Natural Language Processing

The NLP Chapter is the ML6 special division on all things related to Natural Language Processing.

We try to tackle all relevant areas of Natural Language Processing: all things information extraction, speech recognition, sequence to sequence modelling under the roof of a single internal powerhouse.

Think

NLP research is happening at dazzling speed. By exploring trending and relevant AI papers and topics, we keep up to date with the latest and greatest in our field.

No transformer left behind!

Build

We are machine learning engineers at heart.

We build and deploy tools, demos and boilerplates to bootstrap ML6 project work and showcase the value of Natural Language Processing in a wide variety of areas.

Share

We love to show and share our work. Head over to our use cases to see what we’re up to or have fun with one of our NLP-powered demos.

Come check us out on Hugging Face and GitHub as well!

OUR technologies

Case studies

Automating the creation of VRT News summaries by using Natural Language Processing

Building an AI driven model as a powerful tool in assisting journalists in their writing work by speeding up the summary creation as well as suggest alternative wordings. In that way, human creativity and AI can enhance each other’s complementary strengths and produce high quality results.

Building an AI-driven content platform with Funke Mediengruppe

Funke Mediengruppe collaborated with ML6 to explore how to leverage intelligent technology for working with content, in order to enable new ideas and products.

December 30, 2021

Accelerating Keypoint's Digital Real Estate Management Platform with AI

We accelerated the adoption of AI in Keypoint's platforms by automating the processing of their repair offers and doing fraud detection on claims images and weather data.

Increasing efficiency and inspiration at FedNot by using advanced natural language processing techniques

The goal of this project was to increase the efficiency in notary offices by digitizing the wealth of knowledge inside their archives. These archives contain both text-layered documents as well as numerous scans of paperwork.

Demos

Terms & Conditions summarizer

Apply state of the art extractive and abstractive summarization on website Terms of Service to get a quick and concise focus on the main points

Simulate common OCR errors for Dutch texts and use the finetune the character-based ByT5 model to correct them.

HuggingFace Knowledge Graphs

Knowledge graphs in particular are a great tool to flexibly represent complex information, but building one can be a time-consuming process, or can require lots of labelled examples. Enter REBEL designed to tackle these problems, and generate knowledge graphs directly from text.

Demo on converting data to text.

Multi-Article Summarization

Combine multiple news articles on the same subject into a new and novel article using AI.

A custom finetuned GPT2 model capable of generating dutch and german sentences.

Built on OpenAI CLIP, MemeFinder allows you to find a meme template based on a description of the image.

Blogposts

December 15, 2022

Welcome to the multi-language-pipeline universe

Apache Beam has lots of I/O connectors available, but not all connectors are in your language of choice.

October 20, 2022

OpenVINO vs ONNX for Transformers in production

We’ll look at different approaches to optimize transformer models for CPU inference.

October 3, 2022

Who spoke when: Choosing the right speaker diarization tool

A guide on how to choose the right speaker diarization framework for your use case.

Hallucination Detection

Detecting errors in transformers generated abstractive summaries

Decoding Sentence Encoders

Learn more about the similarities and differences between transformers and sentence-transformers.

Triton Ensemble Model for deploying Transformers into production

When deploying a Transformer model, the complexity increases since besides the model one also needs to consider how to deploy the tokenizer. Is it best to deploy the tokenizer on the server or to have the end users handle it and what is the best way to deploy the tokenizer along with the model on the server so that inference latency and costs are minimized?

Domain transfer with GGPL: German Generative Pseudo Labeling

An introduction of semantic search systems, why we need a domain transfer in dense retrieval search engines, the concept of generative pseudo labeling and how you can use generative pseudo labeling to adapt a dense retrieval model on a new domain for non-English texts.

BERT is eating your cash: quantization and ONNXRuntime to save money

Insights in how to run your Transformer in the cloud in a cost-effective way.

Semantic search : A practical Overview

A practical overview on information retrieval, in particular Semantic Search.

January 24, 2022

Data augmentation using few-shot prompting on large Language Models

Proper data augmentation is useful to increase the model performance to annotate a large amount of training data.

November 24, 2021

NLP Trend: A few examples on how to process long documents

Businesses in industries such as media, finance, jurisdiction and many more, tend to accumulate a large quantity of medium to long documents. Processing these documents can be challenging. We'll deep dive into the three main trends here.

September 14, 2021

Multi-Article Summarization: A building blocks approach

In this blogpost, we explain our multi-article summarization approach, where we try to partially eliminate the black-box and make the summarization process more accessible to the user.Our summarization pipeline relies on the idea that sentences are the basic building blocks of any article.

December 18, 2020

Dutch GPT2: Autoregressive Language Modelling on a budget

In this blogpost, we outline our process and code on finetuning an existing GPT2 model towards an entirely different language using a large open Dutch corpus.

November 10, 2020

A Natural Language Processing look at large political documents

A tale of multilingualism, summarization and sentence embeddings

September 7, 2020

Neural text generation from 1 million Belgian real estate deeds

How we trained a large-scale keyword-to-text model for composing real estate deeds in Dutch.

How ML6 built a multilingual COVID-19 assistant powered by NLP

Keeping people well-informed is an important step in flattening the curve. This is why we at ML6 decided to build a multilingual COVID-19 assistant powered by AI, more specifically Natural Language Processing (NLP). Our COVID-19 assistant makes it possible to ask any question in any of 16 languages and search through regulated (Belgian) FAQs for similar questions and answers. You can try it out for yourself at corona.ml6.eu.

Experts

Mats Uytterhoeven

Squad Lead | Senior Machine Learning Engineer

Ghent

Konstantin Buschmeier

Machine Learning Engineer

Amsterdam

Thomas Vrancken

Machine Learning Engineer

Ghent

Jens Bontinck

Head of Labs

Ghent

Interns

No items found.

All rights are reserved, Copyright 2022

Terms and conditions

Data Architecture

ML in production

Structured Data

Software Architecture

Computer Vision

Natural Language Processing