System and a method for generating semantically similar sentences for building a robust SLM
View Patent ↗A system and method are described for generating semantically similar sentences for a statistical language model. A semantic class generator determines for each word in an input utterance a set of corresponding semantically similar words. A sentence generator computes a set of candidate sentences each containing at most one member from each set of semantically similar words. A sentence verifier grammatically tests each candidate sentence to determine a set of grammatically correct sentences semantically similar to the input utterance. Also note that the generated semantically similar sentences are not restricted to be selected from an existing sentence database.
1. A system for generating semantically similar sentences, the system comprising:
at least one processor; and
non-transitory computer-readable media storing executable instructions that, when executed by the at least one processor, cause the system to:
receive input speech;
perform a Fourier transform to extract spectral features to characterize the input speech as a sequence of representative multi-dimensional vectors;
process the sequence of representative multi-dimensional vectors using an automatic-speech-recognition function of a natural-language-processing application to determine a recognition output comprising an input utterance comprising a first word and a second word, the first word being before the second word;
generate, for display, representative text corresponding to the recognition output comprising the input utterance;
determine, for each word in the input utterance, a set of semantically similar words using a domain-specific ontology comprising a domain-specific dictionary;
compute a set of candidate sentences each containing at most one member from each set of semantically similar words, wherein at least one candidate sentence of the set of candidate sentences comprises a third word semantically similar to the first word, and a fourth word semantically similar to the second word, the fourth word being before the third word in the at least one candidate sentence;
grammatically test each candidate sentence of the set of candidate sentences to determine a set of grammatically correct sentences semantically similar to the input utterance and a set of grammatically incorrect sentences,
wherein the grammatically testing each candidate sentence is performed using a syntactic parser, a language model different from the syntactic parser, and a grammar checker;
rank the set of grammatically correct sentences;
add the grammatically correct sentences to an existing statistical language model; and
save the set of grammatically incorrect sentences.
2. The system of claim 1 , wherein the non-transitory computer-readable media stores executable instructions that, when executed by the at least one processor, cause the system to:
receive the input speech through a user-query interface of the natural-language-processing application.
3. The system of claim 1 , wherein the non-transitory computer-readable media stores executable instructions that, when executed by the at least one processor, cause the system to:
build the domain-specific dictionary manually.
4. The system of claim 1 , wherein the non-transitory computer-readable media stores executable instructions that, when executed by the at least one processor, cause the system to:
build the domain-specific dictionary automatically.
5. One or more non-transitory computer-readable media storing executable instructions that, when executed by a processor, cause a device to:
receive input speech;
perform a Fourier transform to extract spectral features to characterize the input speech as a sequence of representative multi-dimensional vectors;
process the sequence of representative multi-dimensional vectors using an automatic-speech-recognition function of a natural-language-processing application to determine a recognition output comprising an input utterance comprising a first word and a second word, the first word being before the second word;
generate, for display, representative text corresponding to the recognition output comprising the input utterance;
determine, for each word in the input utterance, a set of semantically similar words using a domain-specific ontology comprising a domain-specific dictionary;
compute a set of candidate sentences each containing at most one member from each set of semantically similar words, wherein at least one candidate sentence of the set of candidate sentences comprises a third word semantically similar to the first word, and a fourth word semantically similar to the second word, the fourth word being before the third word in the at least one candidate sentence;
grammatically test each candidate sentence of the set of candidate sentences to determine a set of grammatically correct sentences semantically similar to the input utterance and a set of grammatically incorrect sentences, wherein the grammatically testing each candidate sentence is performed using a syntactic parser, a language model different from the syntactic parser, and a grammar checker;
rank the set of grammatically correct sentences;
add the grammatically correct sentences to an existing statistical language model; and
save the set of grammatically incorrect sentences.
6. The one or more non-transitory computer-readable media of claim 5 , storing executable instructions that, when executed by the processor, cause the device to:
receive the input speech through a user-query interface of the natural-language-processing application.
7. The one or more non-transitory computer-readable media of claim 5 , storing executable instructions that, when executed by the processor, cause the device to:
build the domain-specific dictionary manually.
8. The one or more non-transitory computer-readable media of claim 5 , storing executable instructions that, when executed by the processor, cause the device to:
build the domain-specific dictionary automatically.
9. One or more non-transitory computer-readable media storing executable instructions that, when executed by a processor, cause a device to:
receive input speech;
perform a Fourier transform to extract spectral features to characterize the input speech as a sequence of representative multi-dimensional vectors;
process the sequence of representative multi-dimensional vectors using an automatic-speech-recognition function of a natural-language-processing application to determine a recognition output comprising an input utterance comprising a first word and a second word, the first word being before the second word;
generate, for display, representative text corresponding to the recognition output comprising the input utterance;
determine, for each word in the input utterance, a set of semantically similar words using a generic ontology;
compute a set of candidate sentences each containing at most one member from each set of semantically similar words, wherein at least one candidate sentence of the set of candidate sentences comprises a third word semantically similar to the first word, and a fourth word semantically similar to the second word, the fourth word being before the third word in the at least one candidate sentence;
grammatically test each candidate sentence of the set of candidate sentences to determine a set of grammatically correct sentences semantically similar to the input utterance and a set of grammatically incorrect sentences, wherein the grammatically testing each candidate sentence is performed using a syntactic parser, a language model different from the syntactic parser, and a grammar checker;
rank the set of grammatically correct sentences;
add the grammatically correct sentences to the existing natural-language-processing statistical language model; and
save the set of grammatically incorrect sentences.
10. The one or more non-transitory computer-readable media of claim 9 , storing executable instructions that, when executed by the processor, cause the device to:
receive the input speech through a user-query interface of the natural-language-processing application.