Skip to content

Latest commit

 

History

History
350 lines (258 loc) · 12.7 KB

README.md

File metadata and controls

350 lines (258 loc) · 12.7 KB
Raw

Turkish Morphology

Morphology

Turkish is a morphologically rich language. Zemberek provides morphological analysis, morphological ambiguity resolution and word generation functions.

Maven Usage

<dependency>
    <groupId>zemberek-nlp</groupId>
    <artifactId>zemberek-morphology</artifactId>
    <version>0.17.1</version>
</dependency>

Creating TurkishMorphology object

This module provides basic Turkish morphological analysis and generation. Analysis can be done in word and sentence level. For word level analysis and generation, TurkishMorphology class is used. By default, class is instantiated as follows:

TurkishMorphology morphology = TurkishMorphology.createWithDefaults();

After this, Turkish suffix graph is generated, internal dictionaries are loaded and they are connected to related graph nodes. Because creation of this object takes time and consumes memory, a single instance should be used throughout the life of an application.

You can add your own dictionaries or remove existing items during creation of the TurkishMorphology class. For example, you have a dictionary file my-dictionary.txt in this form (albeit terrible words):

show 
relaks [P:Adj]
gugıllamak
Hepsiburada

Dictionary rules are explained here

For adding this dictionary, RootLexicon builder mechanism can be used like this:

RootLexicon lexicon = RootLexicon.builder()
    .addDefaultLexicon()
    .addTextDictionaries(Paths.get("my-dictionary.txt"))
    .build();

TurkishMorphology analyzer = TurkishMorphology.builder()
    .setLexicon(lexicon)
    .build();

Or alternatively:

TurkishMorphology analyzer = TurkishMorphology.create(lexicon);

There are other options available for building the object. Turkish morphology class contain a built in cache, so in time analysis speed will get faster. There is an option to disable the cache if builder mechanism is used. For example:

TurkishMorphology analyzer = TurkishMorphology.builder()
    .setLexicon(RootLexicon.getDefault())
    .disableCache()
    .build();

Generating your own dictionary can be liked this:

  RootLexicon myLexicon = RootLexicon.builder()
      .setLexicon(RootLexicon.getDefault) // start with default
      .addDictionaryLines("foo", "rar") // add two new nouns
      .addTextDictionaries(Paths.get("my-own-dictionary-file")) // add from file
      .build();

Single word morphological analysis

For analyzing a word, analyze method is used. it returns a WordAnalysis object. This object contains the input and zero or more SingleAnalysis objects in it.

  • If a word cannot be parsed, WordAnalysis will have 0 SingleAnalysis item.

  • Even if a word's root does not exist in dictionary, system will try to analyze it anyway. This way numbers or proper nouns that start with capital letters and contain a single quote may get analyzed. Users can disable this behavior by calling disableUnidentifiedTokenAnalyzer() method while building TurkishMorphology object.

    Consider examples "Matsumoto'ya" and "153'ü". System will temporarily generate DictionaryItem objects for "Matsumoto" and "153" and try to parse the words. If successful, returning WordAnalysis will contain SingleAnalysis objects that temporary DictionaryItem object in it. User can check if DictionaryItem is generated temporarily by checking isRuntime() method on returning SingleAnaysis objects of the WordAnalysis result or checking if that DictionaryItem object's root attributes contain RootAttribute.Runtime.

  • Even if input is lower case and without an apostrophe ['], result may contain Proper noun analyses. Analyzer does not care about punctuation rules. User may decide if this is a valid analysis with post processing. In future versions this behavior may change or some helper functionality will be added. Apostrophe is only checked for unknown proper nouns.

