System and method for finding the most likely answer to a natural language question

• US 8,340,955 B2
• Filed: 04/28/2008
• Issued: 12/25/2012
• Est. Priority Date: 11/15/2000
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Brown, Eric W., Coden, Anni R., Prager, John Martin, Radev, Dragomir Radkov, Samn, Valerie
• Primary Examiner(s): Hudspeth, David R
• Assistant Examiner(s): SPOONER, LAMONT M
• Application Number:
  US12/110,481
• Publication Number:
  US 20080201132A1
• Time in Patent Office: 1,702 Days
• Field of Search: 704/1, 704/9, 704/1.9, 707706-708
• US Class Current: 704/9
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

Automated question answering is disclosed that relates to the selection of an answer to a question from a pool of potential answers which are manually or automatically extracted from a large collection of textual documents. The a feature extraction component, a feature combination component, an answer selection component, and an answer presentation component, among others, are included. The input to the system is a set of one or more natural language questions and a collection of textual document The output is a (possibly ranked) set of factual answers to the questions, these answers being extracted from the document collection.

Claims

1. A method for selecting answers to natural language questions from a collection of textual documents comprising the steps of:

extracting scoring features from a candidate list of passages of possible answers, wherein said scoring feature is an average distance between the beginning of a candidate answer in a passage and words in said natural language question that appear in said passage, and wherein said average distance is measured in a count of words;

scoring the possible answers using the extracted scoring features and a features scoring function; and

presenting the best scoring possible answer to the user with context from the passage containing the answer, wherein one or more of said steps are performed by a hardware device.

2. A method as in claim 1, wherein the feature scoring function is a linear combination of weighted features.

3. A method as in claim 2, wherein the parameters of the scoring function are manually determined.

4. A method as in claim 2, wherein the parameters of the scoring function are learned by a machine learning algorithm.

5. A method as in claim 1, wherein the candidate list of passages of possible answers is obtained from the collection of documents using an information retrieval engine.

6. A system for selecting answers to natural language questions from a collection of textual documents, comprising:

a memory; and

at least one processor, coupled to the memory, operative to;

extract scoring features from a candidate list of passages of possible answers, wherein said scoring feature is an average distance between the beginning of a candidate answer in a passage and words in said natural language question that appear in said passage, and wherein said average distance is measured in a count of words;

score the possible answers using the extracted scoring features and a features scoring function; and

present the best scoring possible answer to the user with context from the passage containing the answer.

7. A system as in claim 6, wherein the feature scoring function is a linear combination of weighted features.

8. A system as in claim 7, wherein the parameters of the scoring function are manually determined.

9. A system as in claim 7, wherein the parameters of the scoring function are learned by a machine learning algorithm.

10. A system as in claim 6, wherein the candidate list of passages of possible answers is obtained from the collection of documents using an information retrieval engine.

11. A article of manufacture for selecting answers to natural language questions from a collection of textual documents, comprising a machine readable medium containing one or more programs which when executed implement the steps of:

extracting scoring features from a candidate list of passages of possible answers, wherein said scoring feature is an average distance between the beginning of a candidate answer in a passage and words in said natural language question that appear in said passage, and wherein said average distance is measured in a count of words;

scoring the possible answers using the extracted scoring features and a features scoring function; and

presenting the best scoring possible answer to the user with context from the passage containing the answer.