Our Research

Research by Graeme Hirst and students:

Our research in computational linguistics emphasizes issues in lexical semantics, pragmatics, and text classification that arise when the methods of computational linguistics are applied to real-world language and real-world problems. The ultimate goal of our research is the development of better computational models of language for use in human-computer interaction and in applications such as text analysis, information retrieval, and machine translation. Two applications that have been especially important in our work are intelligent understanding of text and linguistic assistance to disabled users. In addition, we have developed methods for detecting Alzheimer's disease, or cognitive decline, by looking at long-term diachronic changes in people's writing.

Several themes underlie our research. First, we are concerned with fine-grained aspects of language, as it is really used in the world. Second, although our methods are largely statistical and based on machine-learning, we also emphasize problems in the representation of linguistic and semantic knowledge. Third, many of the problems that we work on and the approaches that we take are inherently interdisciplinary; our work draws on research in such disciplines as theoretical linguistics, English literature, philosophy, and speech-language pathology.


Research by Gerald Penn and students:

My research concerns the study of the structure of human languages as a mathematical and computational system. Not only is natural language processing an integral component of the overall vision of artificial intelligence research, but many of the problems that have defined the rest of AI, logic programming and theoretical computer science research in general can be found inside this very rich empirical domain. More recently, the proliferation of electronically available text over the World Wide Web has created an acute demand for machine translation systems, query answering systems, and text summarization tools for using this vast source of information. In order to approach their full potential, the next generations of these systems must be capable of providing far more precise information about meaning and the relations between the people, objects and locations described by these texts. Recent advances in wireless technology also require a natural means of interacting with ever more miniature devices, and this can only be achieved through spoken input and output. My research thus seeks to provide both a formal perspective on language to realise or improve such applications, and the algorithms to support them.

The strategy I have adopted to pursue this goal consists of several related threads of research activity in specialised logics for computational linguistics, other means of grammar specification, and their applications.


Research by Frank Rudzicz and students:

My lab, SPOClab (Signal Processing and Oral Communication), intersects Computer Science and the Toronto Rehabilitation Institute; the nature of our work is accordingly cross-disciplinary and can involve signal processing, statistical analysis, speech recognition and synthesis, machine learning, human-computer interaction (focussing on assistive technology), speech-language pathology, rehabilitation engineering, (psycho-)linguistics, and speech science, for example.

The purpose of SPOClab is to produce software that helps people with disabilities communicate. This includes looking for new and innovative ways of doing speech recognition; for example, using video of a person's face or knowledge of the mechanics of the vocal tract makes the machine learning process much simpler.

We build tools to help people communicate with both humans and machines. One project involves transforming hard-to-understand speech with advanced signal processing to be more intelligible to human listeners. Another project involves building new ways of interacting with intelligent software that combines audio and video input into single commands. A third project involves studying computer models of brain activity during language production and perception that will help people to communicate merely by thinking.


Research by Suzanne Stevenson and students:

I take a highly multidisciplinary approach to computational linguistics, integrating computational theories and techniques with insights from the fields of linguistics and psycholinguistics.

Currently, a primary focus of my work is the automatic acquisition of linguistic knowledge from large text corpora, using machine learning approaches. Especially challenging is the learning of semantic information about predicates, which is only implicitly represented in text. Another main area of interest is work on cognitive models of human language acquisition and processing. In the latter, I am particularly interested in modelling how human beings so effortlessly come to the intended meaning of an utterance in spite of the high degree of ambiguity in everything we say and hear. I am also very interested in analyzing databases combining words and pictures (such as captioned images on the web), to determine how the words can help disambiguate the images, and vice versa.