Projects in Previous Years

2nd Semester 2016/17: Incremental interpretation of anaphoric relations

Jakub Dotlacil
If you are interested in this project, please contact the instructor by email.
In the past two decades, a lot of research in psycholinguistics investigated how syntax affects incremental interpretation of anaphoric relations. Such research shows that there is a (very short-lived) interval at which syntactically illicit interpretations are considered. For example, in the sentence "John thinks that Bill likes himself", the anaphor himself cannot take John as its antecedent. Yet, during reading, it can be shown that English speakers do consider this impossible interpretation (and discard it eventually) (see e.g., Sturt, The time-course of the application of binding constraints in reference resolution, 2003).

Independently, in the semantic literature, it has been uncovered that there are several discourse factors that can block antecedent-anaphor relations (studied in DRT, a.o., cf. Kamp, 1982). For example, while in the discourse "John has a car. It is black", the phrase a car can antecede the pronoun, the same relation is impossible if the indefinite appears in the scope of negation, as in "John does not have a car. It is black". What is not known is how these discourse/semantic factors affect incremental interpretation (i.e., do English speakers consider the illicit interpretation and discard it eventually, in the same way as they consider syntactically illicit relations?).

There are two parts to the project. In one part, (some of) you will design an experiment that targets the question discussed above. The experiment will be built using the self-paced reading method, which is relatively easy to learn, yet it yields reliable results regarding incremental interpretation (Just et al., 1982, Paradigms and processes in reading comprehension). In the other part, (some of) you will design a computational cognitive model in ACT-R that can simulate the experimental findings. This second part (cognitive modelling) is optional (that is, if nobody is interested in this, we can focus only on the experimental part).

Regular meetings in which we discuss literature and your progress in designing the experiment/model of the experimental findings.
If you want to build a cognitive model, you have to know a bit of Python - basic to intermediate knowledge is completely sufficient. You don't need to know ACT-R, but you should be ready to learn it. I'll supply basic materials.
To pass, you should either (i) design and run an experiment, and write a report on it, or (ii) learn ACT-R and build a computational model related to the planned experiment.