System and method for referring to entities in a discourse domain
View Patent ↗Systems, methods, and non-transitory computer-readable media for referring to entities. The method includes receiving domain-specific training data of sentences describing a target entity in a context, extracting a speaker history and a visual context from the training data, selecting attributes of the target entity based on at least one of the speaker history, the visual context, and speaker preferences, generating a text expression referring to the target entity based on at least one of the selected attributes, the speaker history, and the context, and outputting the generated text expression. The weighted finite-state automaton can represent partial orderings of word pairs in the domain-specific training data. The weighted finite-state automaton can be speaker specific or speaker independent. The weighted finite-state automaton can include a set of weighted partial orderings of the training data for each possible realization.
1. A method comprising:
receiving, from a user, a visual reference to a specific entity in a known domain, where an identity of the specific entity is unknown;
extracting, via a processor, a visual context from a domain-specific training data associated with the known domain;
determining, via the processor, the identity of the specific entity using the visual reference in combination with one of a speaker history, the visual context, and speaker preferences; and
generating a text expression referring to the specific entity based on the identity of the specific entity.
2. The method of claim 1 , wherein generating the text expression is further based on a dependency tree model.
3. The method of claim 2 , further comprising generating parsed word strings containing word pairs from the domain-specific training data.
4. The method of claim 1 , wherein generating the text expression is further based on a permute and order model.
5. The method of claim 1 , wherein generating the text expression is further based on a template-based model.
6. The method of claim 1 , wherein generating the text expression is further based on a best-scoring realization.
7. The method of claim 1 , wherein a weighted finite-state automaton represents partial orderings of word pairs in the domain-specific training data.
8. The method of claim 7 , further comprising receiving a set of attributes describing the target entity in context, wherein a weighted finite-state automaton represents the set of attributes.
9. The method of claim 7 , wherein the weighted finite-state automaton comprises a set of weighted partial orderings of the domain-specific training data for each possible realization.
10. The method of claim 7 , wherein the weighted finite-state automaton is generated by computing frequencies for how often a first word precedes a second word in any training data sentence, how often the first word depends on and precedes the second word, and how often the first word depends on and follows the second word.
11. A system comprising:
a processor; and
a computer-readable storage medium having instructions stored which, when executed by the processor, cause the processor to perform operations comprising:
receiving, from a user, a visual reference to a specific entity in a known domain, where an identity of the specific entity is unknown;
extracting a visual context from a domain-specific training data associated with the known domain;
determining the identity of the specific entity using the visual reference in combination with one of a speaker history, the visual context, and speaker preferences; and
generating a text expression referring to the specific entity based on the identity of the specific entity.
12. The system of claim 11 , wherein generating the text expression is further based on one of a dependency tree model, a template based model, a permute and order model, and a best-scoring realization.
13. The system of claim 12 , the computer-readable storage medium having additional instructions stored which result in the operations further comprising generating parsed word strings containing word pairs from the domain-specific training data.
14. The system of claim 11 , wherein a weighted finite-state automaton represents partial orderings of word pairs in the domain-specific training data.
15. The system of claim 11 , the computer-readable storage medium having additional instructions stored which result in the operations further comprising receiving a set of attributes describing the target entity in context, wherein a weighted finite-state automaton represents the set of attributes.
16. The system of claim 14 , wherein the weighted finite-state automaton comprises a set of weighted partial orderings of the domain-specific training data for each possible realization.
17. The system of claim 14 , wherein the weighted finite-state automaton is generated by computing frequencies for how often a first word precedes a second word in any training data sentence, how often the first word depends on and precedes the second word, and how often the first word depends on and follows the second word.
18. A computer-readable storage device having instructions stored which, when executed on a computing device, cause the computing device to perform operations comprising:
receiving, from a user, a visual reference to a specific entity in a known domain, where an identity of the specific entity is unknown;
extracting a visual context from a domain-specific training data associated with the known domain;
determining the identity of the specific entity using the visual reference in combination with one of a speaker history, the visual context, and speaker preferences; and
generating a text expression referring to the specific entity based on the identity of the specific entity.
19. The computer-readable storage device of claim 18 , wherein generating the text expression is further based on one of a dependency tree model, a template based model, a permute and order model, and a best-scoring realization.
20. The computer-readable storage device of claim 18 , wherein a weighted finite-state automaton represents partial orderings of word pairs in the domain-specific training data.