Grant: Discovery Grants Program - Individual

Natural language processing (NLP) is the study of having a computer automatically analyze and interpret human language, which is used in software applications such as next word prediction - which involves predicting the next word when typing on a smartphone and machine translations - which involves translation from one language to another language. My interest in NLP is to broaden the scope of how NLP datasets and models are used so that there can be a broader representation of unique users and their subsequent positive experiences in the day-to-day NLP features found in such software applications. The long-term objective of my research program is to improve the benefits to a broader range of unique and discrete individual users of NLP applications in various software and related NLP tasks. This involves the study and involvement of a wide range of NLP tasks and analyses that prioritize performance for each individual. Each person uses a language in their own unique manner known as an idiolect. Applications in NLP rarely consider an individual's idiolect and often focus on designing applications that benefit the largest group of people of a populace. This is manifested in training the applications on text from multiple people with the objective to achieve a better overall score on a group of people without considering the individual people for whom the applications perform poorly. This program of research focuses on developing NLP applications that prioritize strong performance for each individual to increase the benefits that people can receive regardless of their idiolect. These author-aware applications are known as personalized applications. Personalized applications consider a single individual instead of text from multiple people. Research that focuses on designing personalized applications has been difficult to conduct, partly due to the limited resources available, such as suitable datasets to train and evaluate personalized applications. Therefore, a short-term objective of my program of research is to develop corpora - large collections of text - and datasets that allow for extensive research in this area. These proposed datasets will also allow researchers to study which people are suffering from underperforming NLP applications, which will help prioritize future work in a direction that makes it more impactful. My second short-term objective is to design and develop personalized models for the various considered tasks. The tasks that I will explore include language modeling, (estimating the probability of a sequence of words), word sense disambiguation (assigning a dictionary-like definition called a sense to a word in context), predominant sense detection (determining the most frequently used sense of a word), language acquisition (determining a person's ability to learn a new language), humour detection (determining if a snippet of text is considered humorous), and readability (determining how readable a snippet of text is).