dep2con

dep2con converts Universal Dependencies (UD) parses into constituency representations. It accepts either a CoNLL-U format or text for the Stanza dependency parser.

Requirements

• Python 3.14 or newer
• pyconll 4.0 or newer for CoNLL-U input
• stanza when parsing raw text

Installation

Install the published package:

python -m pip install dep2con

Usage

from dep2con import make_const

Simple usage (for string parsing):

t_s = make_const("I saw a UFO in the backyard.", lang = 'en', use_parser = True)

for dep2con in t_s:
    print(dep2con.sent_parse)
    for i in dep2con.x_phrases:
        print(i)

[VP [NP I] saw [NP a UFO] [PP in [NP the backyard]]] .
...
NPs:
I
a UFO
the backyard
VPs:
I saw a UFO in the backyard
PPs:
in the backyard

More on usage (for a CoNLL-U format):

t_s = make_const("conllu")

for dep2con in t_s:
    print(dep2con.sent_parse)
    print(dep2con.x_phrases)
    print(dep2con.sentence_index)
    ...

Output attributes:

make_const() returns a list with one Dep2Con object per input sentence. Each result has these attributes:

• sent_parse - a constituency tree for the sentence
• x_phrases - constituents for the specified phrases
• sent_dict - a dictionary format for a sentence UD {1: {'idw': 1, 'wf': 'I', 'nf': 'I', 'pos': 'PRON', 'xpos': '_', 'tags': 'Case=Nom|Number=Sing|Person=1|PronType=Prs', 'head_id': 2, 'rel': 'nsubj', 'deps': '_', 'misc': '_'},...
• sent_text - sentence text
• sent_len - sentence length
• sentence_index - an index of a sentence in a text
• source - sentence source if mentioned in a CoNLL-U file

Input arguments:

• input: str, obligatory. Is string if use_parser = True, otherwise input in conllu format
• lang: str, the language parameter ('en',...) in case of text parsing
• use_parser: bool, True - the Stanza parser is at work; False - input in the conllu format, default = False
• delete_orth: bool, delete orthography from the final tree, default = False
• xps: bool, return specific phrases separately, default = True
• emb: bool, returns subconstituents, not only maximal projection, if True, default = True
• h_terminals: List[str], terminals to be put in brackets w/out dependents, default = ['NOUN', 'PRON', 'PROPN', 'ADV']
• selected_cons: List[str], when we want some specific constituents only, default = ['NP', 'VP', 'PP', 'AP', 'AdvP', 'NumP', 'PartP', 'CCONJP', 'INTJP', 'Particle', 'SCONJP', 'SYMP', 'XP']