High-speed corpus-based language model using suffix arrays for variable-length n-gram matching. Instant training, exact matching, O(m log n) queries.
