Abstract
We describe a methodology to recognize efficient controlled natural languages (CLs) that compositionally translate into ontology languages, and as such are suitable to be used in natural language front-ends to ontology-based systems. Efficiency in this setting is defined as the tractability (in the sense of computational complexity theory) of logical reasoning in such fragments, measured in the size of the data they aim to manage. In particular, to identify efficient CLs, we consider fragments corresponding to the DL-Lite family of description logics, known to underpin data intensive ontologies and systems. Our methodology exploits the link between syntax and semantics of natural language captured by categorial grammars, controlling the use of lexical terms that introduce logical structure outside the allowed fragments. A major role is played by the control of function words introducing logical operators in first-order formal semantics meaning representations. Finally, we conducted a preliminary analysis of semantically parsed English written corpora to show how empirical methods may be useful in identifying CLs that provide good trade-offs between coverage and efficiency.