Examples with Feature edu.illinois.cs.cogcomp.edison.features.Feature used on opensource projects

Example 6 with Feature

use of edu.illinois.cs.cogcomp.edison.features.Feature in project cogcomp-nlp by CogComp.

the class PPFeatures method getFeatures.

@Override
public Set<Feature> getFeatures(Constituent c) throws EdisonException {
    TextAnnotation ta = c.getTextAnnotation();
    TreeView parse = (TreeView) ta.getView(parseViewName);
    Set<Feature> feats = new HashSet<>();
    try {
        Constituent phrase = parse.getParsePhrase(c);
        // if the phrase is a PP, then the head word of its
        // rightmost NP child.
        List<Relation> rels = phrase.getOutgoingRelations();
        for (int i = rels.size() - 1; i >= 0; i--) {
            Relation relation = rels.get(i);
            if (relation == null)
                continue;
            Constituent target = relation.getTarget();
            if (ParseTreeProperties.isNominal(target.getLabel())) {
                int head = CollinsHeadFinder.getInstance().getHeadWordPosition(phrase);
                feats.add(DiscreteFeature.create("np-head:" + ta.getToken(head).toLowerCase()));
                feats.add(DiscreteFeature.create("np-head-pos:" + WordHelpers.getPOS(ta, head)));
                break;
            }
        }
        // if the phrase's parent is a PP, then the head of that PP.
        Constituent parent = phrase.getIncomingRelations().get(0).getSource();
        if (parent.getLabel().equals("PP")) {
            int head = CollinsHeadFinder.getInstance().getHeadWordPosition(phrase);
            feats.add(DiscreteFeature.create("p-head:" + ta.getToken(head).toLowerCase()));
        }
    } catch (EdisonException e) {
        throw new RuntimeException(e);
    } catch (Exception e) {
        e.printStackTrace();
    }
    return feats;
}

Example 7 with Feature

use of edu.illinois.cs.cogcomp.edison.features.Feature in project cogcomp-nlp by CogComp.

the class ProjectedPath method getFeatures.

@Override
public Set<Feature> getFeatures(Constituent c) throws EdisonException {
    TextAnnotation ta = c.getTextAnnotation();
    TreeView parse = (TreeView) ta.getView(parseViewName);
    Set<Feature> feats = new HashSet<>();
    // Clone this to avoid concurrency problems
    Constituent c2 = null;
    try {
        c2 = parse.getParsePhrase(c).cloneForNewView("");
    } catch (Exception e) {
        e.printStackTrace();
    }
    assert c2 != null;
    if (!c2.getLabel().equals("VP"))
        return feats;
    boolean found = false;
    boolean done = false;
    while (!done) {
        List<Relation> rels = c2.getIncomingRelations();
        if (rels.size() == 0)
            done = true;
        else {
            Constituent parent = rels.get(0).getSource();
            if (parent.getLabel().equals("VP")) {
                found = true;
                c2 = parent;
            } else {
                done = true;
            }
        }
    }
    if (found) {
        // Clone this to avoid concurrency problems
        Constituent c1 = null;
        try {
            c1 = parse.getParsePhrase(c.getIncomingRelations().get(0).getSource()).cloneForNewView("");
        } catch (Exception e) {
            e.printStackTrace();
        }
        assert c1 != null;
        String path = PathFeatureHelper.getFullParsePathString(c1, c2, 400);
        feats.add(DiscreteFeature.create(path));
    }
    return feats;
}

Example 8 with Feature

use of edu.illinois.cs.cogcomp.edison.features.Feature in project cogcomp-nlp by CogComp.

the class WordConjunctionOneTwoThreeGramWindowTwo method getFeatures.

