Training end-to-end spoken language understanding systems with unordered entities
Training data can be received, which can include pairs of speech and meaning representation associated with the speech as ground truth data. The meaning representation includes at least semantic entities associated with the speech, where the spoken order of the semantic entities is unknown. The semantic entities of the meaning representation in the training data can be reordered into spoken order of the associated speech using an alignment technique. A spoken language understanding machine learning model can be trained using the pairs of speech and meaning representation having the reordered semantic entities. The meaning representation, e.g., semantic entities, in the received training data can be perturbed to create random order sequence variations of the semantic entities associated with speech. Perturbed meaning representation with associated speech can augment the training data.
1. A computer-implemented method comprising:
receiving pairs of speech and meaning representation associated with the speech, the meaning representation including at least semantic entities associated with the speech, wherein spoken order of the semantic entities is unknown;
reordering the semantic entities into spoken order of words associated with the semantic entities in the speech using an alignment technique;
augmenting the received pairs of speech and meaning representation to include random order sequence variations of the semantic entities;
pre-training a spoken language understanding machine learning model using the augmented pairs of speech and meaning representation; and
training the spoken language understanding machine learning model that is pre-trained, using the pairs of speech and meaning representation having the reordered semantic entities.
2. The method of claim 1 , wherein the alignment technique includes acoustic keyword spotting used with a hybrid speech recognition model.
3. The method of claim 1 , wherein the alignment technique includes using time markings derived from an attention model.
4. The method of claim 3 , wherein the speech includes noisy speech data and the attention model is adapted to the noisy speech data.
5. The method of claim 1 , further including fine-tuning the spoken language understanding machine learning model that is pre-trained, using the semantic entities in alphabetical order; and the training includes training the spoken language understanding machine learning model that is fine-tuned, with the reordered semantic entities.
6. The method of claim 1 , wherein the spoken language understanding machine learning model includes a neural network.
7. The method of claim 1 , further including inputting a given speech to the trained spoken language understanding machine learning model, wherein the trained spoken language understanding machine learning model outputs a set prediction including an intent label and semantic entities associated with the given speech.
8. A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions readable by a device to cause the device to:
receive pairs of speech and meaning representation associated with the speech, the meaning representation including at least semantic entities associated with the speech, wherein spoken order of the semantic entities is unknown;
reorder the semantic entities into spoken order of words associated with the semantic entities in the speech using an alignment technique;
augment the received pairs of speech and meaning representation to include random order sequence variations of the semantic entities; and
pre-train the spoken language understanding machine learning model using the augmented pairs of speech and meaning representation; and
train the spoken language understanding machine learning model that is pre-trained, using the pairs of speech and meaning representation having the reordered semantic entities.
9. The computer program product of claim 8 , wherein the alignment technique includes acoustic keyword spotting used with a hybrid speech recognition model.
10. The computer program product of claim 8 , wherein the alignment technique includes using time markings derived from an attention model.
11. The computer program product of claim 8 , wherein the device is further caused to fine-tune the spoken language understanding machine learning model that is pre-trained, using the semantic entities in alphabetical order, wherein the device caused to train the spoken language understanding machine learning model includes the device caused to train the spoken language understanding machine learning model that is fine-tuned, with the reordered semantic entities.
12. A computer-implemented method comprising:
receiving pairs of speech and meaning representation associated with the speech, the meaning representation including at least semantic entities associated with the speech, wherein spoken order of the semantic entities is unknown;
reordering the semantic entities into spoken order of words associated with the semantic entities in the speech using an alignment technique;
augmenting the received pairs of speech and meaning representation to include random order sequence variations of the semantic entities;
pre-training a spoken language understanding machine learning model using the augmented pairs of speech and meaning representation;
fine-tuning the spoken language understanding machine learning model that is pre-trained, using the semantic entities in alphabetical order; and
training the spoken language understanding machine learning model that is fine-tuned, using the pairs of speech and meaning representation having the reordered semantic entities.
13. The method of claim 12 , wherein the alignment technique includes acoustic keyword spotting used with a hybrid speech recognition model.
14. The method of claim 12 , wherein the alignment technique includes using time markings derived from an attention model.
15. The method of claim 14 , wherein the speech includes noisy speech data and the attention model is adapted to the noisy speech data.
16. The method of claim 12 , wherein the spoken language understanding machine learning model includes a neural network.
17. The method of claim 12 , further including inputting a given speech to the trained spoken language understanding machine learning model, wherein the trained spoken language understanding machine learning model outputs a set prediction including an intent label and semantic entities associated with the given speech.