- Implement vocabulary mapper which simply transform text into vocabulary-id given ngram language model or counts
- Implement --min-ngram to perform filtering duing indexing (for bigrams and upward)
- Implement mini-expgram, which supports only loading.
- Support for ngram-suffix traversal in the backward ngram. Use an additional bit to indicate the end of context?
- Port for other batch-queue systems, such as Torque.