Skip to main content

SVG+Python based rendering of linguistics-style (constituent) trees

Project description

svgling: syntax trees in python + svg

Author: Kyle Rawlins, kgr@jhu.edu

Dependencies: svgwrite, python 3

Repository: https://github.com/rawlins/svgling/

Installation: download and use setuptools, or pip install svgling

License: MIT License

Overview

The svgling package is a pure python package for doing single-pass rendering of linguistics-style constituent trees into SVG. It is primarily intended for integrating with Jupyter notebooks, but could be used to generate SVG diagrams for all sorts of other purposes. It involves no javascript and so will work in Jupyter without any plugins.

The basic interface is pretty simple: pass a tree-describing object to svgling.draw_tree (e.g. a tuple consisting of a lable and a sequence of daughter nodes, which mey themselves be trees).

import svgling
svgling.draw_tree(("S", ("NP", ("D", "the"), ("N", "elephant")), ("VP", ("V", "saw"), ("NP", ("D", "the"), ("N", "rhinoceros")))))

This produces an SVG image like the following:

example sentence

The tree drawing code accepts two main tree formats: lisp-style trees made from lists of lists (or tuples of tuples), with node labels as strings, or trees from the nltk package, i.e. objects instantiating the nltk.tree.Tree API. The following nltk code, as long as svgling has been imported, produces an identical tree diagram to the above example, though by a very different route:

import svgling
nltk.Tree.fromstring("(S (NP (D the) (N elephant)) (VP (V saw) (NP (D the) (N rhinoceros))))")

(That is, svgling monkey-patches NLTK to use SVG-based tree drawing code. You may also want to call svgling.disable_nltk_png() to fully disable the default NLTK png renderer, especially if you're on a mac or windows 64, or are running NLTK on a headless device; see nltk issue #1887 for use-cases).)

Beyond basic tree-drawing, the package supports a number of flourishes like movement arrows. For documentation and examples, see the three .ipynb files in the root of this repository: (links below to nbviewer static rendered versions):

Core design principles and goals

  1. Be well suited for programmatic generation of tree diagrams (not just hand-customized diagrams).
  2. Be equally suited for theoretical linguistics and computational linguistics/NLP, at least for cases where the latter is targeting constituent trees. (This package is not aimed at dependency trees/graphs.)
  3. Do as much as possible with pure python (as opposed to python+javascript, or python+tk, or python+dot, or...).

Strengths and limitations

The svgling package does its rendering in one pass -- it takes a tree structure as input, produces an svg output, and that's it. Because of this, it is extremely simple to use in Jupyter, and no messing with plugins or Jupyter settings should be necessary. Because it is SVG-based, scaling and embedding in any web context should work smoothly. It also has minimal dependencies, just one package that provides an abstraction layer over generating svg. (If you're interested in programmatic diagramming in svg for Jupyter, I do recommend svgwrite, it's under active development and has a very pleasant API + good documentation.)

Single-pass rendering also places limitations on what can be done. One of the challenges is that it mostly uses absolute position, and the exact position and width of text elements can't be determined without actually rendering to some device and seeing what happens. In addition, the exact details of rendering are in various ways at the mercy of the rendering device. This all means that svgling uses a bunch of tricks to estimate node size and width, and won't always be perfect on all devices. This situation also places some hard limitations on how far svgling can be extended without adding javascript or other multi-pass rendering techniques. For example, I would eventually like to allow mathjax in nodes, and allow nodes with complex / multi-line shapes, but at the moment this does not seem possible in pure SVG without javascript on the client side. The package does provide basic support for hybrid HTML/SVG tree diagrams that allow complex nodes, including MathJax, but with substantial limitations.

There are many things that it might be nice to add to this package; if you find svgling useful, have any requests, or find any bugs, please let me know.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for svgling, version 0.3.0
Filename, size File type Python version Upload date Hashes
Filename, size svgling-0.3.0-py3-none-any.whl (21.3 kB) File type Wheel Python version py3 Upload date Hashes View hashes
Filename, size svgling-0.3.0.tar.gz (21.1 kB) File type Source Python version None Upload date Hashes View hashes

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page