Multi-layered semiotic mechanism for answering natural language questions using document retrieval combined with information extraction

• US 6,584,470 B2
• Filed: 03/01/2001
• Issued: 06/24/2003
• Est. Priority Date: 03/01/2001
• Status: Expired due to Term

• Current Assignee: Buzzmetrics, Ltd. (Nielsen Holdings plc)
• Original Assignee: Intelliseek (Nielsen Holdings plc)
• Inventors: Veale, Richard A.
• Primary Examiner(s): Metjahic, Safet
• Assistant Examiner(s): Nguyen, Cindy
• Application Number:
  US09/796,961
• Publication Number:
  US 20020188586A1
• Time in Patent Office: 845 Days
• Field of Search: 707/102, 707/3
• US Class Current: 1/1
• CPC Class Codes: G06F 16/367 Ontology
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching
• CPC Class Codes: Y10S 707/99943 Generating database or data structure, e.g. via user interface

Abstract

A system and method are provided for processing a document containing multiple words to extract an answer. The method uses a set of signatures and signature based rules. Each rule includes a name formed by concatenating the signatures included in the rule and a concept which is assignable to the combination of signatures included in the rule. The method parses the words and assigns a signature to each of them. It may also assign a window which includes a parameter defining the number of signatures viewable at one time. The window may be moved so that the window reveals signatures not viewable in its previous position. The signatures viewable in the window are read and a rule name is created by concatenating the signatures. The created rule name is compared to the names of the rules and, if the created name matches any of the rule names, the matching rule is applied to create a concept and the concept is assigned back into the window.

Claims

1. A method for processing a document comprising a plurality of words to extract an answer from the document, the method using a plurality of signatures and a plurality of rules based on the signatures, the method comprising:

establishing the plurality of signatures;

establishing the plurality of rules, each rule including

a name formed by concatenating the signatures included in the rule;

a concept, the concept assignable to the combination of signatures included in the scope of the rule;

parsing the plurality of words;

assigning a signature to each of the plurality of words;

assigning a window, the window including a parameter defining the number of signatures viewable at one time;

moving the window if applicable, so that the window reveals signatures not viewable in its previous position;

reading the signatures viewable in the window;

creating a rule name by concatenating the read signatures;

comparing the created rule name to the names of the plurality of rules;

determining whether the created name matches any of the rule names;

applying the matching rule to create a concept if a matching rule exists; and

assigning the concept back into the window.

2. The method of claim 1 further including a rule operable to assign a conceptual position in a concept ontology to the combination of signatures included in the rule.

3. The method of claim 2 further including a rule operable to assign a signature to the combination of signatures in order to replace the combination with a single signature.

4. The method of claim 1 wherein a word is assigned a conceptual signature, so that the word is associated with a concept in the concept ontology.

5. The method of claim 1 wherein the word is assigned an orthographic signature, so that the word is associated with its orthographic form.

6. The method of claim 1 further including representing each rule as a frame, where the frame is a named object.

7. The method of claim 1 further including a rule comprising a leftmost element and a rightmost element, where at least one of the elements is the product of another rule.

8. A computer system, connectable to a network, the network capable of accessing a search engine and a group of documents, for extracting information using a pipe-lined finite-state architecture including a plurality of signatures and signature based rules, the system including software for processing the documents comprising a plurality of words to extract an answer from the documents, the software comprising instructions for

parsing the plurality of words from a document in the set of documents;

assigning a signature to each of the plurality of words;

reading the signatures;

creating a rule name by concatenating the read signatures;

comparing the created rule name to the names of the plurality of rules;

determining whether the created name matches any of the rule names;

applying the matching rule to create a concept if a matching rule exists; and

assigning the concept back into the window.

9. The system of claim 8 further including instructions for a rule operable to assign a conceptual position in a concept ontology to the combination of signatures included in the rule.

10. The system of claim 9 further including instructions for a rule operable to assign a signature to the combination of signatures in order to replace the combination with a single signature.

11. The system of claim 8 wherein a word is assigned a conceptual signature, so that the word is associated with a concept in the concept ontology.

12. The system of claim 8 wherein the word is assigned an orthographic signature, so that the word is associated with its orthographic form.

13. The system of claim 8 further including instructions for representing each rule as a frame, where the frame is a named object modifiable at run time.

14. The system of claim 13 further including instructions for each frame comprising at least one named slot, where each slot includes a set of reference names to other frames.

15. The system of claim 14 further including instructions for cross linking a plurality of frames via their respective slots to create a concept network, the concept network providing a layer of meaning dependent on the interconnection of the frames.

16. The method of claim 8 further including a rule comprising a leftmost element and a rightmost element, where at least one of the elements is the product of another rule.

17. A method for combining a plurality of partial answers in a group of documents into a single answer, each document comprising a plurality of passages, the method comprising:

retrieving a subset of documents from the group of documents, the subset comprising documents which include at least one partial answer;

identifying a semantic answer in a document in the subset of documents;

identifying a first most relevant passage in the document containing the semantic answer;

identifying a most relevant document in the subset of documents;

identifying a second most relevant passage in the most relevant document;

determining whether the first most relevant passage and the second most relevant passage are the same passage;

determining whether the first most relevant passage includes the semantic answer if the first most relevant passage and the second most relevant passage are the same passage;

forming the single answer by conjoining the semantic answer and the first most relevant passage if the first most relevant passage does not include the semantic answer; and

selecting the first most relevant passage as the single answer if the first most relevant passage includes the semantic answer.

18. The method of claim 17 further including determining whether the first most relevant passage contains the semantic answer;

determining whether the second most relevant passage contains the semantic answer;

forming the single answer by conjoining the semantic answer, the first most relevant passage, and the second most relevant passage if neither the first most relevant passage nor the second most relevant passage contain the semantic answer; and

forming the single answer by conjoining the first most relevant passage and the second most relevant passage if either the first most relevant passage or the second most relevant passage contain the semantic answer.

19. A computer system, connectable to a network, the network capable of accessing a search engine and a group of documents, for extracting information using a pipe-lined finite-state architecture, the system including software for searching combining a plurality of partial answers in the group of documents into a single answer, each document comprising a plurality of passages, the software comprising instructions for

retrieving a subset of documents from the group of documents, the subset comprising documents which include at least one partial answer;

identifying a semantic answer in a document in the subset of documents;

identifying a first most relevant passage in the document containing the semantic answer;

identifying a most relevant document in the subset of documents;

identifying a second most relevant passage in the most relevant document;

determining whether the first most relevant passage and the second most relevant passage are the same passage;

determining whether the first most relevant passage includes the semantic answer if the first most relevant passage and the second most relevant passage are the same passage;

forming the single answer by conjoining the semantic answer and the first most relevant passage if the first most relevant passage does not include the semantic answer; and

selecting the first most relevant passage as the single answer if the first most relevant passage includes the semantic answer.

20. The system of claim 19 further including instructions for

determining whether the first most relevant passage contains the semantic answer;

determining whether the second most relevant passage contains the semantic answer;

forming the single answer by conjoining the semantic answer, the first most relevant passage, and the second most relevant passage if neither the first most relevant passage nor the second most relevant passage contain the semantic answer; and

forming the single answer by conjoining the first most relevant passage and the second most relevant passage if either the first most relevant passage or the second most relevant passage contain the semantic answer.