This is the repo for the code and dataset of the content zone prediction experiment on German source-dependent essays (Textgebundene Erörterung) (GerTE) experiment, to be presented at KONVENS 2025. The accompanying paper, titled Predicting Functional Content Zones in German Source-Dependent Argumentative Essays: Experiments on a Novel Dataset will be published as part of the Proceedings of KONVENS 2025.
The dataset consists of 117 short argumentative essays written in response to 3 different news articles. The essays are therefore source-dependent, and each essay deals with the discussion topic of one article. Each essay has been segmented into sentences, and each sentence is labelled with exactly 1 out of 7 possible content zone labels. Details on the data collection and annotation process are provided in the paper.
The sentence-segmented and content-zone labelled essays are provided in data/gerte_full.tsv. It is a tab-separated file with the following header lines:
| Item | Description |
|---|---|
essay_id |
ID of the essay |
sent_id |
ID of the sentence within the essay |
sent_text |
Sentence text |
sent_label |
Content zone label for the sentence |
topic_id |
ID of the topic that the essay deals with |
The data are also available as a pickled Python object data/full_cz_data.p, which contains the essays as a list object. The Python scripts in this repo use this pickled data file, but the data are the same as in the .tsv file.
The three topics and their IDs are as follows. The respective news articles that served as source text are linked from the discussion topcis:
| Topic | Number of Essays |
|---|---|
| 1 | 50 |
| 2 | 50 |
| 3 | 17 |
| Label | Number of sentence instances |
|---|---|
| own | 551 |
| article_pro | 460 |
| info_intro | 281 |
| article_con | 243 |
| off_topic | 77 |
| other | 71 |
| meta | 30 |
TBD
TBD