Examples with PhoneticAttribute zemberek.core.turkish.PhoneticAttribute used on opensource projects

Example 6 with PhoneticAttribute

use of zemberek.core.turkish.PhoneticAttribute in project zemberek-nlp by ahmetaa.

the class SuffixSurfaceNode method printAttributes.

private void printAttributes(StringBuilder sb) {
    if (!attributes.isEmpty()) {
        sb.append(" [A:");
    } else {
        return;
    }
    int i = 0;
    for (PhoneticAttribute attribute : attributes) {
        sb.append(attribute.getStringForm());
        if (i++ < attributes.size() - 1) {
            sb.append(", ");
        }
    }
    sb.append("]");
}

Example 7 with PhoneticAttribute

use of zemberek.core.turkish.PhoneticAttribute in project zemberek-nlp by ahmetaa.

the class StemNodeGenerator method generateModifiedRootNodes.

private StemNode[] generateModifiedRootNodes(DictionaryItem dicItem) {
    if (dicItem.hasAttribute(Special)) {
        return handleSpecialStems(dicItem);
    }
    TurkishLetterSequence modifiedSeq = new TurkishLetterSequence(dicItem.pronunciation, alphabet);
    EnumSet<PhoneticAttribute> originalAttrs = calculateAttributes(dicItem.pronunciation);
    EnumSet<PhoneticAttribute> modifiedAttrs = originalAttrs.clone();
    EnumSet<PhoneticExpectation> originalExpectations = EnumSet.noneOf(PhoneticExpectation.class);
    EnumSet<PhoneticExpectation> modifiedExpectations = EnumSet.noneOf(PhoneticExpectation.class);
    for (RootAttribute attribute : dicItem.attributes) {
        // generate other boundary attributes and modified root state.
        switch(attribute) {
            case Voicing:
                TurkicLetter last = modifiedSeq.lastLetter();
                TurkicLetter modifiedLetter = alphabet.voice(last);
                if (modifiedLetter == null) {
                    throw new LexiconException("Voicing letter is not proper in:" + dicItem);
                }
                if (dicItem.lemma.endsWith("nk")) {
                    modifiedLetter = TurkishAlphabet.L_g;
                }
                modifiedSeq.changeLetter(modifiedSeq.length() - 1, modifiedLetter);
                modifiedAttrs.remove(PhoneticAttribute.LastLetterVoicelessStop);
                originalExpectations.add(PhoneticExpectation.ConsonantStart);
                modifiedExpectations.add(PhoneticExpectation.VowelStart);
                break;
            case Doubling:
                modifiedSeq.append(modifiedSeq.lastLetter());
                originalExpectations.add(PhoneticExpectation.ConsonantStart);
                modifiedExpectations.add(PhoneticExpectation.VowelStart);
                break;
            case LastVowelDrop:
                if (modifiedSeq.lastLetter().isVowel()) {
                    modifiedSeq.delete(modifiedSeq.length() - 1);
                    modifiedExpectations.add(PhoneticExpectation.ConsonantStart);
                } else {
                    modifiedSeq.delete(modifiedSeq.length() - 2);
                    if (!dicItem.primaryPos.equals(PrimaryPos.Verb)) {
                        originalExpectations.add(PhoneticExpectation.ConsonantStart);
                    }
                    modifiedExpectations.add(PhoneticExpectation.VowelStart);
                }
                break;
            case InverseHarmony:
                originalAttrs.add(PhoneticAttribute.LastVowelFrontal);
                originalAttrs.remove(PhoneticAttribute.LastVowelBack);
                modifiedAttrs.add(PhoneticAttribute.LastVowelFrontal);
                modifiedAttrs.remove(PhoneticAttribute.LastVowelBack);
                break;
            case ProgressiveVowelDrop:
                modifiedSeq.delete(modifiedSeq.length() - 1);
                if (modifiedSeq.hasVowel()) {
                    modifiedAttrs = calculateAttributes(modifiedSeq);
                }
                break;
            default:
                break;
        }
    }
    StemNode original = new StemNode(dicItem.root, dicItem, originalAttrs, originalExpectations);
    StemNode modified = new StemNode(modifiedSeq.toString(), dicItem, modifiedAttrs, modifiedExpectations);
    SuffixData[] roots = suffixProvider.defineSuccessorSuffixes(dicItem);
    original.exclusiveSuffixData = roots[0];
    modified.exclusiveSuffixData = roots[1];
    if (original.equals(modified)) {
        return new StemNode[] { original };
    }
    modified.setTermination(TerminationType.NON_TERMINAL);
    if (dicItem.hasAttribute(RootAttribute.CompoundP3sgRoot)) {
        original.setTermination(TerminationType.NON_TERMINAL);
    }
    return new StemNode[] { original, modified };
}

Example 8 with PhoneticAttribute

use of zemberek.core.turkish.PhoneticAttribute in project zemberek-nlp by ahmetaa.

the class StemNodeGenerator method generate.

