#tree #plot #string-parser #parsed #input #drawing #constituency

parsed_to_plot

plots constituency trees and dependency trees given by strings

3 unstable releases

0.2.0 May 12, 2023
0.1.1 Apr 11, 2023
0.1.0 Apr 10, 2023

#117 in Visualization

MIT license

94KB
1.5K SLoC

parsed_to_plot

example workflow rust version crates.io

Plots constituency trees and dependency trees given by strings.

Overview

Primarily written for inputs like parsed syntactic trees, but can serve other inputs, such as mathematical expressions etc. The program first transforms the input to a conll / tree, then plots the structure, recursively. It is mostly suitable for short parsed sequences of up to 15-20 tokens. The program is a simple drawing program, plots strings that are already parsed. This is not a parser! I wrote this in order to get familiar with Rust and decided to upload it if it can help others. The code uses both the id-tree crate and the plotters crate.

Current content

  • String2Tree + Tree2Plot to move from input constituency strings to plot (API or through command-line).
  • String2Conll + Conll2Plot to move from input dependency strings to plot (API or through command-line).
  • From version 0.2.0 - Tree2String / Conll2String to move back from built structure to original input ().

Input-Output

  • The API expects a string input. Multiple string inputs can be delivered in a file, through the command-line.
  • For constituency trees, the program takes a parsed string given in one line. The string can be syntactic, for example such that represents phrases and parts-of-speech (like the structure in Berkeley Neural Parser in python). Such strings will have "double leaves" (see an example below). Alternatively, the strings can have singular leaves, representing, for example, mathematical expressions.
  • For dependency trees, the programs takes a conll format, in which every token has 10 fields, separated by tab, and presented in a new line. Sentences are separated by an empty line. (see an example below, using an output from spaCy in python).
  • For multiple inputs of the same type, the program expects 3 arguments from the command line :
    • input type ("c" = constituency / "d" = dependency), String
    • input file path, String
    • output path, String

See examples below.

Usage examples

Constituency

How to use the API in order to produce a png from a single parsed constituency string:

// Example parsed sentence:
// (S (NP (det The) (N people)) (VP (V watch) (NP (det the) (N game))))

use parsed_to_plot::Config;
use parsed_to_plot::String2Tree;
use parsed_to_plot::Tree2Plot;
use parsed_to_plot::String2StructureBuilder;
use parsed_to_plot::Structure2PlotBuilder;
 
let mut constituency = String::from("(S (NP (det The) (N people)) (VP (V watch) (NP (det the) (N game))))");
let mut string2tree: String2Tree = String2StructureBuilder::new();
string2tree.build(&mut constituency).unwrap(); // build the tree from the string
let tree = string2tree.get_structure();

// build plot from tree and save
Config::make_out_dir(&"Output".to_string()).unwrap();
let save_to: &str = "Output/constituency_plot.png";
let mut tree2plot: Tree2Plot = Structure2PlotBuilder::new(tree);
tree2plot.build(save_to).unwrap();

Dependency

How to use the API in order to produce a png from a single conll format:

// Example conll:
//  0   The the det _   _   1   det   _   _
//  1	people	people	NOUN	_	_	2	nsubj	_	_
//  2	watch	watch	VERB	_	_	2	ROOT	_	_
//  3	the	the	DET	_	_	4	det	_	_
//  4	game	game	NOUN	_	_	2	dobj	_	_
 
use parsed_to_plot::Config;
use parsed_to_plot::String2Conll;
use parsed_to_plot::Conll2Plot;
use parsed_to_plot::String2StructureBuilder;
use parsed_to_plot::Structure2PlotBuilder;
 
let mut dependency = [
    "0	The	the	DET	_	_	1	det	_	_",
    "1	people	people	NOUN	_	_	2	nsubj	_	_",
    "2	watch	watch	VERB	_	_	2	ROOT	_	_",
    "3	the	the	DET	_	_	4	det	_	_",
    "4	game	game	NOUN	_	_	2	dobj	_	_"
].map(|x| x.to_string()).to_vec();
 
let mut string2conll: String2Conll = String2StructureBuilder::new();
string2conll.build(&mut dependency).unwrap(); // build the conll from the vector of strings
let conll = string2conll.get_structure();

