cm-annotator-corpus / README.md
redmars-01's picture
Upload README.md with huggingface_hub
5d00dad verified
|
Raw History Blame Contribute Delete
1.63 kB
---
license: other
license_name: cm-annotator-nc-nd-derivative
task_categories:
- translation-studies
language:
- en
- zh
tags:
- conceptual-metaphor
- annotation
- parallel-corpus
- lakoff
size_categories:
- 1K<n<10K
---
# CM-Annotator Parallel Annotation Corpus (annotation layer)
English-Chinese parallel sentence pairs from UM-Corpus (Tian et al., LREC 2014; CC BY-NC-ND 4.0)
with conceptual-metaphor span annotations (English and Chinese) and pair-level translation
relations (PRESERVE / REPLACE / LITERALIZE / DELETE / ADD / NONE).
**This is the annotation layer only: `pair_id` + annotations. Sentence text is NOT included**
(UM-Corpus license prohibits redistribution). Restore full text with `scripts/rebuild_fulltext.py`
from your own copy of UM-Corpus. See docs/DATA_CARD.md in the code repository.
## Stats
- pairs: 5000
- EN spans: 3631 / ZH spans: 1652
- null-CM rate: EN 0.686, ZH 0.837
- relations: {'REPLACE': 544, 'DELETE': 808, 'NONE': 3299, 'PRESERVE': 141, 'LITERALIZE': 78, 'ADD': 130}
## Schema
`pair_id`, `domain`, `en_annotation{has_metaphor, metaphor_spans[]}`,
`zh_annotation`, `pair_relation{rel_primary, relations[], links[]}`.
Spans: text/position are omitted in this layer except span metadata (cm_id, cm_name,
source_family, macro_class, conventionality) — see DATA_CARD for the full schema and
provenance. Annotations are model-generated silver labels distilled with human auditing;
they are NOT gold-standard human annotations.
## Citation
UM-Corpus (LREC 2014); Master Metaphor List (Lakoff et al. 1991); CM-Annotator (see citation.cff).