This paper presents a novel time-aware language model, T-gram, to predict the human mobility using location check-in data. While the conventional n-gram language model, which use the contextual co-occurrence to estimate the probability of a sequence of items, are often employed to predict human mobility, the time information of items is merely considered. T-gram exploits the time information associated at each location, and aims to estimate the probability of visiting satisfaction for a given sequence of locations. For a location sequence, if locations are visited at right times and the transitions between locations are proper as well, the T-gram probability gets higher. We also devise a T-gram Search algorithm to predict future locations. Experiments of human mobility prediction conducted on Gowalla check-in data significantly outperform a series of n-gram-based methods and encourage the future usage of T-gram.