/**
 * Generates StemNode objects from the dictionary item.
 * <p>Most of the time a single StemNode is generated.
 *
 * @param item DictionaryItem
 * @return one or more StemNode objects.
 */
public StemNode[] generate(DictionaryItem item) {
    if (hasModifierAttribute(item)) {
        return generateModifiedRootNodes(item);
    } else {
        SuffixData[] roots = suffixProvider.defineSuccessorSuffixes(item);
        EnumSet<PhoneticAttribute> phoneticAttributes = calculateAttributes(item.pronunciation);
        StemNode stemNode = new StemNode(item.root, item, TerminationType.TERMINAL, phoneticAttributes, EnumSet.noneOf(PhoneticExpectation.class));
        stemNode.exclusiveSuffixData = roots[0];
        return new StemNode[] { stemNode };
    }
}

Example 9 with PhoneticAttribute

use of zemberek.core.turkish.PhoneticAttribute in project zemberek-nlp by ahmetaa.

the class WordGenerator method advance.

// for all allowed matching outgoing transitions, new paths are generated.
// Transition conditions are used for checking if a search path is allowed to pass a transition.
private List<GenerationPath> advance(GenerationPath gPath) {
    List<GenerationPath> newPaths = new ArrayList<>(2);
    // for all outgoing transitions.
    for (MorphemeTransition transition : gPath.path.getCurrentState().getOutgoing()) {
        SuffixTransition suffixTransition = (SuffixTransition) transition;
        // if there are no morphemes and this transitions surface is not empty, no need to check.
        if (gPath.morphemes.isEmpty() && suffixTransition.hasSurfaceForm()) {
            if (debugMode) {
                debugData.rejectedTransitions.put(gPath.path, new RejectedTransition(suffixTransition, "Empty surface expected."));
            }
            continue;
        }
        // if transition surface is empty, here will pass.
        if (!gPath.matches(suffixTransition)) {
            if (debugMode) {
                debugData.rejectedTransitions.put(gPath.path, new RejectedTransition(suffixTransition, "Morpheme mismatch." + suffixTransition.to.morpheme));
            }
            continue;
        }
        // if transition condition fails, add it to debug data.
        if (debugMode && suffixTransition.getCondition() != null) {
            Condition condition = suffixTransition.getCondition();
            Condition failed;
            if (condition instanceof CombinedCondition) {
                failed = ((CombinedCondition) condition).getFailingCondition(gPath.path);
            } else {
                failed = condition.accept(gPath.path) ? null : condition;
            }
            if (failed != null) {
                debugData.rejectedTransitions.put(gPath.path, new RejectedTransition(suffixTransition, "Condition → " + failed.toString()));
            }
        }
        // check conditions.
        if (!suffixTransition.canPass(gPath.path)) {
            continue;
        }
        // epsilon transition. Add and continue. Use existing attributes.
        if (!suffixTransition.hasSurfaceForm()) {
            SearchPath pCopy = gPath.path.getCopyForGeneration(new SurfaceTransition("", suffixTransition), gPath.path.getPhoneticAttributes());
            newPaths.add(gPath.copy(pCopy));
            continue;
        }
        String surface = SurfaceTransition.generateSurface(suffixTransition, gPath.path.getPhoneticAttributes());
        SurfaceTransition surfaceTransition = new SurfaceTransition(surface, suffixTransition);
        // if tail is equal to surface, no need to calculate phonetic attributes.
        AttributeSet<PhoneticAttribute> attributes = AttributesHelper.getMorphemicAttributes(surface, gPath.path.getPhoneticAttributes());
        // This is required for suffixes like `cik` and `ciğ`
        // an extra attribute is added if "cik" or "ciğ" is generated and matches the tail.
        // if "cik" is generated, ExpectsConsonant attribute is added, so only a consonant starting
        // suffix can follow. Likewise, if "ciğ" is produced, a vowel starting suffix is allowed.
        attributes.remove(PhoneticAttribute.CannotTerminate);
        SuffixTemplateToken lastToken = suffixTransition.getLastTemplateToken();
        if (lastToken.getType() == TemplateTokenType.LAST_VOICED) {
            attributes.add(PhoneticAttribute.ExpectsConsonant);
        } else if (lastToken.getType() == TemplateTokenType.LAST_NOT_VOICED) {
            attributes.add(PhoneticAttribute.ExpectsVowel);
            attributes.add(PhoneticAttribute.CannotTerminate);
        }
        SearchPath p = gPath.path.getCopyForGeneration(surfaceTransition, attributes);
        newPaths.add(gPath.copy(p));
    }
    return newPaths;
}

