Abstract:

Regression is one of common phenomenon in normal reading. But the eye movements go against normal reading order during regressions. According to the eye moving distance during regressing, researchers divide regressions into long-distance regressions and short-distance regressions. Some studies on English suggest that long-distance regressions are guided by memory and two theories can explain the mechanisms of regression, which are the spatial coding hypothesis and the verbally based reconstruction hypothesis. Both theories have their rationality and get some studies’ supporting, but there are some arguments between them. The spatial coding hypothesis suggests that the eye movements are guided by spatial memory during regressions, but the other suggests that they are guided by verbal memory. And researchers find that initial regression is always followed by some small corrective saccades. There is also no consensus among researchers on the guiding mechanism of corrective saccades. The purpose of this study is to explore the guiding mechanism of long-distance regression in Chinese reading and to provide new evidence for the resolution of this dispute.
In order to explore the effect of memory on regressions, we recruited 20 students as participants in Experiment 1, and divided them into 2 groups: reading group and non-reading group. First, the reading group was presented with a sentence. After reading the sentence, the subjects were presented with a target word. The subjects were required to locate the target word in the sentence. The non-reading group was presented with the target word directly, and a sentence was presented before the target word. The subjects were asked to locate the target word in the sentence. The experiment was a 2×2×3 mixed experimental design. The first variable was the reading condition with 2 levels: reading and non-reading; the second variable was the word frequency with 2 levels: high-frequency words and low-frequency words; the third variable was word location with 3 levels: in the first/middle/last third of the sentence. In Experiment 2, we adopted the procedure similar to Experiment 1 for exploring the effect of text visibility on corrective saccades with another 20 students as participants. It was also a 2×2×3 mixed experimental design, and only the first variable was different. The first variable in Experiment 2 was regression visibility with 2 levels: visible and invisible. It meant that the participants in visible condition, they could see the sentence which they had read before. On the contrary, the participants in invisible condition, when they regressed the goal word, the sentence was masked by ‘※’. In both experiments, we adopted 4 indicators of regressions, including initial regression size, initial regression error, and cumulative regression size and regression reaction time.
The results in Experiment 1 indicated that initial regression error was higher in the non-reading group than in the reading group. Initial regression size was not different between the two groups, but the first regressing location of the reading group was related with prearranged word location rather than the non-reading group. Word location was significant in all indicators except initial regression size. The results in Experiment 2 indicated that initial regression error was shorter in the invisible group than in the visible group and initial regression size was higher in the invisible group than in the visible group. It was because the invisible group might use more time to read. Cumulative regression size was higher in the invisible group than in the visible group. Word location was significant with initial regression size and initial regression error. Word frequency was not significant in all two experiment conditions.
The results proved that initial regression was guided by spatial memory. Verbal attributes such as word frequency had no significant effect on regression. Corrective saccade was affected by text visibility.

Key words: Chinese reading, memory, long-regressions, eye movement