Here is a brief report on Ultrafest III, a conference dedicated to ultrasound as a tool in linguistic research. It was in Tucson this year, hosted by the U of AZ, whose gang did a terrific job organizing the event. There is no website for the conference, but we are all to be looking out for a wiki to be established at some point, and hosted by one of the participating labs. If you want to look up any of these labs in the meantime, QMUC has a list of links for them. As does UBC.
The conference was similar to last year’s at UBC in that it was divided into segments focusing on research, methodological issues, and analytical issues. Judging by the progress in methods of participating labs, and the swelling membership, it appears that good things are going to be coming from the ultrasound linguists.
There is not room enough to post a summary of the whole deal, but I can offer some highlights.
The labs represented were the following:
- Cornell (Amanda Miller, Johanna Brugman)
- Haskins (Doug Whalen, Leonardo Oliveira etc.)
- NYU (Lisa Davidson, Stefan Benus)
- Queen Margaret University College (Jim Scobbie, Natasha Zharkova, Yolanda Vasquez-Alavarez etc.)
- U-Arizona (Diana Archangeli, Jeff Mielke, Adam Baker, Sumayya Racy, Peter Richtsmeier, Bryan Meadows, Gwanhi Yun)
- UBC (Bryan Gick, Fiona Campbell, Ian Wilson etc.)
- U-Maryland Baltimore (Maureen Stone)
- U-Quebec a Montreal (Lucie Menard, Jerome Aubin, Johanna-Pascale Roy)
- U of S. Florida (Stefan Frisch, Sylvie Wodzinski etc.)
- U of Toronto, Speech Pathology (Tim Bressman, Heather Flowers, Yarixa Barillas etc.)
This list appears to be a comprehensive representation of ultrasound linguistic labs – as far as I know, no other institution has an ultrasound lab for linguistics (although Joyce McDonough at Rochester also dabbles in it). For those of you who don’t know, Maureen Stone’s lab at UM-Baltimore is the traditional homeland of ultrasound phonetics, having been the only such facility through most of the 80s and 90s – so the ultrasound linguists are always thrilled to have her come to the conference, and they (almost) always do everything she says.
The conference was also attended by at least 5 ‘scouts’ (including me), representing programs interested enough in the paradigm to consider entering the fray. The keynote was delivered by physiologist and self-described tongue guy, Alan Sokoloff of Emory University.
Current Research falls into several categories: fieldwork, pathology, phonetic modeling, experimental phonology, and pedagogical applications.
- Cornell: Field work with Khoisan languages
- Haskins: Tongue root and pharyngeal chamber modeling
- NYU: positional effects on consonant articulation
- Queen Margaret University College: trough effects in VhV sequences, rapid acquisition of stratified sociophonetic articulatory data
- U-Arizona: variants of /r/, L2 pedagogy, intrusive schwas
- UBC: variants of /r/, bilingual articulatory settings
- U-Maryland Baltimore: 3-dimensional modeling
- U-Quebec a Montreal: hyperarticulation in Quebec French contrastive focus
- U of S. Florida: phonetic pedagogy, POA speech errors
- U of Toronto (Speech Pathology): 3-D modeling, effects of oral/dental (re)construction(s) on articulation
The methodological and analytical questions are issues shared by all labs. Methodological issues include (a) whether to restrict and/or correct for movement of the head relative to the transducer, (b) how to image the palate, (c) how to automate aspects of measurements, and (d) how to maximize temporal sampling.
It seems every lab has means of minimizing or eliminating the movement problem, either by immobilizing the head and transducer, or by tracking the movement of each and transforming images post-hoc to make up for it. Where the goal is qualitative observation, not precise measurement, Gick is pushing for less restrictive recording paradigms. UMd, NYU, QMUC, UofT, and USF have all adopted some degree of immobilization. UBC and Haskins both have infrared Opto-track systems to correct for movement. Arizona has developed a decidely more affordable video paradigm for doing so, which in its current state is automated to work over a set or sequence of stills.
Palate imaging becomes feasible once head-transducer movement is addressed. If you can fixate your images during a session or post-hoc, you can consider each one to be fixed relative to the head. Then, if you also can image the palate, you can trace it and lay it over each image. This really opens up the possibilities of what can be measured and analyzed. Maureen Stone and the AZ lab both presented methods of obtaining palate images, which is done by having the subject ingest water (or soy yogurt!) and swallow while being imaged.
The same groups have methods of automating the detecting the tongue outline in individual images. Stone’s software, Edgetrak, uses time-adjacent still images to help in its (greedy-snake) decision algorithm, so that outlines are tracked dynamically. Arizona uses a non-snake automated edge-detection heuristic called Glossatron.
Temporal sampling is a funny issue, and one that is getting more attention now that the other issues are getting worked out. Every lab faces the problem of downsampling: ultrasound equipment can produce 60 to 200 planar scans per second, but tapping into the video feed puts the user down to video rates of 29.97 images per second (except the QMUC group, who are stuck with PAL rates of 25). The misphasing of ultrasound and video refresh rates also introduces stills in which the “current” ultrasound scan is partway across the image.
Developments here are exciting: Haskins and QMUC are both pursuing direct access to the circuitry of their machines, eliminating the video cable middle-man, and maximizing the sampling rate of current technology. Arizona has discovered that digital video uses interlacing to cram 60 frames into 30, and has bootstrapped the scan line to develop a means of extrapolating the time dimension value of a particular spatial sample.
Oh, then there’s the stats issue. This matters if you’re trying to make comparisons…and your analysis depends on your research question. It turns out that tongue outlines are pretty complicated things to quantify in terms of analyzable dimensions. Lisa reported on a promising advanced (indeed, still developing) ANOVA method for comparing curvilinear functions.
Befuddling for all of us (well, for me particularly) is the question of whether, how, and why statistical significance (of some arbitrarily defined physical dimension) translates to linguistic significance. This is definitely going to be a recurring topic.
Like I said before, the conference has no site of its own, but a wiki is soon to come.