    Input = ankaradan
    Analysis = [Ankara:Noun,Prop] ankara:Noun+A3sg+dan:Abl

Examples

Finds all morphological analyses of word "kalemin"

TurkishMorphology morphology = TurkishMorphology.createWithDefaults();
WordAnalysis results = morphology.analyze("kalemin");
results.forEach(s -> System.out.println(s.formatLong()));

Output:

[kale:Noun] kale:Noun+A3sg+m:P1sg+in:Gen
[Kale:Noun,Prop] kale:Noun+A3sg+m:P1sg+in:Gen
[kalem:Noun] kalem:Noun+A3sg+in:Gen
[kalem:Noun] kalem:Noun+A3sg+in:P2sg

Stemming and Lemmatization Example

Finds all morphological analyses, stems and lemmas of word "kitabımızsa"

TurkishMorphology morphology = TurkishMorphology.createWithDefaults();
WordAnalysis result = morphology.analyze("kitabımızsa");
for (SingleAnalysis analysis : result) {
  System.out.println(analysis.formatLong());
  System.out.println("\tStems = " + analysis.getStems());
  System.out.println("\tLemmas = " + analysis.getLemmas());
}

Output:

[kitap:Noun] kitab:Noun+A3sg+ımız:P1pl|Zero→Verb+sa:Cond+A3sg
  Stems = [kitab, kitabımız]
  Lemmas = [kitap, kitabımız]

Known Issues

  • Some words may not get analyzed correctly.
  • Words with circumflex letters may have problems.
  • Proper noun and Abbreviations may not be analyzed correctly

Informal Turkish Words Analysis

As of version 0.16.0, There is a mechanism for analyzing Turkish informal words. For example, word okuycam, analysis:

[okumak:Verb] oku:Verb+yca:Fut_Informal+m:A1sg

Informal morpheme names (like Fut_Informal) have _Informal suffix.

For enabling informal morphological analysis, TurkishMorphology class should be initialized like this:

TurkishMorphology morphology = TurkishMorphology.builder()        
    .setLexicon(RootLexicon.DEFAULT)
    .useInformalAnalysis()
    .build();

morphology.analyzeAndDisambiguate("vurucam kırbacı")
    .bestAnalysis()
    .forEach(System.out::println);

Output:

[vurmak:Verb] vur:Verb+uca:Fut_Informal+m:A1sg
[kırbaç:Noun] kırbac:Noun+A3sg+ı:P3sg

Note that ambiguity resolution mechanism may not work well if sentence contains informal morphemes. There is also a simple informal to formal conversion mechanism InformalAnalysisConverter that generates formal surface form of an informal word analysis.

For example lets assume we used the TurkishMorhology instance created in the previous example:

List<SingleAnalysis> analyses = morphology
        .analyzeAndDisambiguate("okuycam diyo")
        .bestAnalysis();

for (SingleAnalysis a : analyses) {
  System.out.println(a.surfaceForm() + "-" + a);
}

System.out.println("Converting formal surface form:");

InformalAnalysisConverter converter =
    new InformalAnalysisConverter(morphology.getWordGenerator());

for (SingleAnalysis a : analyses) {
  System.out.println(converter.convert(a.surfaceForm(), a));
}

Result will be:

okuycam-[okumak:Verb] oku:Verb+yca:Fut_Informal+m:A1sg
diyo-[demek:Verb] di:Verb+yo:Prog1_Informal+A3sg

Converting formal surface form:

okuyacağım-[okumak:Verb] oku:Verb+yacağ:Fut+ım:A1sg
diyor-[demek:Verb] di:Verb+yor:Prog1+A3sg

Informal Analysis is still experimental and only some cases are covered.

Diacritics Ignored Analysis

Morphological analysis can be configured to ignore Turkish diacritics marks as used in characters [ç,ğ,i,ö,ü,ş] For that purpose ignoreDiacriticsInAnalysis() method is used. For example:

TurkishMorphology morphology = TurkishMorphology.builder()        
    .setLexicon(RootLexicon.DEFAULT)
    .ignoreDiacriticsInAnalysis()
    .build();

morphology.analyze("kisi").forEach(System.out::println);

Output will be:

[kış:Noun,Time] kış:Noun+A3sg+ı:Acc
[kış:Noun,Time] kış:Noun+A3sg+ı:P3sg
[kişi:Noun] kişi:Noun+A3sg

Note that same output will be generated for inputs "kısı, kışi, kişi, kışı" etc.

