Reading text with randomly rotated letters

Purpose: Previous research (Corballis et al., Mem. & Cog., 1978) has shown that recognition time for isolated letters is virtually independent of orientation (viewpoint invariance). If letter recognition is viewpoint invariant during reading, then randomizing letter orientations within text should have no effect on reading speed. Method: 12 subjects read text passages (100 words approx.) in which each letter was oriented randomly within a range of ±θ. Reading speed was measured with θ = 0° (i.e., normal text), 20°, 40°, 60°, 90°, 120° and 180° (see below). θ=0° We followed those ahead of us and soon θ=60° were putting on rubber coats that dragged θ=120° on the floor. On our heads we wore rubber θ=180° hoods that allowed only our faces to deep Results: Reading speed decreased steadily with increasing orientation range. When θ was 180°, reading speed was 33% of that for normal upright text. We found a similar effect using texts with uppercase print: this condition reduced the potential impact of rotationally confusable letters (e.g., "b" & "q"). Conclusions: These results suggest that, in reading, letters are not recognized independently by viewpoint-invariant mechanisms. Misoriented letters may be mentally rotated to upright prior to recognition (our data would indicate a rotation rate of 1034 deg/sec). Alternatively, letter recognition in reading may depend on spatial arrangement cues such as the relative orientation of adjacent letters, or the relative orientation of letters and words.