Error-tolerant language understanding system and method

• US 20030225579A1
• Filed: 12/18/2002
• Published: 12/04/2003
• Est. Priority Date: 05/31/2002
• Status: Active Grant

• Current Assignee: Industrial Technology Research Institute
• Original Assignee: Industrial Technology Research Institute
• Inventors: Wang, Huei-Ming, Lin, Yi-Chung
• Granted Patent: US 7,333,928 B2
• Time in Patent Office: Days
• US Class Current: 704/251
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning

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.

Claims

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). Comparing said concept sequence set with at least one exemplary concept sequence and with at least one confidence measure 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 the most possible concept sequence into a semantic frame.

2. The error-tolerant language understanding method of claim 1, 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;

$\left(\stackrel{}{W},\stackrel{}{F},\stackrel{}{C},\stackrel{}{M},\stackrel{}{K},\stackrel{}{E}\right)=\underset{\left(W,F,C,M,K,E\right)}{\arg <br><br>\text{\hspace{1em}<br><br>}<br><br>\max }<br><br>\left\{S_W+S_F+S_K+S_E\right\}$

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_W is the acoustic score, S_F is the grammar score, S_K is the example score and S_E is the edit operation score.

3. The error-tolerant language understanding method of claim 2, wherein

4. The error-tolerant language understanding method of claim 2, wherein

S_W, S_F, S_K, S_E will be given proper weight, that is,

^S(W,F,C,M,K,E)≡

w₁×

S_W+w₂×

S_F+w₃×

S_K+w₄×

S_E

wherein the w₁, w₂, w₃, w₄ are the weights that are greater than zero.

5. The error-tolerant language understanding method of claim 2, wherein said edit operation sequence comprises at least one said edit operation, said edit operation can be classified into four categories “

Insert”

, “

Delete”

, “

Substitute” and

“

Accept” and

can be expressed as <

x, y>

, wherein,

when x=ε

, y≠

ε

, said edit operation is “

Insert y”

,

when x≠

ε

, y≠

ε

, said edit operation is “

Delete x”

,

when x≠

ε

, y≠

ε

, x≠

y, said edit operation is “

Substitute x with y”

,

when X≠

ε

, x=y, said edit operation is “

Accept x”

,

wherein ε

represents null value.

6. The error-tolerant language understanding system comprising:

a concept grammar database used to store at least one concept grammar rule;

a concept parser that uses said concept grammar rule to parse at least one word sequence to a concept sequence set;

an exemplary concept sequence database used to store at least one exemplary concept sequence and;

a concept checker used to compare said concept sequence set with said exemplary concept sequence in said exemplary concept sequence database to determine the most possible concept sequence and the most possible edit operation sequence, then said the most possible edit operation sequence will be further used to edit said the most possible concept sequence, and the edited concept sequence will be translated into a semantic frame.