Architecture and processes for computer learning and understanding

• US 10,614,166 B2
• Filed: 03/20/2017
• Issued: 04/07/2020
• Est. Priority Date: 06/24/2016
• Status: Active Grant

• Current Assignee: Mind Lakes, LLC
• Original Assignee: Elemental Cognition LLC
• Inventors: Ferrucci, David, Barborak, Mike, Buchanan, David, Burnham, Greg, Chu-Carroll, Jennifer, Kalyanpur, Aditya, Lally, Adam, Pacifico, Stefano, Wang, Chang
• Primary Examiner(s): He, Jialong
• Application Number:
  US15/464,137
• Publication Number:
  US 20200034424A1
• Time in Patent Office: 1,114 Days
• CPC Class Codes: G06F 16/22 Indexing; Data structures therefor; Storage structures
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/3331 Query processing
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06F 40/35 Discourse or dialogue representation
• CPC Class Codes: G06F 40/40 Processing or translation of natural language natural language analysis G06F40/20; semantic analysis G06F40/30
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/006 based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06N 5/027 Frames
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence

Abstract

An architecture and processes enable computer learning and developing an understanding of arbitrary natural language text through collaboration with humans in the context of joint problem solving. The architecture ingests the text and then syntactically and semantically processes the text to infer an initial understanding of the text. The initial understanding is captured in a story model of semantic and frame structures. The story model is then tested through computer generated questions that are posed to humans through interactive dialog sessions. The knowledge gleaned from the humans is used to update the story model as well as the computing system'"'"'s current world model of understanding. The process is repeated for multiple stories over time, enabling the computing system to grow in knowledge and thereby understand stories of increasingly higher reading comprehension levels.

Claims

1. A computer-implemented method, comprising:

in a computing system for processing a natural language input, performing, by the computing system, a process for generating resources for use in processing the natural language input comprising;

receiving a query to analyze one or more words in the natural language input;

analyzing the one or more words for possible interpretations including, in part, inducing knowledge from language corpora for use in deriving the possible interpretations and performing word sense disambiguation on the one or more words to produce one or more sense vectors, wherein inducing the knowledge includes detecting a setting or a background of that is common to a context of the natural language input;

accumulating one or more knowledge resources containing the knowledge induced from the language corpora, wherein at least one knowledge resource of the one or more knowledge resources includes a three-dimensional tensor structure in which a relationship between two or more parameters has a probability;

storing the one or more knowledge resources, wherein storing the one or more knowledge resources includes storing the at least one knowledge resource as different parameters within the three-dimensional tensor structure; and

returning one or more possible interpretations of the possible interpretations in response to the query.

2. The computer-implemented method of claim 1, wherein performing the word sense disambiguation on the one or more words further produces one or more context vectors.

3. The computer-implemented method of claim 1, wherein analyzing the one or more words further includes detecting at least one of paraphrases, relations, or entity types for the one or more words.

4. The computer-implemented method of claim 1, wherein the process further comprises predicting one or more semantic primitives from the one or more knowledge resources.

5. The computer-implemented method of claim 1, wherein the process further comprises ranking the one or more possible interpretations.

6. The computer-implemented method of claim 1, wherein the process further comprises creating at least one semantic structure to represent the natural language input in part by using the one or more knowledge resources.

18. The computer-implemented method of claim 6, wherein creating the at least one semantic structure includes defining a predicate and one or more roles for the at least one semantic structure, wherein defining the predicate or the one or more roles is based at least in part on the at least one semantic structure having an above threshold probability of being included in a collection of semantic structures that is common to the context of the setting or the background of the natural language input.

