editDist.pdf
-
Upload
amar-kaswan -
Category
Documents
-
view
220 -
download
1
Transcript of editDist.pdf
-
Spelling Correction: Edit Distance
Pawan Goyal
CSE, IITKGP
July 25, 2014
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 1 / 26
-
Spelling Correction
I am writing this email on behaf of ...The user typed behaf.
Which are some close words?behalf
behave
....
Isolated word error correctionPick the one that is closest to behaf
How to define closest?
Need a distance metric
The simplest metric: edit distance
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 2 / 26
-
Spelling Correction
I am writing this email on behaf of ...
The user typed behaf.
Which are some close words?behalf
behave
....
Isolated word error correctionPick the one that is closest to behaf
How to define closest?
Need a distance metric
The simplest metric: edit distance
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 2 / 26
-
Spelling Correction
I am writing this email on behaf of ...The user typed behaf.
Which are some close words?
behalf
behave
....
Isolated word error correctionPick the one that is closest to behaf
How to define closest?
Need a distance metric
The simplest metric: edit distance
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 2 / 26
-
Spelling Correction
I am writing this email on behaf of ...The user typed behaf.
Which are some close words?behalf
behave
....
Isolated word error correctionPick the one that is closest to behaf
How to define closest?
Need a distance metric
The simplest metric: edit distance
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 2 / 26
-
Spelling Correction
I am writing this email on behaf of ...The user typed behaf.
Which are some close words?behalf
behave
....
Isolated word error correctionPick the one that is closest to behaf
How to define closest?
Need a distance metric
The simplest metric: edit distance
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 2 / 26
-
Spelling Correction
I am writing this email on behaf of ...The user typed behaf.
Which are some close words?behalf
behave
....
Isolated word error correctionPick the one that is closest to behaf
How to define closest?
Need a distance metric
The simplest metric: edit distance
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 2 / 26
-
Spelling Correction
I am writing this email on behaf of ...The user typed behaf.
Which are some close words?behalf
behave
....
Isolated word error correctionPick the one that is closest to behaf
How to define closest?
Need a distance metric
The simplest metric: edit distance
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 2 / 26
-
Spelling Correction
I am writing this email on behaf of ...The user typed behaf.
Which are some close words?behalf
behave
....
Isolated word error correctionPick the one that is closest to behaf
How to define closest?
Need a distance metric
The simplest metric: edit distance
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 2 / 26
-
Edit Distance
The minimum edit distance between two strings
Is the minimum number of editing operations
I InsertionI DeletionI Substitution
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 3 / 26
-
Edit Distance
The minimum edit distance between two strings
Is the minimum number of editing operations
I InsertionI DeletionI Substitution
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 3 / 26
-
Edit Distance
The minimum edit distance between two strings
Is the minimum number of editing operations
I InsertionI DeletionI Substitution
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 3 / 26
-
Minimum Edit Distance
ExampleEdit distance from intention to execution
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 4 / 26
-
Minimum Edit Distance
ExampleEdit distance from intention to execution
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 4 / 26
-
Minimum Edit Distance
If each operation has a cost of 1I Distance between these is 5
If substitution costs 2 (Levenshtein)I Distance between these is 8
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 5 / 26
-
Minimum Edit Distance
If each operation has a cost of 1I Distance between these is 5
If substitution costs 2 (Levenshtein)I Distance between these is 8
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 5 / 26
-
Minimum Edit Distance
If each operation has a cost of 1I Distance between these is 5
If substitution costs 2 (Levenshtein)I Distance between these is 8
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 5 / 26
-
How to find the Minimum Edit Distance?
Searching for a path (sequence of edits) from the start string to the final string:
Initial state: the word we are transforming
Operators: insert, delete, substitute
Goal state: the word we are trying to get to
Path cost: what we want to minimize: the number of edits
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 6 / 26
-
How to find the Minimum Edit Distance?
Searching for a path (sequence of edits) from the start string to the final string:
Initial state: the word we are transforming
Operators: insert, delete, substitute
Goal state: the word we are trying to get to
Path cost: what we want to minimize: the number of edits
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 6 / 26
-
How to find the Minimum Edit Distance?
Searching for a path (sequence of edits) from the start string to the final string:
Initial state: the word we are transforming
Operators: insert, delete, substitute
Goal state: the word we are trying to get to
Path cost: what we want to minimize: the number of edits
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 6 / 26
-
How to find the Minimum Edit Distance?
Searching for a path (sequence of edits) from the start string to the final string:
Initial state: the word we are transforming
Operators: insert, delete, substitute
Goal state: the word we are trying to get to
Path cost: what we want to minimize: the number of edits
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 6 / 26
-
How to find the Minimum Edit Distance?
