Abstract: Parafoveal-on-foveal effects occur when lexical properties of word n influence processing on word n-1 during reading. According to the E-Z Reader model (Reichle et. al, 2003, 2006), lexical characteristics of word n should not influence the processing of word n-1 (Rayner et. al, 1998). In contrast, the parallel model SWIFT (Engbert et. al, 2002, 2005) stipulates that multiple words within the perceptual span can be processed in parallel. If true, then parafoveal processing difficulty should slow down foveal processing (Inhoff et. al, 2000; Vitu et. al, 2004). However, Kennedy (2002, 2004) reported an inverted parafoveal-on-foveal effect where greater parafoveal processing difficulty resulted in the reader spending less time on the foveal word. He argued that a difficult configuration of parafoveal words acts as a salient target causing the reader to execute an early saccade. They refer to this process as “magnetic attraction”.
The main goal of the present study was to investigate parafoveal-on-foveal effects in normal Chinese reading. We examined whether the semantic transparency of word n influenced readers’ processing time on word n-1. Semantic transparency is the semantic relationship between a compound word and its constituent morphemes. By definition, a semantically transparent compound is one which is semantically related to the meaning of its constituent morphemes. A compound word is semantically opaque when its meaning cannot be understood by simply combining the constituent morphemes’ individual meanings.
We compared first pass reading times on word n-1 in a condition where word n was semantically transparent with those in a condition where word n was semantically opaque. Additionally, the effect of the processing load of word n-1 was examined by making a word frequency manipulation on word n-1. According to the assumptions of the SWIFT model, individual morpheme processing is part of lexical access, and therefore, processing of word n when semantically transparent would be easier and consequently, the processing of word n-1 would take less time in this condition than when word n was semantically opaque. Contrary to the SWIFT model, “magnetic attraction” would predict an inverted effect, because an opaque word in the parafovea would attract attention away from foveal word processing. If parallel processing exists, as claimed by SWIFT and magnetic attraction theory, an effect of semantic transparency should occur. Also, if foveal processing load modulates parafoveal processing, semantic transparency effects may be greater in the high than the low frequency word n-1 condition. In contrast, according to the E-Z Reader model, the semantic characteristics of word n would not affect processing of word n-1, except in cases where word n was due to be skipped.
Twenty seven university students who were native Chinese speakers participated in this research (average age 21.55 years, SD=1.19). Sentence pairs contained a target word that was manipulated for semantic transparency (word n). The words in each sentence pair prior to word n were identical between these two conditions. A set of 26 semantically transparent and 26 opaque compound words was selected for the target words in the sentences. The two sets were equated for number of strokes, frequency of each character and frequency of the whole word. For both sets, the frequency of the pretarget word (word n-1) was manipulated: one set had a high frequency word n-1 (M=319.49 per million, SD=148.52) and the other set had a low frequency word n-1 (M=0.66 per million, SD=0.23). All pre-target words were two-character words long and were matched for number of strokes in each character. Eye movements were recorded using an Eyelink Ⅱ eye tracker (sample rate =250 Hz). The size of each character was 1.30cm×1.30cm (33 pixel×33 pixel). At a viewing distance of 75cm, one character space subtended approximately 1°of visual angle.
We analyzed reading times on word n-1 as a function of the semantic transparency of word n. The following eye fixation measures were examined: single fixation duration, first fixation duration, gaze duration, final fixation, refixation probability and the probability of skipping. The word frequency effect of word n-1 was robust in terms of gaze duration and refixation probability. The results showed that in both high- and low- frequency word n-1 conditions, the semantic transparency of word n did not affect processing time on word n-1, prior to direct fixation on word n.
In summary, we found no evidence of parafoveal-on-foveal effects in this study. The results support the assumptions of the serial processing model, E-Z Reader model.

Key words: parafoveal-on-foveal effect, semantic transparency, eye movements, serial processing, parallel processing