Integrated programming framework for speech and text understanding with block and statement structure
The technology disclosed relates to authoring of vertical applications of natural language understanding (NLU), which analyze text or utterances and construct their meaning. In particular, it relates to new programming constructs and tools and data structures implementing those new applications.
1. A method of building an automated natural language understanding application, the method including:
receiving at least one electronic record containing programming code that interprets sequences of input tokens by extending a programming language with interpret-block and interpret-statement data structures;
wherein the interpret-block data structures include at least one of the interpret-statements and one or more variables returned by the interpret-block;
wherein at least one of the interpret-statements includes a pattern of one or more tokens, and one or more action instructions;
wherein the action instructions perform logic to construct a meaning of the sequence of input tokens; and
further including parsing the received program code to produce an executable representation of the interpret-block and interpret-statement data structures.
2. The method in claim 1 , wherein outgoing weights are associated with token alternatives to express probabilities of token sequences.
3. The method in claim 2 , wherein the outgoing weights at each token are automatically normalized.
4. The method in claim 3 , wherein the normalization of outgoing weights is performed at sub-expressions.
5. The method in claim 1 , wherein at least one token of at least one interpret-statement is another interpret-block.
6. The method in claim 5 , wherein the variables returned from other interpret-blocks are made available to action statements inside the interpret block.
7. The method in claim 1 , wherein at least one token of the interpret-statement can be an extended pattern table with fixed returned values for each row of the table and without any action statements.
8. The method in claim 1 , wherein the weight of a parsed or partially parsed sequence of tokens is the combination of the weights of all of its possible parses or partial parses.
9. A parser used to build a natural language understanding application, the parser including:
at least one processor, memory coupled to the processor, and computer instructions in the memory that, when executed, cause the processor to take actions including:
receiving at least one electronic record containing programming code that interprets sequences of input tokens by extending a programming language with interpret-block and interpret-statement data structures;
wherein the interpret-block data structures include at least one of the interpret-statements and one or more variables returned by the interpret-block;
wherein at least one of the interpret-statements includes a pattern of one or more tokens, and one or more action instructions;
wherein the action instructions perform logic to construct a meaning of the sequence of input tokens; and
further including parsing the received program code to produce an executable representation of the interpret-block and interpret-statement data structures.
10. The parser of claim 9 , wherein outgoing weights are associated with token alternatives to express probabilities of token sequences.
11. The parser of claim 10 , wherein the outgoing weights at each token are automatically normalized.
12. The parser of claim 11 , wherein the normalization of outgoing weights is performed at sub-expressions.
13. The parser of claim 9 , wherein at least one token of at least one interpret-statement is another interpret-block.
14. The parser of claim 13 , wherein the variables returned from other interpret-blocks are made available to action statements inside the interpret block.
15. The parser of claim 9 , wherein at least one token of the interpret-statement can be an extended pattern table with fixed returned values for each row of the table and without any action statements.
16. The parser of claim 9 , wherein the weight of a parsed or partially parsed sequence of tokens is the combination of the weights of all of its possible parses or partial parses.
17. A method of building a natural language understanding application, the method including:
receiving at least one electronic record containing programming code, the programming code expressed using an extension of a general purpose programming language; and
creating executable code from the programming code,
wherein the executable code, when executed by a processor, causes the processor to create a parse and an interpretation of a sequence of input tokens,
wherein the programming code includes an interpret-block,
wherein the interpret-block includes an interpret-statement,
wherein the interpret-statement includes a pattern expression, and
wherein the interpret-statement includes an action statement expressed in the programming language.
18. The method of claim 17 , wherein the pattern expression includes pattern tokens.
19. The method of claim 18 , wherein pattern tokens are associated with weights.
20. The method of claim 19 , wherein the weights are automatically normalized.
21. The method of claim 18 , wherein at least one pattern token refers to a second interpret-block.
22. The method of claim 21 , wherein the second interpret-block has a return variable, and a value of the return variable is made available to the action statement.
23. The method of claim 17 , wherein an extended pattern token table refers to a table of token expressions, and each row in the table has a fixed returned return value, and has no associated action statement.