/**
* Builds Jet.Tipster.Document object from Penn treebank corpus.
*
* @param in
* @return
* @throws IOException
* @throws InvalidFormatException
*/
public Treebank load(Reader in) throws IOException, InvalidFormatException {
List<ParseTreeNode> trees = new ArrayList<ParseTreeNode>();
PushbackReader input = new PushbackReader(in);
int start = 0;
while (true) {
skipWhitespace(input);
if (lookAhead(input) == -1) {
break;
}
ParseTreeNode tree = readNode(input);
trees.add(tree);
determineSpans(tree, start);
setAnnotations(tree, null);
start = tree.end;
}
String text = buildDocumentString(trees);
Document doc = new Document(text);
for (ParseTreeNode tree : trees) {
doc.annotate("sentence", new Span(tree.start, tree.end), new FeatureSet());
annotate(doc, tree);
}
return new Treebank(doc, trees);
}
/** generate mention annotations (with entity numbers) based on the ACE entities and mentions. */
static void addMentionTags(Document doc, AceDocument aceDoc) {
ArrayList<AceEntity> entities = aceDoc.entities;
for (int i = 0; i < entities.size(); i++) {
AceEntity entity = entities.get(i);
ArrayList<AceEntityMention> mentions = entity.mentions;
for (int j = 0; j < mentions.size(); j++) {
AceEntityMention mention = (AceEntityMention) mentions.get(j);
// we compute a jetSpan not including trailing whitespace
Span aceSpan = mention.head;
// skip mentions in ChEnglish APF not aligned to any English text
if (aceSpan.start() < 0) continue;
Span jetSpan = new Span(aceSpan.start(), aceSpan.end() + 1);
FeatureSet features = new FeatureSet("entity", new Integer(i));
if (flags.contains("types")) {
features.put("type", entity.type.substring(0, 3));
if (entity.subtype != null) features.put("subtype", entity.subtype);
}
if (flags.contains("extents")) {
String cleanExtent = mention.text.replaceAll("\n", " ");
features.put("extent", AceEntityMention.addXmlEscapes(cleanExtent));
}
doc.annotate("mention", jetSpan, features);
}
}
}
/** write 'fileText' out as file 'XMLfileName' with ENAMEX tags for the names in the document */
static void addENAMEXtags(Document doc, AceDocument aceDoc) {
ArrayList<AceEntity> entities = aceDoc.entities;
for (int i = 0; i < entities.size(); i++) {
AceEntity entity = entities.get(i);
ArrayList<AceEntityName> names = entity.names;
for (int j = 0; j < names.size(); j++) {
AceEntityName name = names.get(j);
Span aceSpan = name.extent;
Span jetSpan = new Span(aceSpan.start(), aceSpan.end() + 1);
doc.annotate("ENAMEX", jetSpan, new FeatureSet("TYPE", entity.type));
}
// for 2004 we have to examine PRE mentions and decide which are names
if (year.equals("2004")) {
ArrayList<AceEntityMention> mentions = entity.mentions;
for (int j = 0; j < mentions.size(); j++) {
AceEntityMention mention = mentions.get(j);
String htext = Resolve.normalizeName(mention.headText);
String[] mentionName = Gazetteer.splitAtWS(htext);
String preClass = preDict.get(htext.toLowerCase());
if (mention.type.equals("PRE")) {
if (gazetteer.isNationality(mentionName)
|| gazetteer.isLocation(mentionName)
|| "N".equals(preClass)) {
Span aceSpan = mention.head;
Span jetSpan = new Span(aceSpan.start(), aceSpan.end() + 1);
doc.annotate("ENAMEX", jetSpan, new FeatureSet("TYPE", entity.type));
} else if (preClass != null) {
// do nothing
} else {
System.out.println(
"Unclassified PRE: " + mention.text + " {" + mention.headText + ")");
unknownPre.add(htext.toLowerCase());
}
}
}
}
}
}
static void addTimexTags(Document doc, AceDocument aceDoc) {
List<AceTimex> timeExpressions = aceDoc.timeExpressions;
for (AceTimex timex : timeExpressions) {
AceTimexMention mention = (AceTimexMention) timex.mentions.get(0);
Span aceSpan = mention.extent;
Span jetSpan = new Span(aceSpan.start(), aceSpan.end() + 1);
FeatureSet features = new FeatureSet();
if (timex.val != null && !timex.val.equals("")) features.put("val", timex.val);
if (timex.anchorVal != null && !timex.anchorVal.equals(""))
features.put("anchor_val", timex.anchorVal);
if (timex.anchorDir != null && !timex.anchorDir.equals(""))
features.put("anchor_dir", timex.anchorDir);
if (timex.set != null && !timex.set.equals("")) features.put("set", timex.set);
if (timex.mod != null && !timex.mod.equals("")) features.put("mod", timex.mod);
doc.annotate("timex2", jetSpan, features);
}
}
/** generate mention annotations (with entity numbers) based on the ACE entities and mentions. */
static void addMentionTags(Document doc, AceDocument aceDoc) {
ArrayList<AceEntity> entities = aceDoc.entities;
for (int i = 0; i < entities.size(); i++) {
AceEntity entity = (AceEntity) entities.get(i);
ArrayList<AceEntityMention> mentions = entity.mentions;
for (int j = 0; j < mentions.size(); j++) {
AceEntityMention mention = mentions.get(j);
// we compute a jetSpan not including trailing whitespace
Span aceSpan = mention.head;
Span jetSpan = new Span(aceSpan.start(), aceSpan.end() + 1);
FeatureSet features = new FeatureSet("entity", new Integer(i));
if (showTypes) {
features.put("type", entity.type.substring(0, 3));
if (entity.subtype != null) features.put("subtype", entity.subtype);
}
doc.annotate("mention", jetSpan, features);
}
}
}
private void annotate(Document doc, ParseTreeNode node) {
doc.addAnnotation(node.ann);
if (node.children != null) {
Annotation[] children = new Annotation[node.children.length];
for (int i = 0; i < node.children.length; i++) {
children[i] = node.children[i].ann;
}
node.ann.put("children", children);
for (ParseTreeNode child : node.children) {
annotate(doc, child);
}
}
if (node.children == null && isAddingTokens) {
// TODO: adds `case' property
doc.annotate("token", node.ann.span(), new FeatureSet());
}
}