7. A computer-implemented method, comprising:

for a natural language input, performing, by a computing system, a process comprising;

receiving the natural language input;

performing a syntactic analysis of the natural language input to produce one or more linguistic analysis results;

inducing knowledge from language corpora including detecting a setting or a background that are common to a context of the natural language input;

creating multiple semantic structures to represent the natural language input in part by using the one or more linguistic analysis results and the knowledge induced from the language corpora, wherein the multiple semantic structures include one or more generative semantic primitive (GSP) structures defining one or more roles that are common to the context of the setting or the background of the natural language input, wherein defining the one or more roles includes mapping one or more entities in the natural language input to the one or more roles;

engaging in a dialog session with a human user to receive input from the human user to use by the computing system to evaluate the multiple semantic structures as an understanding of the natural language input;

revising the multiple semantic structures based on one or more responses from the human user to improve the understanding of the natural language input, wherein revising the multiple semantic structures includes defining at least one new GSP structure with at least one or more new roles based on the one or more responses and based at least in part on the at least one new GSP structure having at least a threshold probability of being included in the multiple semantic structures that is associated with the context of the setting or the background of the natural language input; and

storing the knowledge as different parameters within a three-dimensional tensor structure in which relations among the different parameters have associated probabilities.

8. The computer-implemented method of claim 7, wherein inducing the knowledge comprises statistically assessing words in the language corpora.

9. The computer-implemented method of claim 7, wherein inducing the knowledge further includes detecting paraphrases, relations, and entity types for words in the language corpora.

10. The computer-implemented method of claim 7, wherein the knowledge includes multiple possible interpretations and further comprising ranking the multiple possible interpretations.

11. The computer-implemented method of claim 7, wherein creating the multiple semantic structures comprises predicting semantic primitives from the knowledge induced from the language corpora.

12. The computer-implemented method of claim 7, wherein the process further comprises repeating, by the computing system, the process to form additional semantic structures for subsequent natural language inputs.

20. The computer-implemented method of claim 7, wherein the process further comprises engaging, by the computing system, in multiple dialog sessions with computing devices of multiple human users in parallel to evaluate the multiple semantic structures as an understanding of the natural language input.

13. A computing system, comprising:

one or more processors; and

memory coupled to the one or more processors, the memory storing;

a knowledge induction engine to analyze language corpora and to generate a resource repository including one or more knowledge resources for use in deriving possible interpretations of natural language, wherein the knowledge induction engine stores the one or more knowledge resources as different parameters within a three-dimensional tensor structure in which relations among the different parameters have associated probabilities; and

a knowledge integration engine to form one or more semantic structures that provide a semantic representation of a natural language story, the knowledge integration engine forming the one or more semantic structures through use of one or more linguistic analysis results generated from parsing the natural language story and the one or more knowledge resources stored in the resource repository, wherein forming the one or more semantic structures includes forming at least one frame structure that relates a set of generative primitive (GSP) structures under a particular context of a particular setting of the natural language story, an individual GSP structure of the set of GSP structures having at least a threshold probability of being included in the at least one frame structure that is associated with the particular context of the particular setting of the natural language story.

14. The computing system of claim 13 wherein the knowledge induction engine comprises computer-executable instructions to perform multiple tasks comprising word sense disambiguation, relation detection, paraphrase generation, textual entailment, scene analysis, and missing text generation.

15. The computing system of claim 13, wherein the knowledge induction engine ranks the possible interpretations.

16. The computing system of claim 13, wherein the knowledge integration engine predicts one or more semantic primitives from the one or more knowledge resources.

17. The computing system of claim 13, wherein the memory further stores a dialog engine to facilitate a dialog session with a human user to evaluate an extent to which the one or more semantic structures represent the natural language story, wherein the knowledge integration engine updates the one or more semantic structures based in part on responses from the human user.

19. The computing system of claim 13, wherein the memory further stores a dialog engine to facilitate multiple dialog sessions with computing devices of multiple human users in parallel to receive responses from the multiple human users for use by the computing system to evaluate an extent to which the one or more semantic structures represent the natural language story, wherein the knowledge integration engine updates the one or more semantic structures based in part on the responses from the multiple human users.