Authoring system for bayesian networks automatically extracted from text

1. A Bayesian Belief Network (BBN) authoring system comprising:

one or more content sources providing content;

a programmed processing unit for communicating with said content sources and configured to:

access natural language text content stored in a content storage device, said natural language text content includes text units;

identify statements within said accessed natural language text content indicating a probabilistic dependence relation;

extract said statements indicating said probabilistic dependence relation from said natural language text content;

parse said extracted statement to generate a directed typed-dependency graph structure, said graph structure having nodes associated with words of said text unit, and said nodes being interconnected and associated with a label denoting a grammatical relation among said nodes; a parsed text unit including non-restrictive relative clauses and restrictive relative clauses;

break a directed typed-dependency graph into one or more subgraphs, a subgraph corresponding to restrictive relative clauses and a subgraph corresponding to non-restrictive relative clauses; each said subgraph processed separately to obtain evidence statements; and

aggregate said extracted statements and evidence statements to form a BBN network structure.

2. The system as claimed in claim 1 , wherein said processor device is configured to:

obtain a set of typed dependencies among nodes from said directed typed-dependency graph structure.

3. The system as claimed in claim 2 wherein to identify statements indicating a probabilistic dependence relation, said programmed processing unit is configured to:

identify one or more lexical and semantic attributes of words within a text unit indicating a probabilistic dependence relation between two or more variables; and

add identified typed dependencies based on said identified word attributes to the set of typed dependencies obtained from said directed typed-dependency graph structure.

4. The system as claimed in claim 3 , wherein to identify said one or more lexical and semantic attributes, said programmed processing unit is configured to: perform, for each word of said text unit, a table look-up of lexical information upon a database or corpus.

5. The system as claimed in claim 4 , wherein said programmed processing unit is configured to:

correct for linguistic variations amongst statements representing said probabilistic dependence relations.

6. The system as claimed in claim 1 , wherein to aggregate said extracted statements, said programmed processing unit is configured to:

determine statements having a same pair of variables;

reconcile statements having a same pair of variables; and

create a directed acyclic graph associated with the reconciled statements.

7. The system as claimed in claim 6 , wherein to reconcile statements, said programmed processing unit is configured to:

count a number of statements setting forth a dependence relation category and an independence relation category; and

create an output statement of either dependence relation category or independence relation category.

8. The system as claimed in claim 5 , further comprising:

applying transformational rules to map conditional dependence information to a canonical form.

9. The system as claimed in claim 1 , wherein to form said BBN network structure, said programmed processing unit is further configured to:

identify from said text content quantitative probabilistic information associated with one or more variables; and

build one or more quantitative probability statements from said extracted probabilistic information and said variables, a probability statement indicating a probability of a given variable being in a specified state given another variable in a specified state.

10. The system as claimed in claim 9 , wherein to identify from said text content quantitative probabilistic information, said programmed processing unit is configured to perform one or more of:

identify and extract numerical expressions from said text content; or

implement a rule-based approach for extracting quantitative probability statements directed to said one or more variables.

11. The system as claimed in claim 10 , wherein said processor device is configured to:

reconcile said quantitative probability statements with the BBN network structure.

12. The system as claimed in claim 1 , comprising:

a user interface to a computing system implementing said programmed processing unit for authoring said BBN, wherein via said user interface, said authoring includes a user assisting in one or more said identifying, extracting and statement aggregating to form the BBN network structure.

13. A computer program product for authoring a Bayesian Belief Network (BBN), the computer program product comprising a non-transitory tangible storage medium, said medium readable by a processing circuit and storing instructions run by the processing circuit for performing a method, the method comprising:

accessing natural language text content stored in a content storage device, said natural language text content includes text units;

identifying statements within said accessed natural language text content indicating a probabilistic dependence relation;

extracting said statements indicating said probabilistic dependence relation from said natural language text content;

parsing said extracted statement to generate a directed typed-dependency graph structure, said graph structure having nodes associated with words of said text unit, and said nodes being interconnected and associated with a label denoting a grammatical relation among said nodes; a parsed text unit including non-restrictive relative clauses and restrictive relative clauses;

breaking a directed typed-dependency graph into one or more subgraphs, a subgraph corresponding to restrictive relative clauses and a subgraph corresponding to non-restrictive relative clauses; each said subgraph processed separately to obtain evidence statements; and

aggregating said extracted statements and evidence statements to form a BBN network structure, wherein a programmed processing unit performs said identifying, statement extracting, and aggregating.

14. The computer program product of claim 13 , wherein said method further comprises:

obtaining a set of typed dependencies among nodes from said directed typed-dependency graph structure.

15. The computer program product of claim 14 wherein said identifying statements indicating a probabilistic dependence relation comprises:

identifying one or more lexical and semantic attributes of words within a text unit indicating a probabilistic dependence relation between two or more variables; and

adding identified typed dependencies based on said identified word attributes to the set of typed dependencies obtained from said directed typed-dependency graph structure.

16. The computer program product as claimed in claim 15 , wherein said identifying said one or more lexical and semantic attributes comprises: performing, for each word of said text unit, a table look-up of lexical information upon a database or corpus.

17. The computer program product as claimed in claim 13 , wherein said forming said BBN network structure further comprises:

identifying from said text content quantitative probabilistic information associated with one or more variables; and

building one or more quantitative probability statements from said extracted probabilistic information and said variables, said probability statement indicating a probability of a given variable being in a specified state given another variable in a specified state.