Searching for a path (sequence of edits) from the start string to the final string:
Initial state: the word we are transforming
Operators: insert, delete, substitute
Goal state: the word we are trying to get to
Path cost: what we want to minimize: the number of edits
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 6 / 26
-
How to find the Minimum Edit Distance?
Searching for a path (sequence of edits) from the start string to the final string:
Initial state: the word we are transforming
Operators: insert, delete, substitute
Goal state: the word we are trying to get to
Path cost: what we want to minimize: the number of edits
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 6 / 26
-
How to find the Minimum Edit Distance?
Searching for a path (sequence of edits) from the start string to the final string:
Initial state: the word we are transforming
Operators: insert, delete, substitute
Goal state: the word we are trying to get to
Path cost: what we want to minimize: the number of edits
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 6 / 26
-
Minimum Edit as Search
How to navigate?The space of all edit sequences is huge
Lot of distinct paths end up at the same state
Dont have to keep track of all of them
Keep track of the shortest path to each state
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 7 / 26
-
Minimum Edit as Search
How to navigate?The space of all edit sequences is huge
Lot of distinct paths end up at the same state
Dont have to keep track of all of them
Keep track of the shortest path to each state
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 7 / 26
-
Minimum Edit as Search
How to navigate?The space of all edit sequences is huge
Lot of distinct paths end up at the same state
Dont have to keep track of all of them
Keep track of the shortest path to each state
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 7 / 26
-
Minimum Edit as Search
How to navigate?The space of all edit sequences is huge
Lot of distinct paths end up at the same state
Dont have to keep track of all of them
Keep track of the shortest path to each state
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 7 / 26
-
Defining Minimum Edit Distance Matrix
For two stringsX of length n
Y of length m
We define D(i, j)
the edit distance between X[1..i] and Y[1..j]
i.e., the first i characters of X and the first j characters of Y
Thus, the edit distance between X and Y is D(n,m)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 8 / 26
-
Defining Minimum Edit Distance Matrix
For two stringsX of length n
Y of length m
We define D(i, j)
the edit distance between X[1..i] and Y[1..j]
i.e., the first i characters of X and the first j characters of Y
Thus, the edit distance between X and Y is D(n,m)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 8 / 26
-
Defining Minimum Edit Distance Matrix
For two stringsX of length n
Y of length m
We define D(i, j)
the edit distance between X[1..i] and Y[1..j]
i.e., the first i characters of X and the first j characters of Y
Thus, the edit distance between X and Y is D(n,m)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 8 / 26
-
Computing Minimum Edit Distance
Dynamic Programming
A tabular computation of D(n,m)
Solving problems by combining solutions to subproblemsBottom-up
I Compute D(i, j) for small i, jI Compute larger D(i, j) based on previously computed smaller valuesI Compute D(i, j) for all i and j till you get to D(n,m)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 9 / 26
-
Computing Minimum Edit Distance
Dynamic Programming
A tabular computation of D(n,m)
Solving problems by combining solutions to subproblems
Bottom-upI Compute D(i, j) for small i, jI Compute larger D(i, j) based on previously computed smaller valuesI Compute D(i, j) for all i and j till you get to D(n,m)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 9 / 26
-
Computing Minimum Edit Distance
Dynamic Programming
A tabular computation of D(n,m)
Solving problems by combining solutions to subproblemsBottom-up
I Compute D(i, j) for small i, jI Compute larger D(i, j) based on previously computed smaller valuesI Compute D(i, j) for all i and j till you get to D(n,m)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 9 / 26
-
Computing Minimum Edit Distance
Dynamic Programming
A tabular computation of D(n,m)
Solving problems by combining solutions to subproblemsBottom-up
I Compute D(i, j) for small i, j
I Compute larger D(i, j) based on previously computed smaller valuesI Compute D(i, j) for all i and j till you get to D(n,m)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 9 / 26
-
Computing Minimum Edit Distance
Dynamic Programming
A tabular computation of D(n,m)
Solving problems by combining solutions to subproblemsBottom-up
I Compute D(i, j) for small i, jI Compute larger D(i, j) based on previously computed smaller values
I Compute D(i, j) for all i and j till you get to D(n,m)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 9 / 26
-
Computing Minimum Edit Distance
Dynamic Programming
A tabular computation of D(n,m)
Solving problems by combining solutions to subproblemsBottom-up
I Compute D(i, j) for small i, jI Compute larger D(i, j) based on previously computed smaller valuesI Compute D(i, j) for all i and j till you get to D(n,m)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 9 / 26
-
Dynamic Programming Algorithm
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 10 / 26
-
The Edit Distance Table
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 11 / 26
-
The Edit Distance Table
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 11 / 26
-
The Edit Distance Table
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 12 / 26
