Current theories of reading eye movements claim that reading saccades are programmed primarily on the basis of information about the length of the upcoming word, determined by low-level visual processes that detect spaces to the right of fixation. Many studies attempted to test this claim by filling spaces between words with various non-space symbols (fillers). This manipulation, however, confounds the effect of inserting extraneous characters into text with the effect of obscuring word boundaries by filling spaces. We performed the control conditions necessary to unconfound these effects. Skilled readers read continuous stories aloud and silently. Three factors were varied: (i) position of the fillers in the text (at the beginning, the end, or surrounding each word); (ii) the presence or absence of spaces in the text; and (iii) the effect of the type of filler on word recognition (from greatest effect to least effect: Latin letters, Greek letters, digits and shaded boxes). The effect of fillers on reading depended more on the type of filler than on the presence of spaces. The greater effect the fillers had on word recognition, the more they showed reading. Surrounding each word with digits or Greek letters slowed reading as much as filling spaces with these symbols. Surrounding each word with randomly chosen letters, while preserving spaces, slowed reading by 44-75%--as much as, or more than, removing spaces from normal text. Removing spaces from text with Latin-letter fillers slowed reading by only 10-20% more. We conclude that fillers in text disrupt reading by affecting word recognition directly, without necessarily affecting the eye movement pattern.

The role of spaces between words in text was studied in many experiments by filling the spaces with irrelevant symbols. This practice is based on the assumption that these fillers occlude spaces without disturbing word recognition appreciably. There is no empirical evidence to support this assumption. We performed a series of experiments to study the role of fillers and spaces in text. We found that texts in all conditions in which words were surrounded by fillers were read more slowly than normal texts, as long as the fillers shared common features with the letters of the text (digits, irrelevant Latin letters, and Greek letters). Reading was as slow when words were surrounded by fillers (1like2 8this6), as when fillers replaced spaces (1like2this6), showing that reading was disrupted by the presence of fillers, not by the absence of spaces. Reading eye movements were recorded with some of the subjects and text conditions. The placement of fillers and the presence or absence of spaces in the text had no effect on percentage regressions, or on where reading saccades landed within words. However, more saccades per line were made in conditions that contained fillers than either with normal or with unspaced text. Observed differences in eye movement patterns and reading speed can be accounted for by a global adjustment to only one eye movement parameter, viz saccade size, made on the basis of the global appearance of the text. No evidence was found that the presence of fillers or the absence of spaces required new strategies for programming individual saccades. This suggests that word recognition and global strategies are more significant in programming reading saccades than the local physical features of the text.