Researcher Spotlight: Michael Wagner
How we pronounce a sentence can make the difference between making a statement or asking a question. By shifting the emphasis, for example, we can indicate which aspects of the sentence are important; or, again, mispronouncing a sentence can make it hard or impossible to understand—as you will know if you have ever had your computer read something to you. We convey a lot of information with speech by the way in which we pronounce a sentence, by its prosody. We all have internalized conventions about prosody, just like we have internalized conventions about what particular words mean and how they can be combined to form sentences; and languages differ dramatically in what these conventions are. Investigating prosody requires integrating insights from different disciplines, ranging from signal processing to modeling the meaning of prosody. The research of Professor Michael Wagner, Canada Research Chair in Speech and Language Processing, Department of Linguistics does just that. His main research tools are speech production and speech perception experiments that aim at a developing a better understanding of the answers to three questions: what information does a speaker encode in the prosody of a sentence? How is this information decoded by the listener when processing speech? And how do languages differ in their prosodic systems?
In 2009, Professor Wagner’s research was funded by a Canada Foundation for Innovation (CFI) Grant. This infrastructure would allow Professor Wagner to investigate speech prosody in production and perception. Integrating insights from different disciplines in the cognitive sciences in novel ways, this research will make progress on two key questions: first, what information is reflected in the prosody of different languages and how and second, how much of this information is retrieved by the listener, and how this information is used in speech processing. With the help of the CFI Grant, Professor Wagner set-up a perception and eye-tracking lab in the Linguistics Department at McGill University. Today, this lab allows researchers at McGill to run the experiments necessary to understand the apprehension and on-line processing of phonetic cues during speech perception. In addition, the new eye-tracking equipment provided by the lab allows running dialogue experiments while tracking the eye-movements of two subjects at the same time, a unique experimental set up and the first laboratory of its kind at McGill.
Professor Wagner’s prosody research has also funded by the FRQSC New Researcher Grant (Établissement de nouveaux professeurs-chercheurs) and the SSHRC Standard Research Grant. This funding has allowed him to investigate influence of a discourse context on the prosody of utterances and to evaluate whether prosodic phrasing encoded with categorically different prosodic boundaries or with boundaries that differ in strength relative to each other. He has also been investigating how this prosodic constituency maps to syntactic constituency and his research will describe what prosody phrasing can tell us about the incremental building of syntactic structure in speech. These research projects, Professor Wagner explains, have potential implications in a wide range of areas, including speech therapy, and the comprehensibility of synthesized speech and the recognition rate in speech recognition systems.
Professor Wagner, with Professor Mats Rooth, Cornell University, was also one of eight winners of the first round of the prestigious international competition, Digging into Data, that challenged scholars to devise innovative humanities and social science research projects using large-scale data analysis. His project harvested audio and transcribed data from podcasts, news broadcasts, public and educational lectures and other sources to create a massive corpus of speech. To achieve this, Professor Wagner and his team developed tools required to analyze the different uses of prosody (rhythm, stress and intonation). An automatic transcriber for sound data based on a speech recognizer is available from his lab webpage. These tools help analyze distinctions of prosody in spoken language.