7 Conclusions
This paper has presented a state-of-the-art authoring system for natural language
interfaces to relational databases. Internally the system uses semantic
grammars encoded in _-SCFG to map typed user requests to an extended variant
of Codd’s tuple calculus which in turn is automatically mapped to SQL.
The author builds the semantic grammar through a series of naming, tailoring
and defining operations within a web-based GUI. The author is shielded from
the formal complexity of the underlying grammar and as an added benefit, the
given grammar rules may be used in the reverse direction to achieve paraphrases
of logical queries (see [11]).
Using the Geoquery 250 corpus, our results are compared with contemporary
machine learning results and approaches based on light annotation. Our
initial experimental evidence shows quick bootstrapping of the initial interface
can be achieved. Future work will focus on more complete evaluation and experimentation
with more comprehensive grammatical frameworks (e.g. CCG [15]),
especially under regimes that enable enhanced robustness [18]. Future work will
also explore hybrid approaches using initial authoring to bootstrap NLIs, followed
by interactive machine learning that, applied in the limit, will edge such
NLIs toward 100% precision and recall.