Project 3: Perceptual salience of FPD 1 - Segmentation of speech into words and phrases
Speech-segmentation research investigates how listeners identify word boundaries in connected speech. Many perceptual/linguistic mechanisms supporting speech segmentation are documented, but evidence for their application to real-life speech is virtually non-existent. Having established the distributional validity and reliability of segmentation cues from spontaneous-speech corpora (Project 1), S2S will fill this gap by testing listeners’ reliance on such cues when hearing spontaneous speech. The aim is to provide the most ecologically valid, empirically supported account of speech segmentation to date, thereby helping to model efficient speech understanding.
An original new focus will compare the role of stress-based word segmentation in Czech, Dutch, English, and French. Word-initial stress is a useful heuristic for word segmentation in English, presumably because most common English words are stress-initial. At first sight, the contribution of stress to word segmentation should be even greater in Czech because all Czech words are stress-initial. But the phonetics of stress are so distinct in the two languages that, in practice, one would expect many differences in the contribution of stress to segmentation. For example, English stress makes its largest contribution in adverse listening conditions, but Czech stress, less based on prominence, might contribute relatively less to segmentation in noisy conditions. S2S will test trade-offs between reliability and intelligibility cross-linguistically. Dutch and English share similar distributional and acoustic stress properties, so should show comparable results. French, however, is characterised by fixed word-final lengthening. Thus French shows reliable prosodic patterns, but, like Czech, is unlikely to survive noisy conditions. The above predictions are for L1 listeners. Segmentation patterns will be studied for L2 listeners as well, e.g. English speakers learning French or Czech speaker learning English. Effectiveness of L2 word segmentation should depend on the similarity between L1 and L2 stress patterns and FPD (notion of cue transferability), interacting with auditory salience when in noise (Theme II).
Methods S2S has expertise in standard phonetic and psycholinguistic paradigms (e.g. identification, discrimination, word-monitoring, word-spotting, cross-modal priming, intelligibility in noise), and pause-detection, a paradigm developed by Bristol, validated by York, which reflects degree of lexical processing when the pause is heard. It provides fast reaction times, thought to aid understanding of the time course of processing, and is easily transferable across languages, since the event to detect (silence) is defined similarly for all speakers of all languages.
Materials will vary from cross-spliced spontaneous speech, to copy-synthesized speech using PROCSY (allowing fast spectrotemporal manipulation of natural-sounding synthetic speech). Work on English and Dutch will build on that in the literature. Casual non-spontaneous speech may also be used for full control as appropriate.
Young researchers One ESR (Rauch) is based at Bristol. Her visits to other site will be dictated by her needs: she will probably visit Prague, Geneva, Nijmegen and possibly Aix; and Cambridge or York if PROCSY is used.
Links Project 3 links to other Theme 1 projects, and to projects 5 and 8, and it will inform Theme IV. The project entails collaboration between phoneticians, other linguists (phonologists/morphologists), and psychologists, possibly pairing computational psychologists and experimental psychologists too.
Working on this project: » Dr Sven Mattys » Dr Mirjam Ernestus » Prof Sarah Hawkins » Olesya Rauch » Prof Jeff Bowers » Dr Laurence White
