ZemberekDotNet

No Java. No JVM. No sidecar. Pure .NET Standard 2.1 — drop it into any .NET Core 3.0+ or .NET 5/6/7/8 project from NuGet and start processing Turkish text immediately.

ZemberekDotNet started as a C#/.NET port of Zemberek-NLP (Natural Language Processing tools for Turkish) and has since evolved into an actively improved library. While it maintains compatibility with the original Java library's module structure and core algorithms, it is no longer a strict port — new features, correctness fixes, and .NET-specific improvements are introduced where needed for production use.

The goal is to provide a high-quality, production-ready Turkish NLP library for the .NET ecosystem, not merely to replicate the Java implementation.

Improvements over the original Java port

• Reciprocal verb morphology: Re-enabled the reciprocal suffix (Iş) transition in the morphotactics engine, which was disabled in the port. Verbs such as kaçış, dövüşmek are now correctly analyzed.
• Smart apostrophe tokenization: Added dual-path merge logic in the tokenizer to correctly handle foreign brand names containing apostrophes (e.g., L'Oréal is emitted as a single token), while preserving Turkish morphological suffix boundaries (e.g., Ankara'ya remains split for downstream analysis).
• ANTLR runtime upgrade: Upgraded from ANTLR 4.9.3 to 4.13.1, replacing a fragile 460-line custom ATN deserializer with the standard runtime implementation.

This library will maintain the same module structure as Zemberek-NLP using NuGet packages under separate projects, and will continue to track the original library where relevant.

Quick Start

Install the morphology module:

dotnet add package ZemberekDotNet.Morphology

Analyze and disambiguate a sentence:

using ZemberekDotNet.Morphology;
using ZemberekDotNet.Morphology.Analysis;

// Loads the built-in lexicon and disambiguation model (one-time initialization)
TurkishMorphology morphology = TurkishMorphology.CreateWithDefaults();

SentenceAnalysis result = morphology.AnalyzeAndDisambiguate("Kitaplara gidiyorum.");
foreach (SentenceWordAnalysis swa in result)
{
    SingleAnalysis best = swa.GetBestAnalysis();
    Console.WriteLine($"{swa.GetWordAnalysis().GetInput()} → {best.GetLemmas()[0]}  [{best.FormatLexical()}]");
}
// Kitaplara → kitap  [kitap:Noun+A3pl+Dat]
// gidiyorum → git    [git:Verb+Pres+A1sg]

Install all modules at once:

dotnet add package ZemberekDotNet.All

Modules

Module	Package Name	Description	Status
All	ZemberekDotNet.All	Wrapper Package that includes all the modules.
Core	ZemberekDotNet.Core	Special Collections, Hash functions and helpers.
Morphology	ZemberekDotNet.Morphology	Turkish morphological analysis, disambiguation and word generation.
Tokenization	ZemberekDotNet.Tokenization	Turkish Tokenization and sentence boundary detection.
Normalization	ZemberekDotNet.Normalization	Basic spell checker, word suggestion. Noisy text normalization.
NER	ZemberekDotNet.NER	Turkish Named Entity Recognition.
Classification	ZemberekDotNet.Classification	High-performance fastText-based text classification.
Language Identification	ZemberekDotNet.LangID	Fast identification of text language.
Language Modeling	ZemberekDotNet.LM	Provides a language model compression algorithm.
Applications	ZemberekDotNet.Apps	CLI tools for classification, morphology, corpus preprocessing, and NER workflows.	✅ Available
gRPC Server (planned)	ZemberekDotNet.GRPC	gRPC server for access from other languages.	Deferred for future (not required for current port scope)
Examples: Classification	—	News title category classification using fastText.	✅ Available
Examples: Morphology	—	Analysis, disambiguation, LINQ lemma extraction, word generation (conjugation + noun cases).	✅ Available
Examples: Tokenization	—	Sentence splitting, token-type inspection, document processing.	✅ Available
Examples: Language ID	—	Language ID, confidence scores, ContainsLanguage, sentence-level mixed-language scanner.	✅ Available
Examples: NER	—	In-memory NER training and named-entity inference with PERSON/LOCATION/ORGANIZATION labels.	✅ Available
Examples: Normalization	—	Turkish spell check, word suggestions, sentence-level typo highlighting.	✅ Available
Examples: Pipeline	—	End-to-end Tokenization → Morphology → LangID pipeline with POS fingerprinting.	✅ Available

Documentation

The full developer documentation is published on the GitHub Wiki.

Page	Description
Apps CLI Guide	CLI tools for classification, morphology, corpus preprocessing, and NER
Java to .NET Migration Quickstart	Step-by-step guide for teams moving from Java Zemberek
Developer Guide	Build, test, and contribute from source
Morphology Notes	Morphological analysis and disambiguation internals
Morphemes Reference	Full morpheme inventory with examples
Additions and Release Notes	ZemberekDotNet-specific additions and release-level API changes
Classification Training Guide	Training and evaluating fastText text classifiers
Normalization Guide	Noisy text normalization and spell checking
Proper Nouns and Named Entities	NER, proper noun handling, and entity types
Text Dictionary Rules	Lexicon format, rule syntax, and custom dictionaries
Java vs .NET Side-by-Side	API name mapping between the Java and .NET libraries
FAQ	Common questions: licensing, performance, .NET Framework limits
2026 SOTA Roadmap	Phased improvement plan: parity tooling, performance sprint, Native AOT, neural disambiguation

Wiki source lives in the docs/ folder and is auto-synced to the wiki on each local build.

