Head-mounted and remote point-of-gaze estimation methodologies were used in an automated reading assistant to vocalize unknown words in real-time. Points-of-gaze were mapped onto stationary and moving reading material using homographic mappings established from point correspondences between the reading material and images from a scene camera. Points-of-gaze on reading material were used to measure the processing time of each viewed word. Unknown words were detected based on processing time length. Viewed words could be uniquely identified when words on the reading material were separated by more than 15 mm. In a pilot study of four subjects, using remote point-of-gaze estimation and stationary reading material, the detector achieved a detection rate of 89% with a false alarm rate of 11%. In a pilot study with two subjects, using head-mounted point-of-gaze estimation to allow free head movement, the automated reading assistant provided vocalization for 94% of unknown words and 10% of known words.