Generating training data for natural language search systems

1. A computer-implemented method comprising:

generating, by a training data generator, an abstract query representation from a probabilistic context-free grammar (PCFG), the PCFG having a set of rules to which probabilities are assigned, wherein generating the abstract query representation includes performing a top-down expansion of a semantic structure of the PCFG by traversing the PCFG and sampling non-terminals based on the assigned probabilities, wherein generating an abstract query representation includes applying a beam search process;

selecting, by the training data generator, a sample record from a database using the abstract query representation;

generating, by the training data generator, a tagged user query from the abstract query representation and the sample record by replacing tokens of the abstract query representation with one or more values from the sample record;

generating, by the training data generator, one or more structured query language (SQL) statements from the tagged user query and one or more SQL templates;

generating, by the training data generator, the tagged user query by replacing tokens of the abstract query representation with one or more values from the sample record;

incorporating, by the training data generator, variants into the tagged user query, the variants including misspellings; and

performing, by the training data generator, a lookup operation for every token of the abstract query representation against the sample record to find one or more values for each token and assembling the one or more values as the tagged user query.

2. The computer-implemented method of claim 1 , comprising:

applying the tagged user query to a natural language search system, including a named entity recognition (NER) model, to train the NER model.

3. The computer-implemented method of claim 1 , comprising:

generating the abstract query representation by repeated symbol expansion.

4. The computer-implemented method of claim 1 , comprising: selecting the sample record from a target table of the database.

5. The computer-implemented method of claim 1 , comprising:

generating the one or more SQL statements by pattern matching the tagged user query against the one or more SQL templates.

6. A tangible, non-transitory computer-readable storage medium having instructions stored thereon which, when executed by a processing device, cause the processing device to:

generate, by a training data generator, an abstract query representation from a probabilistic context-free grammar (PCFG), the PCFG having a set of rules to which probabilities are assigned, wherein generating the abstract query representation includes performing a top-down expansion of a semantic structure of the PCFG by traversing the PCFG and sampling non-terminals based on the assigned probabilities, wherein generating an abstract query representation includes applying a beam search process;

select, by the training data generator, a sample record from a database using the abstract query representation;

generate, by the training data generator, a tagged user query from the abstract query representation and the sample record by replacing tokens of the abstract query representation with one or more values from the sample record;

generate, by the training data generator, one or more structured query language (SQL) statements from the tagged user query and one or more SQL templates;

incorporating, by the training data generator, variants into the tagged user query, the variants including misspellings; and

perform, by the training data generator, a lookup operation for every token of the abstract query representation against the sample record to find one or more values for each token and assemble the one or more values as the tagged user query.

7. The tangible, non-transitory computer-readable storage medium of claim 6 having instructions stored thereon which, when executed by a processing device, cause the processing device to generate the abstract query representation by repeated symbol expansion.

8. The tangible, non-transitory computer-readable storage medium of claim 6 having instructions stored thereon which, when executed by a processing device, cause the processing device to generate the one or more SQL statements by pattern matching the tagged user query against the one or more SQL templates.

9. A system comprising:

one or more servers, the servers configurable to cause:

generating, by a training data generator, an abstract query representation from a probabilistic context-free grammar (PCFG), the PCFG having a set of rules to which probabilities are assigned, wherein generating the abstract query representation includes performing a top-down expansion of a semantic structure of the PCFG by traversing the PCFG and sampling non-terminals based on the assigned probabilities, wherein generating an abstract query representation includes applying a beam search process;

selecting, by the training data generator, a sample record from a database using the abstract query representation;

generating, by the training data generator, a tagged user query from the abstract query representation and the sample record by replacing tokens of the abstract query representation with one or more values from the sample record;

generating, by the training data generator, one or more structured query language (SQL) statements from the tagged user query and one or more SQL templates; and

introducing, by the training data generator, variants into the tagged user query, the variants including misspellings;

performing, by the training data generator, a lookup operation for every token of the abstract query representation against the sample record to find one or more values for each token and assembling the one or more values as the tagged user query.

10. The system of claim 9 , the servers further configurable to cause generating the one or more SQL statements by pattern matching the tagged user query against the SQL templates.