Argument Aspect Corpus

The Argument Aspect Corpus (AAC) contains argumentative English-language sentences from four different topics with aspect annotations on a token level. It was introduced in this paper:

Mattes Ruckdeschel and Gregor Wiedemann. 2022. Boundary Detection and Categorization of Argument Aspects via Supervised Learning. In Proceedings of the 9th Workshop on Argument Mining, pages 126–136, Online and in Gyeongju, Republic of Korea. International Conference on Computational Linguistics.

The Corpus is based on the argumentative sentences in the UKP SAM[1] dataset for four highly debated topics: nuclear energy, minimum wage, abortion, and marijuana legalization. The corpus contains one conll-formatted file per topic, containing the gold standard annotation. Further the coding guidelines used for annotation are uploaded. of all sentences from that topic. For the reproduction of paper results, check out the corresponding GitHub repository. The gold standard annotation was obtained by chunk-normalization of a token-level gold standard. Using the default chunker from flair[2], sentences were split into chunks, and all tokens of a chunk were labeled with an aspect if at least one token in the chunk was labeled. Any conflicts were resolved by an additional coder.

Coding was done by two trained expert coders with a background in Social science. Conflicts were resolved by a third trained coder with a background in Computer Science.

Topic information

The following tables shows statistics for the different topics. \(\alpha_k\) gives the intercoder-agreement as Krippendorff’s alpha. Arg Occurrences gives the number of arguments containig a specific aspect, while Chunk Occurrences gives the number chunks that have been labeled with a specific aspect.

General Statistics

\(N_{args}\) describes the number of arguments for a topic and \(N_{singles}\) the amount of arguments with only one aspect.

Minimum Wage

Nuclear Energy

Marijuana Legalization

Abortion

[1] Stab, C., Miller, T., Schiller, B., Rai, P., & Gurevych, I. Cross-topic Argument Mining from Heterogeneous Sources. In E. Riloff, D. Chiang, J. Hockenmaier, & J. Tsujii (Eds.), Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (pp. 3664–3674). Association for Computational Linguistics. https://doi.org/10.18653/v1/D18-1402

[2] Alan Akbik, Tanja Bergmann, Duncan Blythe, Kashif Rasul, Stefan Schweter, and Roland Vollgraf. 2019. FLAIR: An Easy-to-Use Framework for State-of-the-Art NLP. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics (Demonstrations), pages 54–59, Minneapolis, Minnesota. Association for Computational Linguistics.