NLP & language models
Transformer and seq2seq models for understanding, transforming and generating natural language across languages.
- Language modeling
- Transformers & seq2seq
- Multilingual models
Natural language processing · LLM
I help machines
understand language.
I am Martin Poláček — an NLP researcher and developer focused on language models, intelligent search and ASR output processing.
- NLP
- Transformers
- Information Retrieval
- Python · PyTorch
01 / About
Language research that goes beyond the paper.
I bridge academic NLP research and the development of language technology designed to work in real-world conditions.
At AILab@TUL, I work on natural language processing — from language models to search and recommender systems.
My main research line explores post-processing for automatic speech recognition: punctuation and capitalization restoration, multilingual models and new seq2seq approaches. I also work with information retrieval, RAG and literary-work recommendation.
When I am not training models or writing code, I am probably taking photographs, 3D-printing something or looking for another technical problem to solve.
02 / Focus
Language as data.
I enjoy the space where language understanding, solid research, clean engineering and measurable outcomes meet.
Search & recommendation
Information retrieval, reranking and recommender systems — including literary-work search research in LibriBot.
- Information retrieval
- Recommender systems
- RAG & reranking
Speech & ASR
ASR output post-processing, punctuation and capitalization restoration, and low-latency streaming processing.
- ASR post-processing
- Low-latency inference
- Python & PyTorch
03 / Publications
Complete list of work.
Conference paper · TSD 2026
Online Punctuation and Capitalization Restoration Using Seq2Seq Approach
Martin Poláček · Petr Červa · Lukáš Matějů
RANLP Student Research Workshop
Study on Automatic Punctuation Restoration in Bilingual Broadcast Stream
Martin PoláčekText, Speech and Dialogue
Efficient Enhancement of Norwegian ASR Model
Lukáš Matějů · Jan Nouza · Martin Poláček · Petr ČervaSpeech Communication · vol. 173
Lightweight online punctuation and capitalization restoration for streaming ASR systems
Martin Poláček · Petr Červa · Jindřich ŽďánskýInterspeech
Online Punctuation Restoration using ELECTRA Model for streaming ASR Systems
Martin Poláček · Petr Červa · Jindřich Žďánský · Lenka Weingartová04 / Contact
Let’s discuss
your idea.
Research collaboration, teaching or an interesting NLP project? Send me a few details and I will get back to you.
in You can also find me on LinkedIn