iate-2016 / README.md
alvations's picture
Initial release: monolingual + en-de/fr/fi/cs configs + README
619d495 verified
---
pretty_name: IATE 2016 Inter-Active Terminology for Europe
license: other
license_name: ec-decision-2011-833-eu
license_link: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32011D0833
task_categories:
- translation
- feature-extraction
- token-classification
language:
- bg
- cs
- da
- de
- el
- en
- es
- et
- fi
- fr
- ga
- hr
- hu
- it
- lt
- lv
- mt
- nl
- pl
- pt
- ro
- sk
- sl
- sv
tags:
- terminology
- european-union
- multilingual
- parallel-terms
- eu-translation
size_categories:
- 1M<n<10M
configs:
- config_name: monolingual
data_files:
- split: train
path: data/monolingual/train.parquet
- config_name: en-de
data_files:
- split: train
path: data/en-de/train.parquet
- config_name: en-fr
data_files:
- split: train
path: data/en-fr/train.parquet
- config_name: en-fi
data_files:
- split: train
path: data/en-fi/train.parquet
- config_name: en-cs
data_files:
- split: train
path: data/en-cs/train.parquet
---
# IATE 2016 — Inter-Active Terminology for Europe
Tidy HuggingFace release of the **IATE** terminology export dump from **2016-04-25** (the snapshot in `iate.zip`). IATE is the official multilingual terminology database of the EU institutions and bodies — see <https://iate.europa.eu>, <https://iate.europa.eu/faq>, and <https://iate.europa.eu/download-iate>.
This is the **2016 snapshot only**. IATE is continuously updated; for the latest dump pull from the official IATE download page.
Mirror of the raw upstream zip: [`alvations/stash · iate/iate.zip`](https://huggingface.co/datasets/alvations/stash/tree/main/iate).
## Configs
| Config | Rows | Schema | Notes |
| ------------- | ---------: | ------ | ----- |
| `monolingual` | 6,148,883 | one row per `(term_id, lang)` entry | full multilingual TSV; spans all EU official languages (~24) |
| `en-de` | 546,974 | one row per aligned EN ↔ DE term pair | pre-aligned bilingual subset |
| `en-fr` | 730,779 | one row per aligned EN ↔ FR term pair | pre-aligned bilingual subset |
| `en-fi` | 226,688 | one row per aligned EN ↔ FI term pair | pre-aligned bilingual subset |
| `en-cs` | 13,524 | one row per aligned EN ↔ CS term pair | pre-aligned bilingual subset |
The upstream zip also contained `iate_en-ru.csv` / `.json` but they were empty (header only) — skipped.
## Usage
```python
from datasets import load_dataset
# Full multilingual dump
mono = load_dataset("alvations/iate-2016", "monolingual", split="train")
print(mono[0])
# {'term_id': 'IATE-14', 'lang': 'ro', 'term': 'Agenție de aprovizionare',
# 'term_type': 'fullForm', 'reliability': 3}
# Pre-aligned bilingual subsets (great as MT terminology gazetteers)
en_de = load_dataset("alvations/iate-2016", "en-de", split="train")
print(en_de[0])
# {'term_id': 'IATE-84',
# 'src_lang': 'en', 'src_term': 'competence of the Member States',
# 'tgt_lang': 'de', 'tgt_term': 'Zuständigkeit der Mitgliedstaaten'}
# Recover all language entries for a single term
mono_df = mono.to_pandas()
mono_df[mono_df["term_id"] == "IATE-84"]
# 23 rows — one per EU official language
```
## Schemas
### `monolingual`
| Column | Type | Description |
| ------------- | ------ | ----------- |
| `term_id` | string | IATE entry id, e.g. `"IATE-14"` |
| `lang` | string | ISO-639-1 code (`en`, `de`, `fr`, `bg`, `el`, …) |
| `term` | string | the term text in that language |
| `term_type` | string | upstream type label: most commonly `fullForm` (the canonical form); sometimes abbreviation / short-form variants |
| `reliability` | int8 | reliability code, 1–5 (see "Reliability codes" below) |
### `en-{de,fr,fi,cs}`
| Column | Type | Description |
| ----------- | ------ | ----------- |
| `term_id` | string | IATE entry id (same one shared across languages) |
| `src_lang` | string | always `"en"` |
| `src_term` | string | English term |
| `tgt_lang` | string | `"de"` / `"fr"` / `"fi"` / `"cs"` |
| `tgt_term` | string | target-language term |
The 4 bilingual configs share one schema, so you can `concatenate_datasets([...])` them with a single column projection.
## Reliability codes (IATE convention)
| Code | Meaning |
| ---: | ------- |
| 1 | Not verified |
| 2 | Minimum reliability |
| 3 | Reliable (terminologist-checked) |
| 4 | Very reliable |
| 5 | Verified / institutional gold standard |
Most rows in this snapshot carry code `3` (reliable).
## License & attribution
IATE data is reusable under the [European Commission Decision 2011/833/EU of 12 December 2011 on the reuse of Commission documents](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32011D0833). Reuse is free for both commercial and non-commercial purposes, provided you preserve the source attribution.
**Required attribution**:
> © European Union, 1995–2016. Source: IATE — Inter-Active Terminology for Europe (<https://iate.europa.eu>).
The data is provided "as is" with no warranty of accuracy or fitness for any particular purpose — the EU institutions accept no liability arising from its use.
## Provenance
- Upstream zip: `iate.zip` containing 12 files (1 monolingual TSV, 4 bilingual `.csv`/`.json` pairs for EN-DE/FR/FI/CS, plus empty EN-RU); 564 MB uncompressed, snapshot 2016-04-25.
- Mirror: <https://huggingface.co/datasets/alvations/stash/tree/main/iate>
- Conversion: stream-build via `zipfile` + pyarrow `ParquetWriter` (no on-disk extraction). The 321 MB master CSV is written in 200K-row batches.
- Build script: `build.py` alongside this README.
## Citation
There is no single canonical paper for IATE itself. Cite it as the EU institutional resource:
```bibtex
@misc{iate,
title = {{IATE} -- Interactive Terminology for {E}urope},
author = {{Translation Centre for the Bodies of the European Union and EU Institutions}},
year = {1995--2016},
note = {Snapshot of 2016-04-25. Released under EC Decision 2011/833/EU.},
url = {https://iate.europa.eu/},
}
```
For newer / continuously-updated dumps, pull from <https://iate.europa.eu/download-iate>.