The fine-tuning datasets were limited to Finnish. Although the original model is multilingual, we did not evaluate its ability to generate definitions in other languages.

Generated definitions can contain all sorts of biases and stereotypes, stemming from the underlying language model and raw dictionary data.

Direct Use

script to run prediction

Training Details

Training Data

axolotl24

dbnary

Training Procedure

script to run training

Evaluation

run evaluation

Testing Data

axolotl24 Finnish test set

Metrics

BLEU, BERTScore

Citation

BibTeX:

@inproceedings{fedorova-etal-2025-explaining,
    title = "Explaining novel senses using definition generation with open language models",
    author = "Fedorova, Mariia  and
      Kutuzov, Andrey  and
      Periti, Francesco  and
      Scherrer, Yves",
    editor = "Christodoulopoulos, Christos  and
      Chakraborty, Tanmoy  and
      Rose, Carolyn  and
      Peng, Violet",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2025",
    month = nov,
    year = "2025",
    address = "Suzhou, China",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.findings-emnlp.1214/",
    pages = "22294--22302",
    ISBN = "979-8-89176-335-7",
    abstract = "We apply definition generators based on open-weights large language models to the task of creating explanations of novel senses, taking target word usages as an input. To this end, we employ the datasets from the AXOLOTL{'}24 shared task on explainable semantic change modeling, which features Finnish, Russian and German languages. We fine-tune and provide publicly the open-source models performing higher than the best submissions of the aforementioned shared task, which employed closed proprietary LLMs. In addition, we find that encoder-decoder definition generators perform on par with their decoder-only counterparts."
}

Framework versions

bert-score==0.3.13
peft==0.14.0
sentencepiece==0.2.0
tokenizers==0.20.1
torch==2.2.2
transformers==4.46.1
trl==0.15.2