-
Computing Alignments
Computing edit distance may not be sufficient for some applicationsI We often need to align characters of the two strings to each other
We do this by keeping a backtrace
Every time we enter a cell, remember where we came fromWhen we reach the end,
I Trace back the path from the upper right corner to read off the alignment
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 13 / 26
-
Computing Alignments
Computing edit distance may not be sufficient for some applications
I We often need to align characters of the two strings to each other
We do this by keeping a backtrace
Every time we enter a cell, remember where we came fromWhen we reach the end,
I Trace back the path from the upper right corner to read off the alignment
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 13 / 26
-
Computing Alignments
Computing edit distance may not be sufficient for some applicationsI We often need to align characters of the two strings to each other
We do this by keeping a backtrace
Every time we enter a cell, remember where we came fromWhen we reach the end,
I Trace back the path from the upper right corner to read off the alignment
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 13 / 26
-
Computing Alignments
Computing edit distance may not be sufficient for some applicationsI We often need to align characters of the two strings to each other
We do this by keeping a backtrace
Every time we enter a cell, remember where we came fromWhen we reach the end,
I Trace back the path from the upper right corner to read off the alignment
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 13 / 26
-
Computing Alignments
Computing edit distance may not be sufficient for some applicationsI We often need to align characters of the two strings to each other
We do this by keeping a backtrace
Every time we enter a cell, remember where we came from
When we reach the end,I Trace back the path from the upper right corner to read off the alignment
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 13 / 26
-
Computing Alignments
Computing edit distance may not be sufficient for some applicationsI We often need to align characters of the two strings to each other
We do this by keeping a backtrace
Every time we enter a cell, remember where we came fromWhen we reach the end,
I Trace back the path from the upper right corner to read off the alignment
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 13 / 26
-
Computing Alignments
Computing edit distance may not be sufficient for some applicationsI We often need to align characters of the two strings to each other
We do this by keeping a backtrace
Every time we enter a cell, remember where we came fromWhen we reach the end,
I Trace back the path from the upper right corner to read off the alignment
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 13 / 26
-
The Edit Distance Table
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 14 / 26
-
The Edit Distance Table
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 15 / 26
-
Minimum Edit with Backtrace
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 16 / 26
-
Adding Backtrace to Minimum Edit
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 17 / 26
-
The distance matrix
Every non-decreasing pathfrom (0,0) to (M,N)corresponds to an alignmentof two sequences.
An optimal alignment iscomposed of optimalsub-alignments.
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 18 / 26
-
The distance matrix
Every non-decreasing pathfrom (0,0) to (M,N)corresponds to an alignmentof two sequences.
An optimal alignment iscomposed of optimalsub-alignments.
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 18 / 26
-
The distance matrix
Every non-decreasing pathfrom (0,0) to (M,N)corresponds to an alignmentof two sequences.
An optimal alignment iscomposed of optimalsub-alignments.
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 18 / 26
-
Result of Backtrace
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 19 / 26
-
Performance
Time
O(nm)
Space
O(nm)
BacktraceO(n+m)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 20 / 26
-
Performance
TimeO(nm)
Space
O(nm)
BacktraceO(n+m)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 20 / 26
-
Performance
TimeO(nm)
Space
O(nm)
Backtrace
O(n+m)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 20 / 26
-
Performance
TimeO(nm)
Space
O(nm)
BacktraceO(n+m)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 20 / 26
-
Weighted Edit Distance
Why to add weights to the computation?Some letters are more likely to be mistyped.
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 21 / 26
-
Confusion Matrix for Spelling Errors
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 22 / 26
-
Keyboard Design
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 23 / 26
-
Weighted Minimum Edit Distance
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 24 / 26
-
Another Application: Cognate Words
Cognate WordsIn linguistics, cognates are words that have a common etymological origin.
Cognate words for night (English)nuit (French), Nacht (German), nacht (Dutch), nag (afrikaans), nakta(Sanskrit), noch (Russian), notte (Italian), noite (Portugese)
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 25 / 26
-
Another Application: Cognate Words
Replacement from gh to k or g should have lesser cost (weight) than fromgh to t, for instance.
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 26 / 26
-
Another Application: Cognate Words
Replacement from gh to k or g should have lesser cost (weight) than fromgh to t, for instance.
Pawan Goyal (IIT Kharagpur) Spelling Correction July 25, 2014 26 / 26