An Object-Oriented Framework for Constructing and Comparing Models of Eye Movement Control in Reading

Ha Nguyen¹ (nguyenha@cs.tu-berlin.de), Ronan G. Reilly², & Mel � Cinn�ide²

¹Department of Computer Science, Technical University of Berlin, Berlin, Germany

²Department of Computer Science, National University of Ireland, Dublin, Belfield, Dublin 4, Ireland

There has been a dramatic growth in recent years in the use of computational models to try to understand the reading process, and specifically the issue of eye movement control in reading (e.g., Reilly & O�Regan, 1998; Reichle, Pollatsek, Fisher & Rayner, 1999; Legge, Klitz, & Tjan, 1997).  With this growth comes the need for software support not just for the modelling process itself, but also for the task of comparing and contrasting competing models. 

Computational modelling does not just involve writing one program.  The program is usually the vehicle for a series of computational experiments by its designer.  As the experiments progress, it is usual for the program to evolve through a number of minor and occasionally major revisions.  In addition to comparing the performance of the author’s model with that of a competing one, it is also desirable, ultimately, to make the model, or family of models, available to other researchers who may wish to test or extend it.  At present, this is far from easy, since the extending another’s code is difficult enough for even skilled programmers.

What is required is a software system for constructing models that encapsulates domain expertise and is applicable to a wide variety of models, but not so specific that extending it requires too many changes in too many places in the code.  If we can find this modelling middle way, it will have the added benefit of making the relationships between different models clearer than they might otherwise be.  By localising where extensions and modifications can occur in the code, the differences and similarities of models becomes a lot easier to see.  The goal of this paper is to demonstrate the utility of the software engineering concept of Frameworks to the task of modelling eye movement control in reading.  The approach we propose, however, is readily generalisable to other cognitive modelling domains.  Frameworks provide a means of capturing features of the modelling process, such as clearly highlighting the family resemblances of models.  More importantly it provides a toolkit for the development of new models by extending and re-using existing model components and architectures.  An implementation and comparison of the models of Reilly and O’Regan, and Reichle et al. will be used to illustrate the usefulness of the frameworks approach.

References

Legge, G.E., Klitz, T.S.& Tjan, B.S (1997). Mr Chips: An ideal-observer model of reading. Psychological Review, 104, 524-553.

Reichle, E.D., Pollatsek, A., Fisher, D.L., & Rayner, K. (1999). Eye movement control in reading: accounting for initial fixation locations and refixations within the E-Z Reader model.  Vision Research, 39, 4403-4411.

Reilly, R.G. & O�Regan, J.K. (1998).  Eye movement control during reading: A simulation of some word-targeting strategies.  Vision Research, 38, 303-317.