Skip to content

TypediffTop

NedLetcher edited this page Dec 10, 2013 · 22 revisions

Typediff

Typediff is a tool that enables rapid exploration of the types used in the processing of input by DELPH-IN grammars. Its intended use case is identifying types associated with the implementation of the analysis of specific linguistic phenomena. Besides phenomena investigation, Typediff could also be useful for grammar documentation, exploring unfamiliar grammars, and comparing different versions of the same grammar. If you find Typediff useful, feedback on your particular use case and how you used Typediff would be very much appreciated.

A live installation of Typediff can be found here, which currently supports the following grammars:

The source for Typediff is available on GitHub.

    git clone https://github.com/ned2/typediff.git

Typediff uses ACE for parsing, and should be compatible with any DELPH-IN grammar that has been configured to work with ACE. Many thanks to Woodley Packard for his assistance with using the ACE engine to extract the data needed to power Typediff.

How it Works

Users enter any number of input items into the A items set and the B items set. Each input item is parsed (with ACE) and then for each selected reading, every type that appears in the full AVM is extracted. The difference of the sets A and B is then computed such that Typediff will return all types used to process the A items that were not used by the B items. For investigating phenomena, this diffing approach works best when you can identify "minimal pairs" of sentences, as otherwise unrelated types from the positive sentences will appear as noise in the output. If the phenomenon you wish to investigate does not lend itself to having such pairs, you can try adding additional sentences to the B items to filter out noisy types from the A items. Some example input that could be used to explore right-node-raising might be:

A: We relied on and hired consultants.
B: We relied on consultants and we hired consultants.

Example Diffs

Here are some example inputs attempting to pinpoint specific phenomena within the ERG:

Using Typediff

Typediff is both a command line tool and also has a browser-based interface. The downside to the command line tool is that you are limited to either using the best parse returned by ACE or all of the best N parses. The web interface will not work on versions of Internet Explorer < 9 and has currently only been tested on Google Chrome and Firefox. You can either use the live version of the web interface (hosted at The University of Melbourne) or install it locally. See the README.md file for installation instructions.

Clone this wiki locally