details may change
Conversation or dialogue is the most natural way in which we humans use language, and arguably the holy grail of language-enabled AI systems. With machine learning advancing at such a rapid pace, we now have powerful tools for modelling interacting agents. Yet, to model human-like conversational abilities remains remarkably difficult. This course will examine what makes dialogue so challenging, delving into classic and contemporary research in linguistics, cognitive science, NLP and AI.
This is a research-oriented course. It will involve reading and critiquing current related work and completing a small-scale research project. The goal of the course is to help you develop the following:
- Familiarity with fundamental concepts in the study of dialogue interaction
- Deep understanding of the theoretical and empirical basis of current computational models and their limitations
- Practice in carrying out novel research related to dialogue modelling
Although we will discuss current NLP and machine learning methods applied to dialogue – and you will have the chance to work with these methods in the final project if you have the right background – the goal of the course is not to teach you how to build dialogue systems or chatbots.
I expect you to attend all classes and presentation slots: there will be lectures and discussions of specific papers, and you won’t be able to fully benefit from the course if you do not participate in person. Most of the action will happen in class, but you will also be expected to do a substantial amount of reading as preparation.
The grades will be based on participation in in-class discussion and a final project.
Besides background literature (often covering classic work – see topics), we will provide you with a list of concrete readings for discussion (these will be mostly recent technical papers). You will be asked to submit at least one discussion question per reading before the class in which the reading will be discussed. These questions should focus on high-level, conceptual issues (e.g. questions regarding assumptions made by a model, or limitations of a technical approach) rather than being simply clarification questions. In addition to submitting questions on a regular basis, each student will serve as a “discussion leader” for one paper during the course. This will require choosing a paper from the list to read in depth and submiting a short (one or two page) write-up describing the main contributions of the work and situating it in relation to the material discussed in the course so far. You will be expected to lead the discussion of that paper in class (you may be able to do a short presentation with slides – the details are to be decided). There may be more than one “discussion leader” per paper.
During the second part of the course, you will work on a small-scale research project of your choice, individually or in a small group. Students will have the choice between different types of project: for example, implementing computational models, analysing the behviour of an existing model, doing a more empirical study based on data analysis, or writing a detailed research proposal. The final deliverable will be a short report in the form of a conference-style paper. The exact contribution of the project to the final grade is yet to be decided, but it will be around 50%.
This course can accommodate a maximum of 30 students. To avoid having to put a cap on the number of registrations (and thus to select on a first-come first-serve basis), students will be accepted into the course on the basis of motivation and technical background. If interested, please register in the usual manner. You will then be invited to take part in the selection process well before the start of the course.
- 27 Januray 2021: All registered students have received an email with instructions for how to prepare for the selection process.
- 29 January 2021: All registered students have been invited to fill in the selection questionnaire. The deadline to submit the responses is 13:00 on the 5th of February.