dep2latex

dep2latex is a Python script that converts dependencies output by natural language parsers into nice graphs in LaTeX.

This script is currently very custom, i.e. it was built for my own needs, but it could still be very useful to others.

Requirements

• TikZ
• LuaTeX

You will need to compile your TeX document with lualatex.

Usage

The script accepts output files from three parsers as input.

• C++ C&C parser: output file from bin/candc --models models --input INPUTFILE
• Stanford parser: output file using switch -outputFormat "penn,typedDependencies"
• Berkeley parser: output file from java ... edu.stanford.nlp.trees.EnglishGrammaticalStructure -treeFile TREEFILE, where TREEFILE is the output file from the Berkeley parser. The Stanford parser is used to convert the trees into dependencies.

Run ./dep2latex.py DEPSFILE, with a data folder already created in the same directory. The following files will be generated:

• DEPSFILE.tree containing all the LaTeX code
• data/sentXYY.tex for each sentence in DEPSFILE; X is the sentence number and YY is the code for the parser

The individual TeX files are for easier management; it is wise to use something like \input{data/sent2cc.tex} in your TeX document, which will also need the following:

\usepackage{tikz}
\usetikzlibrary{graphdrawing}
\usetikzlibrary{graphs}
\usetikzlibrary{quotes}
\usegdlibrary{layered}

The generated graphs will not be perfect. Fortunately, you can shift the positions of the nodes and labels with nudge and pos respectively:

...
\node (can-8) [nudge left=80mm] {can};
\node (savings-15) [nudge right=20mm] {savings};
...
(see-30) edge["dobj", ->, pos=0.3] (trusts-36)
(responsibility-46) edge["nmod:for", ->, pos=0.2] (the-48)
...