PyDelphin version 1.4.0
pip install pydelphin==1.4.0
ACE Parser (only required for parsing, not for data conversion):
Download and unzip the parser and the precompiled grammar image (ERG 1214) from the ACE website. Place the ace binary in the executable path so that PyDelphin can access it. Place the grammar image in same folder as the grammar (by default, data/original/erg1214).
This includes the annotated data (in tsdb/gold) and the ERG grammar.
mkdir -p data/original/erg1214
svn co http://svn.delph-in.net/erg/tags/1214/ data/original/erg1214
Syntactic annotations: HPSG supertags (.tag) and derivation trees (.tree)
python src/extract-convert-mrs.py --redwoods -i data/original/erg1214/tsdb/gold/ -o data/extracted/ --extract_syntax 2> data/extracted/all.err
In the trees, PTB-style normalizations are applied to brackets.
To extract a single profile, use its directory only (without --redwoods)
python src/extract-convert-mrs.py -i data/original/erg1214/tsdb/gold/wsj00a -o data/extracted/wsj00a --extract_syntax
Semantic annotations: Dependency MRS with token-level span normalizations (.dmrs in JSON format)
python src/extract-convert-mrs.py --redwoods -i data/original/erg1214/tsdb/gold/ -o data/extracted/ --convert_semantics --extract_semantics 2> data/extracted/all.err
Get semantics for the profile from MRP EDS's:
python src/extract-convert-mrs.py --mrp -i data/original/erg1214/tsdb/gold/wsj21a/ --mrp_input data/other/mrp-extended/2020/cf/validation/eds/wsj.eds --convert_semantics -o data/extracted/mrp-wsj21a --extract_semantics
'''
## Parse and convert
Parse input sentence with ACE and convert (to syntactic or semantic representation)
python src/parse-convert-mrs.py --extract_syntax --convert_semantics --extract_semantics '''