Our keynote speakers are:
- Michael Ashby (University College London)
- Harald Baayen (University of Alberta/University of Tübingen)
- Chris Culy (University of Tübingen)
- Mark Huckvale (University College London)
Visualizing Vowels: Restoring Some Lost Images
Michael Ashby, University College London
Visual representations are crucial to our understanding of vowels: the vocal-tract section, the speech waveform, the acoustic spectrum, and the multi-dimensional vowel space (phonetic or perceptual). None of these now-familiar representations were obtained easily, and the conventionalized versions which appear in every introductory textbook originally faced competition from forgotten rivals which still deserve investigation and understanding. This paper presents new historiographical findings on a range of topics: the discovery of lost X-ray data of Tsutomu Chiba (1883-1959), the complex story of the interplay between X-ray data and the Cardinal Vowel system, re-discovery and restoration of the first X-ray sound film from 1935, Robert Curry’s pioneering efforts to capture cathode-ray oscillograms on film, and the remarkable elliptical vowel space proposed by Sun-Gee Gim in 1937.
Visualization of Linguistic Data Using Generalized Additive Models
R. Harald Baayen, University of Tübingen & University of Alberta
Generalized additive models (GAMs, see, e.g., Wood, 2006) provide a flexible toolkit for modeling complex prediction surfaces and hypersurfaces. For understanding such surfaces, visualization, for instance with contour plots or perspective plots, is essential. The goal of my presentation is twofold.
First, I will illustrate the potential of GAMs for linguistic research, using as examples data from a dialectometric study (Wieling et al., 2011), a study using evoked response potentials to auditory stimuli (Kryuckova et al., 2012), and work in progress on the analysis of first fixation durations in an eye-tracking study of compound reading (Kuperman & Baayen, in progress).
Second, the complex regression surfaces revealed by GAMs can be quite difficult to interpret. I will argue that rather than ignoring unwelcome complexity, we should embrace it, and search for computational models that correctly predict these complex surfaces that GAMs detect in linguistic data. I will discuss one example data set concerning the lexical processing of compounds, contrasting two very different computational approaches (Baayen, 2010 and Baayen et al., 2011) to understanding the observed regression surfaces.
On the basis of these examples, I will argue that GAMs provide an analytic tool in which visualization and theory-building go hand in hand and have actually become inseparable.
Baayen, R.H. (2010) The directed compound graph of English. An exploration of lexical connectivity and its processing consequences. In S. Olson (ed.), New impulses in word-formation (Linguistische Berichte Sonderheft 17), Buske, Hamburg, 383-402.
Baayen, R. H., Milin, P., Filipovic Durdevic, D., Hendrix, P. and Marelli, M. (2011), An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review 118, 438-482.
Kryuchkova, T., Tucker, B. V., Wurm, L., and Baayen (2012) R. H., Danger and usefulness in auditory lexical processing: evidence from electroencephalography. Brain and Language 122, 81-91.
Wieling, M., Nerbonne, J. and Baayen, R. H. (2011). Quantitative Social Dialectology: Explaining Linguistic Variation Geographically and Socially. PLoS ONE 6(9): e23613. doi:10.1371/journal.pone.0023613.
Wood, S. (2006). Generalized additive models. Chapman/Hall.
Some Challenges and Directions for the Visualization of Language and Linguistic Data
Chris Culy, University of Tübingen
While linguists and language professionsals have long used visual representations of information (e.g. syntax diagrams, language family trees, lexical meaning diagrams, etc.), digital, interactive visualizations are becoming more popular due to their usefulness and flexibility.
At the same time, language and linguistic (L/L) data poses some interesting challenges for visualization, due to the fact that L/L data has some significant differences from other types of data. The most important is that language is not mappable: it cannot in general be represented in a more compact, human understandable way, unlike e.g. numbers, which can be represented by location, or size.
I will show why these properties of L/L data are a challenge for visualization and I will give some working examples of how they might be addressed. I will also discuss how the visualization of L/L data can contribute to further explorations of the connections between cognitive psychology and computer interfaces. Finally, I will present some concrete suggestions for future directions for the development of the field of visualization of L/L data.
Speech as Visible Patterns of Sound
Mark Huckvale, Speech, Hearing & Phonetic Sciences, University College London
Spoken language is built on the reliable communication of information through tiny, rapid fluctuations of air pressure which are both invisible and ephemeral. Our species has learned to exploit its vocal apparatus to encode meaningful utterances into patterns of sound and to exploit its auditory apparatus to decode them. Insights into these encoding and decoding processes can be gained by making these patterns of sound visible and permanent. In this talk I will demonstrate a number of ways in which we can look at speech as patterns of sound. I will cover a broad range of linguistic levels: from pressure variations, spectral analysis and neural firing to phonetic properties, lexical contrast and dialogue structure. On the way I hope to show how visualisation can aid understanding of how speech communication works, and can also create a sense of wonder at the marvel that it works at all.