International Business Machines Corporation

Christopher A Welty

James J Fan

Aditya Anand Kalyanpur

David Angelo Ferrucci

James W Murdock, Iv

Wlodek Wlodzimierz Zadrozny

Mihaela Ancuta Bornea

Sugato Bagchi

Alfio Massimiliano Gliozzo

Pablo Ariel Duboue

Lei Zhang

Chen Wang

Jennifer Chu-Carroll

Li Li Ma

Kenneth James Barker

Yue Pan

Apoorv Agarwal

Zhao Ming Qiu

David A Ferucci

A system, method and computer program product for conducting questions and answers with deferred type evaluation based on any corpus of data. The method includes processing a query including waiting until a 'Type' (i.e. a descriptor) is determined AND a candidate answer is provided. Then, a search is conducted to look (search) for evidence that the candidate answer has the required Lexical Answer Type (e.g., as determined by a matching function that can leverage a parser, a semantic interpreter and/or a simple pattern matcher). Prior to or during candidate answer evaluation, a process is provided for extracting and storing collections of entity-type pairs from semi-structured text documents. During QA processing and candidate answer scoring, a process is implemented to match the query LAT against the lexical type of each provided candidate answer and generate a score judging a degree of match.

Providing question and answers with deferred type evaluation using text with limited structure

According to an aspect, a heterogeneous graph in a data store is accessed. The heterogeneous graph includes a plurality of nodes having a plurality of node types. The nodes are connected by edges having a plurality of relation types. One or more intermediary graphs are created based on the heterogeneous graph. The intermediary graphs include intermediary nodes that are the relation types of the edges of the heterogeneous graph and include intermediary links between the intermediary nodes based on shared instances of the nodes between relation types in the heterogeneous graph. The intermediary graphs are traversed to find sets of relations based on intermediary links according to a template. An inference rule is extracted from the heterogeneous graph based on finding sets of relations in the intermediary graphs. The inference rule defines an inferred relation type between at least two of the nodes of the heterogeneous graph.

Extraction of inference rules from heterogeneous graphs

A system and method for automatically mapping LATs and candidate answers to multiple taxonomies without a need to merge these taxonomies. The method includes using a syntactic analysis of a corpus to extract all type instances of the LAT. The extracted instances are then mapped to a given taxonomy and clustered in a set of supertypes. Each supertype receives a score based on the coverage of LAT instances in the corpus. The method includes mapping the candidate answer to the same taxonomy to determine if the candidate answer is an instance of a significant supertype. Then the score of a candidate answer is obtained by aggregating or taking a maximum of the score of the matched significant supertypes. This score evaluates the type match between the LAT and candidate answer for a taxonomy. Multiple taxonomies can be used to increase the chance of LAT and candidate answer mapping.

Type evaluation in a question-answering system

In a method of answering questions, a question is received, a question LAT is determined, and a candidate answer to the question is identified. Preliminary types for the candidate answer are determined using first components to produce the preliminary types. Each of the first components produces a preliminary type using different methods. A first type-score representing a degree of match between the preliminary type and the question LAT is produced. Each preliminary type and each first type-score is evaluated using second components. Each of the second components produces a second score based on a combination of the first type-score and a measure of degree that the preliminary type matches the question LAT. The second components use different methods to produce the second score. A final score representing a degree of confidence that the candidate answer matches the question LAT is calculated based on the second score.

Combining different type coercion components for deferred type evaluation

According to an aspect, a pair of related entities that includes a first entity and a second entity is received. Distributional relations are detected between the first entity and the second entity. The detecting includes identifying two sets of entities in a corpus, the first set including the first entity and at least one other entity that is semantically similar to the first entity, and the second set including the second entity and at least one other entity that is semantically similar to the second entity. Semantic relations are detected between entities in the first set and entities in the second set. A relation classifier is trained using the pair of related entities and detected semantic relations. The relation classifier model is applied to a new pair of entities to determine a likelihood of a semantic relation between the entities in the new pair of entities.

Extraction of semantic relations using distributional relation detection

A computer-implemented system, method and program product generates answers to questions in an input query text string. The method includes determining, by a programmed processor unit, a lexical answer type (LAT) string associated with an input query; automatically obtaining a candidate answer string to the input query from a data corpus; mapping the query LAT string to a first type string in a structured resource; mapping the candidate answer string to a second type string in the structured resource; and determining if the first type string and the second type string are disjointed; and scoring the candidate answer string based on the determination of the types being disjointed wherein the structured resource includes a semantic database providing ontological content.

Using ontological information in open domain type coercion

A method and an apparatus for selecting an answer to a natural language question. The method includes: detecting a named entity in the natural language question; extracting information related to an answer from the natural language question; searching in linked data according to the detected named entity; generating a candidate answer according to a search result; parsing the candidate answer according to the information related to the answer; and obtaining a value of a feature of the candidate answer; and evaluating each candidate answer by synthesizing the value of the feature of the candidate answer.

Growing community of inventors

Flushing, NY, United States of America

Christopher A Welty