edu.washington.cs.knowitall.sequence

Class LayeredTokenPattern

java.lang.Object

edu.washington.cs.knowitall.sequence.LayeredTokenPattern

public class LayeredTokenPattern
extends Object

A class that defines a regular expression over the tokens appearing in a LayeredSequence object.

For example, suppose we want to find parts of sentences that match the pattern "DT cow", where "DT" is the part-of-speech tag representing a determiner. Assume that sentences are represented as LayeredSequence objects, where the words layer has the name "word" and the part-of-speech layer has the name "pos". Then the above pattern can be constructed by calling new LayeredTokenPattern( "DT_pos cow_word"). Given a test sentence sent, the matcher(LayeredSequence) method will return a LayeredTokenMatcher object that will allow you to access the ranges and groups.

The patterns are expressed using the standard Pattern language, but with the following changes.

The basic unit of match is not a character, but instead a token. A token consists of two parts: a value and a layer name. A token is expressed using an underscore to separate the two. For example Foo_bar will match when the token @{code Foo} appears on the layer with the name bar. In the example above, the token DT_pos will match the word- POS pair (w, p) pair when p = DT. The value of w is allowed to be anything. Currently there is no way to match the value of multiple layers at once (e.g. match all occurrences of "bank" that are nouns).

The value of a token can only have characters from this set: [a-zA-Z0-9\\-.,:;?!"'`]. The layer name can only have characters from this set: [a-zA-Z0-9\\-].

When expressing a pattern, tokens must be space separated.

In the following examples pos refers to a part-of-speech layer, and word refers to a word layer.

• ^John_word lives_word in_word NNP_pos+ - matches sentences that start with "John lives in" and then is followed by at least one proper noun.
• ^(NNP_pos+) lives_word in_word (NNP_pos+) ._pos$ - matches sentences that start with at least one proper noun, followed by "lives in", followed by at least one proper noun, and then ending with a period. Captures the two proper nouns as groups (see LayeredTokenMatcher).

Author:

afader

Constructor Summary

Constructors

Constructor and Description
LayeredTokenPattern(String patternString) Constructs a new instance from the given String pattern

Method Summary

Methods

Modifier and Type	Method and Description
Pattern	getEncodedPattern()
LayeredTokenMatcher	matcher(LayeredSequence seq) Returns a matcher object, which can be used to scan seq for any subsequences that match this pattern.
String	toString()

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Detail

LayeredTokenPattern

public LayeredTokenPattern(String patternString)
                    throws SequenceException

Constructs a new instance from the given String pattern

Parameters:

patternString -

Throws:

SequenceException - if unable to compile patternString

Method Detail

toString

public String toString()

Overrides:

toString in class Object

matcher

public LayeredTokenMatcher matcher(LayeredSequence seq)
                            throws SequenceException

Returns a matcher object, which can be used to scan seq for any subsequences that match this pattern.

Parameters:

seq -

Returns:

the matcher

Throws:

SequenceException - if unable to create a matcher over seq

getEncodedPattern

public Pattern getEncodedPattern()

Returns:

the character-level pattern that this LayeredTokenPattern was compiled into.