You are in: Home » Research at the Unit » Speech and Language
The Bayesian Reader model (Norris, 2006).
Let's pretend that human perception was perfect. How would a perfect, or optimally designed, system read? Might people come pretty close to behaving like an optimal system? Rather surprisingly, it seems that they do. If we make the assumption that perception works by a process of collecting noisy information from the input (in this case, from the earliest stages of the visual system) we can construct a formal model of how people should behave when reading individual words, or when performing common laboratory tasks such as deciding whether letters form real words or nonsense words. This is the principle behind the Bayesian Reader model (Norris, 2006). This simple idea turns out to give a principled explanation of a wide range of experimental data on reading.
One example of this model's success is in explaining why common words are easier to read than rare words. This seems so obvious that it hardly needs an explanation, but there are lots of reasons why it might be so. How easy a word is to read is approximately a logarithmic function of how frequently it occurs in the language. Why should this be a logarithmic function rather than anything else? Well, it turns out that this is exactly how an optimal system has to behave. By assuming that people approximate optimal recognisers, we get the form of this function for free.
One of the big benefits of this style of modelling (see also the Shortlist B model of Norris & McQueen, 2008) is that almost everything about the model follows automatically from the assumption that people are behaving almost optimally. The model has now extended to cover masked priming (Norris & Kinoshita, 2008), modelling reaction-time distributions (Norris, 2009), and understanding how the order to letters is represented during reading (Kinoshita & Norris, in press).