In 2 experiments, a novel experimental paradigm investigated how spatial attention is distributed across a talker's face during auditory-visual speech discourse processing. Dots were superimposed onto several talkers' faces for 17-msec durations on the talker's left side, mouth, right side, and eyebrow area. Participants reported the locations of dots in 2 task conditions varying the presence or absence of concurrent language processing. Concurrent language processing did not interfere with spatial performance. Dot detection performance was greater for the talker's left, compared to their right, side, and for the mouth, compared to the eyebrow, area. Emotional expressiveness of the face, attractiveness, and clarity of voice pronunciation ratings were correlated with detection performance. Results support a 2-pool attention resource model of auditory-visual language processing. Results also suggest that attention engagement during multimodal language processing is primarily internally, rather than externally, driven, because attention is directed to the eye area when the talker displays low facial emotional expressiveness, and to the speaker's left side due to left hemisphere language processing, even though the right side displays greater motion.