Models are not comparable
No standard training/test set.
No "complete" models
No model uses all four components: character, lexical, semantic, syntactic.
Learnability has not been demonstrated
- Parsing (lexical) requires hand-coded rules (split on spaces)
or a hand-built dictionary.
- HMM and PCFG learning requires a hand-built grammar or tagged
The goal is poorly known
- Shannon game - bounds are too wide.
- Cover and King gambling game - probabilities are hard to estimate.