Intelligent integration of information continues to challenge database research for over 35 years. While data integration processes of all kinds are now reasonably well understood and widely used in practice, the growth and heterogeneity of data requires much higher degrees of automation to limit the need for human specialist work. This requires deeper insights in data-centric approaches of Enterprise Information Integration which focus on the semantics of information integration. Recent formalizations and algorithms enable both significant improvement in schema integration, and in its automated transformation to efficient data-level integration, in a wide variety of architectural settings such as data warehouses or peer-to-peer databases. In addition to giving a short overview of developments in this field for the past 20 years, this paper focuses particularly on the challenges posed by heterogeneity in data models.