It is best to impose all your types in csv before importing, and if you have complex fields, decompose them using a regular expression in the query (as you suggested).
However, we are actively studying support for new input formats and are interested in feedback on which formats will be most useful. The base query processor has support ( Dremel ) for types that are similar to the hive map type, but BigQuery, however, does not currently disclose a mechanism for swallowing nested records.
source share