🔤 Pretraining Language Models with LoRA and Artificial Languages
Nalin Kumar, Mateusz Lango, @tuetschek.bsky.social t
aclanthology.org/2025.babylm-...
Constructed artificial languages with LoRA affects language model development.

🎓 You are an LLM teaching a smaller model everything you know: Multi-task pretraining of language models with LLM-designed study plans
Wiktor Kamzela, Mateusz Lango, @tuetschek.bsky.social
aclanthology.org/2025.babylm-...

📚 SRS-Stories: Vocabulary-constrained multilingual story generation for language learning
Wiktor Kamzela, Mateusz Lango & @toonietuesday.bsky.social
aclanthology.org/2025.emnlp-i...
LLM stories teach vocab while reviewing learned words via Spaced Repetition-more grammatical than standard generation

🤖 LLM Agents Implement an NLG System from Scratch
Mateusz Lango, Ondrej Dusek
aclanthology.org/2025.emnlp-i...
LLM agents can autonomously build interpretable, rule-based RDF-to-text generators from scratch, combining the LLMs with the transparency and reliability of traditional rule-based systems.

👥 Can Large Language Models Personalize Dialogues to Generational Styles?
P. Balestrucci, @tuetschek.bsky.social, L. Anselma, A. Mazzei
aclanthology.org/2025.finding...
Can LLMs adapt dialogues to generational styles? We show with P-MultiWoZ that models capture patterns from Boomers to Gen Z.

📊 Real-World Summarization: When Evaluation Reaches Its Limits
@patuchen.bsky.social , @tuetschek.bsky.social , @saad.me.uk
aclanthology.org/2025.finding...
For hotel highlights, metrics like word overlap surprisingly match human judgments better than complex methods. LLMs unreliable as evaluators.

It's fine by me if they generate it, as long as it works and they know how... but I've been getting loads of roughly plausible but non-functional code, with hallucinated API calls etc. 😒. Not that many emojis though (in docs only).

FreshTab: Sourcing Fresh Resources for Table-to-Text Generation Evaluation
by @navitas.bsky.social, ‪@oplatek.bsky.social‬, ‪@zdenekkasner.bsky.social‬, @tuetschek.bsky.social .bsky.social‬

ReproHum #0669-08: Reproducing Sentiment Transfer Evaluation
by @navitas.bsky.social, M. Lango, @patuchen.bsky.social, @tuetschek.bsky.social
Challenge to reproduce human evaluations from NLP papers, testing the reproducibility of evaluation studies

OpeNLGauge: An Explainable Metric for NLG Evaluation with Open-Weights LLMs
by @ivankartac.bsky.social, M. Lango, @tuetschek.bsky.social
arxiv.org/abs/2503.11858
Open-source NLG evaluation metric that explains errors and matches human judgments without proprietary models

Slides and links to papers at bit.ly/mlprague25-od 🤓