Abstract:

Based on sequential processing theory, parallel processing theory and interactive activation theory, some classic models about eye movement control are constructed to simulate the eye movements, experimental effects, and to explore the possible cognitive mechanisms of information processing during reading. A systematic and in-depth comparative analysis of five classic models (E-Z Reader 10th, SWIFT, Glenmore, OB1 Reader and CRM) was made in this paper. Specifically, the similarities and differences among these models were analyzed and discussed.

There are similarities among the five models: visual acuity and word frequency are all considered to influence the recognition of letters/ Chinese characters and words; common eye movement patterns including fixation, regression and saccade are all simulated; typical experimental effects such as word frequency, word length, word prediction, and preview effects are well explained.

The core difference of the five models is whether the distribution of attention in the perceptual span is sequential or parallel, and this difference evoked significant consequences on the five models. First, different claims were made regarding letter/Chinese character recognition and word recognition. The sequential attention shift (SAS) models claimed that multiple words cannot be processed simultaneously, while the parallel graded processing (PG) models raised the opposite argument. Second, different interpretations of common eye movement patterns were made. For example, the SAS models claimed that regressions are derived from post-lexical integration, while the PG models argued that regressions come from lexical recognition. Third, different explanations for some typical effects were made. For example, the SAS models argued that readers cannot obtain semantic meanings from preview, and there is no parafoveal on foveal effect, while the PG models made the opposite argument. Forth, each model can explain some specific effects that other models cannot. For example, based on the special feature of Chinese script (the lack of word boundary demarcation), CRM raised a reasonable explanation for the word segmentation and preferred viewing location in Chinese reading; E-Z Reader simulated some effects of post-lexical integration which were not considered in other models; SWIFT, Glenmore, and OB1 all discussed the effects of extralinguistic factors which were not included in other models.

For the future development of eye movement control models, researchers may need to take the following aspects into consideration. First, post-lexical integration needs to be considered and simulated. Semantic integration plays an important role in reading. But only E-Z Reader contained a module of post-lexical integration and simulated the possible mechanism of semantic integration. More attentions should be laid on the semantic integration procedure for further model development. Second, the word order coding in reading should be considered. Compared to the SAS models which provided a relatively intuitive answer to word order coding, the PG models require a clear answer to this question. OB1 Reader tried to solve this problem by adding a spatiotopic representation module, which however is not suitable for Chinese reading due to the specific features of Chinese. Further studies need to focus on understanding how word order is coded in Chinese reading under the framework of parallel processing. Third, some extralinguistic factors need to be considered. Now, SWIFT, Glenmore, and OB1 Reader discussed the influences of reader specificity or task difficulty in eye movement patterns in reading. In the future, more extralinguistic factors such as age, gender, intelligence, attention and language proficiency level should be considered to make the model more interpretable. Forth, general standards to compare the suitability of models should be made. Each model simulated specific experimental effects based on their corresponding empirical data, which makes it difficult to compare different models quantitatively and directly. Future studies should try to build a unified large-scale database for the convenience of comparing the explanatory power of different models for the same effect. Last, the possibility of cross-language explanations should be explored. Each of the existing models was based on a specific language. In future studies, researchers may try to explore whether the models based on a specific language can be applied to other languages.

Key words: E-Z Reader, SWIFT, Glenmore, OB1 Reader, CRM, eye movement control