International Business Machines Corporation

Benjamin L Brumfield

Stephen Arthur Boxwell

Stanley John Vernier

Timothy A Bishop

Nirav P Desai

Mechanisms are provided for implementing a proximity based candidate answer pre-processor engine that outputs a sub-set of candidate answers to a question and answer (QA) system. The mechanisms receive a lexical answer type (LAT) and an entity specified in an input natural language question as well as an ontology data structure representing a corpus of natural language content. The mechanisms identify a set of candidate answers having associated nodes in the ontology data structure that are within a predetermined proximity of a node corresponding to the entity, and a sub-set of candidate answers in the set of candidate answers having an entity type corresponding to the LAT. The mechanisms output, to the QA system, the sub-set of candidate answers as candidate answers to the input natural language question for evaluation and selection of a final answer to the input natural language question.

LAT based answer generation using anchor entities and proximity

An approach is provided in which a knowledge manager identifies training entities in a training question answer pair and identifies database paths a database that link the training entities. The knowledge manager then generates a generalized query based upon the identified database paths. In addition, the knowledge manager generates a generalized syntactic pattern based upon syntactic relationships between the training entities in the training question answer pair. When the knowledge manager receives a question and matches the syntactic pattern of the received question to the generalized syntactic pattern, the knowledge manager generates a custom query from the generalized query and queries the database using the custom query.

Mapping questions to complex database lookups using synthetic events

An approach is provided in which a knowledge manager identifies training entities in a training question answer pair and identifies multiple database paths a database that link the training entities. The knowledge manager then generates multiple generalized queries based upon the identified multiple database paths and assigns query scores to the multiple generalized queries. In addition, the knowledge manager generates a generalized syntactic pattern based upon syntactic relationships between the training entities in the training question answer pair. When the knowledge manager receives a question and matches the syntactic pattern of the received question to the generalized syntactic pattern, the knowledge manager generates custom queries from the generalized queries and queries the database using the custom queries. The knowledge manager, in turn, scores answers generated from the custom queries based upon the custom queries' corresponding query scores.

Scoring automatically generated language patterns for questions using synthetic events

A method, executed by a computer, includes receiving a triple comprising a predicate, a first entity and an un-typed second entity, scanning a collection of triples for triples that match the predicate of the triple to produce similar triples, determining frequency information for second entity types referenced by the similar triples, and processing the un-typed second entity using the frequency information. A computer system and computer program product corresponding to the method are also disclosed herein.

Processing un-typed triple store data

An approach is provided in which a knowledge manager matches a question to multiple natural language contexts that each correspond to relations associated with entities in a structured resource. The knowledge manager identifies database queries corresponding to the multiple natural language contexts and assigns priority scores to the database queries based upon their relative specificity. In turn, the knowledge manager invokes one of the database queries based upon the assigned priority scores.

Using priority scores for iterative precision reduction in structured lookups for questions

An approach is provided in which a knowledge manager receives a question and selects candidate answers based upon a lexical answer type included in the question. The knowledge manager evaluates the candidate answers against conditions detected in the question and generates condition fulfillment scores accordingly. In turn, the knowledge manager ranks the candidate answers based upon the corresponding condition fulfillment scores.

Identification and evaluation of lexical answer type conditions in a question to generate correct answers

An approach is provided in which a knowledge manager analyzes a corpus of documents based upon relations corresponding to entities in a structured resource and constructs a natural language context associated with a set of the relations. The knowledge manager maps the natural language context to a database query and, in turn, invokes the database query when the knowledge manager matches a question to the natural language context.

Using synthetic events to identify complex relation lookups

A mechanism is provided in a data processing system configured with a natural language processing pipeline executing on at least one processor of the data processing system, for class narrowing for answer lookups. The natural language processing pipeline annotates an input question with a verb in the input question and a syntactic relation between the verb and a focus of the question. A class narrowing component of the natural language processing pipeline finds an entry in a verb/type lookup resource corresponding to the verb and the syntactic relation. The entry comprises the verb, the syntactic relation, and a set of classes. Each given class in the set of classes has a corresponding frequency value representing a frequency the given class occurs with the syntactic relation to the verb. The class narrowing component performs class-narrowed type-restricted answer lookup from a topology based on the entry in the verb/type lookup resource to generate a set of candidate answers.

Class-narrowing for type-restricted answer lookups

One or more words are received. A set of frequency of occurrence values of the received word(s) within a set of domain tables is determined. A domain table in the set of domain tables is associated to the received word(s), based on the set of frequency of occurrence values meeting a threshold value. A word-sense of the received word(s) is determined based on a corresponding word-sense in the associated domain table and/or corresponding domain dictionary.

Identifying word-senses based on linguistic variations

An approach is provided in which a knowledge manager analyzes parent/child hierarchies between semi-structured items included in a semi-structured resource. The knowledge manager generates synthetic events based upon the parent/child hierarchy analysis and creates a structured resource utilizing the synthetic events. In one embodiment, the structured resource includes structured resource entries, which include a first semi-structured item, a second semi-structured item, and one of the synthetic events. In turn, the knowledge manager queries the structured resource based upon a question received from a user interface.

Growing community of inventors

Cedar Park, TX, United States of America

Benjamin L Brumfield