@Override
public /**
     * This feature extractor assumes that the TOKEN View has been generated in the Constituents TextAnnotation. 
     * It generate a feature for a window [-2, +2] of Forms (original text) for each constituent.
     *
     **/
Set<Feature> getFeatures(Constituent c) throws EdisonException {
    TextAnnotation ta = c.getTextAnnotation();
    TOKENS = ta.getView(ViewNames.TOKENS);
    // We can assume that the constituent in this case is a Word(Token)
    int startspan = c.getStartSpan();
    int endspan = c.getEndSpan();
    // k is 3 since we need up to 3-grams
    int k = 3;
    int window = 2;
    // All our constituents are words(tokens)
    String[] forms = getWindowK(TOKENS, startspan, endspan, window);
    String id, value;
    String classifier = "WordConjunctionOneTwoThreeGramWindowTwo";
    Set<Feature> result = new LinkedHashSet<>();
    for (int j = 0; j < k; j++) {
        for (int i = 0; i < forms.length; i++) {
            // forms.length = 5, So i goes from 0 to 4, for each String in the forms array.
            StringBuilder f = new StringBuilder();
            // and three word combinations within [-2,2] window or words.
            for (int context = 0; context <= j && i + context < forms.length; context++) {
                // add a '_' between words to conjoin them together
                if (context != 0) {
                    f.append("_");
                }
                f.append(forms[i + context]);
            }
            // 2 is the center object in the array so i should go from -2 to +2 (with 0 being
            // the center)
            // j is the size of the n-gram so it goes 1 to 3
            id = classifier + ":" + ((i - window) + "_" + (j + 1));
            value = "(" + (f.toString()) + ")";
            result.add(new DiscreteFeature(id + value));
        }
    }
    return result;
}

Example 9 with Feature

use of edu.illinois.cs.cogcomp.edison.features.Feature in project cogcomp-nlp by CogComp.

the class HasVerb method getFeatures.

@Override
public Set<Feature> getFeatures(Constituent c) throws EdisonException {
    boolean hasVerb = false;
    TextAnnotation ta = c.getTextAnnotation();
    for (int i = c.getStartSpan(); i < c.getEndSpan(); i++) {
        if (POSUtils.isPOSVerb(WordHelpers.getPOS(ta, i))) {
            hasVerb = true;
            break;
        }
    }
    Set<Feature> feats = new HashSet<>();
    if (hasVerb) {
        feats.add(DiscreteFeature.create(getName()));
    }
    return feats;
}

Example 10 with Feature

use of edu.illinois.cs.cogcomp.edison.features.Feature in project cogcomp-nlp by CogComp.

the class ParsePath method getFeatures.

@Override
public Set<Feature> getFeatures(Constituent c) throws EdisonException {
    TextAnnotation ta = c.getTextAnnotation();
    TreeView parse = (TreeView) ta.getView(parseViewName);
    Set<Feature> features = new LinkedHashSet<>();
    List<Relation> incomingRelations = c.getIncomingRelations();
    if (incomingRelations.size() > 0) {
        Constituent c1, c2;
        try {
            c1 = parse.getParsePhrase(incomingRelations.get(0).getSource());
            c2 = parse.getParsePhrase(c);
        } catch (Exception e) {
            throw new EdisonException(e);
        }
        Pair<List<Constituent>, List<Constituent>> paths = PathFeatureHelper.getPathsToCommonAncestor(c1, c2, 400);
        List<Constituent> list = new ArrayList<>();
        for (int i = 0; i < paths.getFirst().size() - 1; i++) {
            list.add(paths.getFirst().get(i));
        }
        Constituent top = paths.getFirst().get(paths.getFirst().size() - 1);
        list.add(top);
        for (int i = paths.getSecond().size() - 2; i >= 0; i--) {
            list.add(paths.getSecond().get(i));
        }
        StringBuilder sb = new StringBuilder();
        for (int i = 0; i < paths.getFirst().size() - 1; i++) {
            Constituent cc = paths.getFirst().get(i);
            sb.append(cc.getLabel());
            sb.append(PathFeatureHelper.PATH_UP_STRING);
        }
        String pathToAncestor = sb.toString();
        String pathString = PathFeatureHelper.getPathString(paths, true, false);
        features.add(DiscreteFeature.create(pathString));
        features.add(DiscreteFeature.create(pathToAncestor));
        features.add(RealFeature.create("l", list.size()));
    }
    return features;
}