Semantic parser including a coarse semantic parser and a fine semantic parser
Techniques for improving a semantic parser of a dialog system, by breaking the semantic parser into a coarse semantic parser and a fine semantic parser, are described. A method described herein includes accessing an utterance received in a dialog system. The utterance is a text-based natural language expression. The method further includes applying a coarse semantic parser to the utterance to determine an intermediate logical form for the utterance. The intermediate logical form indicates one or more intents in the utterance. The method further includes applying a fine semantic parser to the intermediate logical form to determine a logical form for the utterance. The logical form is a syntactic expression of the utterance according to an established grammar, and the logical form includes one or more parameters of the one or more intents. The logical form can be used to conduct a dialog with a user of the dialog system.
1. A method comprising:
accessing training data comprising utterances, intermediate logical forms, and logical forms;
using a coarse semantic parser to determine predicted intermediate logical forms based on the utterances in the training data;
training the coarse semantic parser by updating the coarse semantic parser based on comparing the predicted intermediate logical forms to the intermediate logical forms in the training data;
using a fine semantic parser to determine predicted logical forms based on the utterances and the intermediate logical forms in the training data; and
training the fine semantic parser by updating the fine semantic parser based on comparing the predicted logical forms to the logical forms in the training data;
using the coarse semantic parser to determine an intermediate logical form for an utterance of a dialog system, wherein the utterance is a text-based natural language expression, and wherein the intermediate logical form indicates one or more intents in the utterance; and
using the fine semantic parser to determine a logical form for the utterance and the intermediate logical form, wherein the logical form is a syntactic expression of the utterance according to an established grammar, and wherein the logical form comprises one or more parameters of the one or more intents; and
using the logical form to conduct a dialog through the dialog system.
2. The method of claim 1 , wherein applying the fine semantic parser to the intermediate logical form to determine the logical form for the utterance comprises applying the fine semantic parser to the utterance and the intermediate logical form to determine the logical form for the utterance.
3. The method of claim 1 , wherein the coarse semantic parser is a machine learning model.
4. The method of claim 3 , wherein the fine semantic parser is a machine learning model.
5. The method of claim 1 , wherein the coarse semantic parser is a sequence-to-sequence neural network, and wherein the fine semantic parser is an additional sequence-to-sequence neural network.
6. The method of claim 1 , further comprising:
receiving speech input in the dialog system; and
converting the speech input to the utterance using automatic speech recognition.
7. The method of claim 6 , further comprising:
analyzing the logical form to generate a response text responsive to the speech input; and
determining an audio output of the dialog system by applying text-to-speech to the response text.
8. A system comprising:
a coarse semantic parser configured to:
access an utterance received in a dialog system, wherein the utterance is a text-based natural language expression; and
determine, based on the utterance, an intermediate logical form for the utterance, wherein the intermediate logical form indicates one or more intents in the utterance;
a fine semantic parser configured to:
determine, based on the intermediate logical form, a logical form for the utterance and the intermediate logical form, wherein the logical form is a syntactic expression of the utterance according to an established grammar, and wherein the logical form comprises one or more parameters of the one or more intents,
wherein the logical form is useable to conduct a dialog through the dialog system; and
a training system configured to:
access training data comprising utterances, intermediate logical forms, and logical forms;
using a coarse semantic parser to determine predicted intermediate logical forms based on the utterances in the training data;
training the coarse semantic parser by updating the coarse semantic parser based on comparing the predicted intermediate logical forms to the intermediate logical forms in the training data;
using a fine semantic parser to determine predicted logical forms based on the utterances and the intermediate logical forms in the training data; and
training the fine semantic parser by updating the fine semantic parser based on comparing the predicted logical forms to the logical forms in the training data.
9. The system of claim 8 , wherein the fine semantic parser determines the logical form based on both the utterance and the intermediate logical form.
10. The system of claim 8 , wherein the coarse semantic parser is a sequence-to-sequence neural network.
11. The system of claim 10 , and wherein the fine semantic parser is an additional sequence-to-sequence neural network.
12. The system of claim 8 , further comprising an automatic speech recognition subsystem configured to:
receive speech input; and
convert the speech input to the utterance using automatic speech recognition.
13. The system of claim 12 , further comprising:
a dialog manager subsystem configured to analyze the logical form to generate a response text responsive to the speech input; and
a text-to-speech subsystem configured to determine an audio output of the system by applying text-to-speech to the response text.