Generating model training data from a domain specification
Examples described herein generate training data for machine learning (ML) for natural language (NL) processing (such as semantic parsing for translating NL). A formula tree is generated based on sampling both a formula grammar and NL templates. Using the formula tree, an ML training data instance pair is generated comprising a formula example and an NL example. A context example may also be used during instantiation of the formula tree. An ML model is trained with training data including the ML training data instance pair, and ML output is generated from NL input. The ML output includes, for example, a machine-interpretable formula, a database querying language command, or a general programming language instruction. Some examples support context-free grammar, probabilistic context-free grammar, and/or non-context-free production rules.
1. A system comprising:
a processor; and
a computer-readable medium storing instructions that are operative upon execution by the processor to:
receive a formula grammar and natural language (NL) templates;
generate a formula tree, wherein generating the formula tree comprises generating at least one of a flattened NL sketch or a flattened formula sketch, and wherein generating the formula tree comprises sampling the formula grammar and sampling the NL templates;
using the formula tree, generate a machine learning (ML) training data instance pair that associates an NL example utterance with a formula converting the NL example utterance into a machine-interpretable representation or meaning representation;
receive an NL input from a user accessing an application;
generate an ML output from the NL input using an ML model trained with the ML training data instance pair; and
send the ML output to the application being accessed by the user.
2. The system of claim 1 , wherein the ML output comprises a machine-interpretable formula.
3. The system of claim 1 , wherein generating the formula tree further comprises generating a formula sketch.
4. The system of claim 1 , wherein the instructions are further operative to:
receive context examples; and
instantiate the formula tree, wherein instantiating the formula tree comprises sampling the context examples.
5. The system of claim 1 , wherein the instructions are further operative to:
repeat sampling the formula grammar and sampling the NL templates to generate a plurality of formula trees; and
using the plurality of formula trees, generate a plurality of training data instance pairs, each training data instance pair comprising a formula example and an NL example.
6. The system of claim 1 , wherein the formula grammar includes probabilistic context-free grammar.
7. The system of claim 1 , wherein the formula grammar includes non-context-free production rules.
8. A computerized method comprising:
receiving a formula grammar and natural language (NL) templates;
generating a formula tree, wherein generating the formula tree comprises sampling the formula grammar and sampling the NL templates and wherein at least one of sampling the formula grammar or sampling the NL templates comprises identifying a sampling factor that increases a likelihood of a first sample occurring more often than a second sample;
using the formula tree, generating a machine learning (ML) training data instance pair that associates an NL example utterance with a formula converting the NL example utterance into a machine-interpretable representation or meaning representation;
receiving an NL input from a user accessing an application;
generating an ML output from the NL input using an ML model trained with the ML training data instance pair; and
sending the ML output to the application being accessed by the user.
9. The computerized method of claim 8 , wherein the ML output comprises a machine-interpretable formula.
10. The computerized method of claim 8 , wherein generating the formula tree further comprises generating a formula sketch.
11. The computerized method of claim 8 , further comprising:
receiving context examples; and
instantiating the formula tree, wherein instantiating the formula tree comprises sampling the context examples.
12. The computerized method of claim 8 , further comprising:
repeating sampling the formula grammar and sampling the NL templates to generate a plurality of formula trees; and
using the plurality of formula trees, generating a plurality of training data instance pairs, each training data instance pair comprising a formula example and an NL example.
13. The computerized method of claim 8 , wherein the formula grammar includes probabilistic context-free grammar.
14. The computerized method of claim 8 , wherein the formula grammar includes non-context-free production rules.
15. One or more computer storage devices having computer-executable instructions stored thereon, which, on execution by a computer, cause the computer to perform operations comprising:
receiving a formula grammar and natural language (NL) templates;
generating a formula tree, wherein generating the formula tree comprises generating at least one of a flattened NL sketch or a flattened formula sketch, and wherein generating the formula tree comprises sampling the formula grammar and sampling the NL templates;
using the formula tree, generating a machine learning (ML) training data instance pair that associates an NL example utterance with a formula converting the NL example utterance into a machine-interpretable representation or meaning representation;
receiving an NL input from a user accessing an application;
generating an ML output from the NL input using an ML model trained with the ML training data instance pair; and
sending the ML output to the application being accessed by the user.
16. The one or more computer storage devices of claim 15 , wherein the ML output comprises a machine-interpretable formula.
17. The one or more computer storage devices of claim 15 , wherein the operations further comprise:
receiving context examples; and
instantiating the formula tree, wherein instantiating the formula tree comprises sampling the context examples.
18. The one or more computer storage devices of claim 15 , wherein the operations further comprise:
repeating sampling the formula grammar and sampling the NL templates to generate a plurality of formula trees; and
using the plurality of formula trees, generating a plurality of training data instance pairs, each training data instance pair comprising a formula example and an NL example.
19. The one or more computer storage devices of claim 15 , wherein the formula grammar includes probabilistic context-free grammar.
20. The one or more computer storage devices of claim 15 , wherein the formula grammar includes non-context-free production rules.