Swiss court files are curated
Legal area and judgment are both filtered and written to their own files. The audit passed.
Original languages, train / validation / test:
- German judgments: 32,036 / 4,077 / 8,270
- French judgments: 18,595 / 2,597 / 5,569
- Italian judgments: 2,843 / 367 / 724
- Legal area, 13 sub-areas, all three languages together: 9,854 / 2,872 / 7,577
The English machine translation is a separate file: 32,363 train rows. It has no official validation or test split. It fits the 2,048-token window. The original-language file does not go into that English pack.
Together the original-language train file has 63,328 rows. The other curated pack stays separate and is still much larger.