Typically developing (TD) infants could capitalize on social eye gaze and social contexts to aid word learning. Although children with autism disorder (AD) are known to exhibit atypicality in word learning via social eye gaze, their ability to utilize social contexts for word learning is not well understood. We investigated whether verbal AD children exhibit word learning ability via social contextual cues by late childhood. We found that AD children, unlike TD controls, failed to infer the speaker’s referential intention through information gathered from the social context. This suggests that TD children can learn words in diverse social pragmatic contexts in as early as toddlerhood whereas AD children are still unable to do so by late childhood.