Welcome to DEPGLOVE Small Server

A basic service to try the embedding vectors built from dependency paths using DepGlove, a home-made variant of Glove extended to handle syntactic paths.

• The vectors were extracted from a large corpora (including an old dump of the French wikipedia), parsed with FRMG
• The vectors are associated to pairs (lemma, part of speech), expressed as lemma_pos (eg. homme_nc or indiquer_v).
• The main POS labels are nc (common nouns), np (proper nouns), v (verb), adj (adjectives), adv (adverbs)
• When providing a lemma, or a space-separated lemma sequence, the server will return the 10 most similar lemmas
• A lemma may be prefixed by a minus sign to try analogies, for instance -homme_nc femme_nc oncle_nc to get tante_nc (e.g. aunt is to uncle what woman is to man)
• The vectors are of dimension 200 and the list of vectors may be downloaded from here. The vector file follows the format of word2vec.