Notes:

• Current port-completion scope prioritizes core library and example parity.
• gRPC documentation is deferred for future because the gRPC module is deferred for future in this repository.

Recent Additions

Recent morphology additions are documented in docs/additions.md, including:

• TurkishCase enum and SingleAnalysis case helpers in ZemberekDotNet.Morphology.Extended
• Numeric apostrophe analysis (ExtdAnalyzeNumeralWithSuffix)
• Confidence-ranked analyses (ExtdAnalyzeWithRanking, ExtdGetRankedAnalyses)
• BK-tree fuzzy analysis (ExtendedMorphologyContext.FuzzyAnalyze)
• Morphology synthesis helper (ExtdSynthesize)

Known Missing / Deferred Items

• ZemberekDotNet.GRPC project is not in the repository yet. The module name is reserved in the table for future implementation.
• Apps CLI parity is focused on high-use workflows (classification, morphology, corpus preprocessing, NER). Additional app surface from the original Java ecosystem may be expanded over time.

Examples

Each example project is a self-contained runnable console app - clone the repo, dotnet run, and see real output. Examples are validated primarily through runnable sample projects and module-level test suites.

Project	What it shows	Entry point
Examples.Morphology	Single-word analysis, sentence disambiguation, LINQ lemma extraction, word generation	MorphologyExamples.cs
Examples.Tokenization	Sentence splitting, token-type inspection, document processing	TokenizationExamples.cs
Examples.LangID	Language detection, confidence scores, ContainsLanguage, mixed-language sentence scanner	LangIDExamples.cs
Examples.NER	Train a small NER model and run PERSON/LOCATION/ORGANIZATION extraction	NERExamples.cs
Examples.Normalization	Spell checking, ranked word suggestions, sentence typo highlighting	NormalizationExamples.cs
Examples.Pipeline	End-to-end Tokenization + Morphology + LangID workflow	PipelineExamples.cs
Examples.Classification	News title category classification (fastText, no model file needed to browse code)	SimpleClassification.cs

API at a Glance

ZemberekDotNet uses idiomatic C# — no Java-style builder chains, no Guava dependencies. The table below shows the most common use cases side-by-side.

Morphological Analysis

Java (zemberek-nlp)	C# (ZemberekDotNet)
TurkishMorphology m = TurkishMorphology.createWithDefaults();	TurkishMorphology m = TurkishMorphology.CreateWithDefaults();
SentenceAnalysis a = m.analyzeAndDisambiguate(s);	SentenceAnalysis a = m.AnalyzeAndDisambiguate(s);
a.forEach(e -> e.getBestAnalysis().getLemmas())	a.Select(e => e.GetBestAnalysis().GetLemmas())

// LINQ-style: extract all root lemmas from a sentence
TurkishMorphology morphology = TurkishMorphology.CreateWithDefaults();

List<string> lemmas = morphology
    .AnalyzeAndDisambiguate("Güzel bir gün bugün.")
    .Where(swa => !swa.GetBestAnalysis().IsUnknown())
    .Select(swa => swa.GetBestAnalysis().GetLemmas()[0])
    .ToList();
// ["güzel", "bir", "gün", "bu"]

Tokenization & Sentence Splitting

using ZemberekDotNet.Tokenization;

// Split a paragraph into sentences
List<string> sentences = TurkishSentenceExtractor.Default
    .FromParagraph("Merhaba dünya. Bugün iyi bir gün.");
// ["Merhaba dünya.", "Bugün iyi bir gün."]

// Tokenize a sentence
List<string> tokens = TurkishTokenizer.Default
    .TokenizeToStrings(sentences[0]);
// ["Merhaba", "dünya", "."]

Language Identification

using ZemberekDotNet.LangID;

LanguageIdentifier lid = LanguageIdentifier.FromInternalModels();

Console.WriteLine(lid.Identify("merhaba dünya ve tüm gezegenler")); // tr
Console.WriteLine(lid.Identify("hello world and all the planets")); // en
Console.WriteLine(lid.Identify("Hola mundo y todos los planetas")); // es

// With confidence scores
List<LanguageIdentifier.IdResult> scores =
    lid.GetScores("merhaba dünya", maxSampleCount: -1);
scores.ForEach(r => Console.WriteLine($"{r.id}: {r.score:F4}"));

Current targets are:

• Library packages target netstandard2.1 (cross-platform for modern .NET runtimes).
• Test projects target net8.0.
• Apps and examples target net8.0.

Compatibility notes:

• netstandard2.1 libraries can be consumed by .NET Core 3.0+ and .NET 5+.
• .NET Framework is not supported by netstandard2.1.

CI/CD

Repository is configured to continuously trigger a build, test and release cycle using Azure DevOps. At the end of a successful release, it automatically publishes the artifacts to NuGet.org.