Granted Patent B2

US 8,086,028 · App. 12/648,933 · Granted Dec 27, 2011

Abstract

Claims (49)

1. A method of generating a set of tuple-term pairs from a corpus of text, comprising:

compiling via a processor concordance lines associated with terms in the corpus;

identifying a set of verb-argument tuples and associated terms from the concordance lines;

selecting from the set each verb-argument tuple and associated term having a computed numerical quantification of strength of association greater than a threshold, wherein the selected verb-argument tuples and associated terms represent most likely actions associated with the terms; and

storing in a database each tuple-term pair in the set of tuple-term pairs that matches a verb in a sentence input to a text-to-scene conversion system that constructs an arbitrary three-dimensional scene from input text without using previously stored images.

2. The method of claim 1 , wherein each term describes an aspect of the environment of a scene.

3. The method of claim 2 , wherein the aspect of the environment of the scene denotes one of a name of: a location, an occupied space, an unoccupied space, a room, a time of day, and a season.

4. The method of claim 1 , wherein the verb-argument tuples are one of a verb-object tuple or a verb-preposition-object tuple.

5. The method of claim 1 , wherein the numerical quantification of the strength of association is computing according to steps comprising:

computing a ratio of a likelihood of a first and a second hypothesis, wherein:

the first hypothesis indicates that a probability of the verb-argument tuple occurring given the associated term is indistinguishable from a probability of the verb-argument tuple occurring given something other than the associated term, and

the second hypothesis indicates that the probability of the verb-argument tuple occurring given the associated term is distinguishable from the probability of the verb-argument tuple occurring given something other than the associated term.

6. The method of claim 1 , further comprising deleting duplicate concordance lines.

7. The method of claim 1 , further comprising filtering the set of concordance lines to remove collocations.

8. The method of claim 1 , further comprising computing a lemma of each verb in the set of tuple-term pairs.

9. The method of claim 1 , wherein verbs are in one of a base form and an inflected form.

10. The method of claim 1 , further comprising depicting an action described by the verb in the tuple-term pair in a context of an environment denoted by the term of the tuple-term pair.

11. An system for generating a set of tuple-term pairs from a corpus of text, the system comprising:

a processor;

a first module controlling the processor to compile via a processor concordance lines associated with terms in the corpus;

a second module controlling the processor to identify a set of verb-argument tuples and associated terms from the concordance lines;

a third module controlling the processor to select from the set each verb-argument tuple and associated term having a computed numerical quantification of strength of association greater than a threshold, wherein the selected verb-argument tuples and associated terms represent most likely actions associated with the terms; and

a fourth module controlling the processor to store in a database each tuple-term pair in the set of tuple-term pairs that matches a verb in a sentence input to a text-to-scene conversion system that constructs an arbitrary three-dimensional scene from input text without using previously stored images.

12. The system of claim 11 , wherein the numerical quantification of the strength of association is computing according to steps comprising:

computing a ratio of a likelihood of a first and a second hypothesis, wherein:

the first hypothesis indicates that a probability of the verb-argument tuple occurring given the associated term is indistinguishable from a probability of the verb-argument tuple occurring given something other than the associated term, and

the second hypothesis indicates that the probability of the verb-argument tuple occurring given the associated term is distinguishable from the probability of the verb-argument tuple occurring given something other than the associated term.

13. A non-transitory computer-readable medium storing instructions which, when executed by a computing device, cause the computing device to generate a set of tuple-term pairs from a corpus of text, the instructions comprising:

compiling via a processor concordance lines associated with terms in the corpus;

identifying a set of verb-argument tuples and associated terms from the concordance lines;

selecting from the set each verb-argument tuple and associated term having a computed numerical quantification of strength of association greater than a threshold, wherein the selected verb-argument tuples and associated terms represent most likely actions associated with the terms; and

storing in a database each tuple-term pair in the set of tuple-term pairs that matches a verb in a sentence input to a text-to-scene conversion system that constructs an arbitrary three-dimensional scene from input text without using previously stored images.

14. The non-transitory computer-readable medium of claim 13 , wherein the numerical quantification of the strength of association is computed according to steps comprising:

computing a ratio of likelihood of a first and a second hypothesis, wherein:

the first hypothesis indicates that a probability of the verb-argument tuple occurring given the associated term is indistinguishable from a probability of the verb-argument tuple occurring given something other than the associated term, and

the second hypothesis indicates that the probability of the verb-argument tuple occurring given the associated term is distinguishable from the probability of the verb-argument tuple occurring given something other than the associated term.

15. A method of inferring an environment from a sentence input to a text-to-scene conversion system, the method comprising:

constructing, via a processor, an arbitrary three-dimensional scene from a received sentence without using previously stored images, the sentence having a verb-argument tuple and a subject of an action included in the verb-argument tuple;

identifying an equivalent verb-argument tuple in a database of verb-argument tuples and environmental terms associated with the verb-argument tuples; and

placing the subject in a rendered environment via the text-to-scene conversion system according to the environmental term associated with the equivalent verb-argument tuple.

16. The method of claim 15 , wherein the verb-argument tuple is one of a verb-object tuple or a verb-preposition-object tuple.

17. The method of claim 15 , wherein the equivalent verb-argument tuple includes one of a lemmatized form of a verb in the verb-argument tuple, a base form of the verb in the verb-argument tuple, and an inflected form of the verb in the verb-argument tuple.

18. The method of claim 15 , wherein the environmental term includes one of a name of: a location, an occupied space, an unoccupied space, a room, a time of day, and a season.

19. A text-to-scene conversion system that infers an environment from a received sentence input, the system comprising:

a processor;

constructing, without using previously stored images, an arbitrary three-dimensional scene from a received sentence input, the sentence having a verb-argument tuple and a subject of an action included in the verb-argument tuple;

identifying an equivalent verb-argument tuple in a database of verb-argument tuples and environmental terms associated with the verb-argument tuples; and

placing the subject in a rendered environment via the text-to-scene conversion system according to the environmental term associated with the equivalent verb-argument tuple.

20. The system of claim 19 , wherein the environmental term includes one of a name of: a location, an occupied space, an unoccupied space, a room, a time of day, and a season.

Assignments (7)