README.md

---
license: mit
---
## Dataset of plausibility and stance annotations of the generated definitions.


The dataset was produced as part of the annotation study described in the paper:
[Stance-aware Definition Generation for Argumentative Texts]([URL](https://aclanthology.org/2025.argmining-1.16/)).

The dataset can be used for studies in the plausibility and stance evaluation of the generated output. 

This dataset contains only arguments and definitions on the topic of *abortion*. 
The dataset contains an original argument, the stance of the original argument, the generated definition from an extracted argumentative sequence that contains the keyword, 
the model used for generation, the topic (keyword), and annotations for stance and plausibility by two annotators. 

The dataset contains original arguments from Webis args.me cor-
pus (Ajjour et al., 2019b) and IBM Keypoint Dataset (Friedman et al., 2021). The definitions were generated by one of the following models:

|   Model           | Training data          |
|-------------------|--------------------------|
| LT3/definitions-oxford-llama-8B-instruct        | Oxford                  |
| LT3/definitions-all-noslang-llama-8B-instruct   | WordNet, Wiki, Oxford                  | 
| LT3/definitions-all-llama-8B-instruct    | WordNet, Wiki, Oxford, Urban                  | 
| LT3/definitions-wordnet-llama-8B-instruct   |WordNet                | 
| LT3/definitions-slang-llama-8B-instruct    | Urban                 | 



## How to use 

Use this code to extract the argumentative sequence from the original arguments that was used to generate definitions:

```python
def extract_keyword_sentence(text, keyword, max_length=256):
    """Extract and truncate the sentence containing the keyword."""
    if pd.isna(text) or pd.isna(keyword):
        return None
    keyword_index = text.lower().find(keyword.lower())
    if keyword_index == -1:
        return None  # Keyword not found
    half_length = max_length // 2
    start_index = max(0, keyword_index - half_length)
    end_index = min(len(text), keyword_index + half_length)
    return text[start_index:end_index]

def add_example_column(df, argument_col="argument", topic_col="topic", max_length=256):
    """
    Create an 'example' column by extracting the snippet from the argument text
    that contains the topic keyword.
    """
    df = df.copy()
    df["example"] = df.apply(
        lambda row: extract_keyword_sentence(str(row[argument_col]), str(row[topic_col]), max_length),
        axis=1
    )
    return df


```


### BibTeX entry and citation info

If you would like to use or cite this dataset, feel free to use the following BibTeX code:

```bibtex
@inproceedings{evgrafova-etal-2025-stance,
    title = "Stance-aware Definition Generation for Argumentative Texts",
    author = "Evgrafova, Natalia  and
      De Langhe, Loic
and
 Hoste, Veronique and
      Lefever, Els  ",
    editor = "Chistova, Elena  and
      Cimiano, Philipp  and
      Haddadan, Shohreh  and
      Lapesa, Gabriella  and
      Ruiz-Dolz, Ramon",
    booktitle = "Proceedings of the 12th Argument mining Workshop",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.argmining-1.16/",
    doi = "10.18653/v1/2025.argmining-1.16",
    pages = "168--180",
    ISBN = "979-8-89176-258-9",
    abstract = "Definition generation models trained on dictionary data are generally expected to produce neutral and unbiased output while capturing the contextual nuances. However, previous studies have shown that generated definitions can inherit biases from both the underlying models and the input context. This paper examines the extent to which stance-related bias in argumentative data influences the generated definitions. In particular, we train a model on a slang-based dictionary to explore the feasibility of generating persuasive definitions that concisely reflect opposing parties' understandings of contested terms. Through this study, we provide new insights into bias propagation in definition generation and its implications for definition generation applications and argument mining."
}