For sentence simplification part of DEplain, we aligned both document simplification corpora, DEplain-APA and DEplain-WEB, also on the sentence level. All 483 available parallel documents of DEplain-APA and 147 documents of DEplain-web are manually aligned on the sentence level with the assistance of a TS annotation tool (Stodden & Kallmeyer, 2022). Overall, 14,968 sentence pairs of 630 document pairs are manually aligned.
To exemplify the usage of the manual alignments and to provide sentence-wise alignments for the unaligned documents of DEplain-WEB we evaluate different alignment algorithms (see ../C__Alignment_Algorithms) on the manually aligned data.
This directory contains:
- a link to the manual sentence-wise alignments of DEplain-APA,
- the manual sentence-wise alignments of DEplain-web, and
- the automatic sentence-wise alignments of DEplain-web.
| Corpus Name Sent. | # Original Sents | # Simple Sents. | Alignment | License | # Sent. Pairs (train/dev/test) | |
|---|---|---|---|---|---|---|
| 1 | DEplain-APA-sent | 25,607 | 26,471 | manual | upon request | 13,122 (10,660/1,231/1,231) |
| 2 | DEplain-web-sent-manual-open | 6,138 | 6,402 | manual | open | 1,846 (-/-/1846) |
| 3 | DEplain-web-sent-auto-open | 7,087 | 7,760 | auto | open | 652 (514/138/-) |
| 4 | DEplain-web-sent-auto-closed | 12,847 | 18,068 | auto | closed | 942 (767/175/-) |
The documents were aligned by two annotators. For the inter-annotator agreement see the table below. In the dataset, we provide only the alignments of one annotator per document, because otherwise sentence pairs would be repeated.
| domain | avg. | std. | interpretation | # sents | # docs |
|---|---|---|---|---|---|
| news | 0.7497 | 0.28 | moderate | 25224 | 10 |
| bible | 0.7011 | 0.31 | moderate | 6903 | 3 |
| fiction | 0.6131 | 0.39 | moderate | 23289 | 3 |
| health | 0.5147 | 0.28 | weak | 13736 | 6 |
| language learner | 0.9149 | 0.17 | almost perfect | 18493 | 65 |
| all | 0.8505 | 0.23 | strong | 87645 | 87 |
Further annotation on the sentence simplification data, e.g., regarding grammaticality, simpliciy or simplification operations can be found here: ../F__Human_Annotations.
We provide statistics regarding the alignment types in the manual aligned corpora. In the first table, we show the statistics on n:m alignments, where n and m are > 0.
| Name | # pairs | 1:1 (rephrase) | 1:n (split) | n:1 (merge) | n:m (fusion) |
|---|---|---|---|---|---|
| DEplain-apa | 13122 | 9912 | 2360 | 382 | 468 |
| DEplain-web | 1846 | 863 | 796 | 77 | 110 |
In the second table, we show the statsitics on n:m aligned pairs, where n and m are ≤ 1.
| Name | # pairs | 1:1(identical) | 0:1(addition) | 1:0(deletion) |
|---|---|---|---|---|
| DEplain-apa | 12353 | 2712 | 3964 | 5677 |
| DEplain-web | 5482 | 887 | 1572 | 3050 |
The DEplain-web corpus is available in the current repository. Furthermore, the dataset is also available on Huggingface: https://huggingface.co/datasets/DEplain/DEplain-web-sent.
- DEplain-APA: The dataset is provided for research purposes only. Please request access using the following form: https://zenodo.org/record/7674560
- DEplain-web: The corpus includes the following licenses: CC-BY-SA-3, CC-BY-4, and CC-BY-NC-ND-4. The corpus also include a "save_use_share" license, for these documents the data provider permitted us to share the data for research purposes.