faq

FAQ (.NET Port)

This page ports and adapts the Java Zemberek FAQ for ZemberekDotNet users.

What is ZemberekDotNet?

ZemberekDotNet is a .NET implementation of Turkish NLP tooling inspired by Java Zemberek. It provides modules for:

• morphology
• tokenization
• normalization
• NER
• classification
• language identification
• language modeling

See module matrix in the root README.

What is the current project state?

Active .NET development with practical parity goals for core modules and examples. The repository is not a strict line-by-line Java mirror; .NET-specific improvements are allowed.

Should I use it?

Yes, especially for:

• Turkish preprocessing pipelines
• tokenization and sentence splitting
• lemmatization/stemming workflows
• baseline NLP systems in .NET

For state-of-the-art tasks, modern transformer-based pipelines may outperform classical approaches, but ZemberekDotNet remains very useful for deterministic preprocessing and hybrid systems.

What is morphological analysis?

Morphological analysis decomposes words into stem/lemma and morphemes. For example, a word may include plural, possessive, and case morphemes that are surfaced in analysis output.

Use:

TurkishMorphology morphology = TurkishMorphology.CreateWithDefaults();
WordAnalysis analyses = morphology.Analyze("kalemlerimden");

What is morphological disambiguation?

Many Turkish words are ambiguous out of context. Disambiguation selects the most likely parse using sentence context.

Use:

SentenceAnalysis sentence = morphology.AnalyzeAndDisambiguate("yarın okula gideceğim");

Why does disambiguation fail sometimes?

Disambiguation is statistical/contextual and can fail on sparse, noisy, or domain-specific text. Common causes:

• insufficient domain adaptation
• informal/noisy language
• out-of-vocabulary forms

Can I use it as stemmer/lemmatizer?

Yes. You can extract stems/lemmas from best analyses. For sentence-level quality, prefer disambiguated results.

Can I add dictionary items programmatically?

Yes. You can build custom lexicons and load dictionary lines. See:

• Text Dictionary Rules
• Migration Quickstart

Can I generate words?

Yes. Word generation is available in morphology module. See morphology examples in this repository.

Where is spell checking / suggestion functionality?

Use the normalization module. It contains spell-checking and suggestion functionality. For noisy sentence normalization, additional data/resources may be needed depending on scenario.

Can I detect languages?

Yes. Use the LangID module:

LanguageIdentifier lid = LanguageIdentifier.FromInternalModels();
string lang = lid.Identify("merhaba dünya");

For very short strings (especially single words), language detection is less reliable.

Is Python integration available?

There is no first-class Python package in this repository. Typical integration patterns are:

• host .NET services and call from Python
• inter-process communication in your own architecture

gRPC support in this repository is deferred for future and not part of current port-completion scope.

Is Android supported?

No official Android support is provided in this repository. Primary target is modern .NET runtime usage (server/desktop/tooling scenarios).

Can I use it in commercial products?

Check this repository license terms and dependency licenses. As a general rule, commercial usage is possible when license obligations are met.

What are alternatives?

Alternatives include other Turkish NLP toolchains and modern deep-learning frameworks. In many real systems, ZemberekDotNet works well as a preprocessing component even when a neural model is used later in the pipeline.

Why are docs/code in English?

English documentation/code conventions improve accessibility for broader developer audiences and cross-team collaboration.