The Internet of Things is democratizing computer usage, with computers metamorphosing into communication devices making their way from offices to inside people’s homes, cars, factories, and so on. Data access on these computers is still a frustrating process for many; therefore, democratizing data access is of utmost importance.

Much of these data are held in relational databases, inside web servers and other important storage systems. Natural languages, such as English, provide a natural way of human-computer communication, with various efforts having being expended over the years, offering varying degrees of success, in the quest to build natural language interfaces to computers.

This paper describes the architecture of an interactive natural language query interface for relational databases. Through carefully circumscribed interactions with the user, the system is able to interpret complex natural language queries. It consists of three main phases: a transformation of a natural language query to a query tree; an interactive verification, with the user, of the fidelity of the query tree vis à vis the query; and a component that translates the query tree to SQL statements. Using SQL to query databases is often challenging for experts, and will be overly complicated for the general user. Natural languages have redundancies in them, making them effective for human-to-human communication. These ambiguities need to be solved to make natural language computer communication effective and efficient.

The system tries to provide a high correspondence between the queries being asked and the transformed query by having a user interaction mechanism for user validation and ambiguity resolution, forming the basis of the correct translation into SQL. The architecture shows a good separation of concerns between the different phases of the system. By making sure the system provided explanations for how it processed the queries, it helped to build user trust and understanding.

The paper contains many overarching statements with few supporting back-up points. For example, when the system faced an ambiguity it wanted to resolve, it generated multiple interpretations and chose the “best” interpretation from these choices. The reader was not told how “best” was computed. The paper mentioned “unreasonable” query logics. How did they quantify “unreasonableness”? Although they designed a different metric to validate their system, average times for running the system were not given, making it difficult to make system comparisons. It should also be noted that the heuristics employed, although they appeared to be effective in the domain of literature search used, might not work in other (less structured) domains.

Although their validation needs improvement, this paper has contributed positively to this area, especially with its system architecture and novel query tree data structure.