Translate Word Docs (EN→FI) with FME + Poro (Ollama)

Category: AI / NLP • Type: Workspace Template (.fmwt)

FME Version: 2024.2+

Tags: translation, Word, DOCX, Finnish, LLM, Ollama, Poro

Overview

This template translates Microsoft Word (.docx) documents from English to Finnish using the FME Word Reader/Writer and a small Finnish LLM (Llama-Poro-2-8B-Instruct) served locally via Ollama.

It keeps element order and layout by translating text-bearing features in place, then writing a new Finnish DOCX.

Why this template?

- No XML wrangling—works directly with the Word Reader/Writer.

- Runs locally (good for sensitive docs).

- Chunked context per block for more consistent translations.

When to Use

You have DOCX files with standard paragraphs, lists, tables, captions.

You want local inference and decent coherence without cloud services.

You don’t need enterprise-scale throughput (see Notes & Limits).

Features

Word Reader → filter text elements → contextual translation → merge by ID → Word Writer.

Glossary / “do-not-translate” masking (optional).

Basic translation memory (optional; CSV/SQLite) to avoid re-translating repeats.

Parameterized model name, chunk size, temperature, and timeouts.

Requirements

FME Form 2024.x or newer with Word Reader/Writer available.

Ollama installed and running (http://localhost:11434 by default).

Pull the model once:

```bash

ollama pull hf.co/mradermacher/Llama-Poro-2-8B-Instruct-GGUF:latest