SwiftPOSTagger

A Swift package for Part-of-Speech (POS) tagging using BERT-based machine learning models with CoreML integration.

Features

• 🧠 BERT-based Model: Utilizes a quantized BERT model for accurate POS tagging
• 📱 iOS/macOS Support: Compatible with iOS 16+ and macOS 12+
• ⚡ CoreML Integration: Optimized for Apple Silicon and GPU acceleration
• 🔄 Automatic Model Management: Built-in model download and extraction
• 🎯 Simple API: Clean, easy-to-use interface
• 📦 Swift Package Manager: Easy integration into your projects

Installation

Swift Package Manager

Add SwiftPOSTagger to your project using Xcode:

1. File → Add Package Dependencies
2. Enter the repository URL: https://github.com/Otosaku/OtosakuPOSTagger-iOS
3. Select the version you want to use

Or add it to your Package.swift:

dependencies: [
    .package(url: "https://github.com/Otosaku/OtosakuPOSTagger-iOS", from: "1.0.0")
]

Usage

Basic Usage

import SwiftPOSTagger

// Initialize the tagger with model directory
let modelDirectoryURL = // URL to directory containing model files
let tagger = try SwiftPOSTagger(modelDirectoryURL: modelDirectoryURL)

// Perform POS tagging
let text = "The quick brown fox jumps over the lazy dog."
let results = try tagger.predict(text: text)

// Results is an array of (word, tag) tuples
for (word, tag) in results {
    print("\(word) -> \(tag)")
}

Example Output

The -> DT
quick -> JJ
brown -> JJ
fox -> NN
jumps -> VBZ
over -> IN
the -> DT
lazy -> JJ
dog -> NN
. -> .

Model Setup

The library expects a directory containing:

• ModelQuantized.mlmodelc - The CoreML model file
• vocab.txt - BERT vocabulary file
• outTokens.txt - POS tag labels

You can download the pre-trained model or use your own compatible BERT-based POS tagging model.

Compute Units

You can specify which compute units to use for inference:

// Use Neural Engine (default)
let tagger = try SwiftPOSTagger(modelDirectoryURL: modelURL, computeUnits: .all)

// Use CPU only
let tagger = try SwiftPOSTagger(modelDirectoryURL: modelURL, computeUnits: .cpuOnly)

// Use GPU only
let tagger = try SwiftPOSTagger(modelDirectoryURL: modelURL, computeUnits: .cpuAndGPU)

Model Information

This library uses a quantized BERT model specifically fine-tuned for Part-of-Speech tagging. The model:

• Architecture: BERT-base with classification head
• Quantization: INT8 quantized for efficient mobile inference
• Input: Tokenized text with BERT tokenization
• Output: POS tags following Penn Treebank tagset
• Context Length: Supports up to 128 tokens per inference

Supported POS Tags

The model outputs tags from the Penn Treebank tagset, including:

• NN (Noun, singular), NNS (Noun, plural)
• VB (Verb, base form), VBZ (Verb, 3rd person singular)
• JJ (Adjective), RB (Adverb)
• DT (Determiner), IN (Preposition)
• CC (Coordinating conjunction)
• And many more...

Example App

The repository includes a complete example iOS app demonstrating:

• Automatic model download and setup
• Real-time POS tagging with visual results
• Error handling and user feedback

To run the example:

1. Open Example/Example.xcodeproj
2. Build and run on iOS Simulator or device
3. Download the model and test with your own text

Error Handling

The library provides comprehensive error handling:

do {
    let results = try tagger.predict(text: text)
    // Process results
} catch SwiftPOSTaggerError.modelLoadingFailed(let message) {
    print("Model loading failed: \(message)")
} catch SwiftPOSTaggerError.outputExtractionFailed(let message) {
    print("Output extraction failed: \(message)")
} catch {
    print("Other error: \(error)")
}

Requirements

• iOS 16.0+ / macOS 12.0+
• Xcode 14.0+
• Swift 6.0+

License

This project is licensed under the MIT License - see the LICENSE file for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Acknowledgments

• Built with Apple's CoreML framework
• Uses BERT architecture for natural language understanding
• Tokenization based on BERT WordPiece tokenizer