Does linguistic distance matter for internal migrations? To answer this question, we study India over the course of the 20th century. We use historical and present-day census data to compute bilateral migration flows and linguistic distance between pairs of districts in 1901 and in 2011. We estimate a gravity model of migration and find that linguistic distance mattered twice more in 1901 than in 2011. A regression discontinuity design at the border of linguistic regions confirms the findings.