Method and system for learning ontological relations from documents
Embodiments of an ontological determination method for use in natural language processing applications are described. In one embodiment, shallow lexico-syntactic patterns are applied to identify relations by extracting term features to distinguish relation terms from non-relation terms, identifying coordinate relations for every adjacent terms; identifying short-distance ontological (e.g., hypernym or part-whole relations) for other adjacent terms based on term features and lexico-syntactic patterns; and then inferring long-distance hypernym and part-whole relations based on the identified coordinate relations and the short-distance relations.
1. A computer-implemented method of learning ontological relations in a document stored in a computer memory, the method executed in a processor-based platform and comprising:
extracting terms from noun-phrases in the document, wherein the terms are included in word sequences comprising the noun-phrases;
generating short-distance patterns for both coordinate terms and ontological relations in the document, wherein short-distance patterns comprise terms separated by at most one additional term, and the coordinate terms comprise terms that share the same hypernym/holonym parent;
generalizing the short-distance patterns for both the coordinate terms and the ontological relations;
identifying short-distance coordinate relations and ontological relations by grouping the short-distance patterns by verbs or prepositions, extracting the longest common substring within each group of pattern strings, and deriving generalized patterns for every verb or preposition; respectively from the generalized short-distance coordinate terms and short-distance ontological relations; and
deriving long-distance ontological relations from the identified short-distance coordinate relations and ontological relations, wherein the long-distance ontological relations comprise ontological relations between terms separated by at least two additional terms.
2. The method of claim 1 , wherein the document comprises a plurality of terms characterized by syntactic relationships among the terms.
3. The method of claim 1 , wherein at least one ontological relation of the ontological relations is selected from the group consisting of: hypernym, part-whole, synonym, antonym, coordinate, entailment, troponym, and causal relations.
4. The method of claim 1 , wherein identifying short-distance coordinate terms comprises performing statistical classification on the coordinate terms.
5. The method of claim 1 , wherein identifying short-distance ontological relations comprises performing statistical classification on the ontological relations.
6. The method of claim 1 , wherein the long-distance ontological relations are used in a natural language processing system to build a knowledge base for use with a natural language processing system.
7. An apparatus, comprising:
an input stage for inputting noun phrase portions of sentences in the document;
a database functionally coupled to the input stage and configured to store a knowledge base including words from the document; and
an ontological learning component coupled to the input stage, and including
a first processor identifying relations by extracting terms from the noun phrase portions to distinguish relation terms from non-relation terms,
a second processor identifying coordinate relations for every pair of adjacent terms,
a third processor identifying short-distance ontological relations for other adjacent terms based on term features and lexico-syntactic patterns, and inferring long-distance ontological relations based on the identified coordinate relations and the short-distance relations, wherein short-distance ontological relations comprise ontological relations between adjacent terms separated by at most one intermediate term, and the long-distance ontological relations comprise ontological relations between terms separated by at least two additional terms, and further wherein the coordinate terms comprise terms that share the same hypernym/holonym parent, the third processor further generalizing the short-distance pattern for both the coordinate terms and the ontological relations by grouping the short-distance patterns by verbs or prepositions, extracting the longest common substring within each group of pattern strings, and deriving generalized patterns for every verb or preposition.
8. The apparatus of claim 7 , wherein at least one ontological relation of the ontological relations is selected from the group consisting of hypernym, part-whole, synonym, antonym, coordinate, entailment, troponym, and casual relations.
9. The apparatus of claim 7 , wherein the long-distance ontological relations comprise ontological relations between terms separated by at least two terms.
10. The apparatus of claim 7 , further comprising a natural language processing system functionally coupled to the database and utilizing words stored in the knowledge base.
11. The apparatus of claim 10 , wherein the natural language processing system comprises a dialog system.
12. The apparatus of claim 10 , wherein the natural language processing system is selected from the group consisting of a reasoning system, a question and answer system, an information retrieval system, a machine translation system, and a semantic inference system.
13. The method of claim 1 wherein the terms from the noun-phrases in the document are extracted using Conditional Maximum Entropy modeling.