Archive for December 9th, 2013
Speaker: Morgan Sonderegger (McGill)
Title: Phonetic and phonological variation on reality television: dynamics and interspeaker variation
Date/Time: Monday, Dec 9, 5:30p
This talk examines two types of variability in phonetics and phonology about which relatively little is known: (1) the dynamics of the accents of individuals from day to day, and (2) differences among speakers of the same language in the structure of variability. We examine a number of variables (focusing on VOT and t/d deletion) in a corpus of spontaneous speech from a setting which is particularly well-suited to examining (1) and (2) — a British reality television show (Big Brother UK) where individuals live in a house with no outside contact for three months — using statistical models of synchronic variability across speakers, and dynamics within individual speakers. Speakers show several qualitatively different types of dynamics; the most common types are day-by-day variability and absolute stability in the use of a variable. There is a surprising degree of variability across speakers in the quantitative strength of each variable’s conditioning factors, e.g. the effect of place of articulation on VOT. However, nearly all speakers show the same qualitative effects of each conditioning factor (e.g., bilabials < velars for VOT). We discuss the relevance of these findings for theories of language change, phonetics, and phonology, as well as directions for future work with this corpus.
With Fall semester classes ending this week, we will go on our regular Winter semi-hiatus until classes resume at the beginning of February. As always, we will report exciting breaking news as it happens (that’s why it’s just a semi-hiatus), but otherwise — see you in 2014!
Speaker: Sandrien van Ommen (Utrecht Institute of Linguistics, OTS) (joint study with Rene Kager)
Date/Time: Friday, Dec 13, 3 pm (please note unusual time)
Title: Language-specific metrical segmentation in Dutch, Turkish, Polish and Hungarian
With the current study we investigated the role of stress as a boundary marker in processing. Previous studies have shown that listeners interpret stressed or strong syllables as potential word-beginnings in a.o. English (Cutler & Norris, 1988), and Dutch (Quené & Koster, 1998; Vroomen & de Gelder, 1995). This is interpreted as evidence for the Metrical Segmentation Hypothesis, which predicts that listeners have and use a parsing ability based on edge-aligned stress. Unfortunately, most empirical evidence supporting this hypothesis comes from languages with (statistically dominant) word-initial stress. Evidence for a facilitatory effect of right-edge aligned stress is sparse and inconclusive (see a.o. Toro-Soto et al., 2007, Cunillera et al. 2008, Kabak et al., 2010). We designed a cross-linguistic experiment to address the question of language-specificity in metrical segmentation. In this experiment, we measured response latencies in a non-word spotting task with six different metrical conditions. The participants were speakers of Dutch (penultimate word-stress, variable), Polish (penultimate word-stress, fixed), Turkish (word-final stress, variable) and Hungarian (word-initial stress, fixed).
Besides finding the expected overall effect of facilitation of the native canonical stress pattern in (non-)word segmentation, we conclude to have found a language-specific anticipitory use of stress in segmentation. Furthermore, the results invite us to further investigate the role of peripherality and variability of stress in processing. To gain more insight into what role these factors may have, we recently started designing a computational model for the acquisition and use of metrical patterns. This is a very recent and tentative project that we welcome discussion on.