Initial upload: 268K Spanish-English offline dictionary + 6K phrases

README.md

@@ -0,0 +1,204 @@
+---
+license: gpl-3.0
+tags:
+- translation
+- spanish
+- english
+- offline
+- on-device
+- dictionary
+- idioms
+- phrases
+- rule-based
+- ios
+- mobile
+- no-cloud
+language:
+- es
+- en
+task_categories:
+- translation
+size_categories:
+- 100K<n<1M
+---
+
+# Spanish-English Offline Translation Dictionary
+
+A **268,622-entry** Spanish→English lookup dictionary with **6,056 idiomatic phrases** and translation rules — built for **offline, on-device** mobile use. No API calls, no cloud, no internet required.
+
+This is the production dictionary shipped inside [**RealTime AI Camera**](https://apps.apple.com/us/app/realtime-ai-cam/id6751230739), powering the app's offline Spanish→English translation feature.
+
+## Why this exists
+
+Most Spanish-English translation resources on the Hub are training datasets for seq2seq models. This is different — it's a **lookup dictionary** designed for **rule-based translation on a phone**:
+
+- Runs in pure code (Swift, Python, JS — anything that reads JSON). No model inference, no GPU, no neural network.
+- Sub-millisecond lookups. Key-value, not beam search.
+- 4 MB compressed. Ships inside an iOS app binary without inflating the download.
+- Works in airplane mode, in a basement, in a national park with no signal.
+- Includes idiomatic phrases that word-by-word translation gets wrong ("a hurtadillas" → "stealthily", not "to little stolen things").
+
+If you need a lightweight, offline Spanish↔English dictionary for a mobile app, embedded device, or edge deployment — this is it.
+
+## Contents
+
+Single file: `es_final_with_rules.json.gz` (4.2 MB compressed, ~25 MB uncompressed)
+
+```json
+{
+  "version": "3.1-merged",
+  "generated": "2025-09-02T05:14:26Z",
+  "dictionary": { ... },   // 268,622 entries
+  "phrases": { ... },      // 6,056 idiomatic phrases
+  "rules": { ... }         // 10 categories of translation guidelines
+}
+```
+
+### Dictionary — 268,622 entries
+
+Each entry maps a Spanish string to an English translation with an optional part-of-speech tag:
+
+```json
+{
+  "abandonar": { "translation": "to abandon", "pos": "VERB" },
+  "abogado": { "translation": "lawyer", "pos": "NOUN" },
+  "a decir verdad, esto es un poco demasiado picante para mi.": {
+    "translation": "to tell you the truth, this is a little too spicy for me.",
+    "pos": "UNKNOWN"
+  }
+}
+```
+
+Entries range from single words to full contextual sentences. Many entries include conjugated verb forms, common phrases with prepositions, and regional variations.
+
+**Part-of-speech distribution:**
+
+| POS | Count | Examples |
+|---|---|---|
+| UNKNOWN | 213,954 | Sentences, multi-word expressions, unlabeled entries |
+| VERB | 22,299 | abandonar, abrir, aceptar |
+| NOUN | 21,262 | abogado, agua, amigo |
+| ADJ | 9,133 | alto, bueno, grande |
+| ADV | 703 | además, bien, cerca |
+| AUX | 336 | haber, ser, estar |
+| PRON | 277 | yo, tú, él |
+| INTJ | 214 | hola, oye, caramba |
+| Other | 408 | DET, NUM, ADP, SCONJ, etc. |
+
+### Phrases — 6,056 idiomatic expressions
+
+Phrase-level translations that word-by-word lookup would get wrong:
+
+```json
+{
+  "a caballo regalado no le mires el diente": "don't look a gift horse in the mouth",
+  "a flor de piel": "skin-deep",
+  "a gatas": "on all fours",
+  "a grandes rasgos": "in broad strokes",
+  "a hurtadillas": "stealthily",
+  "dar en el clavo": "to hit the nail on the head",
+  "echar de menos": "to miss someone"
+}
+```
+
+### Rules — translation guidelines
+
+10 categories of structured translation rules including:
+
+- **translationGuidelines** — general principles ("prioritize meaning over literalism; preserve tone and intent")
