nounPosTagCollection =
PosTagsCollectionEnum.getPosTagCollection(lang, PosTypeCollectionType.NOUN);
if(nounPosTagCollection != null && !nounPosTagCollection.isEmpty()){
return new PosTypeChunker(nounPosTagCollection,
PosTagsCollectionEnum.getPosTagCollection(
lang,PosTypeCollectionType.FOLLOW),minPosTagProbaility);
} else {
return null;
}
}
/**
* Initialise a new PosTypeChunker for the parsed POS tag collections. This
* Constructor can be used if no predefined Configuration for a given
* language is available in the {@link PosTagsCollectionEnum}
* Note that buildPosTypes are added to the followed once. Therefore the
* followPosTypes may or may not include some/all buildPosTypes.
* @param buildPosTypes the POS types that trigger a new Chunk (MUST NOT be
* null nor {@link Set#isEmpty() empty}).
* @param followPosTypes additional POS types followed to extend Chunks (MAY
* BE null or empty).
*/
public PosTypeChunker(Set buildPosTypes,Set followPosTypes,double minPosProb){
if(buildPosTypes == null || buildPosTypes.isEmpty()){
throw new IllegalArgumentException("The set of POS types used to" +
"build Chunks MUST NOT be NULL nor empty!");
}
this.buildTypes = Collections.unmodifiableSet(new TreeSet(buildPosTypes));
Set follow = new TreeSet();
follow.addAll(buildTypes);
if(followPosTypes != null){
follow.addAll(followPosTypes);
}
this.followTypes = Collections.unmodifiableSet(follow);
if(minPosProb > 1){
throw new IllegalArgumentException("The minimum POS tag probalility MUST BE set to a value [0..1] or values < 0 to deactivate this feature (parsed="+minPosProb+")!");
} else {
this.minPosProb = minPosProb;
}
}
/**
* @param props the probabilities of the pos tags or null if
* not available
* @param pos the POS tags
* @return true if follow
*/
private boolean followPOS(double[] props,String... pos){
boolean reject = false;
for(int i=0;i= minPosProb){
if(followTypes.contains(pos[i])){
return true;
} else {
reject = true;
}
} //else prob to low ... do not process
}
//in case we have not found a POS tag with a prob > minPosProb
//return TRUE
return !reject;
}
private boolean includePOS(double[] props,String... pos){
boolean reject = false;
for(int i=0;i= minPosProb){
if(buildTypes.contains(pos[i])){
return true;
} else {
reject = true;
}
}
}
//in case we have not found a POS tag with a prob > minPosProb
//return TRUE
return !reject;
}
/**
* The set of POS types followed to extend Chunks. This includes the
* {@link #getChunkPosTypes()} values
* @return the followTypes
*/
public final Set getFollowedPosTypes() {
return followTypes;
}
/**
* The set of POS types used to create Chunks
* @return the buildTypes
*/
public final Set getChunkPosTypes() {
return buildTypes;
}
/**
* Build the chunks based on the parsed tokens and POS tags.
* This method is the equivalent to
* {@link opennlp.tools.chunker.Chunker#chunkAsSpans(String[], String[])}
* @param tokens the tokens
* @param tags the POS tags for the tokens
* @return the chunks as spans over the parsed tokens
*/
public Span[] chunkAsSpans(String[] tokens, String[] tags) {
int consumed = -1;
List chunks = new ArrayList();
for(int i=0;i consumed && followPOS(null,tags[start-1])){
start--; //follow backwards until consumed
}
int followEnd = i;
int end = i;
while(followEnd+1 < tokens.length && followPOS(null,tags[followEnd+1])){
followEnd++; //follow
if(includePOS(null,tags[followEnd])){
end = followEnd; //extend end only if act is include
}
}
chunks.add(new Span(start,end));
// consumed = end;
i = followEnd;
}//build no chunk for this token
}
return chunks.toArray(new Span[chunks.size()]);
}
/**
* Build the chunks based on the parsed tokens and the one or more detected
* POS tags alternatives for the tokens.
* @param tokens the tokens
* @param tags the POS tags for the tokens (1D:tokens; 2D:POS tags)
* @return the chunks as spans over the parsed tokens
*/
public Span[] chunkAsSpans(String[] tokens, String[][] tags,double[][]props) {
//NOTE: this is a 1:1 copy of the above method!! However this is the
// only solution, because merging them into a single one would
// need to copy the Stirng[] of the other into a String[][1] as
// used by this one :(
// If someone has a better Idea feel free to change!
// Rupert Westenthaler (28.Sep.2011)
int consumed = -1;
List chunks = new ArrayList();
for(int i=0;i consumed && followPOS(props[start-1],tags[start-1])){
start--; //follow backwards until consumed
}
int followEnd = i;
int end = i;
while(followEnd+1 < tokens.length && followPOS(props[followEnd+1],tags[followEnd+1])){
followEnd++; //follow
if(includePOS(props[followEnd],tags[followEnd])){
end = followEnd; //extend end only if act is include
}
}
chunks.add(new Span(start,end));
// consumed = end;
i = followEnd;
}//build no chunk for this token
}
return chunks.toArray(new Span[chunks.size()]);
}
}