Natural language question answering systems

1. A method comprising:

determining a set of candidate database queries, each including a respective sequence of tokens of a database syntax, based on a string;

determining a first score for a first candidate database query from the set of candidate database queries, wherein the first score is based on a match between one or more words of the string and a token of the respective sequence of tokens of the first candidate database query;

determining a second score for the first candidate database query, wherein the second score is based on natural language syntax data determined for words of the string;

selecting, based on the first score and the second score, the first candidate database query from the set of candidate database queries; and

invoking a search of the database using a query based on the first candidate database query to obtain search results.

2. The method of claim 1 , comprising:

extracting an inference pattern from the string and the first candidate database query;

classifying the inference pattern to determine an inference type;

determining a resolution, wherein the resolution includes one or more tokens of the database syntax;

identifying a set of context features for the inference pattern, wherein the set of context features includes words from the string and tokens from the first candidate database query;

determining a confidence score based on the set of context features; and

storing an inference record in an inference store, wherein the inference record includes the set of context features, the resolution, and the confidence score.

3. The method of claim 2 , wherein the confidence score is determined based in part upon receiving feedback via a user interface that was used modify the first candidate database query.

4. The method of claim 2 , wherein the confidence score is determined based in part upon receiving a feedback representing user interaction with a like icon in a user interface while the user interface includes representations of the string and the first candidate database query.

5. The method of claim 2 , comprising:

maintaining a contains probability score associated with a column of the database, wherein the contains probability score is based on a count of times the column has been used to filter data with a contains query; and

using the contains probability score for the column to determine the resolution.

6. The method of claim 1 , comprising:

identifying a current set of context features for the first candidate database query, wherein the current set of context features includes words from the string and tokens from the first candidate database query;

selecting a first inference record from an inference store based on comparison of the current set of context features to context features of inference records in the inference store; and

modifying the first candidate database query using a resolution of the first inference record to obtain an inferred database query, wherein the resolution includes one or more tokens of the database syntax, and wherein the query is based on the inferred database query.

7. The method of claim 6 , wherein the first inference record includes a confidence score and the first inference record is selected based on the confidence score.

8. The method of claim 6 , wherein the current set of context features include part-of-speech tags for words of the string, and comprising:

comparing the part-of-speech tags of the current set of context features to part-of-speech tags of inference records in the inference store, wherein the first inference record is selected based on a match of part-of-speech tags of the first inference record to the part-of-speech tags of the current set of context features.

9. The method of claim 1 , comprising:

receiving an indication that the string matches the first candidate database query;

responsive to the indication, determining a pattern based on the string and the first candidate database query, wherein the pattern includes a collection of token constraints, a rewrite rule that maps one or more input tokens to a sequence of output tokens, and a ranking score adjustment that can be applied to a ranking score for a candidate database query; and

storing the pattern.

10. The method of claim 9 , wherein the indication is based on a user interaction with a like icon in a user interface while the user interface includes representations of the string and the first candidate database query.

11. The method of claim 9 , wherein the indication includes user feedback data concerning the database query that was used to modify the first candidate database query.

12. The method of claim 9 , wherein the collection of token constraints is a sequence of token constraints.

13. The method of claim 1 , comprising:

checking whether the first candidate database query from the set of candidate database queries satisfies a collection of token constraints of a pattern, wherein the pattern includes the collection of token constraints, a rewrite rule that maps one or more input tokens to a sequence of output tokens, and a ranking score adjustment that can be applied to a ranking score for a candidate database query; and

responsive to satisfaction of the collection of token constraints of the pattern, modifying the first candidate database query using the rewrite rule of the pattern, and adjusting the second score based on the ranking score adjustment of the pattern.

14. The method of claim 1 , wherein the first score is determined based on a first inverse document frequency of an unmatched word of the string and a second inverse document frequency of a matched word of the string.

15. The method of claim 1 , comprising:

presenting, via a user interface, respective text representations for tokens in the sequence of tokens;

receiving feedback data concerning the first candidate database query via the user interface;

modifying the first candidate database query based on the feedback data;

invoking a search of a database using a query based on the modified first candidate database query to obtain modified search results; and

presenting data based on the modified search results in the user interface.

16. The method of claim 1 , wherein the first score is determined based on usage data for the token of the respective sequence of tokens of the first candidate database query.

17. A system for providing a search interface for databases, comprising:

a network interface,

a processor, and

a memory, wherein the memory stores instructions executable by the processor to:

determine a set of candidate database queries, each including a respective sequence of tokens of a database syntax, based on a string;

determine a first score for a first candidate database query from the set of candidate database queries, wherein the first score is based on a match between one or more words of the string and a token of the respective sequence of tokens of the first candidate database query;

determine a second score for the first candidate database query, wherein the second score is based on natural language syntax data determined for words of the string;

select, based on the first score and the second score, the first candidate database query from the set of candidate database queries; and

invoke a search of the database using a query based on the first candidate database query to obtain search results.

18. The system of claim 17 , wherein the first score is determined based on usage data for the token of the respective sequence of tokens of the first candidate database query.

19. A non-transitory computer-readable storage medium that includes instructions that, when executed by a processor, facilitate performance of operations comprising:

determining a set of candidate database queries, each including a respective sequence of tokens of a database syntax, based on a string;

determining a first score for a first candidate database query from the set of candidate database queries, wherein the first score is based on a match between one or more words of the string and a token of the respective sequence of tokens of the first candidate database query;

determining a second score for the first candidate database query, wherein the second score is based on natural language syntax data determined for words of the string;

selecting, based on the first score and the second score, the first candidate database query from the set of candidate database queries; and

invoking a search of the database using a query based on the first candidate database query to obtain search results.

20. The non-transitory computer-readable storage medium of claim 19 , wherein the first score is determined based on usage data for the token of the respective sequence of tokens of the first candidate database query.