Example 10 with PhoneticAttribute

use of zemberek.core.turkish.PhoneticAttribute in project zemberek-nlp by ahmetaa.

the class StemTransitionsBase method generateModifiedRootNodes.

private List<StemTransition> generateModifiedRootNodes(DictionaryItem dicItem) {
    StringBuilder modifiedSeq = new StringBuilder(dicItem.pronunciation);
    AttributeSet<PhoneticAttribute> originalAttrs = calculateAttributes(dicItem.pronunciation);
    AttributeSet<PhoneticAttribute> modifiedAttrs = originalAttrs.copy();
    MorphemeState modifiedRootState = null;
    MorphemeState unmodifiedRootState = null;
    for (RootAttribute attribute : dicItem.attributes) {
        // generate other boundary attributes and modified root state.
        switch(attribute) {
            case Voicing:
                char last = alphabet.lastChar(modifiedSeq);
                char voiced = alphabet.voice(last);
                if (last == voiced) {
                    throw new LexiconException("Voicing letter is not proper in:" + dicItem);
                }
                if (dicItem.lemma.endsWith("nk")) {
                    voiced = 'g';
                }
                modifiedSeq.setCharAt(modifiedSeq.length() - 1, voiced);
                modifiedAttrs.remove(PhoneticAttribute.LastLetterVoicelessStop);
                originalAttrs.add(PhoneticAttribute.ExpectsConsonant);
                modifiedAttrs.add(PhoneticAttribute.ExpectsVowel);
                // TODO: find a better way for this.
                modifiedAttrs.add(PhoneticAttribute.CannotTerminate);
                break;
            case Doubling:
                modifiedSeq.append(alphabet.lastChar(modifiedSeq));
                originalAttrs.add(PhoneticAttribute.ExpectsConsonant);
                modifiedAttrs.add(PhoneticAttribute.ExpectsVowel);
                modifiedAttrs.add(PhoneticAttribute.CannotTerminate);
                break;
            case LastVowelDrop:
                TurkicLetter lastLetter = alphabet.getLastLetter(modifiedSeq);
                if (lastLetter.isVowel()) {
                    modifiedSeq.deleteCharAt(modifiedSeq.length() - 1);
                    modifiedAttrs.add(PhoneticAttribute.ExpectsConsonant);
                    modifiedAttrs.add(PhoneticAttribute.CannotTerminate);
                } else {
                    modifiedSeq.deleteCharAt(modifiedSeq.length() - 2);
                    if (!dicItem.primaryPos.equals(PrimaryPos.Verb)) {
                        originalAttrs.add(PhoneticAttribute.ExpectsConsonant);
                    } else {
                        unmodifiedRootState = morphotactics.verbLastVowelDropUnmodRoot_S;
                        modifiedRootState = morphotactics.verbLastVowelDropModRoot_S;
                    }
                    modifiedAttrs.add(PhoneticAttribute.ExpectsVowel);
                    modifiedAttrs.add(PhoneticAttribute.CannotTerminate);
                }
                break;
            case InverseHarmony:
                originalAttrs.add(PhoneticAttribute.LastVowelFrontal);
                originalAttrs.remove(PhoneticAttribute.LastVowelBack);
                modifiedAttrs.add(PhoneticAttribute.LastVowelFrontal);
                modifiedAttrs.remove(PhoneticAttribute.LastVowelBack);
                break;
            case ProgressiveVowelDrop:
                if (modifiedSeq.length() > 1) {
                    modifiedSeq.deleteCharAt(modifiedSeq.length() - 1);
                    if (alphabet.containsVowel(modifiedSeq)) {
                        modifiedAttrs = calculateAttributes(modifiedSeq);
                    }
                    modifiedAttrs.add(PhoneticAttribute.LastLetterDropped);
                }
                break;
            default:
                break;
        }
    }
    if (unmodifiedRootState == null) {
        unmodifiedRootState = morphotactics.getRootState(dicItem, originalAttrs);
    }
    StemTransition original = new StemTransition(dicItem.root, dicItem, originalAttrs, unmodifiedRootState);
    // if modified root state is not defined in the switch block, get it from morphotactics.
    if (modifiedRootState == null) {
        modifiedRootState = morphotactics.getRootState(dicItem, modifiedAttrs);
    }
    StemTransition modified = new StemTransition(modifiedSeq.toString(), dicItem, modifiedAttrs, modifiedRootState);
    if (original.equals(modified)) {
        return Collections.singletonList(original);
    }
    return Lists.newArrayList(original, modified);
}