+- **glossaryCore** — consistent translations for UI terms, technical vocabulary
+- **Metadata** — provenance and schema versioning
+
+## Usage
+
+### Python
+
+```python
+import gzip
+import json
+
+with gzip.open("es_final_with_rules.json.gz", "rt", encoding="utf-8") as f:
+    data = json.load(f)
+
+dictionary = data["dictionary"]
+phrases = data["phrases"]
+
+# Word lookup
+entry = dictionary.get("abogado")
+if entry:
+    print(entry["translation"])  # "lawyer"
+
+# Phrase lookup (check phrases first, fall back to dictionary)
+text = "a hurtadillas"
+if text in phrases:
+    print(phrases[text])  # "stealthily"
+elif text in dictionary:
+    print(dictionary[text]["translation"])
+```
+
+### Swift (iOS / macOS)
+
+```swift
+import Foundation
+
+guard let url = Bundle.main.url(forResource: "es_final_with_rules", withExtension: "json.gz"),
+      let compressed = try? Data(contentsOf: url),
+      let json = try? (compressed as NSData).decompressed(using: .zlib),
+      let dict = try? JSONSerialization.jsonObject(with: json) as? [String: Any],
+      let dictionary = dict["dictionary"] as? [String: [String: String]],
+      let phrases = dict["phrases"] as? [String: String]
+else { return }
+
+// Phrase lookup
+if let translation = phrases["a hurtadillas"] {
+    print(translation)  // "stealthily"
+}
+
+// Word lookup
+if let entry = dictionary["abogado"], let translation = entry["translation"] {
+    print(translation)  // "lawyer"
+}
+```
+
+For the full production implementation with rule-based grammar, see [`SpanishTranslationEngine.swift`](https://github.com/nicedreamzapp/RealTimeAICam/blob/main/SpanishTranslationEngine.swift) in the RealTime AI Camera repo.
+
+### JavaScript / Node.js
+
+```javascript
+const zlib = require("zlib");
+const fs = require("fs");
+
+const compressed = fs.readFileSync("es_final_with_rules.json.gz");
+const data = JSON.parse(zlib.gunzipSync(compressed).toString());
+
+const { dictionary, phrases } = data;
+
+console.log(phrases["a hurtadillas"]);              // "stealthily"
+console.log(dictionary["abogado"].translation);      // "lawyer"
+```
+
+## How it was built
+
+The dictionary was compiled from multiple open sources, merged, deduplicated, and POS-tagged. Idiomatic phrases were curated separately — these are expressions where the whole is not the sum of the parts, so they need dedicated entries. Translation rules were structured to guide consistent output in a rule-based engine (no neural model needed).
+
+The goal was a dictionary that's:
+1. **Complete enough** for real-world camera OCR text (signs, menus, documents, labels)
+2. **Small enough** to ship inside an iOS app (4 MB compressed)
+3. **Fast enough** for real-time use (key-value lookup, not inference)
+
+## License
+
+**GPL-3.0** — same as the parent project [RealTime AI Camera](https://github.com/nicedreamzapp/RealTimeAICam).
+
+Commercial licensing available — contact **Matt Macosko** via [nicedreamzwholesale.com](https://nicedreamzwholesale.com/github-realtime-ai-camera/).
+
+## Links
+
+- 📱 **App Store:** [RealTime AI Camera (free)](https://apps.apple.com/us/app/realtime-ai-cam/id6751230739)
+- 💻 **Source code:** [github.com/nicedreamzapp/RealTimeAICam](https://github.com/nicedreamzapp/RealTimeAICam)
+- 🌐 **Product page:** [NiceDreamzApps](https://nicedreamzwholesale.com/github-realtime-ai-camera/)
+- 🤖 **YOLOv8 CoreML model:** [divinetribe/yolov8n-oiv7-coreml](https://huggingface.co/divinetribe/yolov8n-oiv7-coreml)
+
+---
+
+Built by [Matt Macosko / NiceDreamzApps](https://nicedreamzwholesale.com) · Offline-first · Privacy-first

es_final_with_rules.json.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb246e7ddb63b65b43c444db7130b65386d6e35fe793e05f5fe728673cf74e3f
+size 4193125