TANGO: Table Analysis for Generating OntologiesProject VisionContrary to prior assumptions, TANGOresearch is showing that nearly automaticontology generation is possible.Ontologies, which are machine-readabledescriptions of a knowledge domain,resolve many issues of semantics, mutualunderstanding, concept matching, andinteroperability, which are at the heart oftoday's information explosion problems. The TANGO project, supported by NSF'sInformation and Intelligent Systems Division and being carried out by David W.Embley and Deryle Lonsdale at BrighamYoung University and George NagyRensselaer Polytechnic University, workswith tables, a neglected yet excellentsource of basic material for building ontologies.The TANGO (Table ANalysis for Generating Ontologies) approach applies ideas in table interpretation, conceptual modeling, and schema matching and merging in new and innovative ways to: (i) understand a table's structure and its conceptual content; (ii) discover the constraints that hold between concepts extracted from the table; (iii) match the recognized concepts with those recognized in other tables; (iv) merge the resulting structures to create a domain ontology; and (v) adjust the created domain ontology so that it is a clean, complete, and accurate conceptualization of the source tables.Project ChallengesThe difficulties of table interpretation are readily apparent to anyone who has consulted a foreign railroad schedule, or a table in an alien technical field (some of us are still wary of the Periodic Table of Elements). The first problem is that of establishing the two-dimensional alignment of row and column headers and content cells, which is often compromised in typesetting by differences in word or phrase lengths, and is only partially preserved in many web documents. Next, the logical structure must be discovered and represented in canonical form. Although an ideal table can be viewed as row and column header hierarchies whose intersection point to a unique content cell, or as a set of entity-value relationships like those in relational databases, the interpretation of actual tables is often hampered by non-standardized nomenclature and headers that are often omitted altogether when obvious to a humanreader. Once a table is in canonical form, it must be reverse-engineered into a conceptual structure suitable for integration into a growing ontology. Both reverse engineering of canonical tables and conceptual-model integration have challenged researchers for years. A properly interpreted table in canonical form, however, provides the leverage needed to generate hypergraph-based conceptual models, and recent work in information-extraction approaches to integrating hypergraph-based conceptual models offer confidence that the integration can succeed.Project ResultsThe algorithms developed for table interpretation let the machine learn from human correction of its mistakes. Although still being investigated, the tool appears to be about twenty times faster than manual entry. A companion tool that works for the special case in which sets of machine-generated tables, such as those found on the hidden web, interprets tables nearly flawlessly. Experiments show that the new tool can process large sets of complex here.Students working on the TANGO integration tool.tables, such as those in molecular biology, with 97% accuracy. Additional tools are being built for the remaining steps of TANGO’s ontology generation