Ambiguity Resolution

Turkish is a highly ambiguous language. As shown in the example below, Word yarın has 9 morphological analyses. Zemberek uses an Averaged Perceptron based mechanism to resolve ambiguity.

Example

Below is an example of analyzing and applying ambiguity resolution to a sentence.

TurkishMorphology morphology = TurkishMorphology.createWithDefaults();
String sentence = "Yarın kar yağacak.";
System.out.println("Sentence  = " + sentence);
List<WordAnalysis> analysis = morphology.analyzeSentence(sentence);

System.out.println("Before disambiguation.");
for (WordAnalysis entry : analysis) {
  System.out.println("Word = " + entry.getInput());
  for (SingleAnalysis single : entry) {
    System.out.println(single.formatLong());
  }
}

System.out.println("\nAfter disambiguation.");
SentenceAnalysis after = morphology.disambiguate(sentence, analysis);
after.bestAnalysis().forEach(s-> System.out.println(s.formatLong()));

Output:

Sentence  = Yarın kar yağacak.
Before disambiguation.
Word = Yarın
[yarın:Adv] yarın:Adv
[yarmak:Verb] yar:Verb+Imp+ın:A2pl
[Yar:Noun,Prop] yar:Noun+A3sg+ın:Gen
[Yar:Noun,Prop] yar:Noun+A3sg+ın:P2sg
[yar:Noun] yar:Noun+A3sg+ın:Gen
[yar:Noun] yar:Noun+A3sg+ın:P2sg
[yarı:Noun] yarı:Noun+A3sg+n:P2sg
[yarın:Noun,Time] yarın:Noun+A3sg
[yarı:Adj] yarı:Adj|Zero→Noun+A3sg+n:P2sg
Word = kar
[karmak:Verb] kar:Verb+Imp+A2sg
[kar:Noun] kar:Noun+A3sg
[kâr:Noun] kar:Noun+A3sg
Word = yağacak
[yağmak:Verb] yağ:Verb+acak:Fut+A3sg
[yağmak:Verb] yağ:Verb|acak:FutPart→Adj
Word = .
[.:Punc] .:Punc

After disambiguation.
[yarın:Noun,Time] yarın:Noun+A3sg
[kar:Noun] kar:Noun+A3sg
[yağmak:Verb] yağ:Verb+acak:Fut+A3sg
[.:Punc] .:Punc

Known Issues

As of version 0.12, ambiguity resolution mechanism is trained with automatically generated training data. In later versions, manually generated data will be added to improve the accuracy. Previous language model based system is retired for the time being.

Word Generation

Zemberek offers simple word generation functionality. Generation requires root form of a word or a Dictionary item and morphemes. Generation mechanism is very similar to analysis mechanism. But it passes through empty Morphemes in the search graph even if they are not provided in the input. For example, user does not need to provide A3sg morpheme as input as it's surface is empty. Generation returns an inner static class Result instance. There may be multiple results. Form that object, generated word (surface form) and analysis results of the generated word can be accessed.

Example

In the example below, a dictionary with a single word "armut" is used for creating TurkishMorphology instance. After that, possessive and case suffix combinations are used for generating inflections of that word. You can run this example from zemberek.examples.morphology.GenerateWords in examples module.

String[] number = {"A3sg", "A3pl"};
String[] possessives = {"P1sg", "P2sg", "P3sg"};
String[] cases = {"Dat", "Loc", "Abl"};

TurkishMorphology morphology =
    TurkishMorphology.builder().addDictionaryLines("armut").disableCache().build();

DictionaryItem item = morphology.getLexicon().getMatchingItems("armut").get(0);
for (String numberM : number) {
  for (String possessiveM : possessives) {
    for (String caseM : cases) {
      List<Result> results =
          morphology.getWordGenerator().generate(item, numberM, possessiveM, caseM);
      results.forEach(s->System.out.println(s.surface));
    }
  }
}

Output:

armuduma
armudumda
armudumdan
armuduna
armudunda
armudundan
armuduna
armudunda
armudundan
armutlarıma
armutlarımda
armutlarımdan
armutlarına
armutlarında
armutlarından
armutlarına
armutlarında
armutlarından