Fact recognition system

• US 6,609,087 B1
• Filed: 04/28/1999
• Issued: 08/19/2003
• Est. Priority Date: 04/28/1999
• Status: Expired due to Term

• Current Assignee: Cxense ASA
• Original Assignee: Genuity Inc. (Lumen Technologies, Inc.), Verizon Corporate Services Group Incorporated (Verizon Communications Inc.)
• Inventors: Miller, Scott, Ramshaw, Lance Arthur, Weischedel, Ralph Mark, Fox, Heidi Jennifer
• Primary Examiner(s): Jones, Hugh
• Assistant Examiner(s): FERRIS III, FRED O
• Application Number:
  US09/300,922
• Time in Patent Office: 1,574 Days
• Field of Search: 707/5, 707/1, 704/9, 704/1, 382/155, 382/177, 703/22, 706/45
• US Class Current: 703/22
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: Y10S 707/99931 Database or file accessing
• CPC Class Codes: Y10S 707/99935 Query augmenting and refining, e.g. inexact access

Abstract

In accordance with methods and systems consistent with the present invention, an improved fact recognition system is provided that automatically learns from syntactic language examples and semantic language examples, thus facilitating development of the system. The language examples are rather simplistic and can be provided by a lay person with little training, thus relieving the need for knowledge engineers. Furthermore, the learning performed by the improved fact recognition system results in a collection of probabilities that is used by the system to recognize facts in a typically more accurate manner than conventional systems.

Claims

1. A method in a data processing system, comprising the steps of:

receiving a set of syntactic trees reflecting a syntactic structure of a first plurality of sentences;

generating a syntactic model reflecting a likelihood of the syntactic structure;

receiving semantically annotated sentences reflecting semantic information for a second plurality of sentences;

identifying a most likely syntactic structure for each of the semantically annotated sentences by using the syntactic model;

augmenting the identified syntactic structures to include the semantic information for each of the second plurality of sentences;

generating a statistical model reflecting a likelihood of both the semantic information and the identified syntactic structure of the second plurality of sentences;

receiving a new sentence containing a fact; and

recognizing the fact in the new sentence by using the statistical model.

2. The method of claim 1, wherein the recognizing step includes:

storing the fact in a data store.

3. The method of claim 1, wherein the recognizing step includes:

generating a plurality of parse trees for the new sentence, each parse tree reflecting likely semantic information and likely syntactic structure for at least a portion of the new sentence;

selecting from among the plurality of parse trees the parse tree having the greatest likelihood of matching the semantic information and the syntactic structure of the new sentence; and

examining the selected parse tree to recognize the fact.

4. A method in a data processing system, comprising the steps of:

receiving syntactic language examples, wherein receiving syntactic language examples includes receiving first syntactic trees reflecting syntactic structure for a first plurality of sentences and generating a syntactic model containing probabilities indicating a likelihood of the syntactic structure for each of the first plurality of sentences;

receiving semantic language examples, wherein receiving semantic language examples includes receiving a second plurality of sentences with semantic annotations and generating second syntactic trees reflecting syntactic structure of the second plurality of sentences by using the syntactic model;

creating a model from both the syntactic language examples and the semantic language examples, wherein creating a model includes augmenting the second syntactic trees for the second plurality of sentences to include semantic information derived from the semantic annotations, wherein the augmented syntactic trees have nodes and augmenting the second syntactic trees includes generating probabilities for each of the nodes of the augmented syntactic trees;

using the model to determine a meaning of a sequence of words, wherein determining the meaning of a sequence of words includes recognizing at least one fact in the sequence of words; and

storing the recognized fact in a data store.

5. A computer-readable medium containing instructions for controlling a data processing system to perform a method comprising the steps of:

receiving syntactic language examples, wherein receiving syntactic language examples includes receiving first syntactic trees reflecting syntactic structure for a first plurality of sentences and generating a syntactic model containing probabilities indicating a likelihood of the syntactic structure;

receiving semantic language examples, wherein receiving semantic language examples includes receiving a second plurality of sentences with semantic annotations and generating second syntactic trees reflecting syntactic structure of the second plurality of sentences by using the syntactic model;

creating a model from both the syntactic language examples and the semantic language examples, wherein creating a model includes augmenting the second syntactic trees for the second plurality of sentences to include semantic information derived from the semantic annotations, wherein the augmented syntactic trees have nodes and augmenting the second syntactic trees includes generating probabilities for each of the nodes of the augmented syntactic trees;

using the model to determine a meaning of a sequence of words, wherein determining the meaning of a sequence of words includes recognizing at least one fact in the sequence of words; and

storing the recognized fact into a data store.