Error-tolerant language understanding system and method

US 7,333,928 B2
Filed: 12/18/2002
Issued: 02/19/2008
Est. Priority Date: 05/31/2002
Status: Expired due to Fees

First Claim

Patent Images

1. An error-tolerant language understanding method comprising the following steps:

(a) Inputting at least one word sequence and its corresponding acoustic score;

(b) Parsing said word sequence to obtain a corresponding concept sequence set;

(c) Attach at least one confidence measure sequence to each concept sequence in the said concept sequence set and compare the concept sequences together with their associated confidence measure sequences against at least one exemplary concept sequence to obtain at least one edit operation sequence;

(d) According to said acoustic score of said word sequence, the corresponding grammar score of a concept sequence in said concept sequence set, the corresponding example score of said exemplary concept sequence and the corresponding edit operation score of said edit operation sequence to determine the most possible concept sequence; and

(e) Translating said most possible concept sequence into a semantic frame,wherein the step (d) further comprising;

Using a probabilistic scoring function to determine said the most possible concept sequence, and said probabilistic scoring function is formulated as follows;

$(W, F, C, M, K, E) = \underset{(W, F, C, M, K, E)}{\arg \max} {S_{W} + S_{F} + S_{K} + S_{E}}$ wherein is the most possible word sequence in the sentence list that outputs from the speech recognition module, is the most possible concept parse forest, is the corresponding concept sequence, is the corresponding confidence measure sequences, is the most possible exemplary concept sequence and is the most possible edit operation sequence, S_Wis the acoustic score, S_Fis the grammar score, S_Kis the example score and S_Eis the edit operation score, wherein $S_{W} = \log P (U | W), S_{F} = \sum_{T \in F, A \to α \in T_{i}} \log P (α | A), S_{K} = \sum_{i = l}^{m} \log P (k_{i} | k_{i - 1}), S_{E} = \sum_{e \in E, e = < ɛ, k_{q} >} \log P (e) + \sum_{e \in E, e = < c_{p}, g >} \sum_{h = 1}^{X} \log P (e | δ_{h, p}),$ U represents a utterance signal, W represents said possible word sequence in the sentence list that outputs from the speech recognition module, F represents a possible concept parses forest of W T is a concept parse tree of said concept parse forest F, A→

α

is a concept grammar that generates said T, A is a left-hand-side symbol and α

is right-hand-side symbols, m is the number of concept in exemplary concept sequence K, k₁^mis a brief note of k₁. . . k_m, k_iis the i^thconcept, e is an edit operation in edit operation sequence E, said utterance signal U is processed with X number of confidence measure modules and X number of confidence measure sequences are generated, one of said confidence measure sequences M_hcorresponding to the r number of c₁. . . c_rconcepts is expressed as $M_{h} = δ_{h, 1} \dots δ_{h, r} = δ_{h, 1}^{h, r}, h \sim [1, X] .$

View all claims

1 Assignment

Timeline View

Assignment View

0 Petitions

Accused Products

Abstract

The present invention relates to an error-tolerant language understanding, system and method. The system and the method is using example sentences to provide the clues for detecting and recovering errors. The procedure of detection and recovery is guided by a probabilistic scoring function which integrated the scores from the speech recognizer, concept parser, the scores of concept-bigram and edit operations, such as deleting, inserting and substituting concepts. Meanwhile, the score of edit operations refers the confidence measure achieving more precise detection and recovery of the speech recognition errors. That said, a concept with lower confidence measure tends to be deleted or substituted, while a concept with higher one tends to be retained.

Citations

1 Claim

1. An error-tolerant language understanding method comprising the following steps:
- (a) Inputting at least one word sequence and its corresponding acoustic score;
  
  (b) Parsing said word sequence to obtain a corresponding concept sequence set;
  
  (c) Attach at least one confidence measure sequence to each concept sequence in the said concept sequence set and compare the concept sequences together with their associated confidence measure sequences against at least one exemplary concept sequence to obtain at least one edit operation sequence;
  
  (d) According to said acoustic score of said word sequence, the corresponding grammar score of a concept sequence in said concept sequence set, the corresponding example score of said exemplary concept sequence and the corresponding edit operation score of said edit operation sequence to determine the most possible concept sequence; and
  
  (e) Translating said most possible concept sequence into a semantic frame,wherein the step (d) further comprising;
  
  Using a probabilistic scoring function to determine said the most possible concept sequence, and said probabilistic scoring function is formulated as follows;
  
  $(W, F, C, M, K, E) = \underset{(W, F, C, M, K, E)}{\arg \max} {S_{W} + S_{F} + S_{K} + S_{E}}$ wherein is the most possible word sequence in the sentence list that outputs from the speech recognition module, is the most possible concept parse forest, is the corresponding concept sequence, is the corresponding confidence measure sequences, is the most possible exemplary concept sequence and is the most possible edit operation sequence, S_Wis the acoustic score, S_Fis the grammar score, S_Kis the example score and S_Eis the edit operation score, wherein $S_{W} = \log P (U | W), S_{F} = \sum_{T \in F, A \to α \in T_{i}} \log P (α | A), S_{K} = \sum_{i = l}^{m} \log P (k_{i} | k_{i - 1}), S_{E} = \sum_{e \in E, e = < ɛ, k_{q} >} \log P (e) + \sum_{e \in E, e = < c_{p}, g >} \sum_{h = 1}^{X} \log P (e | δ_{h, p}),$ U represents a utterance signal, W represents said possible word sequence in the sentence list that outputs from the speech recognition module, F represents a possible concept parses forest of W T is a concept parse tree of said concept parse forest F, A→
  
  α
  
  is a concept grammar that generates said T, A is a left-hand-side symbol and α
  
  is right-hand-side symbols, m is the number of concept in exemplary concept sequence K, k₁^mis a brief note of k₁. . . k_m, k_iis the i^thconcept, e is an edit operation in edit operation sequence E, said utterance signal U is processed with X number of confidence measure modules and X number of confidence measure sequences are generated, one of said confidence measure sequences M_hcorresponding to the r number of c₁. . . c_rconcepts is expressed as $M_{h} = δ_{h, 1} \dots δ_{h, r} = δ_{h, 1}^{h, r}, h \sim [1, X] .$

Specification

Resources

Litigation Campaign Assessment

Current Assignee
Industrial Technology Research Institute
Original Assignee
Industrial Technology Research Institute
Inventors
Wang, Huei-Ming, Lin, Yi-Chung
Primary Examiner(s)
Edouard; Patrick N.
Assistant Examiner(s)
YEN, ERIC L

Application Number

US10/321,492
Publication Number

US 20030225579A1
Time in Patent Office

1,889 Days
Field of Search

None
US Class Current

704/9
CPC Class Codes

G10L 15/1815 Semantic context, e.g. disa...

Error-tolerant language understanding system and method

First Claim

1 Assignment

0 Petitions

Accused Products

Abstract

Citations

1 Claim

Specification

Solutions

Use Cases

Quick Links

Error-tolerant language understanding system and method

First Claim

1 Assignment

Subscription Required

Subscription Required

0 Petitions

Subscription Required

Accused Products

Subscription Required

Abstract

Citations

1 Claim

Specification

Subscription Required

Solutions

Use Cases

Quick Links