Type evaluation in a question-answering system

A system for scoring candidate answers to questions comprising:

• a memory storage device;
• a hardware processor, coupled to said memory storage device, for receiving instructions therefrom to configure said hardware processor to perform a method comprising:
• receiving a natural language question;
• applying syntactic analysis on said question to determine a lexical answer type (LAT) for said question;
• generating a second question based on said LAT;
• identifying one or more LAT instances by obtaining answers to said generated second question from a knowledge base;
• mapping said one or more LAT instances to nodes in said one or more taxonomies;
• identifying, in each said one or more taxonomies, one or more supertype nodes associated with nodes of said mapped LAT instances, each said one or more supertype nodes identified based on a coverage value indicating a percentage of said one or more LAT instances mapped to nodes that are child nodes of said supertype node;
• generating, using a question answering system, one or more candidate answers to said question,
• mapping each said candidate answer to nodes in said one or more taxonomies;
• for each respective taxonomy, generating a type evaluation score for a candidate answer based on a linkage between said candidate answer node and a supertype node within the taxonomy;
• for each candidate answer, aggregating their type evaluation scores found in the respective taxonomies;
• and
• generating one or more answers to said question, wherein each answer has an associated confidence score based at least in part on said aggregated type evaluation score.

A computer program product for scoring candidate answers to questions, said computer program product comprising a computer readable storage medium having program instructions embodied therewith, said program instruction executable by a computer to cause said computer to:

• receive, from an input device, a natural language question;
• perform, by a processor, syntactic analysis on said question to determine a lexical answer type (LAT) for said question;
• generate, by a processor, a second question based on said LAT;
• identify, by a processor, one or more LAT instances by obtaining answers to said generated second question from a knowledge base;
• map, by a processor, said one or more LAT instances to nodes of one or more taxonomies;
• identify, by a processor, in said one or more taxonomies, one or more supertype nodes associated with nodes of said mapped LAT instances, each said one or more supertype nodes identified based on a coverage value indicating a percentage of said one or more LAT instances mapped to nodes that are child nodes of said supertype node;
• generate, using a question answering system, one or more candidate answers to said question,
• map each said candidate answer to nodes in said one or more taxonomies;
• for each respective taxonomy, generate, by a processor, a type evaluation score for a candidate answer based on a linkage between said candidate answer node and a supertype node within the taxonomy;
• for each candidate answer, aggregate their type evaluation scores found in the respective taxonomies;
• and
• generate, by a processor, one or more answers to said question, wherein each answer has an associated confidence score based at least in part on said aggregated type evaluation score.