nova-twi
v1.0.1
Published
A lightweight, hybrid English-to-Twi translation engine for Node.js. It combines lightning-fast dictionary/phrase mapping with LLM context-refinement via Groq (Llama-3) to produce fluid, natural Twi translations instead of direct mechanical word-for-word
Maintainers
Readme
NOVA Twi Translator SDK 🇬🇭
A lightweight, hybrid English-to-Twi translation engine for Node.js. It combines lightning-fast dictionary/phrase mapping with LLM context-refinement via Groq (Llama-3) to produce fluid, natural Twi translations instead of direct mechanical word-for-word transliterations.
Features
- Hybrid Translation Pipeline: Uses a static keyword/phrase dictionary first, followed by a smart contextual fixer via Groq.
- Context-Aware Refinement: Automatically replaces raw transliterations with sound, idiomatic Twi.
- Preserves Technical Keywords: Smarts enough to retain essential words that don't have direct Twi equivalents.
- Asynchronous Execution: Built from the ground up using modern ES Modules and
async/await.
Installation
npm install twi-translator-sdkSetup & Configuration
Ensure you have your Groq API key set up. The internal AI engine defaults to using an OpenAI-compatible interface pointed at Groq's high-speed inference endpoints.
In your environment configuration, ensure you have access to your key or update it inside your inference config module.
Usage
Direct Translation
Import the unified translate function. It takes care of mapping phrases from your dictionary and automatically cleansing the output using the LLM context fixer.
import { translate } from "nova-twi.js";
const EnglishText = "He smiled, watching a street hawker dash under the concrete awning of a closed shop.";
try {
const twiResult = await translate(EnglishText);
console.log("Translation:", twiResult);
} catch (error) {
console.error("Translation failed:", error);
}Flow Breakdown
translateToTwi(sentence): Validates the input, sweeps for multi-word phrases defined inkeywords.js, and protects spaces using temporary formatting to guarantee accurate key matches.expert(message): Forwards the processed string directly to the Groq inference model along with system instructions to filter out grammatical flaws and smooth out rough phrasing.
File Structure
app.js- Contains the base dictionary mapping logic (translateToTwi).nova-twi.js- Houses the unifiedtranslate(sentence)entry point.models/grok-inference.js- Handles the connection wrapper around the Groq/OpenAI client instance.src/keywords.js- Exports your dictionary array or object pairings.
License
MIT#� �C�o�d�e�-�T�r�i�a�l�s� � �#� �C�o�d�e�-�T�r�i�a�l�s� � �
