Datasets:

Gallicagram
/

ngram-libe

pretty_name: Gallicagram n-grams - libe
license: other
language:
  - fr
configs:
  - config_name: _meta
    data_files: _meta/*.parquet
  - config_name: bigram
    data_files: bigram/*.parquet
  - config_name: token
    data_files: token/*.parquet
  - config_name: total_bigram
    data_files: total_bigram/*.parquet
  - config_name: total_trigram
    data_files: total_trigram/*.parquet
  - config_name: total_unigram
    data_files: total_unigram/*.parquet
  - config_name: trigram
    data_files: trigram/*.parquet
  - config_name: unigram
    data_files: unigram/*.parquet

Gallicagram - n-grammes du corpus "libe"

Tables n-grammes encodées en identifiants entiers (w1, w2, ...) ; le dictionnaire est dans token (id, word). Les tables total_* donnent les effectifs totaux pour normaliser les fréquences.

Exemple (DuckDB) :

SELECT t.word, u.annee, sum(u.n) AS n
FROM 'hf://datasets/Gallicagram/ngram-libe/unigram/*.parquet' u
JOIN 'hf://datasets/Gallicagram/ngram-libe/token/*.parquet' t ON t.id = u.w1
WHERE t.word = 'république'
GROUP BY ALL ORDER BY annee;