Miðeind's Neural Constituency Parser - v. 1.0
The Miðeind neural constituency parser is an experimental variant of the Berkeley neural parser architecture. It is self-contained and conveniently plug-and-play via a docker image. Currently POS tags are not part of its constituency trees.
The input to the parser is a full path to a text file (${INPUT_FILE}) where each line contains a sentence that will be parsed. No prior tokenization is required. The output file will be located in ${OUTPUT_DIR}/output.txt and the output format is line-separated bracketed trees .
To run the parser use the following:
docker run --volume ${INPUT_FILE}:/data/input.txt --volume ${OUTPUT_DIR}:/data/ mideind/neural-parser:${TAG}
The output follows the bracketed tree format described at https://www.ling.upenn.edu/~janabeck/tutorial.html
---
Tauganetsþáttari Miðeindar er tilraunaafbrigði af Berkeley tauganetsþáttaranum. Þáttarinn skilar stofnliðatrjám án POS-marka (eins og er).
Inntakið í þáttarann er full algjör slóð texta að skrá (${INPUT_FILE}) þar sem hver lína geymir eina málsgrein. Eftir keyrslu má finna úttakið í skránni ${OUTPUT_DIR}/output.txt þar sem úttakssniðið er tré á svigaformi með auðri línu á milli .
Til að keyra þáttarann skal nota:
docker run --volume ${INPUT_FILE}:/data/input.txt --volume ${OUTPUT_DIR}:/data/ mideind/neural-parser:${TAG} (edited)