Online social networks (e.g., Twitter) offer an open platform for people to interact and connect without restrictions of language usage or geographic borders. Because of their pervasiveness, online social networks provide data and become real-time sensors of society. This work looks at Twitter to reveal the hidden relationship of languages that stems from users' language preference for writing their tweets. We show that the language relationships are dependent of place by comparing 12 large-scale datasets with different locality levels. For instance, the secondary language of French speakers in Canada is different from French speakers in France. We used network science and clustering techniques to find that languages groups are more driven by spatial than syntactic proximity. The characterization of language relationships is key to the understanding of information spread in social media and the detection of cultural shifts.