Localex for iOS

Localex is a lightweight SwiftUI showcase app demonstrating Apple’s NaturalLanguage framework for classical Natural Language Processing (NLP) tasks such as tokenization, lemmatization, language identification, and part-of-speech tagging — all fully on-device.

⚠️ Disclaimer: This app is intended for demonstration and learning purposes only and is not production-ready.

✨ Features

• 🔤 Tokenization: Split text into words, sentences, and paragraphs
• 🌍 Language Identification: Detect the dominant language of text
• 🧬 Word Embeddings: Generate and inspect the vector representation of a word
• ⚡ On-device & Offline: No internet connection required
• 🎨 Minimal SwiftUI UI: Clean interface to visualize NLP results

🛠 Manual

The app has three tabs, each backed by its own view model:

• TokenizerViewModel for tokenization
• RecognizerViewModel for language detection
• EmbeddingViewModel for word vectors

1) Tokenizer Case (NLTokenizer)

This case demonstrates how a text is split into discrete word units.

1. The view model creates NLTokenizer(unit: .word).
2. When the user taps Tokenize, inputText is assigned to tokenizer.string.
3. enumerateTokens(in:) walks through the full string range.
4. For each token range, the app slices inputText[range] and appends it to tokens.
5. The UI renders each token as a separate Text row.

2) Language Recognizer Case (NLLanguageRecognizer)

This case demonstrates how Apple estimates which language(s) a text belongs to and with what confidence.

1. The view model creates NLLanguageRecognizer().
2. On Recognize, processString(inputText) feeds the entire input into the recognizer.
3. The app calls languageHypotheses(withMaximum: 10) to get up to 10 candidates.
4. Candidates are sorted descending by probability (Double confidence value from 0.0 to 1.0).
5. The UI shows each language code (rawValue), a percentage label, and a ProgressView.

3) Embedding Case (NLEmbedding)

This case demonstrates how a word is mapped to a high-dimensional numeric vector that represents semantic meaning.

1. The input is split by spaces and only the first word is used.
2. The app loads NLEmbedding.wordEmbedding(for: .english).
3. It requests a vector with vector(for: firstWord).
4. If successful, the resulting [Double] is stored in vector.
5. The UI shows the vector dimension (vector.count) and each component value.