NLP KInIT
Part of KInIT
About the team
O tíme: Výskumný tím zameraný na oblasť spracovania prirodzeného jazyka (NLP). Popri základnom a aplikovanom výskume sa aktívne venuje aj prenosu najnovších vedeckých poznatkov do praxe. V spolupráci s partnermi z akademického aj súkromného sektora pomáha pretavovať pokročilé jazykové technológie, ktoré pomáhajú riešiť reálne výzvy v spoločnosti a priemysle, do praxe.
Hlavné témy výskumu: Metódy spracovania jazyka s malým množstvom zdrojov (low-resource NLP), jazykové modelovanie (language modeling), multilingválne spracovanie jazyka (multilingual NLP), multimodálne prístupy k spracovaniu jazyka (multimodal NLP) a metódy dôveryhodného spracovania jazyka (trustworthy NLP).
Zloženie tímu: Aktuálne pozostáva z 22 vedeckých alebo odborných pracovníkov, 2 doktorandov a niekoľkých dobrovoľníkov a stážistov.
Resources in the catalog17
- Parakeet TDT 0.6b v3 SK
Model, speech and audio, CC BY 4.0
Slovak fine-tune of Parakeet TDT 0.6b v3 from Nvidia.
Transcribe speechNLP KInIT, KInIT
- Canary 1B v2 SK
Model, speech and audio, CC BY 4.0
Slovak fine-tune of Canary 1B v2 from Nvidia.
Transcribe speechNLP KInIT, KInIT
- Whisper-tiny-sk
Model, speech and audio, Apache 2.0
Slovak fine-tune of Whisper-tiny.
Transcribe speechNLP KInIT, KInIT
- Whisper-base-sk
Model, speech and audio, Apache 2.0
Slovak fine-tune of Whisper-base.
Transcribe speechNLP KInIT, KInIT
- Whisper-small-sk
Model, speech and audio, Apache 2.0
Slovak fine-tune of Whisper-small.
Transcribe speechNLP KInIT, KInIT
- Whisper-medium-sk
Model, speech and audio, Apache 2.0
Slovak fine-tune of Whisper-medium.
Transcribe speechNLP KInIT, KInIT
- ANLI SK
Dataset, text, CC BY-NC-SA 4.0
Translated version of the Czech ANLI dataset.
Understand textNLP KInIT, DFKI
- GEST
Dataset, text, Apache 2.0
GEST – a manually created dataset designed to measure gender-stereotypical reasoning in language models and machine translation systems.
Check text and modelsNLP KInIT, WUDAP KInIT, Ethics KInIT
- Whisper-large-v3-sk
Model, speech and audio, Apache 2.0
Slovak fine-tune of Whisper-large-v3.
Transcribe speechNLP KInIT, KInIT
- Whisper-large-v3-turbo-sk
Model, speech and audio, MIT
Slovak fine-tune of Whisper-large-v3-turbo.
Transcribe speechNLP KInIT, KInIT
Connections
Organization
- Kempelen Institute of Intelligent Technologies
Other teams: WUDAP KInIT
Benchmarks2
Partner teams3
- LRKT FEI TUKE
Shared resources: E5-sk-large, E5-sk-small, and skLEP
- NaiveNeuron FMPI CU
Shared resources: E5-sk-large, E5-sk-small, and skLEP
- WUDAP KInIT
Shared resources: GEST
Articles4
- NLP Summer School 2026: Three Days Exploring Language, AI and the Future of Intelligent Systems
, mentions skLEP and SkMTEB
- How (not) to read benchmarks and leaderboards
, mentions SkMTEB and skLEP
- What's New in the SLAIH Asset Catalog? Five Resources Worth Exploring
, mentions SkMTEB and GEST
- SkMTEB: The First Benchmark for Slovak Embeddings
, mentions SkMTEB
Connections are based only on catalog data: who created a resource, which datasets a benchmark uses, which models it evaluates, and what articles link to.