Lambda Notebook: Formal Semantics with Jupyter and Python
This project is a framework for linguists and especially semanticists developing analyses in compositional semantics. It aims to provide a means of developing 'digital fragments', following from the method of fragments in Montague grammar.
The project is in an alpha state. While code is publicly up on github, expect bugs and rapid (or occasionally, not-so-rapid) changes. Contributions, requests, and suggestions are welcome. To see an example, a pre-rendered demo notebook can be found at http://nbviewer.jupyter.org/github/rawlins/lambda-notebook/blob/master/notebooks/Lambda%20Notebook%20Demo%20%28executed%29.ipynb.
The lead developer for this project is Kyle Rawlins, kgr at jhu dot edu. I'm an associate professor in the Cognitive Science Department at Johns Hopkins University, and I do both theoretical linguistic semantics and more computational semantics. My eventual goal is for any theoretical linguistics I do to come with a lambda notebook file. (I'm a long way from that dream.)
While this repository comes with many demo and documentation notebooks, they are mostly saved unexecuted. If you would like to see some samples in nbviewer without downloading anything, the notebook-collection repository mirrors the (non-documentation) notebooks, but they are all pre-executed. Here are four samples. The SVG MathJax renderer is highly recommended for viewing any of these (to select it, right click on any formula).
- Continuations and Quantifier scope, implementing the 2002 proposal by Chris Barker.
- Intensional scope: three treatments of scope with respect to intensional items, building on the von Fintel & Heim lecture notes, work by Orin Percus, and by Ezra Keshet.
- Intro to neo-Davidsonian event semantics.
- Compositional DRT (notebook by Dee Ann Reisinger) showing how to implement the basics of compositional DRT.
Basically, download the repository and ensure you have Jupyter installed (probably via anaconda).
To run the notebook relative to the repository file structure:
- On a mac, double click
- Or, from a shell, run
lambda_notebook.py. On windows you may need to explicitly call something like
- To user a notebook directory other than the default
notebooks, you can call something like
./lambda_notebook.py --notebook-dir=~/Documents/notebooks/. I recommend not keeping your working notebooks in the repository copy.
- Once the lambda-notebook kernel is installed, you can open lambda notebook files from any instance of Jupyter Notebook.
This will start a server in the terminal and open your web browser to the notebook directory. Then, look through the various notebooks to see examples of what can be done. I recommend starting (for now) with:
- Lambda Notebook Intro (start here).ipynb
- Lambda Notebook Demo.ipynb
- look through the various fragments and tutorials
To stop the server from a terminal, hit Ctrl-C twice in the terminal window. (To stop it from the 0.5 app, hit "Cancel".)
In most instances, you can upgrade by simply downloading a new repository version and running the lambda notebook from within there (or, minimally, running
install_lambda_kernel.py from the new repository). Your old notebooks will typically work -- the notebook format is forward compatible, and I try to avoid metalanguage regressions, but because of the alpha state, the API may change. (There may be metalanguage changes before beta, though.)
A note on the lambda notebook UI
- Lambda notebook is running within the Jupyter notebook, and inherits much of the AI.
- As of IPython 2.0, the Jupyter UI is now modal; this can take some getting used to if you aren't familiar with modal editors such as VI. Basically, you are either in edit mode (for editing cells) or command mode (for running cells). Use Enter/esc (or mouse click inside/outside of an edit area) to switch between those. There will be a little pen marker in the upper right corner if you are in edit mode, and the selected cell border will be green. If something isn't doing what you expect, check if you are in the correct mode.
- Try running through the UI tour (from the help menu of any notebook) to get a sense of the Jupyter UI.
There are three main parts to the code, structured into
meta.py ("meta" for metalanguage),
types.pytogether provide a typed logical metalanguage somewhat comparable to
lang.pyprovides machinery for doing composition on an object language.
Two additional files, magics.py and parsing.py provide support for using cell magics in the notebook to directly type expressions in the metalanguage. See the notebooks for demos of what this looks like; better documentation coming soon!
The file tree_mini.py provides nltk.tree, modified to work with the lambda notebook. The long-term plan is to depend directly on nltk, but this isn't there yet.
See here for NLTK license information (Apache license).