// build plot from conll and save
Config::make_out_dir(&"Output".to_string()).unwrap();
let save_to: &str = "Output/dependency_plot.png";
let mut conll2plot: Conll2Plot = Structure2PlotBuilder::new(conll);
conll2plot.build(save_to).unwrap();

Multiple inputs via file

You can use a combination of the API and command-line to process multiple inputs of the same type through a file. The command-line format is as follows:

cargo run INPUT_TYPE INPUT_FILE OUTPUT_PATH

when:

  • INPUT_TYPE should be replaced with "c" for constituency or "d" for dependency.
  • INPUT_FILE should be replaced with a path to a txt file with inputs.
  • OUTPUT_PATH should be replaced with a path to a requested output dir.

For example, you can enter multiple constituencies by using the following command:

cargo run c constituencies.txt Output 

With the following usage:

use parsed_to_plot::Config;
use parsed_to_plot::String2Tree;
use parsed_to_plot::Tree2Plot;
use parsed_to_plot::String2StructureBuilder;
use parsed_to_plot::Structure2PlotBuilder;
use std::env;
 
// collect arguments from command line 
let args: Vec<String> = env::args().collect();
// note: your command line args should translate to something similar to the following:
let args: Vec<String> = ["PROGRAM_NAME", "c", "Input/constituencies.txt", "Output"].map(|x| x.to_string()).to_vec();
 
// run configuration protocol and inspectations
let sequences = match Config::new(&args) {
    Ok(sequences) => Vec::<String>::try_from(sequences).unwrap(),
    Err(config) => panic!("{}", config) 
};
 
for (i, mut constituency) in sequences.into_iter().enumerate() {
            
     println!("working on input number {} ...", i);
     let save_to = &Config::get_out_file(&args[3], i.to_string().as_str());
     
     // build tree from consituency
     let mut string2tree: String2Tree = String2StructureBuilder::new();
     string2tree.build(&mut constituency).unwrap();
     let tree = string2tree.get_structure();
     
     // build plot from tree
     let mut tree2plot: Tree2Plot = Structure2PlotBuilder::new(tree);
     tree2plot.build(save_to).unwrap();
}

Those will save png images of constituency trees drawn for the inputs in constituencies.txt, in an Output dir. The dependency equivalent is similar.

String reconstruction

As of version 0.2.0 you can create a string from a built structure, tree or Vec. This can be useful, for example, to assert the built tree made from a string x, by making sure that x = Structure2String(String2Structure(x)). For example, on a dependency string (constituency equivalent is similar):

//  0   The the det _   _   1   det   _   _
//  1	people	people	NOUN	_	_	2	nsubj	_	_
//  2	watch	watch	VERB	_	_	2	ROOT	_	_
//  3	the	the	DET	_	_	4	det	_	_
//  4	game	game	NOUN	_	_	2	dobj	_	_

use parsed_to_plot::Config;
use parsed_to_plot::String2Conll;
use parsed_to_plot::Conll2String;
use parsed_to_plot::String2StructureBuilder;
use parsed_to_plot::Structure2PlotBuilder;
 
let example = [
    "0	The	the	DET	_	_	1	det	_	_",
    "1	people	people	NOUN	_	_	2	nsubj	_	_",
    "2	watch	watch	VERB	_	_	2	ROOT	_	_",
    "3	the	the	DET	_	_	4	det	_	_",
    "4	game	game	NOUN	_	_	2	dobj	_	_"
].map(|x| x.to_string()).to_vec();
let mut dependency = example.clone();
 
let mut string2conll: String2Conll = String2StructureBuilder::new();
string2conll.build(&mut dependency).unwrap(); // build the conll from the vector of strings
let conll = string2conll.get_structure();

// from v0.2.0 - reconstruction of the original dependency from the built conll
Config::make_out_dir(&"Output".to_string()).unwrap();
let save_to: &str = "Output/dependency_reconstruction.txt";
let mut conll2string: Conll2String = Structure2PlotBuilder::new(conll);
conll2string.build(save_to).unwrap();
let dependency_reproduction = conll2string.get_conll();
assert_eq!(dependency_reproduction, example);

References

  • I used the crates: id-tree, plotters.
  • I used spaCy to create a couple of dependency-parsed examples for illustration.

License

Under MIT license.

Dependencies

~4.5–6.5MB
~110K SLoC