Examples with StemTransition zemberek.morphology.morphotactics.StemTransition used on opensource projects

Example 1 with StemTransition

use of zemberek.morphology.morphotactics.StemTransition in project zemberek-nlp by ahmetaa.

the class WordGenerator method generate.

private List<Result> generate(String input, List<StemTransition> candidates, List<Morpheme> morphemes) {
    if (debugMode) {
        debugData = new AnalysisDebugData();
        debugData.input = input;
        debugData.candidateStemTransitions.addAll(candidates);
    }
    // generate initial search paths.
    List<GenerationPath> paths = new ArrayList<>();
    for (StemTransition candidate : candidates) {
        // we set the tail as " " because in morphotactics, some conditions look for tail's size
        // during graph walk. Because this is generation we let that condition pass always.
        SearchPath searchPath = SearchPath.initialPath(candidate, " ");
        List<Morpheme> morphemesInPath;
        // we skip it if it matches with the initial morpheme of the graph visiting SearchPath object.
        if (morphemes.size() > 0) {
            if (morphemes.get(0).equals(searchPath.getCurrentState().morpheme)) {
                morphemesInPath = morphemes.subList(1, morphemes.size());
            } else {
                morphemesInPath = new ArrayList<>(morphemes);
            }
        } else {
            morphemesInPath = new ArrayList<>(0);
        }
        paths.add(new GenerationPath(searchPath, morphemesInPath));
    }
    // search graph.
    List<GenerationPath> resultPaths = search(paths);
    // generate results from successful paths.
    List<Result> result = new ArrayList<>(resultPaths.size());
    for (GenerationPath path : resultPaths) {
        SingleAnalysis analysis = SingleAnalysis.fromSearchPath(path.path);
        result.add(new Result(analysis.surfaceForm(), analysis));
        if (debugMode) {
            debugData.results.add(analysis);
        }
    }
    return result;
}

Example 2 with StemTransition

use of zemberek.morphology.morphotactics.StemTransition in project zemberek-nlp by ahmetaa.

the class StemTransitionTrieBasedTest method testPrefix.

@Test
public void testPrefix() {
    RootLexicon lexicon = getLexicon();
    StemTransitionsTrieBased t = new StemTransitionsTrieBased(lexicon, new TurkishMorphotactics(lexicon));
    List<StemTransition> matches = t.getPrefixMatches("kabağa", false);
    Assert.assertEquals(3, matches.size());
    Set<String> lemmas = matches.stream().map(s -> s.item.lemma).collect(Collectors.toSet());
    Assert.assertTrue(TestUtil.containsAll(lemmas, "kaba", "kabağ", "kabak"));
    matches = t.getPrefixMatches("kabak", false);
    Assert.assertEquals(2, matches.size());
    lemmas = matches.stream().map(s -> s.item.lemma).collect(Collectors.toSet());
    Assert.assertTrue(TestUtil.containsAll(lemmas, "kaba", "kabak"));
    matches = t.getPrefixMatches("kapak", false);
    Assert.assertEquals(3, matches.size());
    lemmas = matches.stream().map(s -> s.item.lemma).collect(Collectors.toSet());
    Assert.assertTrue(TestUtil.containsAll(lemmas, "kapak"));
}

Example 3 with StemTransition

use of zemberek.morphology.morphotactics.StemTransition in project zemberek-nlp by ahmetaa.

the class StemTransitionTrieBasedTest method testItem.

@Test
public void testItem() {
    RootLexicon lexicon = getLexicon();
    StemTransitionsTrieBased t = new StemTransitionsTrieBased(lexicon, new TurkishMorphotactics(lexicon));
    DictionaryItem item = lexicon.getItemById("kapak_Noun");
    List<StemTransition> transitions = t.getTransitions(item);
    Assert.assertEquals(2, transitions.size());
    Set<String> surfaces = transitions.stream().map(s -> s.surface).collect(Collectors.toSet());
    Assert.assertTrue(TestUtil.containsAll(surfaces, "kapak", "kapağ"));
}

Example 4 with StemTransition

use of zemberek.morphology.morphotactics.StemTransition in project zemberek-nlp by ahmetaa.

the class StemTransitionsMapBased method generateAsciiTolerantMap.

// TODO: this is kind of a hack. Because StemTransitions may be shared between
// analyzer classes, this may be necessary when one of them happens to be ascii tolerant
// and other is not.
private void generateAsciiTolerantMap() {
    lock.writeLock().lock();
    asciiKeys = HashMultimap.create(1000, 2);
    try {
        // generate MultiMap for ascii tolerant keys
        for (String s : singleStems.keySet()) {
            String ascii = TurkishAlphabet.INSTANCE.toAscii(s);
            if (TurkishAlphabet.INSTANCE.containsAsciiRelated(s)) {
                asciiKeys.put(ascii, s);
            }
        }
        for (StemTransition st : multiStems.values()) {
            String s = st.surface;
            String ascii = TurkishAlphabet.INSTANCE.toAscii(s);
            if (TurkishAlphabet.INSTANCE.containsAsciiRelated(s)) {
                asciiKeys.put(ascii, s);
            }
        }
    } finally {
        lock.writeLock().unlock();
    }
}

Example 5 with StemTransition

use of zemberek.morphology.morphotactics.StemTransition 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);
}