamic character of the data. Among the methods explored for the analysis and applied to the data were Mixed Models and Functional Data Analysis.
Further objectives of the project are related to perception and specifically to a) the influence of external sandhi and of glottalization on word recognition by English and German listeners, and b) the influence of external sandhi and of glottalization on the perception of phrase boundaries by English and German listeners. Regarding objective a) a more robust method than word recognition was developed. In order to avoid the collection of reaction times, which can be less robust than responses, and in order to avoid the use of real words as stimuli, which can trigger a native language effect, the perceptual data consisted into responses to nonsense stimuli which contain glottalisations. Since also for the task in objective b) the native language effect would have been problematic, it was decided to focus on two perceptual experiments with glottalised nonsense words.
In the first experiment, the duration of the glottalized stretches in the stimuli was varied along a 10-step continuum, and the stimuli were associated to pitch contours by the perceptual experiment participants. The assumption behind was that German participants, unlike English participants, will interpret word-initial glottalization as a segmental word-boundary phenomenon, rather than an intonation phenomenon to be associated to falling pitch, and this would lead to a bias towards level pitch responses. The data of German listeners was analysed, and the experiment was also carried out with 23 English native speakers at Queen Margaret University. The second experiment replicates a model previously employed by Dr. Bissiri to investigate the perception of pitch in longer stretches of glottalisations. For this experiment English native speakers were recruited at Queen Margaret University. Data analysis has been carried out and the preparation of two articles with the perceptual results for a phonetic journal is in process.
As part of Dr. Bissiri’s training, in order to increase her teaching skills and experience, Dr. Bissiri has been teaching the class “Practical Phonetics” at Queen Margaret University.
The paper presented at the SPECOM conference in Novi Sad contains some useful results for manipulating glottalizations, which can be employed to create stimuli for perceptual experiments. The paper investigated the analysis and synthesis of glottalisation phenomena in German–accented English. Word-initial glottalization was manually annotated in a subset of a German–accented English speech corpus. For each glottalized segment, time-normalized F0 and log–energy contours were produced and principal component analysis was performed on the contour sets in order to reduce their dimensionality. Centroid contours of the PC clusters were used for contour reconstruction in the resynthesis experiments. The prototype intonation and intensity contours were superimposed over non-glottalized wordinitial vowels in order to resynthesize creaky voice.
The investigation on the extent and nature of external sandhi and of glottalization of word-initial vowels in English-accented German as compared to native English productions has so far shown the following main results: 1) /r/-sandhi occurs only in English in phrase-medial deaccented condition, 2) English native speakers do not transfer sandhi to their German L2; 3) co-occurrence of /r/-sandhi and glottalisation has not been found; 4) functional data analysis of tongue tip velocity has shown that the tongue tip moves faster in phrase-medial than in phrase boundary condition, probably due to phrase final lengthening; 3) Tongue tip velocity is not influenced by word accent nor by the presence of glottalisation.
Regarding the perceptual results, we found that the duration of the glottalized stretches affected the categorization of the stimuli by the German listeners, and that the German listeners were not influenced by the preceding pitch context, unlike in a previous study on longer stretches of glottalization of constant duration. The responses by the English natives, on the contrary, were influenced by the preceding pitch context. Possibly German listeners interpret shorter stretches of glottalisation as segmental word-boundary-phenomena rather than as intonation.
We expect further data analysis to confirm the results so far. The project results implicate that word boundary phenomena, such as glottalizations and external sandhi, are relevant and complex and should be explicitly addressed in language teaching and training. Glottalizations and external sandhi are influenced by prosodic structure. Differently from glottalizations, external sandhi is not transferred from the native language to second language productions, and glottalizations are perceived differently depending on their duration and the previous pitch context. The extensive articulatory database collected for the present project will be employed further to analyse other aspects of glottalisations and external sandhi. The results of the present project build the base of further investigations which Dr. Bissiri will carry out for her upcoming Habilitation project at the Technical University of Dresden, during which Dr. Bissiri will also carry out articulatory synthesis of word boundary phenomena.
The present project combined the analysis of prosody and articulation, which is infrequent in phonetic research, and also created an effective method based on Functional Data Analysis to investigate tongue dynamics. Furthermore, the experiment on the authomatic analysis and synthesis of glottalizations could create naturally sounding stimuli with manipulated glottalisation, and such procedures can be effectively employed for further investigations on the production and perception of glottalizations. The project created and/or strenghtened international research collaborations, e.g. of the host institution Queen Margaret University with the Technical University in Dresden and with the Scuola Normale Superiore in Pisa. These collaborations will have a long-term positive effect. Moreover, research on German and English as second languages is a quite relevant and useful topic for today’s need of learning foreign languages in the context of global communication. This fact makes research on second languages particularly attractive for researchers and funding bodies, thus contributing to European excellence and competitiveness.
Project website: https://sites.google.com/a/qmu.ac.uk/maria-paola-bissiri/home/trasanglott