Replace dependency parser

[thjbdvlt]

Improve the dependency parsing quality by training a new parser to replace the current one (2025-01-27).

• Get or find a good corpus:
  • enough sentences.
  • containing all verb tenses and moods.
  • containing all personal pronouns.
  • GPLv3 compatible.
• Create some tests to evaluate it.

The current parser is trained on many corpora, which could lead to inconsistent annotations (UD_French-Sequoia, UD_French-Rhapsodie and UD_French-ParisStories).

Current parser configuration:

[components.parser]
factory = "parser"

[components.parser.model]
@architectures = "spacy.TransitionBasedParser.v2"
state_type = "parser"
extra_state_tokens = false
hidden_width = 128
maxout_pieces = 3
use_upper = true
nO = null

[components.parser.model.tok2vec]
@architectures = "spacy.Tok2Vec.v2"

[components.parser.model.tok2vec.embed]
@architectures = "SolipcysmeMultiHash"
width = ${components.parser.model.tok2vec.encode.width}
features = ["NORM","POS", "MORPH"]
u_features = []
rows = [4000, 2000, 4000]
include_static_vectors = true

[components.parser.model.tok2vec.encode]
@architectures = "spacy.MaxoutWindowEncoder.v2"
width = 128
depth = 4
window_size = 1
maxout_pieces = 3